initial import of nouveau kernel module core, some simple tools, and drm

Ignore (or, if you're really keen, fix) any of the horrors that are part
of the emulate-linux-on-libpciaccess (lib/) layer.  When I started down
this path I was only aiming to prototype the reworked driver core, and
never actually intended on going quite this far with things, but it turns
out that being able to develop and test the nouveau core from userspace
is *very* useful.

This tree is based on the code available as of Linux 3.8-rc2.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

299 files changed, 59084 insertions, 0 deletions

diff --git a/nvkm/Makefile.am b/nvkm/Makefile.am
new file mode 100644
index 000000000..81f349cea
--- /dev/null
+++ b/nvkm/Makefile.am
@@ -0,0 +1,8 @@
+SUBDIRS = core subdev engine
+
+noinst_LTLIBRARIES = libnouveau.la
+
+libnouveau_la_SOURCES =
+libnouveau_la_LIBADD  = core/libcore.la \
+			subdev/libsubdev.la \
+			engine/libengine.la
diff --git a/nvkm/core/Makefile.am b/nvkm/core/Makefile.am
new file mode 100644
index 000000000..e0c47c958
--- /dev/null
+++ b/nvkm/core/Makefile.am
@@ -0,0 +1,21 @@
+noinst_LTLIBRARIES = libcore.la
+
+libcore_la_CPPFLAGS = -I${top_srcdir}/nvkm \
+		      -I${top_srcdir}/nvkm/include \
+		      -I${top_srcdir}/lib
+
+libcore_la_SOURCES = client.c \
+		     engctx.c \
+		     engine.c \
+		     enum.c \
+		     falcon.c \
+		     gpuobj.c \
+		     handle.c \
+		     mm.c \
+		     namedb.c \
+		     object.c \
+		     option.c \
+		     parent.c \
+		     printk.c \
+		     ramht.c \
+		     subdev.c
diff --git a/nvkm/core/client.c b/nvkm/core/client.c
new file mode 100644
index 000000000..c617f0480
--- /dev/null
+++ b/nvkm/core/client.c
@@ -0,0 +1,103 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/client.h>
+#include <core/handle.h>
+#include <core/option.h>
+
+#include <subdev/device.h>
+
+static void
+nouveau_client_dtor(struct nouveau_object *object)
+{
+	struct nouveau_client *client = (void *)object;
+	nouveau_object_ref(NULL, &client->device);
+	nouveau_handle_destroy(client->root);
+	nouveau_namedb_destroy(&client->base);
+}
+
+static struct nouveau_oclass
+nouveau_client_oclass = {
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.dtor = nouveau_client_dtor,
+	},
+};
+
+int
+nouveau_client_create_(const char *name, u64 devname, const char *cfg,
+		       const char *dbg, int length, void **pobject)
+{
+	struct nouveau_object *device;
+	struct nouveau_client *client;
+	int ret;
+
+	device = (void *)nouveau_device_find(devname);
+	if (!device)
+		return -ENODEV;
+
+	ret = nouveau_namedb_create_(NULL, NULL, &nouveau_client_oclass,
+				     NV_CLIENT_CLASS, nouveau_device_sclass,
+				     0, length, pobject);
+	client = *pobject;
+	if (ret)
+		return ret;
+
+	ret = nouveau_handle_create(nv_object(client), ~0, ~0,
+				    nv_object(client), &client->root);
+	if (ret) {
+		nouveau_namedb_destroy(&client->base);
+		return ret;
+	}
+
+	/* prevent init/fini being called, os in in charge of this */
+	atomic_set(&nv_object(client)->usecount, 2);
+
+	nouveau_object_ref(device, &client->device);
+	snprintf(client->name, sizeof(client->name), "%s", name);
+	client->debug = nouveau_dbgopt(dbg, "CLIENT");
+	return 0;
+}
+
+int
+nouveau_client_init(struct nouveau_client *client)
+{
+	int ret;
+	nv_debug(client, "init running\n");
+	ret = nouveau_handle_init(client->root);
+	nv_debug(client, "init completed with %d\n", ret);
+	return ret;
+}
+
+int
+nouveau_client_fini(struct nouveau_client *client, bool suspend)
+{
+	const char *name[2] = { "fini", "suspend" };
+	int ret;
+
+	nv_debug(client, "%s running\n", name[suspend]);
+	ret = nouveau_handle_fini(client->root, suspend);
+	nv_debug(client, "%s completed with %d\n", name[suspend], ret);
+	return ret;
+}
diff --git a/nvkm/core/engctx.c b/nvkm/core/engctx.c
new file mode 100644
index 000000000..84c71fad2
--- /dev/null
+++ b/nvkm/core/engctx.c
@@ -0,0 +1,251 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/namedb.h>
+#include <core/handle.h>
+#include <core/client.h>
+#include <core/engctx.h>
+
+#include <subdev/vm.h>
+
+static inline int
+nouveau_engctx_exists(struct nouveau_object *parent,
+		      struct nouveau_engine *engine, void **pobject)
+{
+	struct nouveau_engctx *engctx;
+	struct nouveau_object *parctx;
+
+	list_for_each_entry(engctx, &engine->contexts, head) {
+		parctx = nv_pclass(nv_object(engctx), NV_PARENT_CLASS);
+		if (parctx == parent) {
+			atomic_inc(&nv_object(engctx)->refcount);
+			*pobject = engctx;
+			return 1;
+		}
+	}
+
+	return 0;
+}
+
+int
+nouveau_engctx_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engobj,
+		       struct nouveau_oclass *oclass,
+		       struct nouveau_object *pargpu,
+		       u32 size, u32 align, u32 flags,
+		       int length, void **pobject)
+{
+	struct nouveau_client *client = nouveau_client(parent);
+	struct nouveau_engine *engine = nv_engine(engobj);
+	struct nouveau_object *engctx;
+	unsigned long save;
+	int ret;
+
+	/* check if this engine already has a context for the parent object,
+	 * and reference it instead of creating a new one
+	 */
+	spin_lock_irqsave(&engine->lock, save);
+	ret = nouveau_engctx_exists(parent, engine, pobject);
+	spin_unlock_irqrestore(&engine->lock, save);
+	if (ret)
+		return ret;
+
+	/* create the new context, supports creating both raw objects and
+	 * objects backed by instance memory
+	 */
+	if (size) {
+		ret = nouveau_gpuobj_create_(parent, engobj, oclass,
+					     NV_ENGCTX_CLASS,
+					     pargpu, size, align, flags,
+					     length, pobject);
+	} else {
+		ret = nouveau_object_create_(parent, engobj, oclass,
+					     NV_ENGCTX_CLASS, length, pobject);
+	}
+
+	engctx = *pobject;
+	if (ret)
+		return ret;
+
+	/* must take the lock again and re-check a context doesn't already
+	 * exist (in case of a race) - the lock had to be dropped before as
+	 * it's not possible to allocate the object with it held.
+	 */
+	spin_lock_irqsave(&engine->lock, save);
+	ret = nouveau_engctx_exists(parent, engine, pobject);
+	if (ret) {
+		spin_unlock_irqrestore(&engine->lock, save);
+		nouveau_object_ref(NULL, &engctx);
+		return ret;
+	}
+
+	if (client->vm)
+		atomic_inc(&client->vm->engref[nv_engidx(engobj)]);
+	list_add(&nv_engctx(engctx)->head, &engine->contexts);
+	nv_engctx(engctx)->addr = ~0ULL;
+	spin_unlock_irqrestore(&engine->lock, save);
+	return 0;
+}
+
+void
+nouveau_engctx_destroy(struct nouveau_engctx *engctx)
+{
+	struct nouveau_object *engobj = nv_object(engctx)->engine;
+	struct nouveau_engine *engine = nv_engine(engobj);
+	struct nouveau_client *client = nouveau_client(engctx);
+	unsigned long save;
+
+	nouveau_gpuobj_unmap(&engctx->vma);
+	spin_lock_irqsave(&engine->lock, save);
+	list_del(&engctx->head);
+	spin_unlock_irqrestore(&engine->lock, save);
+
+	if (client->vm)
+		atomic_dec(&client->vm->engref[nv_engidx(engobj)]);
+
+	if (engctx->base.size)
+		nouveau_gpuobj_destroy(&engctx->base);
+	else
+		nouveau_object_destroy(&engctx->base.base);
+}
+
+int
+nouveau_engctx_init(struct nouveau_engctx *engctx)
+{
+	struct nouveau_object *object = nv_object(engctx);
+	struct nouveau_subdev *subdev = nv_subdev(object->engine);
+	struct nouveau_object *parent;
+	struct nouveau_subdev *pardev;
+	int ret;
+
+	ret = nouveau_gpuobj_init(&engctx->base);
+	if (ret)
+		return ret;
+
+	parent = nv_pclass(object->parent, NV_PARENT_CLASS);
+	pardev = nv_subdev(parent->engine);
+	if (nv_parent(parent)->context_attach) {
+		mutex_lock(&pardev->mutex);
+		ret = nv_parent(parent)->context_attach(parent, object);
+		mutex_unlock(&pardev->mutex);
+	}
+
+	if (ret) {
+		nv_error(parent, "failed to attach %s context, %d\n",
+			 subdev->name, ret);
+		return ret;
+	}
+
+	nv_debug(parent, "attached %s context\n", subdev->name);
+	return 0;
+}
+
+int
+nouveau_engctx_fini(struct nouveau_engctx *engctx, bool suspend)
+{
+	struct nouveau_object *object = nv_object(engctx);
+	struct nouveau_subdev *subdev = nv_subdev(object->engine);
+	struct nouveau_object *parent;
+	struct nouveau_subdev *pardev;
+	int ret = 0;
+
+	parent = nv_pclass(object->parent, NV_PARENT_CLASS);
+	pardev = nv_subdev(parent->engine);
+	if (nv_parent(parent)->context_detach) {
+		mutex_lock(&pardev->mutex);
+		ret = nv_parent(parent)->context_detach(parent, suspend, object);
+		mutex_unlock(&pardev->mutex);
+	}
+
+	if (ret) {
+		nv_error(parent, "failed to detach %s context, %d\n",
+			 subdev->name, ret);
+		return ret;
+	}
+
+	nv_debug(parent, "detached %s context\n", subdev->name);
+	return nouveau_gpuobj_fini(&engctx->base, suspend);
+}
+
+int
+_nouveau_engctx_ctor(struct nouveau_object *parent,
+		     struct nouveau_object *engine,
+		     struct nouveau_oclass *oclass, void *data, u32 size,
+		     struct nouveau_object **pobject)
+{
+	struct nouveau_engctx *engctx;
+	int ret;
+
+	ret = nouveau_engctx_create(parent, engine, oclass, NULL, 256, 256,
+				    NVOBJ_FLAG_ZERO_ALLOC, &engctx);
+	*pobject = nv_object(engctx);
+	return ret;
+}
+
+void
+_nouveau_engctx_dtor(struct nouveau_object *object)
+{
+	nouveau_engctx_destroy(nv_engctx(object));
+}
+
+int
+_nouveau_engctx_init(struct nouveau_object *object)
+{
+	return nouveau_engctx_init(nv_engctx(object));
+}
+
+
+int
+_nouveau_engctx_fini(struct nouveau_object *object, bool suspend)
+{
+	return nouveau_engctx_fini(nv_engctx(object), suspend);
+}
+
+struct nouveau_object *
+nouveau_engctx_get(struct nouveau_engine *engine, u64 addr)
+{
+	struct nouveau_engctx *engctx;
+	unsigned long flags;
+
+	spin_lock_irqsave(&engine->lock, flags);
+	list_for_each_entry(engctx, &engine->contexts, head) {
+		if (engctx->addr == addr) {
+			engctx->save = flags;
+			return nv_object(engctx);
+		}
+	}
+	spin_unlock_irqrestore(&engine->lock, flags);
+	return NULL;
+}
+
+void
+nouveau_engctx_put(struct nouveau_object *object)
+{
+	if (object) {
+		struct nouveau_engine *engine = nv_engine(object->engine);
+		struct nouveau_engctx *engctx = nv_engctx(object);
+		spin_unlock_irqrestore(&engine->lock, engctx->save);
+	}
+}
diff --git a/nvkm/core/engine.c b/nvkm/core/engine.c
new file mode 100644
index 000000000..09b3bd502
--- /dev/null
+++ b/nvkm/core/engine.c
@@ -0,0 +1,55 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/device.h>
+#include <core/engine.h>
+#include <core/option.h>
+
+int
+nouveau_engine_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engobj,
+		       struct nouveau_oclass *oclass, bool enable,
+		       const char *iname, const char *fname,
+		       int length, void **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nouveau_engine *engine;
+	int ret;
+
+	ret = nouveau_subdev_create_(parent, engobj, oclass, NV_ENGINE_CLASS,
+				     iname, fname, length, pobject);
+	engine = *pobject;
+	if (ret)
+		return ret;
+
+	if (!nouveau_boolopt(device->cfgopt, iname, enable)) {
+		if (!enable)
+			nv_warn(engine, "disabled, %s=1 to enable\n", iname);
+		return -ENODEV;
+	}
+
+	INIT_LIST_HEAD(&engine->contexts);
+	spin_lock_init(&engine->lock);
+	return 0;
+}
diff --git a/nvkm/core/enum.c b/nvkm/core/enum.c
new file mode 100644
index 000000000..7cc7133d8
--- /dev/null
+++ b/nvkm/core/enum.c
@@ -0,0 +1,67 @@
+/*
+ * Copyright (C) 2010 Nouveau Project
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <core/os.h>
+#include <core/enum.h>
+
+const struct nouveau_enum *
+nouveau_enum_find(const struct nouveau_enum *en, u32 value)
+{
+	while (en->name) {
+		if (en->value == value)
+			return en;
+		en++;
+	}
+
+	return NULL;
+}
+
+void
+nouveau_enum_print(const struct nouveau_enum *en, u32 value)
+{
+	en = nouveau_enum_find(en, value);
+	if (en)
+		printk("%s", en->name);
+	else
+		printk("(unknown enum 0x%08x)", value);
+}
+
+void
+nouveau_bitfield_print(const struct nouveau_bitfield *bf, u32 value)
+{
+	while (bf->name) {
+		if (value & bf->mask) {
+			printk(" %s", bf->name);
+			value &= ~bf->mask;
+		}
+
+		bf++;
+	}
+
+	if (value)
+		printk(" (unknown bits 0x%08x)", value);
+}
diff --git a/nvkm/core/falcon.c b/nvkm/core/falcon.c
new file mode 100644
index 000000000..6b0843c33
--- /dev/null
+++ b/nvkm/core/falcon.c
@@ -0,0 +1,247 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <core/falcon.h>
+
+#include <subdev/timer.h>
+
+u32
+_nouveau_falcon_rd32(struct nouveau_object *object, u64 addr)
+{
+	struct nouveau_falcon *falcon = (void *)object;
+	return nv_rd32(falcon, falcon->addr + addr);
+}
+
+void
+_nouveau_falcon_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	struct nouveau_falcon *falcon = (void *)object;
+	nv_wr32(falcon, falcon->addr + addr, data);
+}
+
+int
+_nouveau_falcon_init(struct nouveau_object *object)
+{
+	struct nouveau_device *device = nv_device(object);
+	struct nouveau_falcon *falcon = (void *)object;
+	const struct firmware *fw;
+	char name[32] = "internal";
+	int ret, i;
+	u32 caps;
+
+	/* enable engine, and determine its capabilities */
+	ret = nouveau_engine_init(&falcon->base);
+	if (ret)
+		return ret;
+
+	if (device->chipset <  0xa3 ||
+	    device->chipset == 0xaa || device->chipset == 0xac) {
+		falcon->version = 0;
+		falcon->secret  = (falcon->addr == 0x087000) ? 1 : 0;
+	} else {
+		caps = nv_ro32(falcon, 0x12c);
+		falcon->version = (caps & 0x0000000f);
+		falcon->secret  = (caps & 0x00000030) >> 4;
+	}
+
+	caps = nv_ro32(falcon, 0x108);
+	falcon->code.limit = (caps & 0x000001ff) << 8;
+	falcon->data.limit = (caps & 0x0003fe00) >> 1;
+
+	nv_debug(falcon, "falcon version: %d\n", falcon->version);
+	nv_debug(falcon, "secret level: %d\n", falcon->secret);
+	nv_debug(falcon, "code limit: %d\n", falcon->code.limit);
+	nv_debug(falcon, "data limit: %d\n", falcon->data.limit);
+
+	/* wait for 'uc halted' to be signalled before continuing */
+	if (falcon->secret) {
+		nv_wait(falcon, 0x008, 0x00000010, 0x00000010);
+		nv_wo32(falcon, 0x004, 0x00000010);
+	}
+
+	/* disable all interrupts */
+	nv_wo32(falcon, 0x014, 0xffffffff);
+
+	/* no default ucode provided by the engine implementation, try and
+	 * locate a "self-bootstrapping" firmware image for the engine
+	 */
+	if (!falcon->code.data) {
+		snprintf(name, sizeof(name), "nouveau/nv%02x_fuc%03x",
+			 device->chipset, falcon->addr >> 12);
+
+		ret = request_firmware(&fw, name, &device->pdev->dev);
+		if (ret == 0) {
+			falcon->code.data = kmemdup(fw->data, fw->size, GFP_KERNEL);
+			falcon->code.size = fw->size;
+			falcon->data.data = NULL;
+			falcon->data.size = 0;
+			release_firmware(fw);
+		}
+
+		falcon->external = true;
+	}
+
+	/* next step is to try and load "static code/data segment" firmware
+	 * images for the engine
+	 */
+	if (!falcon->code.data) {
+		snprintf(name, sizeof(name), "nouveau/nv%02x_fuc%03xd",
+			 device->chipset, falcon->addr >> 12);
+
+		ret = request_firmware(&fw, name, &device->pdev->dev);
+		if (ret) {
+			nv_error(falcon, "unable to load firmware data\n");
+			return ret;
+		}
+
+		falcon->data.data = kmemdup(fw->data, fw->size, GFP_KERNEL);
+		falcon->data.size = fw->size;
+		release_firmware(fw);
+		if (!falcon->data.data)
+			return -ENOMEM;
+
+		snprintf(name, sizeof(name), "nouveau/nv%02x_fuc%03xc",
+			 device->chipset, falcon->addr >> 12);
+
+		ret = request_firmware(&fw, name, &device->pdev->dev);
+		if (ret) {
+			nv_error(falcon, "unable to load firmware code\n");
+			return ret;
+		}
+
+		falcon->code.data = kmemdup(fw->data, fw->size, GFP_KERNEL);
+		falcon->code.size = fw->size;
+		release_firmware(fw);
+		if (!falcon->code.data)
+			return -ENOMEM;
+	}
+
+	nv_debug(falcon, "firmware: %s (%s)\n", name, falcon->data.data ?
+		 "static code/data segments" : "self-bootstrapping");
+
+	/* ensure any "self-bootstrapping" firmware image is in vram */
+	if (!falcon->data.data && !falcon->core) {
+		ret = nouveau_gpuobj_new(object->parent, NULL,
+					 falcon->code.size, 256, 0,
+					&falcon->core);
+		if (ret) {
+			nv_error(falcon, "core allocation failed, %d\n", ret);
+			return ret;
+		}
+
+		for (i = 0; i < falcon->code.size; i += 4)
+			nv_wo32(falcon->core, i, falcon->code.data[i / 4]);
+	}
+
+	/* upload firmware bootloader (or the full code segments) */
+	if (falcon->core) {
+		if (device->card_type < NV_C0)
+			nv_wo32(falcon, 0x618, 0x04000000);
+		else
+			nv_wo32(falcon, 0x618, 0x00000114);
+		nv_wo32(falcon, 0x11c, 0);
+		nv_wo32(falcon, 0x110, falcon->core->addr >> 8);
+		nv_wo32(falcon, 0x114, 0);
+		nv_wo32(falcon, 0x118, 0x00006610);
+	} else {
+		if (falcon->code.size > falcon->code.limit ||
+		    falcon->data.size > falcon->data.limit) {
+			nv_error(falcon, "ucode exceeds falcon limit(s)\n");
+			return -EINVAL;
+		}
+
+		if (falcon->version < 3) {
+			nv_wo32(falcon, 0xff8, 0x00100000);
+			for (i = 0; i < falcon->code.size / 4; i++)
+				nv_wo32(falcon, 0xff4, falcon->code.data[i]);
+		} else {
+			nv_wo32(falcon, 0x180, 0x01000000);
+			for (i = 0; i < falcon->code.size / 4; i++) {
+				if ((i & 0x3f) == 0)
+					nv_wo32(falcon, 0x188, i >> 6);
+				nv_wo32(falcon, 0x184, falcon->code.data[i]);
+			}
+		}
+	}
+
+	/* upload data segment (if necessary), zeroing the remainder */
+	if (falcon->version < 3) {
+		nv_wo32(falcon, 0xff8, 0x00000000);
+		for (i = 0; !falcon->core && i < falcon->data.size / 4; i++)
+			nv_wo32(falcon, 0xff4, falcon->data.data[i]);
+		for (; i < falcon->data.limit; i += 4)
+			nv_wo32(falcon, 0xff4, 0x00000000);
+	} else {
+		nv_wo32(falcon, 0x1c0, 0x01000000);
+		for (i = 0; !falcon->core && i < falcon->data.size / 4; i++)
+			nv_wo32(falcon, 0x1c4, falcon->data.data[i]);
+		for (; i < falcon->data.limit / 4; i++)
+			nv_wo32(falcon, 0x1c4, 0x00000000);
+	}
+
+	/* start it running */
+	nv_wo32(falcon, 0x10c, 0x00000001); /* BLOCK_ON_FIFO */
+	nv_wo32(falcon, 0x104, 0x00000000); /* ENTRY */
+	nv_wo32(falcon, 0x100, 0x00000002); /* TRIGGER */
+	nv_wo32(falcon, 0x048, 0x00000003); /* FIFO | CHSW */
+	return 0;
+}
+
+int
+_nouveau_falcon_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nouveau_falcon *falcon = (void *)object;
+
+	if (!suspend) {
+		nouveau_gpuobj_ref(NULL, &falcon->core);
+		if (falcon->external) {
+			kfree(falcon->data.data);
+			kfree(falcon->code.data);
+			falcon->code.data = NULL;
+		}
+	}
+
+	nv_mo32(falcon, 0x048, 0x00000003, 0x00000000);
+	nv_wo32(falcon, 0x014, 0xffffffff);
+
+	return nouveau_engine_fini(&falcon->base, suspend);
+}
+
+int
+nouveau_falcon_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, u32 addr, bool enable,
+		       const char *iname, const char *fname,
+		       int length, void **pobject)
+{
+	struct nouveau_falcon *falcon;
+	int ret;
+
+	ret = nouveau_engine_create_(parent, engine, oclass, enable, iname,
+				     fname, length, pobject);
+	falcon = *pobject;
+	if (ret)
+		return ret;
+
+	falcon->addr = addr;
+	return 0;
+}
diff --git a/nvkm/core/gpuobj.c b/nvkm/core/gpuobj.c
new file mode 100644
index 000000000..560b2214c
--- /dev/null
+++ b/nvkm/core/gpuobj.c
@@ -0,0 +1,323 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/gpuobj.h>
+
+#include <subdev/instmem.h>
+#include <subdev/bar.h>
+#include <subdev/vm.h>
+
+void
+nouveau_gpuobj_destroy(struct nouveau_gpuobj *gpuobj)
+{
+	int i;
+
+	if (gpuobj->flags & NVOBJ_FLAG_ZERO_FREE) {
+		for (i = 0; i < gpuobj->size; i += 4)
+			nv_wo32(gpuobj, i, 0x00000000);
+	}
+
+	if (gpuobj->node) {
+		nouveau_mm_free(&nv_gpuobj(gpuobj->parent)->heap,
+				&gpuobj->node);
+	}
+
+	if (gpuobj->heap.block_size)
+		nouveau_mm_fini(&gpuobj->heap);
+
+	nouveau_object_destroy(&gpuobj->base);
+}
+
+int
+nouveau_gpuobj_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, u32 pclass,
+		       struct nouveau_object *pargpu,
+		       u32 size, u32 align, u32 flags,
+		       int length, void **pobject)
+{
+	struct nouveau_instmem *imem = nouveau_instmem(parent);
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nouveau_gpuobj *gpuobj;
+	struct nouveau_mm *heap = NULL;
+	int ret, i;
+	u64 addr;
+
+	*pobject = NULL;
+
+	if (pargpu) {
+		while ((pargpu = nv_pclass(pargpu, NV_GPUOBJ_CLASS))) {
+			if (nv_gpuobj(pargpu)->heap.block_size)
+				break;
+			pargpu = pargpu->parent;
+		}
+
+		if (unlikely(pargpu == NULL)) {
+			nv_error(parent, "no gpuobj heap\n");
+			return -EINVAL;
+		}
+
+		addr =  nv_gpuobj(pargpu)->addr;
+		heap = &nv_gpuobj(pargpu)->heap;
+		atomic_inc(&parent->refcount);
+	} else {
+		ret = imem->alloc(imem, parent, size, align, &parent);
+		pargpu = parent;
+		if (ret)
+			return ret;
+
+		addr = nv_memobj(pargpu)->addr;
+		size = nv_memobj(pargpu)->size;
+
+		if (bar && bar->alloc) {
+			struct nouveau_instobj *iobj = (void *)parent;
+			struct nouveau_mem **mem = (void *)(iobj + 1);
+			struct nouveau_mem *node = *mem;
+			if (!bar->alloc(bar, parent, node, &pargpu)) {
+				nouveau_object_ref(NULL, &parent);
+				parent = pargpu;
+			}
+		}
+	}
+
+	ret = nouveau_object_create_(parent, engine, oclass, pclass |
+				     NV_GPUOBJ_CLASS, length, pobject);
+	nouveau_object_ref(NULL, &parent);
+	gpuobj = *pobject;
+	if (ret)
+		return ret;
+
+	gpuobj->parent = pargpu;
+	gpuobj->flags = flags;
+	gpuobj->addr = addr;
+	gpuobj->size = size;
+
+	if (heap) {
+		ret = nouveau_mm_head(heap, 1, size, size,
+				      max(align, (u32)1), &gpuobj->node);
+		if (ret)
+			return ret;
+
+		gpuobj->addr += gpuobj->node->offset;
+	}
+
+	if (gpuobj->flags & NVOBJ_FLAG_HEAP) {
+		ret = nouveau_mm_init(&gpuobj->heap, 0, gpuobj->size, 1);
+		if (ret)
+			return ret;
+	}
+
+	if (flags & NVOBJ_FLAG_ZERO_ALLOC) {
+		for (i = 0; i < gpuobj->size; i += 4)
+			nv_wo32(gpuobj, i, 0x00000000);
+	}
+
+	return ret;
+}
+
+struct nouveau_gpuobj_class {
+	struct nouveau_object *pargpu;
+	u64 size;
+	u32 align;
+	u32 flags;
+};
+
+static int
+_nouveau_gpuobj_ctor(struct nouveau_object *parent,
+		     struct nouveau_object *engine,
+		     struct nouveau_oclass *oclass, void *data, u32 size,
+		     struct nouveau_object **pobject)
+{
+	struct nouveau_gpuobj_class *args = data;
+	struct nouveau_gpuobj *object;
+	int ret;
+
+	ret = nouveau_gpuobj_create(parent, engine, oclass, 0, args->pargpu,
+				    args->size, args->align, args->flags,
+				    &object);
+	*pobject = nv_object(object);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+void
+_nouveau_gpuobj_dtor(struct nouveau_object *object)
+{
+	nouveau_gpuobj_destroy(nv_gpuobj(object));
+}
+
+int
+_nouveau_gpuobj_init(struct nouveau_object *object)
+{
+	return nouveau_gpuobj_init(nv_gpuobj(object));
+}
+
+int
+_nouveau_gpuobj_fini(struct nouveau_object *object, bool suspend)
+{
+	return nouveau_gpuobj_fini(nv_gpuobj(object), suspend);
+}
+
+u32
+_nouveau_gpuobj_rd32(struct nouveau_object *object, u64 addr)
+{
+	struct nouveau_gpuobj *gpuobj = nv_gpuobj(object);
+	struct nouveau_ofuncs *pfuncs = nv_ofuncs(gpuobj->parent);
+	if (gpuobj->node)
+		addr += gpuobj->node->offset;
+	return pfuncs->rd32(gpuobj->parent, addr);
+}
+
+void
+_nouveau_gpuobj_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	struct nouveau_gpuobj *gpuobj = nv_gpuobj(object);
+	struct nouveau_ofuncs *pfuncs = nv_ofuncs(gpuobj->parent);
+	if (gpuobj->node)
+		addr += gpuobj->node->offset;
+	pfuncs->wr32(gpuobj->parent, addr, data);
+}
+
+static struct nouveau_oclass
+_nouveau_gpuobj_oclass = {
+	.handle = 0x00000000,
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_gpuobj_ctor,
+		.dtor = _nouveau_gpuobj_dtor,
+		.init = _nouveau_gpuobj_init,
+		.fini = _nouveau_gpuobj_fini,
+		.rd32 = _nouveau_gpuobj_rd32,
+		.wr32 = _nouveau_gpuobj_wr32,
+	},
+};
+
+int
+nouveau_gpuobj_new(struct nouveau_object *parent, struct nouveau_object *pargpu,
+		   u32 size, u32 align, u32 flags,
+		   struct nouveau_gpuobj **pgpuobj)
+{
+	struct nouveau_object *engine = parent;
+	struct nouveau_gpuobj_class args = {
+		.pargpu = pargpu,
+		.size = size,
+		.align = align,
+		.flags = flags,
+	};
+
+	if (!nv_iclass(engine, NV_SUBDEV_CLASS))
+		engine = engine->engine;
+	BUG_ON(engine == NULL);
+
+	return nouveau_object_ctor(parent, engine, &_nouveau_gpuobj_oclass,
+				   &args, sizeof(args),
+				   (struct nouveau_object **)pgpuobj);
+}
+
+int
+nouveau_gpuobj_map(struct nouveau_gpuobj *gpuobj, u32 access,
+		   struct nouveau_vma *vma)
+{
+	struct nouveau_bar *bar = nouveau_bar(gpuobj);
+	int ret = -EINVAL;
+
+	if (bar && bar->umap) {
+		struct nouveau_instobj *iobj = (void *)
+			nv_pclass(nv_object(gpuobj), NV_MEMOBJ_CLASS);
+		struct nouveau_mem **mem = (void *)(iobj + 1);
+		ret = bar->umap(bar, *mem, access, vma);
+	}
+
+	return ret;
+}
+
+int
+nouveau_gpuobj_map_vm(struct nouveau_gpuobj *gpuobj, struct nouveau_vm *vm,
+		      u32 access, struct nouveau_vma *vma)
+{
+	struct nouveau_instobj *iobj = (void *)
+		nv_pclass(nv_object(gpuobj), NV_MEMOBJ_CLASS);
+	struct nouveau_mem **mem = (void *)(iobj + 1);
+	int ret;
+
+	ret = nouveau_vm_get(vm, gpuobj->size, 12, access, vma);
+	if (ret)
+		return ret;
+
+	nouveau_vm_map(vma, *mem);
+	return 0;
+}
+
+void
+nouveau_gpuobj_unmap(struct nouveau_vma *vma)
+{
+	if (vma->node) {
+		nouveau_vm_unmap(vma);
+		nouveau_vm_put(vma);
+	}
+}
+
+/* the below is basically only here to support sharing the paged dma object
+ * for PCI(E)GART on <=nv4x chipsets, and should *not* be expected to work
+ * anywhere else.
+ */
+
+static void
+nouveau_gpudup_dtor(struct nouveau_object *object)
+{
+	struct nouveau_gpuobj *gpuobj = (void *)object;
+	nouveau_object_ref(NULL, &gpuobj->parent);
+	nouveau_object_destroy(&gpuobj->base);
+}
+
+static struct nouveau_oclass
+nouveau_gpudup_oclass = {
+	.handle = NV_GPUOBJ_CLASS,
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.dtor = nouveau_gpudup_dtor,
+		.init = nouveau_object_init,
+		.fini = nouveau_object_fini,
+	},
+};
+
+int
+nouveau_gpuobj_dup(struct nouveau_object *parent, struct nouveau_gpuobj *base,
+		   struct nouveau_gpuobj **pgpuobj)
+{
+	struct nouveau_gpuobj *gpuobj;
+	int ret;
+
+	ret = nouveau_object_create(parent, parent->engine,
+				   &nouveau_gpudup_oclass, 0, &gpuobj);
+	*pgpuobj = gpuobj;
+	if (ret)
+		return ret;
+
+	nouveau_object_ref(nv_object(base), &gpuobj->parent);
+	gpuobj->addr = base->addr;
+	gpuobj->size = base->size;
+	return 0;
+}
diff --git a/nvkm/core/handle.c b/nvkm/core/handle.c
new file mode 100644
index 000000000..b8d2cbf8a
--- /dev/null
+++ b/nvkm/core/handle.c
@@ -0,0 +1,223 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/handle.h>
+#include <core/client.h>
+
+#define hprintk(h,l,f,a...) do {                                               \
+	struct nouveau_client *c = nouveau_client((h)->object);                \
+	struct nouveau_handle *p = (h)->parent; u32 n = p ? p->name : ~0;      \
+	nv_printk((c), l, "0x%08x:0x%08x "f, n, (h)->name, ##a);               \
+} while(0)
+
+int
+nouveau_handle_init(struct nouveau_handle *handle)
+{
+	struct nouveau_handle *item;
+	int ret;
+
+	hprintk(handle, TRACE, "init running\n");
+	ret = nouveau_object_inc(handle->object);
+	if (ret)
+		return ret;
+
+	hprintk(handle, TRACE, "init children\n");
+	list_for_each_entry(item, &handle->tree, head) {
+		ret = nouveau_handle_init(item);
+		if (ret)
+			goto fail;
+	}
+
+	hprintk(handle, TRACE, "init completed\n");
+	return 0;
+fail:
+	hprintk(handle, ERROR, "init failed with %d\n", ret);
+	list_for_each_entry_continue_reverse(item, &handle->tree, head) {
+		nouveau_handle_fini(item, false);
+	}
+
+	nouveau_object_dec(handle->object, false);
+	return ret;
+}
+
+int
+nouveau_handle_fini(struct nouveau_handle *handle, bool suspend)
+{
+	static char *name[2] = { "fini", "suspend" };
+	struct nouveau_handle *item;
+	int ret;
+
+	hprintk(handle, TRACE, "%s children\n", name[suspend]);
+	list_for_each_entry(item, &handle->tree, head) {
+		ret = nouveau_handle_fini(item, suspend);
+		if (ret && suspend)
+			goto fail;
+	}
+
+	hprintk(handle, TRACE, "%s running\n", name[suspend]);
+	if (handle->object) {
+		ret = nouveau_object_dec(handle->object, suspend);
+		if (ret && suspend)
+			goto fail;
+	}
+
+	hprintk(handle, TRACE, "%s completed\n", name[suspend]);
+	return 0;
+fail:
+	hprintk(handle, ERROR, "%s failed with %d\n", name[suspend], ret);
+	list_for_each_entry_continue_reverse(item, &handle->tree, head) {
+		int rret = nouveau_handle_init(item);
+		if (rret)
+			hprintk(handle, FATAL, "failed to restart, %d\n", rret);
+	}
+
+	return ret;
+}
+
+int
+nouveau_handle_create(struct nouveau_object *parent, u32 _parent, u32 _handle,
+		      struct nouveau_object *object,
+		      struct nouveau_handle **phandle)
+{
+	struct nouveau_object *namedb;
+	struct nouveau_handle *handle;
+	int ret;
+
+	namedb = parent;
+	while (!nv_iclass(namedb, NV_NAMEDB_CLASS))
+		namedb = namedb->parent;
+
+	handle = *phandle = kzalloc(sizeof(*handle), GFP_KERNEL);
+	if (!handle)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&handle->head);
+	INIT_LIST_HEAD(&handle->tree);
+	handle->name = _handle;
+	handle->priv = ~0;
+
+	ret = nouveau_namedb_insert(nv_namedb(namedb), _handle, object, handle);
+	if (ret) {
+		kfree(handle);
+		return ret;
+	}
+
+	if (nv_parent(parent)->object_attach) {
+		ret = nv_parent(parent)->object_attach(parent, object, _handle);
+		if (ret < 0) {
+			nouveau_handle_destroy(handle);
+			return ret;
+		}
+
+		handle->priv = ret;
+	}
+
+	if (object != namedb) {
+		while (!nv_iclass(namedb, NV_CLIENT_CLASS))
+			namedb = namedb->parent;
+
+		handle->parent = nouveau_namedb_get(nv_namedb(namedb), _parent);
+		if (handle->parent) {
+			list_add(&handle->head, &handle->parent->tree);
+			nouveau_namedb_put(handle->parent);
+		}
+	}
+
+	hprintk(handle, TRACE, "created\n");
+	return 0;
+}
+
+void
+nouveau_handle_destroy(struct nouveau_handle *handle)
+{
+	struct nouveau_handle *item, *temp;
+
+	hprintk(handle, TRACE, "destroy running\n");
+	list_for_each_entry_safe(item, temp, &handle->tree, head) {
+		nouveau_handle_destroy(item);
+	}
+	list_del(&handle->head);
+
+	if (handle->priv != ~0) {
+		struct nouveau_object *parent = handle->parent->object;
+		nv_parent(parent)->object_detach(parent, handle->priv);
+	}
+
+	hprintk(handle, TRACE, "destroy completed\n");
+	nouveau_namedb_remove(handle);
+	kfree(handle);
+}
+
+struct nouveau_object *
+nouveau_handle_ref(struct nouveau_object *parent, u32 name)
+{
+	struct nouveau_object *object = NULL;
+	struct nouveau_handle *handle;
+
+	while (!nv_iclass(parent, NV_NAMEDB_CLASS))
+		parent = parent->parent;
+
+	handle = nouveau_namedb_get(nv_namedb(parent), name);
+	if (handle) {
+		nouveau_object_ref(handle->object, &object);
+		nouveau_namedb_put(handle);
+	}
+
+	return object;
+}
+
+struct nouveau_handle *
+nouveau_handle_get_class(struct nouveau_object *engctx, u16 oclass)
+{
+	struct nouveau_namedb *namedb;
+	if (engctx && (namedb = (void *)nv_pclass(engctx, NV_NAMEDB_CLASS)))
+		return nouveau_namedb_get_class(namedb, oclass);
+	return NULL;
+}
+
+struct nouveau_handle *
+nouveau_handle_get_vinst(struct nouveau_object *engctx, u64 vinst)
+{
+	struct nouveau_namedb *namedb;
+	if (engctx && (namedb = (void *)nv_pclass(engctx, NV_NAMEDB_CLASS)))
+		return nouveau_namedb_get_vinst(namedb, vinst);
+	return NULL;
+}
+
+struct nouveau_handle *
+nouveau_handle_get_cinst(struct nouveau_object *engctx, u32 cinst)
+{
+	struct nouveau_namedb *namedb;
+	if (engctx && (namedb = (void *)nv_pclass(engctx, NV_NAMEDB_CLASS)))
+		return nouveau_namedb_get_cinst(namedb, cinst);
+	return NULL;
+}
+
+void
+nouveau_handle_put(struct nouveau_handle *handle)
+{
+	if (handle)
+		nouveau_namedb_put(handle);
+}
diff --git a/nvkm/core/mm.c b/nvkm/core/mm.c
new file mode 100644
index 000000000..0261a11b2
--- /dev/null
+++ b/nvkm/core/mm.c
@@ -0,0 +1,251 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "core/os.h"
+#include "core/mm.h"
+
+#define node(root, dir) ((root)->nl_entry.dir == &mm->nodes) ? NULL : \
+	list_entry((root)->nl_entry.dir, struct nouveau_mm_node, nl_entry)
+
+void
+nouveau_mm_free(struct nouveau_mm *mm, struct nouveau_mm_node **pthis)
+{
+	struct nouveau_mm_node *this = *pthis;
+
+	if (this) {
+		struct nouveau_mm_node *prev = node(this, prev);
+		struct nouveau_mm_node *next = node(this, next);
+
+		if (prev && prev->type == 0) {
+			prev->length += this->length;
+			list_del(&this->nl_entry);
+			kfree(this); this = prev;
+		}
+
+		if (next && next->type == 0) {
+			next->offset  = this->offset;
+			next->length += this->length;
+			if (this->type == 0)
+				list_del(&this->fl_entry);
+			list_del(&this->nl_entry);
+			kfree(this); this = NULL;
+		}
+
+		if (this && this->type != 0) {
+			list_for_each_entry(prev, &mm->free, fl_entry) {
+				if (this->offset < prev->offset)
+					break;
+			}
+
+			list_add_tail(&this->fl_entry, &prev->fl_entry);
+			this->type = 0;
+		}
+	}
+
+	*pthis = NULL;
+}
+
+static struct nouveau_mm_node *
+region_head(struct nouveau_mm *mm, struct nouveau_mm_node *a, u32 size)
+{
+	struct nouveau_mm_node *b;
+
+	if (a->length == size)
+		return a;
+
+	b = kmalloc(sizeof(*b), GFP_KERNEL);
+	if (unlikely(b == NULL))
+		return NULL;
+
+	b->offset = a->offset;
+	b->length = size;
+	b->type   = a->type;
+	a->offset += size;
+	a->length -= size;
+	list_add_tail(&b->nl_entry, &a->nl_entry);
+	if (b->type == 0)
+		list_add_tail(&b->fl_entry, &a->fl_entry);
+	return b;
+}
+
+int
+nouveau_mm_head(struct nouveau_mm *mm, u8 type, u32 size_max, u32 size_min,
+		u32 align, struct nouveau_mm_node **pnode)
+{
+	struct nouveau_mm_node *prev, *this, *next;
+	u32 mask = align - 1;
+	u32 splitoff;
+	u32 s, e;
+
+	list_for_each_entry(this, &mm->free, fl_entry) {
+		e = this->offset + this->length;
+		s = this->offset;
+
+		prev = node(this, prev);
+		if (prev && prev->type != type)
+			s = roundup(s, mm->block_size);
+
+		next = node(this, next);
+		if (next && next->type != type)
+			e = rounddown(e, mm->block_size);
+
+		s  = (s + mask) & ~mask;
+		e &= ~mask;
+		if (s > e || e - s < size_min)
+			continue;
+
+		splitoff = s - this->offset;
+		if (splitoff && !region_head(mm, this, splitoff))
+			return -ENOMEM;
+
+		this = region_head(mm, this, min(size_max, e - s));
+		if (!this)
+			return -ENOMEM;
+
+		this->type = type;
+		list_del(&this->fl_entry);
+		*pnode = this;
+		return 0;
+	}
+
+	return -ENOSPC;
+}
+
+static struct nouveau_mm_node *
+region_tail(struct nouveau_mm *mm, struct nouveau_mm_node *a, u32 size)
+{
+	struct nouveau_mm_node *b;
+
+	if (a->length == size)
+		return a;
+
+	b = kmalloc(sizeof(*b), GFP_KERNEL);
+	if (unlikely(b == NULL))
+		return NULL;
+
+	a->length -= size;
+	b->offset  = a->offset + a->length;
+	b->length  = size;
+	b->type    = a->type;
+
+	list_add(&b->nl_entry, &a->nl_entry);
+	if (b->type == 0)
+		list_add(&b->fl_entry, &a->fl_entry);
+	return b;
+}
+
+int
+nouveau_mm_tail(struct nouveau_mm *mm, u8 type, u32 size_max, u32 size_min,
+		u32 align, struct nouveau_mm_node **pnode)
+{
+	struct nouveau_mm_node *prev, *this, *next;
+	u32 mask = align - 1;
+
+	list_for_each_entry_reverse(this, &mm->free, fl_entry) {
+		u32 e = this->offset + this->length;
+		u32 s = this->offset;
+		u32 c = 0, a;
+
+		prev = node(this, prev);
+		if (prev && prev->type != type)
+			s = roundup(s, mm->block_size);
+
+		next = node(this, next);
+		if (next && next->type != type) {
+			e = rounddown(e, mm->block_size);
+			c = next->offset - e;
+		}
+
+		s = (s + mask) & ~mask;
+		a = e - s;
+		if (s > e || a < size_min)
+			continue;
+
+		a  = min(a, size_max);
+		s  = (e - a) & ~mask;
+		c += (e - s) - a;
+
+		if (c && !region_tail(mm, this, c))
+			return -ENOMEM;
+
+		this = region_tail(mm, this, a);
+		if (!this)
+			return -ENOMEM;
+
+		this->type = type;
+		list_del(&this->fl_entry);
+		*pnode = this;
+		return 0;
+	}
+
+	return -ENOSPC;
+}
+
+int
+nouveau_mm_init(struct nouveau_mm *mm, u32 offset, u32 length, u32 block)
+{
+	struct nouveau_mm_node *node;
+
+	if (block) {
+		mutex_init(&mm->mutex);
+		INIT_LIST_HEAD(&mm->nodes);
+		INIT_LIST_HEAD(&mm->free);
+		mm->block_size = block;
+		mm->heap_nodes = 0;
+	}
+
+	node = kzalloc(sizeof(*node), GFP_KERNEL);
+	if (!node)
+		return -ENOMEM;
+
+	if (length) {
+		node->offset  = roundup(offset, mm->block_size);
+		node->length  = rounddown(offset + length, mm->block_size);
+		node->length -= node->offset;
+	}
+
+	list_add_tail(&node->nl_entry, &mm->nodes);
+	list_add_tail(&node->fl_entry, &mm->free);
+	mm->heap_nodes++;
+	return 0;
+}
+
+int
+nouveau_mm_fini(struct nouveau_mm *mm)
+{
+	if (nouveau_mm_initialised(mm)) {
+		struct nouveau_mm_node *node, *heap =
+			list_first_entry(&mm->nodes, typeof(*heap), nl_entry);
+		int nodes = 0;
+
+		list_for_each_entry(node, &mm->nodes, nl_entry) {
+			if (WARN_ON(nodes++ == mm->heap_nodes))
+				return -EBUSY;
+		}
+
+		kfree(heap);
+	}
+
+	return 0;
+}
diff --git a/nvkm/core/namedb.c b/nvkm/core/namedb.c
new file mode 100644
index 000000000..1ce95a870
--- /dev/null
+++ b/nvkm/core/namedb.c
@@ -0,0 +1,203 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/namedb.h>
+#include <core/handle.h>
+#include <core/gpuobj.h>
+
+static struct nouveau_handle *
+nouveau_namedb_lookup(struct nouveau_namedb *namedb, u32 name)
+{
+	struct nouveau_handle *handle;
+
+	list_for_each_entry(handle, &namedb->list, node) {
+		if (handle->name == name)
+			return handle;
+	}
+
+	return NULL;
+}
+
+static struct nouveau_handle *
+nouveau_namedb_lookup_class(struct nouveau_namedb *namedb, u16 oclass)
+{
+	struct nouveau_handle *handle;
+
+	list_for_each_entry(handle, &namedb->list, node) {
+		if (nv_mclass(handle->object) == oclass)
+			return handle;
+	}
+
+	return NULL;
+}
+
+static struct nouveau_handle *
+nouveau_namedb_lookup_vinst(struct nouveau_namedb *namedb, u64 vinst)
+{
+	struct nouveau_handle *handle;
+
+	list_for_each_entry(handle, &namedb->list, node) {
+		if (nv_iclass(handle->object, NV_GPUOBJ_CLASS)) {
+			if (nv_gpuobj(handle->object)->addr == vinst)
+				return handle;
+		}
+	}
+
+	return NULL;
+}
+
+static struct nouveau_handle *
+nouveau_namedb_lookup_cinst(struct nouveau_namedb *namedb, u32 cinst)
+{
+	struct nouveau_handle *handle;
+
+	list_for_each_entry(handle, &namedb->list, node) {
+		if (nv_iclass(handle->object, NV_GPUOBJ_CLASS)) {
+			if (nv_gpuobj(handle->object)->node &&
+			    nv_gpuobj(handle->object)->node->offset == cinst)
+				return handle;
+		}
+	}
+
+	return NULL;
+}
+
+int
+nouveau_namedb_insert(struct nouveau_namedb *namedb, u32 name,
+		      struct nouveau_object *object,
+		      struct nouveau_handle *handle)
+{
+	int ret = -EEXIST;
+	write_lock_irq(&namedb->lock);
+	if (!nouveau_namedb_lookup(namedb, name)) {
+		nouveau_object_ref(object, &handle->object);
+		handle->namedb = namedb;
+		list_add(&handle->node, &namedb->list);
+		ret = 0;
+	}
+	write_unlock_irq(&namedb->lock);
+	return ret;
+}
+
+void
+nouveau_namedb_remove(struct nouveau_handle *handle)
+{
+	struct nouveau_namedb *namedb = handle->namedb;
+	struct nouveau_object *object = handle->object;
+	write_lock_irq(&namedb->lock);
+	list_del(&handle->node);
+	write_unlock_irq(&namedb->lock);
+	nouveau_object_ref(NULL, &object);
+}
+
+struct nouveau_handle *
+nouveau_namedb_get(struct nouveau_namedb *namedb, u32 name)
+{
+	struct nouveau_handle *handle;
+	read_lock(&namedb->lock);
+	handle = nouveau_namedb_lookup(namedb, name);
+	if (handle == NULL)
+		read_unlock(&namedb->lock);
+	return handle;
+}
+
+struct nouveau_handle *
+nouveau_namedb_get_class(struct nouveau_namedb *namedb, u16 oclass)
+{
+	struct nouveau_handle *handle;
+	read_lock(&namedb->lock);
+	handle = nouveau_namedb_lookup_class(namedb, oclass);
+	if (handle == NULL)
+		read_unlock(&namedb->lock);
+	return handle;
+}
+
+struct nouveau_handle *
+nouveau_namedb_get_vinst(struct nouveau_namedb *namedb, u64 vinst)
+{
+	struct nouveau_handle *handle;
+	read_lock(&namedb->lock);
+	handle = nouveau_namedb_lookup_vinst(namedb, vinst);
+	if (handle == NULL)
+		read_unlock(&namedb->lock);
+	return handle;
+}
+
+struct nouveau_handle *
+nouveau_namedb_get_cinst(struct nouveau_namedb *namedb, u32 cinst)
+{
+	struct nouveau_handle *handle;
+	read_lock(&namedb->lock);
+	handle = nouveau_namedb_lookup_cinst(namedb, cinst);
+	if (handle == NULL)
+		read_unlock(&namedb->lock);
+	return handle;
+}
+
+void
+nouveau_namedb_put(struct nouveau_handle *handle)
+{
+	if (handle)
+		read_unlock(&handle->namedb->lock);
+}
+
+int
+nouveau_namedb_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, u32 pclass,
+		       struct nouveau_oclass *sclass, u32 engcls,
+		       int length, void **pobject)
+{
+	struct nouveau_namedb *namedb;
+	int ret;
+
+	ret = nouveau_parent_create_(parent, engine, oclass, pclass |
+				     NV_NAMEDB_CLASS, sclass, engcls,
+				     length, pobject);
+	namedb = *pobject;
+	if (ret)
+		return ret;
+
+	rwlock_init(&namedb->lock);
+	INIT_LIST_HEAD(&namedb->list);
+	return 0;
+}
+
+int
+_nouveau_namedb_ctor(struct nouveau_object *parent,
+		     struct nouveau_object *engine,
+		     struct nouveau_oclass *oclass, void *data, u32 size,
+		     struct nouveau_object **pobject)
+{
+	struct nouveau_namedb *object;
+	int ret;
+
+	ret = nouveau_namedb_create(parent, engine, oclass, 0, NULL, 0, &object);
+	*pobject = nv_object(object);
+	if (ret)
+		return ret;
+
+	return 0;
+}
diff --git a/nvkm/core/object.c b/nvkm/core/object.c
new file mode 100644
index 000000000..0daab62ea
--- /dev/null
+++ b/nvkm/core/object.c
@@ -0,0 +1,468 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/parent.h>
+#include <core/namedb.h>
+#include <core/handle.h>
+#include <core/engine.h>
+
+#ifdef NOUVEAU_OBJECT_MAGIC
+static struct list_head _objlist = LIST_HEAD_INIT(_objlist);
+static DEFINE_SPINLOCK(_objlist_lock);
+#endif
+
+int
+nouveau_object_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, u32 pclass,
+		       int size, void **pobject)
+{
+	struct nouveau_object *object;
+
+	object = *pobject = kzalloc(size, GFP_KERNEL);
+	if (!object)
+		return -ENOMEM;
+
+	nouveau_object_ref(parent, &object->parent);
+	nouveau_object_ref(engine, &object->engine);
+	object->oclass = oclass;
+	object->oclass->handle |= pclass;
+	atomic_set(&object->refcount, 1);
+	atomic_set(&object->usecount, 0);
+
+#ifdef NOUVEAU_OBJECT_MAGIC
+	object->_magic = NOUVEAU_OBJECT_MAGIC;
+	spin_lock(&_objlist_lock);
+	list_add(&object->list, &_objlist);
+	spin_unlock(&_objlist_lock);
+#endif
+	return 0;
+}
+
+static int
+_nouveau_object_ctor(struct nouveau_object *parent,
+		     struct nouveau_object *engine,
+		     struct nouveau_oclass *oclass, void *data, u32 size,
+		     struct nouveau_object **pobject)
+{
+	struct nouveau_object *object;
+	int ret;
+
+	ret = nouveau_object_create(parent, engine, oclass, 0, &object);
+	*pobject = nv_object(object);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+void
+nouveau_object_destroy(struct nouveau_object *object)
+{
+#ifdef NOUVEAU_OBJECT_MAGIC
+	spin_lock(&_objlist_lock);
+	list_del(&object->list);
+	spin_unlock(&_objlist_lock);
+#endif
+	nouveau_object_ref(NULL, &object->engine);
+	nouveau_object_ref(NULL, &object->parent);
+	kfree(object);
+}
+
+static void
+_nouveau_object_dtor(struct nouveau_object *object)
+{
+	nouveau_object_destroy(object);
+}
+
+int
+nouveau_object_init(struct nouveau_object *object)
+{
+	return 0;
+}
+
+static int
+_nouveau_object_init(struct nouveau_object *object)
+{
+	return nouveau_object_init(object);
+}
+
+int
+nouveau_object_fini(struct nouveau_object *object, bool suspend)
+{
+	return 0;
+}
+
+static int
+_nouveau_object_fini(struct nouveau_object *object, bool suspend)
+{
+	return nouveau_object_fini(object, suspend);
+}
+
+struct nouveau_ofuncs
+nouveau_object_ofuncs = {
+	.ctor = _nouveau_object_ctor,
+	.dtor = _nouveau_object_dtor,
+	.init = _nouveau_object_init,
+	.fini = _nouveau_object_fini,
+};
+
+int
+nouveau_object_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nouveau_ofuncs *ofuncs = oclass->ofuncs;
+	int ret;
+
+	*pobject = NULL;
+
+	ret = ofuncs->ctor(parent, engine, oclass, data, size, pobject);
+	if (ret < 0) {
+		if (ret != -ENODEV) {
+			nv_error(parent, "failed to create 0x%08x, %d\n",
+				 oclass->handle, ret);
+		}
+
+		if (*pobject) {
+			ofuncs->dtor(*pobject);
+			*pobject = NULL;
+		}
+
+		return ret;
+	}
+
+	nv_debug(*pobject, "created\n");
+	return 0;
+}
+
+static void
+nouveau_object_dtor(struct nouveau_object *object)
+{
+	nv_debug(object, "destroying\n");
+	nv_ofuncs(object)->dtor(object);
+}
+
+void
+nouveau_object_ref(struct nouveau_object *obj, struct nouveau_object **ref)
+{
+	if (obj) {
+		atomic_inc(&obj->refcount);
+		nv_trace(obj, "inc() == %d\n", atomic_read(&obj->refcount));
+	}
+
+	if (*ref) {
+		int dead = atomic_dec_and_test(&(*ref)->refcount);
+		nv_trace(*ref, "dec() == %d\n", atomic_read(&(*ref)->refcount));
+		if (dead)
+			nouveau_object_dtor(*ref);
+	}
+
+	*ref = obj;
+}
+
+int
+nouveau_object_new(struct nouveau_object *client, u32 _parent, u32 _handle,
+		   u16 _oclass, void *data, u32 size,
+		   struct nouveau_object **pobject)
+{
+	struct nouveau_object *parent = NULL;
+	struct nouveau_object *engctx = NULL;
+	struct nouveau_object *object = NULL;
+	struct nouveau_object *engine;
+	struct nouveau_oclass *oclass;
+	struct nouveau_handle *handle;
+	int ret;
+
+	/* lookup parent object and ensure it *is* a parent */
+	parent = nouveau_handle_ref(client, _parent);
+	if (!parent) {
+		nv_error(client, "parent 0x%08x not found\n", _parent);
+		return -ENOENT;
+	}
+
+	if (!nv_iclass(parent, NV_PARENT_CLASS)) {
+		nv_error(parent, "cannot have children\n");
+		ret = -EINVAL;
+		goto fail_class;
+	}
+
+	/* check that parent supports the requested subclass */
+	ret = nouveau_parent_sclass(parent, _oclass, &engine, &oclass);
+	if (ret) {
+		nv_debug(parent, "illegal class 0x%04x\n", _oclass);
+		goto fail_class;
+	}
+
+	/* make sure engine init has been completed *before* any objects
+	 * it controls are created - the constructors may depend on
+	 * state calculated at init (ie. default context construction)
+	 */
+	if (engine) {
+		ret = nouveau_object_inc(engine);
+		if (ret)
+			goto fail_class;
+	}
+
+	/* if engine requires it, create a context object to insert
+	 * between the parent and its children (eg. PGRAPH context)
+	 */
+	if (engine && nv_engine(engine)->cclass) {
+		ret = nouveau_object_ctor(parent, engine,
+					  nv_engine(engine)->cclass,
+					  data, size, &engctx);
+		if (ret)
+			goto fail_engctx;
+	} else {
+		nouveau_object_ref(parent, &engctx);
+	}
+
+	/* finally, create new object and bind it to its handle */
+	ret = nouveau_object_ctor(engctx, engine, oclass, data, size, &object);
+	*pobject = object;
+	if (ret)
+		goto fail_ctor;
+
+	ret = nouveau_object_inc(object);
+	if (ret)
+		goto fail_init;
+
+	ret = nouveau_handle_create(parent, _parent, _handle, object, &handle);
+	if (ret)
+		goto fail_handle;
+
+	ret = nouveau_handle_init(handle);
+	if (ret)
+		nouveau_handle_destroy(handle);
+
+fail_handle:
+	nouveau_object_dec(object, false);
+fail_init:
+	nouveau_object_ref(NULL, &object);
+fail_ctor:
+	nouveau_object_ref(NULL, &engctx);
+fail_engctx:
+	if (engine)
+		nouveau_object_dec(engine, false);
+fail_class:
+	nouveau_object_ref(NULL, &parent);
+	return ret;
+}
+
+int
+nouveau_object_del(struct nouveau_object *client, u32 _parent, u32 _handle)
+{
+	struct nouveau_object *parent = NULL;
+	struct nouveau_object *namedb = NULL;
+	struct nouveau_handle *handle = NULL;
+	int ret = -EINVAL;
+
+	parent = nouveau_handle_ref(client, _parent);
+	if (!parent)
+		return -ENOENT;
+
+	namedb = nv_pclass(parent, NV_NAMEDB_CLASS);
+	if (namedb) {
+		handle = nouveau_namedb_get(nv_namedb(namedb), _handle);
+		if (handle) {
+			nouveau_namedb_put(handle);
+			nouveau_handle_fini(handle, false);
+			nouveau_handle_destroy(handle);
+		}
+	}
+
+	nouveau_object_ref(NULL, &parent);
+	return ret;
+}
+
+int
+nouveau_object_inc(struct nouveau_object *object)
+{
+	int ref = atomic_add_return(1, &object->usecount);
+	int ret;
+
+	nv_trace(object, "use(+1) == %d\n", atomic_read(&object->usecount));
+	if (ref != 1)
+		return 0;
+
+	nv_trace(object, "initialising...\n");
+	if (object->parent) {
+		ret = nouveau_object_inc(object->parent);
+		if (ret) {
+			nv_error(object, "parent failed, %d\n", ret);
+			goto fail_parent;
+		}
+	}
+
+	if (object->engine) {
+		mutex_lock(&nv_subdev(object->engine)->mutex);
+		ret = nouveau_object_inc(object->engine);
+		mutex_unlock(&nv_subdev(object->engine)->mutex);
+		if (ret) {
+			nv_error(object, "engine failed, %d\n", ret);
+			goto fail_engine;
+		}
+	}
+
+	ret = nv_ofuncs(object)->init(object);
+	if (ret) {
+		nv_error(object, "init failed, %d\n", ret);
+		goto fail_self;
+	}
+
+	nv_debug(object, "initialised\n");
+	return 0;
+
+fail_self:
+	if (object->engine) {
+		mutex_lock(&nv_subdev(object->engine)->mutex);
+		nouveau_object_dec(object->engine, false);
+		mutex_unlock(&nv_subdev(object->engine)->mutex);
+	}
+fail_engine:
+	if (object->parent)
+		 nouveau_object_dec(object->parent, false);
+fail_parent:
+	atomic_dec(&object->usecount);
+	return ret;
+}
+
+static int
+nouveau_object_decf(struct nouveau_object *object)
+{
+	int ret;
+
+	nv_trace(object, "stopping...\n");
+
+	ret = nv_ofuncs(object)->fini(object, false);
+	if (ret)
+		nv_warn(object, "failed fini, %d\n", ret);
+
+	if (object->engine) {
+		mutex_lock(&nv_subdev(object->engine)->mutex);
+		nouveau_object_dec(object->engine, false);
+		mutex_unlock(&nv_subdev(object->engine)->mutex);
+	}
+
+	if (object->parent)
+		nouveau_object_dec(object->parent, false);
+
+	nv_debug(object, "stopped\n");
+	return 0;
+}
+
+static int
+nouveau_object_decs(struct nouveau_object *object)
+{
+	int ret, rret;
+
+	nv_trace(object, "suspending...\n");
+
+	ret = nv_ofuncs(object)->fini(object, true);
+	if (ret) {
+		nv_error(object, "failed suspend, %d\n", ret);
+		return ret;
+	}
+
+	if (object->engine) {
+		mutex_lock(&nv_subdev(object->engine)->mutex);
+		ret = nouveau_object_dec(object->engine, true);
+		mutex_unlock(&nv_subdev(object->engine)->mutex);
+		if (ret) {
+			nv_warn(object, "engine failed suspend, %d\n", ret);
+			goto fail_engine;
+		}
+	}
+
+	if (object->parent) {
+		ret = nouveau_object_dec(object->parent, true);
+		if (ret) {
+			nv_warn(object, "parent failed suspend, %d\n", ret);
+			goto fail_parent;
+		}
+	}
+
+	nv_debug(object, "suspended\n");
+	return 0;
+
+fail_parent:
+	if (object->engine) {
+		mutex_lock(&nv_subdev(object->engine)->mutex);
+		rret = nouveau_object_inc(object->engine);
+		mutex_unlock(&nv_subdev(object->engine)->mutex);
+		if (rret)
+			nv_fatal(object, "engine failed to reinit, %d\n", rret);
+	}
+
+fail_engine:
+	rret = nv_ofuncs(object)->init(object);
+	if (rret)
+		nv_fatal(object, "failed to reinit, %d\n", rret);
+
+	return ret;
+}
+
+int
+nouveau_object_dec(struct nouveau_object *object, bool suspend)
+{
+	int ref = atomic_add_return(-1, &object->usecount);
+	int ret;
+
+	nv_trace(object, "use(-1) == %d\n", atomic_read(&object->usecount));
+
+	if (ref == 0) {
+		if (suspend)
+			ret = nouveau_object_decs(object);
+		else
+			ret = nouveau_object_decf(object);
+
+		if (ret) {
+			atomic_inc(&object->usecount);
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+void
+nouveau_object_debug(void)
+{
+#ifdef NOUVEAU_OBJECT_MAGIC
+	struct nouveau_object *object;
+	if (!list_empty(&_objlist)) {
+		nv_fatal(NULL, "*******************************************\n");
+		nv_fatal(NULL, "* AIIIII! object(s) still exist!!!\n");
+		nv_fatal(NULL, "*******************************************\n");
+		list_for_each_entry(object, &_objlist, list) {
+			nv_fatal(object, "%p/%p/%d/%d\n",
+				 object->parent, object->engine,
+				 atomic_read(&object->refcount),
+				 atomic_read(&object->usecount));
+		}
+	}
+#endif
+}
diff --git a/nvkm/core/option.c b/nvkm/core/option.c
new file mode 100644
index 000000000..62a432ea3
--- /dev/null
+++ b/nvkm/core/option.c
@@ -0,0 +1,131 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/option.h>
+#include <core/debug.h>
+
+/* compares unterminated string 'str' with zero-terminated string 'cmp' */
+static inline int
+strncasecmpz(const char *str, const char *cmp, size_t len)
+{
+	if (strlen(cmp) != len)
+		return len;
+	return strncasecmp(str, cmp, len);
+}
+
+const char *
+nouveau_stropt(const char *optstr, const char *opt, int *arglen)
+{
+	while (optstr && *optstr != '\0') {
+		int len = strcspn(optstr, ",=");
+		switch (optstr[len]) {
+		case '=':
+			if (!strncasecmpz(optstr, opt, len)) {
+				optstr += len + 1;
+				*arglen = strcspn(optstr, ",=");
+				return *arglen ? optstr : NULL;
+			}
+			optstr++;
+			break;
+		case ',':
+			optstr++;
+			break;
+		default:
+			break;
+		}
+		optstr += len;
+	}
+
+	return NULL;
+}
+
+bool
+nouveau_boolopt(const char *optstr, const char *opt, bool value)
+{
+	int arglen;
+
+	optstr = nouveau_stropt(optstr, opt, &arglen);
+	if (optstr) {
+		if (!strncasecmpz(optstr, "0", arglen) ||
+		    !strncasecmpz(optstr, "no", arglen) ||
+		    !strncasecmpz(optstr, "off", arglen) ||
+		    !strncasecmpz(optstr, "false", arglen))
+			value = false;
+		else
+		if (!strncasecmpz(optstr, "1", arglen) ||
+		    !strncasecmpz(optstr, "yes", arglen) ||
+		    !strncasecmpz(optstr, "on", arglen) ||
+		    !strncasecmpz(optstr, "true", arglen))
+			value = true;
+	}
+
+	return value;
+}
+
+int
+nouveau_dbgopt(const char *optstr, const char *sub)
+{
+	int mode = 1, level = CONFIG_NOUVEAU_DEBUG_DEFAULT;
+
+	while (optstr) {
+		int len = strcspn(optstr, ",=");
+		switch (optstr[len]) {
+		case '=':
+			if (strncasecmpz(optstr, sub, len))
+				mode = 0;
+			optstr++;
+			break;
+		default:
+			if (mode) {
+				if (!strncasecmpz(optstr, "fatal", len))
+					level = NV_DBG_FATAL;
+				else if (!strncasecmpz(optstr, "error", len))
+					level = NV_DBG_ERROR;
+				else if (!strncasecmpz(optstr, "warn", len))
+					level = NV_DBG_WARN;
+				else if (!strncasecmpz(optstr, "info", len))
+					level = NV_DBG_INFO;
+				else if (!strncasecmpz(optstr, "debug", len))
+					level = NV_DBG_DEBUG;
+				else if (!strncasecmpz(optstr, "trace", len))
+					level = NV_DBG_TRACE;
+				else if (!strncasecmpz(optstr, "paranoia", len))
+					level = NV_DBG_PARANOIA;
+				else if (!strncasecmpz(optstr, "spam", len))
+					level = NV_DBG_SPAM;
+			}
+
+			if (optstr[len] != '\0') {
+				optstr++;
+				mode = 1;
+				break;
+			}
+
+			return level;
+		}
+		optstr += len;
+	}
+
+	return level;
+}
diff --git a/nvkm/core/parent.c b/nvkm/core/parent.c
new file mode 100644
index 000000000..db7c54943
--- /dev/null
+++ b/nvkm/core/parent.c
@@ -0,0 +1,122 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/parent.h>
+
+int
+nouveau_parent_sclass(struct nouveau_object *parent, u16 handle,
+		      struct nouveau_object **pengine,
+		      struct nouveau_oclass **poclass)
+{
+	struct nouveau_sclass *sclass;
+	struct nouveau_engine *engine;
+	struct nouveau_oclass *oclass;
+	u64 mask;
+
+	sclass = nv_parent(parent)->sclass;
+	while (sclass) {
+		if ((sclass->oclass->handle & 0xffff) == handle) {
+			*pengine = parent->engine;
+			*poclass = sclass->oclass;
+			return 0;
+		}
+
+		sclass = sclass->sclass;
+	}
+
+	mask = nv_parent(parent)->engine;
+	while (mask) {
+		int i = ffsll(mask) - 1;
+
+		if ((engine = nouveau_engine(parent, i))) {
+			oclass = engine->sclass;
+			while (oclass->ofuncs) {
+				if ((oclass->handle & 0xffff) == handle) {
+					*pengine = nv_object(engine);
+					*poclass = oclass;
+					return 0;
+				}
+				oclass++;
+			}
+		}
+
+		mask &= ~(1ULL << i);
+	}
+
+	return -EINVAL;
+}
+
+int
+nouveau_parent_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, u32 pclass,
+		       struct nouveau_oclass *sclass, u64 engcls,
+		       int size, void **pobject)
+{
+	struct nouveau_parent *object;
+	struct nouveau_sclass *nclass;
+	int ret;
+
+	ret = nouveau_object_create_(parent, engine, oclass, pclass |
+				     NV_PARENT_CLASS, size, pobject);
+	object = *pobject;
+	if (ret)
+		return ret;
+
+	while (sclass && sclass->ofuncs) {
+		nclass = kzalloc(sizeof(*nclass), GFP_KERNEL);
+		if (!nclass)
+			return -ENOMEM;
+
+		nclass->sclass = object->sclass;
+		object->sclass = nclass;
+		nclass->engine = engine ? nv_engine(engine) : NULL;
+		nclass->oclass = sclass;
+		sclass++;
+	}
+
+	object->engine = engcls;
+	return 0;
+}
+
+void
+nouveau_parent_destroy(struct nouveau_parent *parent)
+{
+	struct nouveau_sclass *sclass;
+
+	while ((sclass = parent->sclass)) {
+		parent->sclass = sclass->sclass;
+		kfree(sclass);
+	}
+
+	nouveau_object_destroy(&parent->base);
+}
+
+
+void
+_nouveau_parent_dtor(struct nouveau_object *object)
+{
+	nouveau_parent_destroy(nv_parent(object));
+}
diff --git a/nvkm/core/printk.c b/nvkm/core/printk.c
new file mode 100644
index 000000000..6161eaf54
--- /dev/null
+++ b/nvkm/core/printk.c
@@ -0,0 +1,74 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/client.h>
+#include <core/subdev.h>
+#include <core/printk.h>
+
+void
+nv_printk_(struct nouveau_object *object, const char *pfx, int level,
+	   const char *fmt, ...)
+{
+	static const char name[] = { '!', 'E', 'W', ' ', 'D', 'T', 'P', 'S' };
+	char mfmt[256];
+	va_list args;
+
+	if (object && !nv_iclass(object, NV_CLIENT_CLASS)) {
+		struct nouveau_object *device = object;
+		struct nouveau_object *subdev = object;
+		char obuf[64], *ofmt = "";
+
+		if (object->engine) {
+			snprintf(obuf, sizeof(obuf), "[0x%08x][%p]",
+				 nv_hclass(object), object);
+			ofmt = obuf;
+			subdev = object->engine;
+			device = object->engine;
+		}
+
+		if (subdev->parent)
+			device = subdev->parent;
+
+		if (level > nv_subdev(subdev)->debug)
+			return;
+
+		snprintf(mfmt, sizeof(mfmt), "%snouveau %c[%8s][%s]%s %s", pfx,
+			 name[level], nv_subdev(subdev)->name,
+			 nv_device(device)->name, ofmt, fmt);
+	} else
+	if (object && nv_iclass(object, NV_CLIENT_CLASS)) {
+		if (level > nv_client(object)->debug)
+			return;
+
+		snprintf(mfmt, sizeof(mfmt), "%snouveau %c[%8s] %s", pfx,
+			 name[level], nv_client(object)->name, fmt);
+	} else {
+		snprintf(mfmt, sizeof(mfmt), "%snouveau: %s", pfx, fmt);
+	}
+
+	va_start(args, fmt);
+	vprintk(mfmt, args);
+	va_end(args);
+}
diff --git a/nvkm/core/ramht.c b/nvkm/core/ramht.c
new file mode 100644
index 000000000..86a64045d
--- /dev/null
+++ b/nvkm/core/ramht.c
@@ -0,0 +1,109 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <core/object.h>
+#include <core/ramht.h>
+#include <core/math.h>
+
+#include <subdev/bar.h>
+
+static u32
+nouveau_ramht_hash(struct nouveau_ramht *ramht, int chid, u32 handle)
+{
+	u32 hash = 0;
+
+	while (handle) {
+		hash ^= (handle & ((1 << ramht->bits) - 1));
+		handle >>= ramht->bits;
+	}
+
+	hash ^= chid << (ramht->bits - 4);
+	hash  = hash << 3;
+	return hash;
+}
+
+int
+nouveau_ramht_insert(struct nouveau_ramht *ramht, int chid,
+		     u32 handle, u32 context)
+{
+	struct nouveau_bar *bar = nouveau_bar(ramht);
+	u32 co, ho;
+
+	co = ho = nouveau_ramht_hash(ramht, chid, handle);
+	do {
+		if (!nv_ro32(ramht, co + 4)) {
+			nv_wo32(ramht, co + 0, handle);
+			nv_wo32(ramht, co + 4, context);
+			if (bar)
+				bar->flush(bar);
+			return co;
+		}
+
+		co += 8;
+		if (co >= nv_gpuobj(ramht)->size)
+			co = 0;
+	} while (co != ho);
+
+	return -ENOMEM;
+}
+
+void
+nouveau_ramht_remove(struct nouveau_ramht *ramht, int cookie)
+{
+	struct nouveau_bar *bar = nouveau_bar(ramht);
+	nv_wo32(ramht, cookie + 0, 0x00000000);
+	nv_wo32(ramht, cookie + 4, 0x00000000);
+	if (bar)
+		bar->flush(bar);
+}
+
+static struct nouveau_oclass
+nouveau_ramht_oclass = {
+	.handle = 0x0000abcd,
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = NULL,
+		.dtor = _nouveau_gpuobj_dtor,
+		.init = _nouveau_gpuobj_init,
+		.fini = _nouveau_gpuobj_fini,
+		.rd32 = _nouveau_gpuobj_rd32,
+		.wr32 = _nouveau_gpuobj_wr32,
+	},
+};
+
+int
+nouveau_ramht_new(struct nouveau_object *parent, struct nouveau_object *pargpu,
+		  u32 size, u32 align, struct nouveau_ramht **pramht)
+{
+	struct nouveau_ramht *ramht;
+	int ret;
+
+	ret = nouveau_gpuobj_create(parent, parent->engine ?
+				    parent->engine : parent, /* <nv50 ramht */
+				    &nouveau_ramht_oclass, 0, pargpu, size,
+				    align, NVOBJ_FLAG_ZERO_ALLOC, &ramht);
+	*pramht = ramht;
+	if (ret)
+		return ret;
+
+	ramht->bits = log2i(nv_gpuobj(ramht)->size >> 3);
+	return 0;
+}
diff --git a/nvkm/core/subdev.c b/nvkm/core/subdev.c
new file mode 100644
index 000000000..f74c30aa3
--- /dev/null
+++ b/nvkm/core/subdev.c
@@ -0,0 +1,115 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/subdev.h>
+#include <core/device.h>
+#include <core/option.h>
+
+void
+nouveau_subdev_reset(struct nouveau_object *subdev)
+{
+	nv_trace(subdev, "resetting...\n");
+	nv_ofuncs(subdev)->fini(subdev, false);
+	nv_debug(subdev, "reset\n");
+}
+
+int
+nouveau_subdev_init(struct nouveau_subdev *subdev)
+{
+	int ret = nouveau_object_init(&subdev->base);
+	if (ret)
+		return ret;
+
+	nouveau_subdev_reset(&subdev->base);
+	return 0;
+}
+
+int
+_nouveau_subdev_init(struct nouveau_object *object)
+{
+	return nouveau_subdev_init(nv_subdev(object));
+}
+
+int
+nouveau_subdev_fini(struct nouveau_subdev *subdev, bool suspend)
+{
+	if (subdev->unit) {
+		nv_mask(subdev, 0x000200, subdev->unit, 0x00000000);
+		nv_mask(subdev, 0x000200, subdev->unit, subdev->unit);
+	}
+
+	return nouveau_object_fini(&subdev->base, suspend);
+}
+
+int
+_nouveau_subdev_fini(struct nouveau_object *object, bool suspend)
+{
+	return nouveau_subdev_fini(nv_subdev(object), suspend);
+}
+
+void
+nouveau_subdev_destroy(struct nouveau_subdev *subdev)
+{
+	int subidx = nv_hclass(subdev) & 0xff;
+	nv_device(subdev)->subdev[subidx] = NULL;
+	nouveau_object_destroy(&subdev->base);
+}
+
+void
+_nouveau_subdev_dtor(struct nouveau_object *object)
+{
+	nouveau_subdev_destroy(nv_subdev(object));
+}
+
+int
+nouveau_subdev_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, u32 pclass,
+		       const char *subname, const char *sysname,
+		       int size, void **pobject)
+{
+	struct nouveau_subdev *subdev;
+	int ret;
+
+	ret = nouveau_object_create_(parent, engine, oclass, pclass |
+				     NV_SUBDEV_CLASS, size, pobject);
+	subdev = *pobject;
+	if (ret)
+		return ret;
+
+	mutex_init(&subdev->mutex);
+	subdev->name = subname;
+
+	if (parent) {
+		struct nouveau_device *device = nv_device(parent);
+		int subidx = nv_hclass(subdev) & 0xff;
+
+		subdev->debug = nouveau_dbgopt(device->dbgopt, subname);
+		subdev->mmio  = nv_subdev(device)->mmio;
+		device->subdev[subidx] = *pobject;
+	}
+
+	return 0;
+}
diff --git a/nvkm/engine/Makefile.am b/nvkm/engine/Makefile.am
new file mode 100644
index 000000000..743767606
--- /dev/null
+++ b/nvkm/engine/Makefile.am
@@ -0,0 +1,16 @@
+SUBDIRS = dmaobj bsp copy crypt disp fifo graph mpeg ppp software vp
+
+noinst_LTLIBRARIES = libengine.la
+
+libengine_la_SOURCES =
+libengine_la_LIBADD  = dmaobj/libdmaobj.la \
+		       bsp/libbsp.la \
+		       copy/libcopy.la \
+		       crypt/libcrypt.la \
+		       disp/libdisp.la \
+		       fifo/libfifo.la \
+		       graph/libgraph.la \
+		       mpeg/libmpeg.la \
+		       ppp/libppp.la \
+		       software/libsoftware.la \
+		       vp/libvp.la
diff --git a/nvkm/engine/Makefile.engine b/nvkm/engine/Makefile.engine
new file mode 100644
index 000000000..513c6c87b
--- /dev/null
+++ b/nvkm/engine/Makefile.engine
@@ -0,0 +1,3 @@
+AM_CPPFLAGS = -I${top_srcdir}/nvkm \
+	      -I${top_srcdir}/nvkm/include \
+	      -I${top_srcdir}/lib
diff --git a/nvkm/engine/bsp/Makefile.am b/nvkm/engine/bsp/Makefile.am
new file mode 100644
index 000000000..c731b3afc
--- /dev/null
+++ b/nvkm/engine/bsp/Makefile.am
@@ -0,0 +1,7 @@
+noinst_LTLIBRARIES = libbsp.la
+
+libbsp_la_SOURCES = nv84.c \
+		    nvc0.c \
+		    nve0.c
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/bsp/nv84.c b/nvkm/engine/bsp/nv84.c
new file mode 100644
index 000000000..1d9f614cb
--- /dev/null
+++ b/nvkm/engine/bsp/nv84.c
@@ -0,0 +1,93 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/engctx.h>
+#include <core/class.h>
+
+#include <engine/bsp.h>
+
+struct nv84_bsp_priv {
+	struct nouveau_engine base;
+};
+
+/*******************************************************************************
+ * BSP object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv84_bsp_sclass[] = {
+	{},
+};
+
+/*******************************************************************************
+ * BSP context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv84_bsp_cclass = {
+	.handle = NV_ENGCTX(BSP, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_engctx_ctor,
+		.dtor = _nouveau_engctx_dtor,
+		.init = _nouveau_engctx_init,
+		.fini = _nouveau_engctx_fini,
+		.rd32 = _nouveau_engctx_rd32,
+		.wr32 = _nouveau_engctx_wr32,
+	},
+};
+
+/*******************************************************************************
+ * BSP engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv84_bsp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nv84_bsp_priv *priv;
+	int ret;
+
+	ret = nouveau_engine_create(parent, engine, oclass, true,
+				    "PBSP", "bsp", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x04008000;
+	nv_engine(priv)->cclass = &nv84_bsp_cclass;
+	nv_engine(priv)->sclass = nv84_bsp_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nv84_bsp_oclass = {
+	.handle = NV_ENGINE(BSP, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv84_bsp_ctor,
+		.dtor = _nouveau_engine_dtor,
+		.init = _nouveau_engine_init,
+		.fini = _nouveau_engine_fini,
+	},
+};
diff --git a/nvkm/engine/bsp/nvc0.c b/nvkm/engine/bsp/nvc0.c
new file mode 100644
index 000000000..0a5aa6bb0
--- /dev/null
+++ b/nvkm/engine/bsp/nvc0.c
@@ -0,0 +1,110 @@
+/*
+ * Copyright 2012 Maarten Lankhorst
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Maarten Lankhorst
+ */
+
+#include <core/falcon.h>
+
+#include <engine/bsp.h>
+
+struct nvc0_bsp_priv {
+	struct nouveau_falcon base;
+};
+
+/*******************************************************************************
+ * BSP object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nvc0_bsp_sclass[] = {
+	{ 0x90b1, &nouveau_object_ofuncs },
+	{},
+};
+
+/*******************************************************************************
+ * PBSP context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nvc0_bsp_cclass = {
+	.handle = NV_ENGCTX(BSP, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_falcon_context_ctor,
+		.dtor = _nouveau_falcon_context_dtor,
+		.init = _nouveau_falcon_context_init,
+		.fini = _nouveau_falcon_context_fini,
+		.rd32 = _nouveau_falcon_context_rd32,
+		.wr32 = _nouveau_falcon_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PBSP engine/subdev functions
+ ******************************************************************************/
+
+static int
+nvc0_bsp_init(struct nouveau_object *object)
+{
+	struct nvc0_bsp_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_falcon_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x084010, 0x0000fff2);
+	nv_wr32(priv, 0x08401c, 0x0000fff2);
+	return 0;
+}
+
+static int
+nvc0_bsp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nvc0_bsp_priv *priv;
+	int ret;
+
+	ret = nouveau_falcon_create(parent, engine, oclass, 0x084000, true,
+				    "PBSP", "bsp", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00008000;
+	nv_engine(priv)->cclass = &nvc0_bsp_cclass;
+	nv_engine(priv)->sclass = nvc0_bsp_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_bsp_oclass = {
+	.handle = NV_ENGINE(BSP, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_bsp_ctor,
+		.dtor = _nouveau_falcon_dtor,
+		.init = nvc0_bsp_init,
+		.fini = _nouveau_falcon_fini,
+		.rd32 = _nouveau_falcon_rd32,
+		.wr32 = _nouveau_falcon_wr32,
+	},
+};
diff --git a/nvkm/engine/bsp/nve0.c b/nvkm/engine/bsp/nve0.c
new file mode 100644
index 000000000..d4f23bbd7
--- /dev/null
+++ b/nvkm/engine/bsp/nve0.c
@@ -0,0 +1,110 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/falcon.h>
+
+#include <engine/bsp.h>
+
+struct nve0_bsp_priv {
+	struct nouveau_falcon base;
+};
+
+/*******************************************************************************
+ * BSP object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nve0_bsp_sclass[] = {
+	{ 0x95b1, &nouveau_object_ofuncs },
+	{},
+};
+
+/*******************************************************************************
+ * PBSP context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nve0_bsp_cclass = {
+	.handle = NV_ENGCTX(BSP, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_falcon_context_ctor,
+		.dtor = _nouveau_falcon_context_dtor,
+		.init = _nouveau_falcon_context_init,
+		.fini = _nouveau_falcon_context_fini,
+		.rd32 = _nouveau_falcon_context_rd32,
+		.wr32 = _nouveau_falcon_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PBSP engine/subdev functions
+ ******************************************************************************/
+
+static int
+nve0_bsp_init(struct nouveau_object *object)
+{
+	struct nve0_bsp_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_falcon_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x084010, 0x0000fff2);
+	nv_wr32(priv, 0x08401c, 0x0000fff2);
+	return 0;
+}
+
+static int
+nve0_bsp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nve0_bsp_priv *priv;
+	int ret;
+
+	ret = nouveau_falcon_create(parent, engine, oclass, 0x084000, true,
+				    "PBSP", "bsp", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00008000;
+	nv_engine(priv)->cclass = &nve0_bsp_cclass;
+	nv_engine(priv)->sclass = nve0_bsp_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nve0_bsp_oclass = {
+	.handle = NV_ENGINE(BSP, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nve0_bsp_ctor,
+		.dtor = _nouveau_falcon_dtor,
+		.init = nve0_bsp_init,
+		.fini = _nouveau_falcon_fini,
+		.rd32 = _nouveau_falcon_rd32,
+		.wr32 = _nouveau_falcon_wr32,
+	},
+};
diff --git a/nvkm/engine/copy/Makefile.am b/nvkm/engine/copy/Makefile.am
new file mode 100644
index 000000000..63d5ac8ca
--- /dev/null
+++ b/nvkm/engine/copy/Makefile.am
@@ -0,0 +1,17 @@
+noinst_LTLIBRARIES = libcopy.la
+
+BUILT_SOURCES = fuc/nva3.fuc.h \
+		fuc/nvc0.fuc.h
+CLEANFILES = ${BUILT_SOURCES}
+
+libcopy_la_SOURCES = ${BUILT_SOURCES} \
+		     nva3.c \
+		     nvc0.c \
+		     nve0.c
+
+fuc/nva3.fuc.h: fuc/nva3.fuc
+	m4 -DNVA3 fuc/nva3.fuc | envyas -a -w -m fuc -V fuc3 -o fuc/nva3.fuc.h
+fuc/nvc0.fuc.h: fuc/nva3.fuc
+	m4 -DNVC0 fuc/nva3.fuc | envyas -a -w -m fuc -V fuc3 -o fuc/nvc0.fuc.h
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/copy/fuc/nva3.fuc b/nvkm/engine/copy/fuc/nva3.fuc
new file mode 100644
index 000000000..219850d53
--- /dev/null
+++ b/nvkm/engine/copy/fuc/nva3.fuc
@@ -0,0 +1,872 @@
+/* fuc microcode for copy engine on nva3- chipsets
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+/* To build for nva3:nvc0
+ *    m4 -DNVA3 nva3_copy.fuc | envyas -a -w -m fuc -V nva3 -o nva3_copy.fuc.h
+ *
+ * To build for nvc0-
+ *    m4 -DNVC0 nva3_copy.fuc | envyas -a -w -m fuc -V nva3 -o nvc0_copy.fuc.h
+ */
+
+ifdef(`NVA3',
+.section #nva3_pcopy_data
+,
+.section #nvc0_pcopy_data
+)
+
+ctx_object:                   .b32 0
+ifdef(`NVA3',
+ctx_dma:
+ctx_dma_query:                .b32 0
+ctx_dma_src:                  .b32 0
+ctx_dma_dst:                  .b32 0
+,)
+.equ #ctx_dma_count 3
+ctx_query_address_high:       .b32 0
+ctx_query_address_low:        .b32 0
+ctx_query_counter:            .b32 0
+ctx_src_address_high:         .b32 0
+ctx_src_address_low:          .b32 0
+ctx_src_pitch:                .b32 0
+ctx_src_tile_mode:            .b32 0
+ctx_src_xsize:                .b32 0
+ctx_src_ysize:                .b32 0
+ctx_src_zsize:                .b32 0
+ctx_src_zoff:                 .b32 0
+ctx_src_xoff:                 .b32 0
+ctx_src_yoff:                 .b32 0
+ctx_src_cpp:                  .b32 0
+ctx_dst_address_high:         .b32 0
+ctx_dst_address_low:          .b32 0
+ctx_dst_pitch:                .b32 0
+ctx_dst_tile_mode:            .b32 0
+ctx_dst_xsize:                .b32 0
+ctx_dst_ysize:                .b32 0
+ctx_dst_zsize:                .b32 0
+ctx_dst_zoff:                 .b32 0
+ctx_dst_xoff:                 .b32 0
+ctx_dst_yoff:                 .b32 0
+ctx_dst_cpp:                  .b32 0
+ctx_format:                   .b32 0
+ctx_swz_const0:               .b32 0
+ctx_swz_const1:               .b32 0
+ctx_xcnt:                     .b32 0
+ctx_ycnt:                     .b32 0
+.align 256
+
+dispatch_table:
+// mthd 0x0000, NAME
+.b16 0x000 1
+.b32 #ctx_object                     ~0xffffffff
+// mthd 0x0100, NOP
+.b16 0x040 1
+.b32 0x00010000 + #cmd_nop           ~0xffffffff
+// mthd 0x0140, PM_TRIGGER
+.b16 0x050 1
+.b32 0x00010000 + #cmd_pm_trigger    ~0xffffffff
+ifdef(`NVA3', `
+// mthd 0x0180-0x018c, DMA_
+.b16 0x060 #ctx_dma_count
+dispatch_dma:
+.b32 0x00010000 + #cmd_dma           ~0xffffffff
+.b32 0x00010000 + #cmd_dma           ~0xffffffff
+.b32 0x00010000 + #cmd_dma           ~0xffffffff
+',)
+// mthd 0x0200-0x0218, SRC_TILE
+.b16 0x80 7
+.b32 #ctx_src_tile_mode              ~0x00000fff
+.b32 #ctx_src_xsize                  ~0x0007ffff
+.b32 #ctx_src_ysize                  ~0x00001fff
+.b32 #ctx_src_zsize                  ~0x000007ff
+.b32 #ctx_src_zoff                   ~0x00000fff
+.b32 #ctx_src_xoff                   ~0x0007ffff
+.b32 #ctx_src_yoff                   ~0x00001fff
+// mthd 0x0220-0x0238, DST_TILE
+.b16 0x88 7
+.b32 #ctx_dst_tile_mode              ~0x00000fff
+.b32 #ctx_dst_xsize                  ~0x0007ffff
+.b32 #ctx_dst_ysize                  ~0x00001fff
+.b32 #ctx_dst_zsize                  ~0x000007ff
+.b32 #ctx_dst_zoff                   ~0x00000fff
+.b32 #ctx_dst_xoff                   ~0x0007ffff
+.b32 #ctx_dst_yoff                   ~0x00001fff
+// mthd 0x0300-0x0304, EXEC, WRCACHE_FLUSH
+.b16 0xc0 2
+.b32 0x00010000 + #cmd_exec          ~0xffffffff
+.b32 0x00010000 + #cmd_wrcache_flush ~0xffffffff
+// mthd 0x030c-0x0340, various stuff
+.b16 0xc3 14
+.b32 #ctx_src_address_high           ~0x000000ff
+.b32 #ctx_src_address_low            ~0xffffffff
+.b32 #ctx_dst_address_high           ~0x000000ff
+.b32 #ctx_dst_address_low            ~0xffffffff
+.b32 #ctx_src_pitch                  ~0x0007ffff
+.b32 #ctx_dst_pitch                  ~0x0007ffff
+.b32 #ctx_xcnt                       ~0x0000ffff
+.b32 #ctx_ycnt                       ~0x00001fff
+.b32 #ctx_format                     ~0x0333ffff
+.b32 #ctx_swz_const0                 ~0xffffffff
+.b32 #ctx_swz_const1                 ~0xffffffff
+.b32 #ctx_query_address_high         ~0x000000ff
+.b32 #ctx_query_address_low          ~0xffffffff
+.b32 #ctx_query_counter              ~0xffffffff
+.b16 0x800 0
+
+ifdef(`NVA3',
+.section #nva3_pcopy_code
+,
+.section #nvc0_pcopy_code
+)
+
+main:
+   clear b32 $r0
+   mov $sp $r0
+
+   // setup i0 handler and route fifo and ctxswitch to it
+   mov $r1 #ih
+   mov $iv0 $r1
+   mov $r1 0x400
+   movw $r2 0xfff3
+   sethi $r2 0
+   iowr I[$r1 + 0x300] $r2
+
+   // enable interrupts
+   or $r2 0xc
+   iowr I[$r1] $r2
+   bset $flags ie0
+
+   // enable fifo access and context switching
+   mov $r1 0x1200
+   mov $r2 3
+   iowr I[$r1] $r2
+
+   // sleep forever, waking for interrupts
+   bset $flags $p0
+   spin:
+      sleep $p0
+      bra #spin
+
+// i0 handler
+ih:
+   iord $r1 I[$r0 + 0x200]
+
+   and $r2 $r1 0x00000008
+   bra e #ih_no_chsw
+      call #chsw
+   ih_no_chsw:
+   and $r2 $r1 0x00000004
+   bra e #ih_no_cmd
+      call #dispatch
+
+   ih_no_cmd:
+   and $r1 $r1 0x0000000c
+   iowr I[$r0 + 0x100] $r1
+   iret
+
+// $p1 direction (0 = unload, 1 = load)
+// $r3 channel
+swctx:
+   mov $r4 0x7700
+   mov $xtargets $r4
+ifdef(`NVA3', `
+   // target 7 hardcoded to ctx dma object
+   mov $xdbase $r0
+', ` // NVC0
+   // read SCRATCH3 to decide if we are PCOPY0 or PCOPY1
+   mov $r4 0x2100
+   iord $r4 I[$r4 + 0]
+   and $r4 1
+   shl b32 $r4 4
+   add b32 $r4 0x30
+
+   // channel is in vram
+   mov $r15 0x61c
+   shl b32 $r15 6
+   mov $r5 0x114
+   iowrs I[$r15] $r5
+
+   // read 16-byte PCOPYn info, containing context pointer, from channel
+   shl b32 $r5 $r3 4
+   add b32 $r5 2
+   mov $xdbase $r5
+   mov $r5 $sp
+   // get a chunk of stack space, aligned to 256 byte boundary
+   sub b32 $r5 0x100
+   mov $r6 0xff
+   not b32 $r6
+   and $r5 $r6
+   sethi $r5 0x00020000
+   xdld $r4 $r5
+   xdwait
+   sethi $r5 0
+
+   // set context pointer, from within channel VM
+   mov $r14 0
+   iowrs I[$r15] $r14
+   ld b32 $r4 D[$r5 + 0]
+   shr b32 $r4 8
+   ld b32 $r6 D[$r5 + 4]
+   shl b32 $r6 24
+   or $r4 $r6
+   mov $xdbase $r4
+')
+   // 256-byte context, at start of data segment
+   mov b32 $r4 $r0
+   sethi $r4 0x60000
+
+   // swap!
+   bra $p1 #swctx_load
+      xdst $r0 $r4
+      bra #swctx_done
+   swctx_load:
+      xdld $r0 $r4
+   swctx_done:
+   xdwait
+   ret
+
+chsw:
+   // read current channel
+   mov $r2 0x1400
+   iord $r3 I[$r2]
+
+   // if it's active, unload it and return
+   xbit $r15 $r3 0x1e
+   bra e #chsw_no_unload
+      bclr $flags $p1
+      call #swctx
+      bclr $r3 0x1e
+      iowr I[$r2] $r3
+      mov $r4 1
+      iowr I[$r2 + 0x200] $r4
+      ret
+
+   // read next channel
+   chsw_no_unload:
+   iord $r3 I[$r2 + 0x100]
+
+   // is there a channel waiting to be loaded?
+   xbit $r13 $r3 0x1e
+   bra e #chsw_finish_load
+      bset $flags $p1
+      call #swctx
+ifdef(`NVA3',
+      // load dma objects back into TARGET regs
+      mov $r5 #ctx_dma
+      mov $r6 #ctx_dma_count
+      chsw_load_ctx_dma:
+         ld b32 $r7 D[$r5 + $r6 * 4]
+         add b32 $r8 $r6 0x180
+         shl b32 $r8 8
+         iowr I[$r8] $r7
+         sub b32 $r6 1
+         bra nc #chsw_load_ctx_dma
+,)
+
+   chsw_finish_load:
+   mov $r3 2
+   iowr I[$r2 + 0x200] $r3
+   ret
+
+dispatch:
+   // read incoming fifo command
+   mov $r3 0x1900
+   iord $r2 I[$r3 + 0x100]
+   iord $r3 I[$r3 + 0x000]
+   and $r4 $r2 0x7ff
+   // $r2 will be used to store exception data
+   shl b32 $r2 0x10
+
+   // lookup method in the dispatch table, ILLEGAL_MTHD if not found
+   mov $r5 #dispatch_table
+   clear b32 $r6
+   clear b32 $r7
+   dispatch_loop:
+      ld b16 $r6 D[$r5 + 0]
+      ld b16 $r7 D[$r5 + 2]
+      add b32 $r5 4
+      cmpu b32 $r4 $r6
+      bra c #dispatch_illegal_mthd
+      add b32 $r7 $r6
+      cmpu b32 $r4 $r7
+      bra c #dispatch_valid_mthd
+      sub b32 $r7 $r6
+      shl b32 $r7 3
+      add b32 $r5 $r7
+      bra #dispatch_loop
+
+   // ensure no bits set in reserved fields, INVALID_BITFIELD
+   dispatch_valid_mthd:
+   sub b32 $r4 $r6
+   shl b32 $r4 3
+   add b32 $r4 $r5
+   ld b32 $r5 D[$r4 + 4]
+   and $r5 $r3
+   cmpu b32 $r5 0
+   bra ne #dispatch_invalid_bitfield
+
+   // depending on dispatch flags: execute method, or save data as state
+   ld b16 $r5 D[$r4 + 0]
+   ld b16 $r6 D[$r4 + 2]
+   cmpu b32 $r6 0
+   bra ne #dispatch_cmd
+      st b32 D[$r5] $r3
+      bra #dispatch_done
+   dispatch_cmd:
+      bclr $flags $p1
+      call $r5
+      bra $p1 #dispatch_error
+      bra #dispatch_done
+
+   dispatch_invalid_bitfield:
+   or $r2 2
+   dispatch_illegal_mthd:
+   or $r2 1
+
+   // store exception data in SCRATCH0/SCRATCH1, signal hostirq
+   dispatch_error:
+   mov $r4 0x1000
+   iowr I[$r4 + 0x000] $r2
+   iowr I[$r4 + 0x100] $r3
+   mov $r2 0x40
+   iowr I[$r0] $r2
+   hostirq_wait:
+      iord $r2 I[$r0 + 0x200]
+      and $r2 0x40
+      cmpu b32 $r2 0
+      bra ne #hostirq_wait
+
+   dispatch_done:
+   mov $r2 0x1d00
+   mov $r3 1
+   iowr I[$r2] $r3
+   ret
+
+// No-operation
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_nop:
+   ret
+
+// PM_TRIGGER
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_pm_trigger:
+   mov $r2 0x2200
+   clear b32 $r3
+   sethi $r3 0x20000
+   iowr I[$r2] $r3
+   ret
+
+ifdef(`NVA3',
+// SET_DMA_* method handler
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_dma:
+   sub b32 $r4 #dispatch_dma
+   shr b32 $r4 1
+   bset $r3 0x1e
+   st b32 D[$r4 + #ctx_dma] $r3
+   add b32 $r4 0x600
+   shl b32 $r4 6
+   iowr I[$r4] $r3
+   ret
+,)
+
+// Calculates the hw swizzle mask and adjusts the surface's xcnt to match
+//
+cmd_exec_set_format:
+   // zero out a chunk of the stack to store the swizzle into
+   add $sp -0x10
+   st b32 D[$sp + 0x00] $r0
+   st b32 D[$sp + 0x04] $r0
+   st b32 D[$sp + 0x08] $r0
+   st b32 D[$sp + 0x0c] $r0
+
+   // extract cpp, src_ncomp and dst_ncomp from FORMAT
+   ld b32 $r4 D[$r0 + #ctx_format]
+   extr $r5 $r4 16:17
+   add b32 $r5 1
+   extr $r6 $r4 20:21
+   add b32 $r6 1
+   extr $r7 $r4 24:25
+   add b32 $r7 1
+
+   // convert FORMAT swizzle mask to hw swizzle mask
+   bclr $flags $p2
+   clear b32 $r8
+   clear b32 $r9
+   ncomp_loop:
+      and $r10 $r4 0xf
+      shr b32 $r4 4
+      clear b32 $r11
+      bpc_loop:
+         cmpu b8 $r10 4
+         bra nc #cmp_c0
+            mulu $r12 $r10 $r5
+            add b32 $r12 $r11
+            bset $flags $p2
+            bra #bpc_next
+         cmp_c0:
+         bra ne #cmp_c1
+            mov $r12 0x10
+            add b32 $r12 $r11
+            bra #bpc_next
+         cmp_c1:
+         cmpu b8 $r10 6
+         bra nc #cmp_zero
+            mov $r12 0x14
+            add b32 $r12 $r11
+            bra #bpc_next
+         cmp_zero:
+            mov $r12 0x80
+         bpc_next:
+         st b8 D[$sp + $r8] $r12
+         add b32 $r8 1
+         add b32 $r11 1
+         cmpu b32 $r11 $r5
+         bra c #bpc_loop
+      add b32 $r9 1
+      cmpu b32 $r9 $r7
+      bra c #ncomp_loop
+
+   // SRC_XCNT = (xcnt * src_cpp), or 0 if no src ref in swz (hw will hang)
+   mulu $r6 $r5
+   st b32 D[$r0 + #ctx_src_cpp] $r6
+   ld b32 $r8 D[$r0 + #ctx_xcnt]
+   mulu $r6 $r8
+   bra $p2 #dst_xcnt
+   clear b32 $r6
+
+   dst_xcnt:
+   mulu $r7 $r5
+   st b32 D[$r0 + #ctx_dst_cpp] $r7
+   mulu $r7 $r8
+
+   mov $r5 0x810
+   shl b32 $r5 6
+   iowr I[$r5 + 0x000] $r6
+   iowr I[$r5 + 0x100] $r7
+   add b32 $r5 0x800
+   ld b32 $r6 D[$r0 + #ctx_dst_cpp]
+   sub b32 $r6 1
+   shl b32 $r6 8
+   ld b32 $r7 D[$r0 + #ctx_src_cpp]
+   sub b32 $r7 1
+   or $r6 $r7
+   iowr I[$r5 + 0x000] $r6
+   add b32 $r5 0x100
+   ld b32 $r6 D[$sp + 0x00]
+   iowr I[$r5 + 0x000] $r6
+   ld b32 $r6 D[$sp + 0x04]
+   iowr I[$r5 + 0x100] $r6
+   ld b32 $r6 D[$sp + 0x08]
+   iowr I[$r5 + 0x200] $r6
+   ld b32 $r6 D[$sp + 0x0c]
+   iowr I[$r5 + 0x300] $r6
+   add b32 $r5 0x400
+   ld b32 $r6 D[$r0 + #ctx_swz_const0]
+   iowr I[$r5 + 0x000] $r6
+   ld b32 $r6 D[$r0 + #ctx_swz_const1]
+   iowr I[$r5 + 0x100] $r6
+   add $sp 0x10
+   ret
+
+// Setup to handle a tiled surface
+//
+// Calculates a number of parameters the hardware requires in order
+// to correctly handle tiling.
+//
+// Offset calculation is performed as follows (Tp/Th/Td from TILE_MODE):
+//    nTx = round_up(w * cpp, 1 << Tp) >> Tp
+//    nTy = round_up(h, 1 << Th) >> Th
+//    Txo = (x * cpp) & ((1 << Tp) - 1)
+//     Tx = (x * cpp) >> Tp
+//    Tyo = y & ((1 << Th) - 1)
+//     Ty = y >> Th
+//    Tzo = z & ((1 << Td) - 1)
+//     Tz = z >> Td
+//
+//    off  = (Tzo << Tp << Th) + (Tyo << Tp) + Txo
+//    off += ((Tz * nTy * nTx)) + (Ty * nTx) + Tx) << Td << Th << Tp;
+//
+// Inputs:
+//    $r4: hw command (0x104800)
+//    $r5: ctx offset adjustment for src/dst selection
+//    $p2: set if dst surface
+//
+cmd_exec_set_surface_tiled:
+   // translate TILE_MODE into Tp, Th, Td shift values
+   ld b32 $r7 D[$r5 + #ctx_src_tile_mode]
+   extr $r9 $r7 8:11
+   extr $r8 $r7 4:7
+ifdef(`NVA3',
+   add b32 $r8 2
+,
+   add b32 $r8 3
+)
+   extr $r7 $r7 0:3
+   cmp b32 $r7 0xe
+   bra ne #xtile64
+   mov $r7 4
+   bra #xtileok
+   xtile64:
+   xbit $r7 $flags $p2
+   add b32 $r7 17
+   bset $r4 $r7
+   mov $r7 6
+   xtileok:
+
+   // Op = (x * cpp) & ((1 << Tp) - 1)
+   // Tx = (x * cpp) >> Tp
+   ld b32 $r10 D[$r5 + #ctx_src_xoff]
+   ld b32 $r11 D[$r5 + #ctx_src_cpp]
+   mulu $r10 $r11
+   mov $r11 1
+   shl b32 $r11 $r7
+   sub b32 $r11 1
+   and $r12 $r10 $r11
+   shr b32 $r10 $r7
+
+   // Tyo = y & ((1 << Th) - 1)
+   // Ty  = y >> Th
+   ld b32 $r13 D[$r5 + #ctx_src_yoff]
+   mov $r14 1
+   shl b32 $r14 $r8
+   sub b32 $r14 1
+   and $r11 $r13 $r14
+   shr b32 $r13 $r8
+
+   // YTILE = ((1 << Th) << 12) | ((1 << Th) - Tyo)
+   add b32 $r14 1
+   shl b32 $r15 $r14 12
+   sub b32 $r14 $r11
+   or $r15 $r14
+   xbit $r6 $flags $p2
+   add b32 $r6 0x208
+   shl b32 $r6 8
+   iowr I[$r6 + 0x000] $r15
+
+   // Op += Tyo << Tp
+   shl b32 $r11 $r7
+   add b32 $r12 $r11
+
+   // nTx = ((w * cpp) + ((1 << Tp) - 1) >> Tp)
+   ld b32 $r15 D[$r5 + #ctx_src_xsize]
+   ld b32 $r11 D[$r5 + #ctx_src_cpp]
+   mulu $r15 $r11
+   mov $r11 1
+   shl b32 $r11 $r7
+   sub b32 $r11 1
+   add b32 $r15 $r11
+   shr b32 $r15 $r7
+   push $r15
+
+   // nTy = (h + ((1 << Th) - 1)) >> Th
+   ld b32 $r15 D[$r5 + #ctx_src_ysize]
+   mov $r11 1
+   shl b32 $r11 $r8
+   sub b32 $r11 1
+   add b32 $r15 $r11
+   shr b32 $r15 $r8
+   push $r15
+
+   // Tys = Tp + Th
+   // CFG_YZ_TILE_SIZE = ((1 << Th) >> 2) << Td
+   add b32 $r7 $r8
+   sub b32 $r8 2
+   mov $r11 1
+   shl b32 $r11 $r8
+   shl b32 $r11 $r9
+
+   // Tzo = z & ((1 << Td) - 1)
+   // Tz  = z >> Td
+   // Op += Tzo << Tys
+   // Ts  = Tys + Td
+   ld b32 $r8 D[$r5 + #ctx_src_zoff]
+   mov $r14 1
+   shl b32 $r14 $r9
+   sub b32 $r14 1
+   and $r15 $r8 $r14
+   shl b32 $r15 $r7
+   add b32 $r12 $r15
+   add b32 $r7 $r9
+   shr b32 $r8 $r9
+
+   // Ot = ((Tz * nTy * nTx) + (Ty * nTx) + Tx) << Ts
+   pop $r15
+   pop $r9
+   mulu $r13 $r9
+   add b32 $r10 $r13
+   mulu $r8 $r9
+   mulu $r8 $r15
+   add b32 $r10 $r8
+   shl b32 $r10 $r7
+
+   // PITCH = (nTx - 1) << Ts
+   sub b32 $r9 1
+   shl b32 $r9 $r7
+   iowr I[$r6 + 0x200] $r9
+
+   // SRC_ADDRESS_LOW   = (Ot + Op) & 0xffffffff
+   // CFG_ADDRESS_HIGH |= ((Ot + Op) >> 32) << 16
+   ld b32 $r7 D[$r5 + #ctx_src_address_low]
+   ld b32 $r8 D[$r5 + #ctx_src_address_high]
+   add b32 $r10 $r12
+   add b32 $r7 $r10
+   adc b32 $r8 0
+   shl b32 $r8 16
+   or $r8 $r11
+   sub b32 $r6 0x600
+   iowr I[$r6 + 0x000] $r7
+   add b32 $r6 0x400
+   iowr I[$r6 + 0x000] $r8
+   ret
+
+// Setup to handle a linear surface
+//
+// Nothing to see here.. Sets ADDRESS and PITCH, pretty non-exciting
+//
+cmd_exec_set_surface_linear:
+   xbit $r6 $flags $p2
+   add b32 $r6 0x202
+   shl b32 $r6 8
+   ld b32 $r7 D[$r5 + #ctx_src_address_low]
+   iowr I[$r6 + 0x000] $r7
+   add b32 $r6 0x400
+   ld b32 $r7 D[$r5 + #ctx_src_address_high]
+   shl b32 $r7 16
+   iowr I[$r6 + 0x000] $r7
+   add b32 $r6 0x400
+   ld b32 $r7 D[$r5 + #ctx_src_pitch]
+   iowr I[$r6 + 0x000] $r7
+   ret
+
+// wait for regs to be available for use
+cmd_exec_wait:
+   push $r0
+   push $r1
+   mov $r0 0x800
+   shl b32 $r0 6
+   loop:
+      iord $r1 I[$r0]
+      and $r1 1
+      bra ne #loop
+   pop $r1
+   pop $r0
+   ret
+
+cmd_exec_query:
+   // if QUERY_SHORT not set, write out { -, 0, TIME_LO, TIME_HI }
+   xbit $r4 $r3 13
+   bra ne #query_counter
+      call #cmd_exec_wait
+      mov $r4 0x80c
+      shl b32 $r4 6
+      ld b32 $r5 D[$r0 + #ctx_query_address_low]
+      add b32 $r5 4
+      iowr I[$r4 + 0x000] $r5
+      iowr I[$r4 + 0x100] $r0
+      mov $r5 0xc
+      iowr I[$r4 + 0x200] $r5
+      add b32 $r4 0x400
+      ld b32 $r5 D[$r0 + #ctx_query_address_high]
+      shl b32 $r5 16
+      iowr I[$r4 + 0x000] $r5
+      add b32 $r4 0x500
+      mov $r5 0x00000b00
+      sethi $r5 0x00010000
+      iowr I[$r4 + 0x000] $r5
+      mov $r5 0x00004040
+      shl b32 $r5 1
+      sethi $r5 0x80800000
+      iowr I[$r4 + 0x100] $r5
+      mov $r5 0x00001110
+      sethi $r5 0x13120000
+      iowr I[$r4 + 0x200] $r5
+      mov $r5 0x00001514
+      sethi $r5 0x17160000
+      iowr I[$r4 + 0x300] $r5
+      mov $r5 0x00002601
+      sethi $r5 0x00010000
+      mov $r4 0x800
+      shl b32 $r4 6
+      iowr I[$r4 + 0x000] $r5
+
+   // write COUNTER
+   query_counter:
+   call #cmd_exec_wait
+   mov $r4 0x80c
+   shl b32 $r4 6
+   ld b32 $r5 D[$r0 + #ctx_query_address_low]
+   iowr I[$r4 + 0x000] $r5
+   iowr I[$r4 + 0x100] $r0
+   mov $r5 0x4
+   iowr I[$r4 + 0x200] $r5
+   add b32 $r4 0x400
+   ld b32 $r5 D[$r0 + #ctx_query_address_high]
+   shl b32 $r5 16
+   iowr I[$r4 + 0x000] $r5
+   add b32 $r4 0x500
+   mov $r5 0x00000300
+   iowr I[$r4 + 0x000] $r5
+   mov $r5 0x00001110
+   sethi $r5 0x13120000
+   iowr I[$r4 + 0x100] $r5
+   ld b32 $r5 D[$r0 + #ctx_query_counter]
+   add b32 $r4 0x500
+   iowr I[$r4 + 0x000] $r5
+   mov $r5 0x00002601
+   sethi $r5 0x00010000
+   mov $r4 0x800
+   shl b32 $r4 6
+   iowr I[$r4 + 0x000] $r5
+   ret
+
+// Execute a copy operation
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//       000002000 QUERY_SHORT
+//       000001000 QUERY
+//       000000100 DST_LINEAR
+//       000000010 SRC_LINEAR
+//       000000001 FORMAT
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_exec:
+   call #cmd_exec_wait
+
+   // if format requested, call function to calculate it, otherwise
+   // fill in cpp/xcnt for both surfaces as if (cpp == 1)
+   xbit $r15 $r3 0
+   bra e #cmd_exec_no_format
+      call #cmd_exec_set_format
+      mov $r4 0x200
+      bra #cmd_exec_init_src_surface
+   cmd_exec_no_format:
+      mov $r6 0x810
+      shl b32 $r6 6
+      mov $r7 1
+      st b32 D[$r0 + #ctx_src_cpp] $r7
+      st b32 D[$r0 + #ctx_dst_cpp] $r7
+      ld b32 $r7 D[$r0 + #ctx_xcnt]
+      iowr I[$r6 + 0x000] $r7
+      iowr I[$r6 + 0x100] $r7
+      clear b32 $r4
+
+   cmd_exec_init_src_surface:
+   bclr $flags $p2
+   clear b32 $r5
+   xbit $r15 $r3 4
+   bra e #src_tiled
+      call #cmd_exec_set_surface_linear
+      bra #cmd_exec_init_dst_surface
+   src_tiled:
+      call #cmd_exec_set_surface_tiled
+      bset $r4 7
+
+   cmd_exec_init_dst_surface:
+   bset $flags $p2
+   mov $r5 #ctx_dst_address_high - #ctx_src_address_high
+   xbit $r15 $r3 8
+   bra e #dst_tiled
+      call #cmd_exec_set_surface_linear
+      bra #cmd_exec_kick
+   dst_tiled:
+      call #cmd_exec_set_surface_tiled
+      bset $r4 8
+
+   cmd_exec_kick:
+   mov $r5 0x800
+   shl b32 $r5 6
+   ld b32 $r6 D[$r0 + #ctx_ycnt]
+   iowr I[$r5 + 0x100] $r6
+   mov $r6 0x0041
+   // SRC_TARGET = 1, DST_TARGET = 2
+   sethi $r6 0x44000000
+   or $r4 $r6
+   iowr I[$r5] $r4
+
+   // if requested, queue up a QUERY write after the copy has completed
+   xbit $r15 $r3 12
+   bra e #cmd_exec_done
+      call #cmd_exec_query
+
+   cmd_exec_done:
+   ret
+
+// Flush write cache
+//
+// Inputs:
+//    $r1: irqh state
+//    $r2: hostirq state
+//    $r3: data
+//    $r4: dispatch table entry
+// Outputs:
+//    $r1: irqh state
+//    $p1: set on error
+//       $r2: hostirq state
+//       $r3: data
+cmd_wrcache_flush:
+   mov $r2 0x2200
+   clear b32 $r3
+   sethi $r3 0x10000
+   iowr I[$r2] $r3
+   ret
+
+.align 0x100
diff --git a/nvkm/engine/copy/nva3.c b/nvkm/engine/copy/nva3.c
new file mode 100644
index 000000000..283248c7b
--- /dev/null
+++ b/nvkm/engine/copy/nva3.c
@@ -0,0 +1,162 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/falcon.h>
+#include <core/class.h>
+#include <core/enum.h>
+
+#include <subdev/fb.h>
+#include <subdev/vm.h>
+
+#include <engine/fifo.h>
+#include <engine/copy.h>
+
+#include "fuc/nva3.fuc.h"
+
+struct nva3_copy_priv {
+	struct nouveau_falcon base;
+};
+
+/*******************************************************************************
+ * Copy object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nva3_copy_sclass[] = {
+	{ 0x85b5, &nouveau_object_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * PCOPY context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nva3_copy_cclass = {
+	.handle = NV_ENGCTX(COPY0, 0xa3),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_falcon_context_ctor,
+		.dtor = _nouveau_falcon_context_dtor,
+		.init = _nouveau_falcon_context_init,
+		.fini = _nouveau_falcon_context_fini,
+		.rd32 = _nouveau_falcon_context_rd32,
+		.wr32 = _nouveau_falcon_context_wr32,
+
+	},
+};
+
+/*******************************************************************************
+ * PCOPY engine/subdev functions
+ ******************************************************************************/
+
+static const struct nouveau_enum nva3_copy_isr_error_name[] = {
+	{ 0x0001, "ILLEGAL_MTHD" },
+	{ 0x0002, "INVALID_ENUM" },
+	{ 0x0003, "INVALID_BITFIELD" },
+	{}
+};
+
+void
+nva3_copy_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_fifo *pfifo = nouveau_fifo(subdev);
+	struct nouveau_engine *engine = nv_engine(subdev);
+	struct nouveau_falcon *falcon = (void *)subdev;
+	struct nouveau_object *engctx;
+	u32 dispatch = nv_ro32(falcon, 0x01c);
+	u32 stat = nv_ro32(falcon, 0x008) & dispatch & ~(dispatch >> 16);
+	u64 inst = nv_ro32(falcon, 0x050) & 0x3fffffff;
+	u32 ssta = nv_ro32(falcon, 0x040) & 0x0000ffff;
+	u32 addr = nv_ro32(falcon, 0x040) >> 16;
+	u32 mthd = (addr & 0x07ff) << 2;
+	u32 subc = (addr & 0x3800) >> 11;
+	u32 data = nv_ro32(falcon, 0x044);
+	int chid;
+
+	engctx = nouveau_engctx_get(engine, inst);
+	chid   = pfifo->chid(pfifo, engctx);
+
+	if (stat & 0x00000040) {
+		nv_error(falcon, "DISPATCH_ERROR [");
+		nouveau_enum_print(nva3_copy_isr_error_name, ssta);
+		printk("] ch %d [0x%010llx] subc %d mthd 0x%04x data 0x%08x\n",
+		       chid, inst << 12, subc, mthd, data);
+		nv_wo32(falcon, 0x004, 0x00000040);
+		stat &= ~0x00000040;
+	}
+
+	if (stat) {
+		nv_error(falcon, "unhandled intr 0x%08x\n", stat);
+		nv_wo32(falcon, 0x004, stat);
+	}
+
+	nouveau_engctx_put(engctx);
+}
+
+static int
+nva3_copy_tlb_flush(struct nouveau_engine *engine)
+{
+	nv50_vm_flush_engine(&engine->base, 0x0d);
+	return 0;
+}
+
+static int
+nva3_copy_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	bool enable = (nv_device(parent)->chipset != 0xaf);
+	struct nva3_copy_priv *priv;
+	int ret;
+
+	ret = nouveau_falcon_create(parent, engine, oclass, 0x104000, enable,
+				    "PCE0", "copy0", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00802000;
+	nv_subdev(priv)->intr = nva3_copy_intr;
+	nv_engine(priv)->cclass = &nva3_copy_cclass;
+	nv_engine(priv)->sclass = nva3_copy_sclass;
+	nv_engine(priv)->tlb_flush = nva3_copy_tlb_flush;
+	nv_falcon(priv)->code.data = nva3_pcopy_code;
+	nv_falcon(priv)->code.size = sizeof(nva3_pcopy_code);
+	nv_falcon(priv)->data.data = nva3_pcopy_data;
+	nv_falcon(priv)->data.size = sizeof(nva3_pcopy_data);
+	return 0;
+}
+
+struct nouveau_oclass
+nva3_copy_oclass = {
+	.handle = NV_ENGINE(COPY0, 0xa3),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nva3_copy_ctor,
+		.dtor = _nouveau_falcon_dtor,
+		.init = _nouveau_falcon_init,
+		.fini = _nouveau_falcon_fini,
+		.rd32 = _nouveau_falcon_rd32,
+		.wr32 = _nouveau_falcon_wr32,
+	},
+};
diff --git a/nvkm/engine/copy/nvc0.c b/nvkm/engine/copy/nvc0.c
new file mode 100644
index 000000000..b3ed2737e
--- /dev/null
+++ b/nvkm/engine/copy/nvc0.c
@@ -0,0 +1,178 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/falcon.h>
+#include <core/class.h>
+#include <core/enum.h>
+
+#include <engine/fifo.h>
+#include <engine/copy.h>
+
+#include "fuc/nvc0.fuc.h"
+
+struct nvc0_copy_priv {
+	struct nouveau_falcon base;
+};
+
+/*******************************************************************************
+ * Copy object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nvc0_copy0_sclass[] = {
+	{ 0x90b5, &nouveau_object_ofuncs },
+	{},
+};
+
+static struct nouveau_oclass
+nvc0_copy1_sclass[] = {
+	{ 0x90b8, &nouveau_object_ofuncs },
+	{},
+};
+
+/*******************************************************************************
+ * PCOPY context
+ ******************************************************************************/
+
+static struct nouveau_ofuncs
+nvc0_copy_context_ofuncs = {
+	.ctor = _nouveau_falcon_context_ctor,
+	.dtor = _nouveau_falcon_context_dtor,
+	.init = _nouveau_falcon_context_init,
+	.fini = _nouveau_falcon_context_fini,
+	.rd32 = _nouveau_falcon_context_rd32,
+	.wr32 = _nouveau_falcon_context_wr32,
+};
+
+static struct nouveau_oclass
+nvc0_copy0_cclass = {
+	.handle = NV_ENGCTX(COPY0, 0xc0),
+	.ofuncs = &nvc0_copy_context_ofuncs,
+};
+
+static struct nouveau_oclass
+nvc0_copy1_cclass = {
+	.handle = NV_ENGCTX(COPY1, 0xc0),
+	.ofuncs = &nvc0_copy_context_ofuncs,
+};
+
+/*******************************************************************************
+ * PCOPY engine/subdev functions
+ ******************************************************************************/
+
+static int
+nvc0_copy_init(struct nouveau_object *object)
+{
+	struct nvc0_copy_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_falcon_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wo32(priv, 0x084, nv_engidx(object) - NVDEV_ENGINE_COPY0);
+	return 0;
+}
+
+static int
+nvc0_copy0_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nvc0_copy_priv *priv;
+	int ret;
+
+	if (nv_rd32(parent, 0x022500) & 0x00000100)
+		return -ENODEV;
+
+	ret = nouveau_falcon_create(parent, engine, oclass, 0x104000, true,
+				    "PCE0", "copy0", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000040;
+	nv_subdev(priv)->intr = nva3_copy_intr;
+	nv_engine(priv)->cclass = &nvc0_copy0_cclass;
+	nv_engine(priv)->sclass = nvc0_copy0_sclass;
+	nv_falcon(priv)->code.data = nvc0_pcopy_code;
+	nv_falcon(priv)->code.size = sizeof(nvc0_pcopy_code);
+	nv_falcon(priv)->data.data = nvc0_pcopy_data;
+	nv_falcon(priv)->data.size = sizeof(nvc0_pcopy_data);
+	return 0;
+}
+
+static int
+nvc0_copy1_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nvc0_copy_priv *priv;
+	int ret;
+
+	if (nv_rd32(parent, 0x022500) & 0x00000200)
+		return -ENODEV;
+
+	ret = nouveau_falcon_create(parent, engine, oclass, 0x105000, true,
+				    "PCE1", "copy1", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000080;
+	nv_subdev(priv)->intr = nva3_copy_intr;
+	nv_engine(priv)->cclass = &nvc0_copy1_cclass;
+	nv_engine(priv)->sclass = nvc0_copy1_sclass;
+	nv_falcon(priv)->code.data = nvc0_pcopy_code;
+	nv_falcon(priv)->code.size = sizeof(nvc0_pcopy_code);
+	nv_falcon(priv)->data.data = nvc0_pcopy_data;
+	nv_falcon(priv)->data.size = sizeof(nvc0_pcopy_data);
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_copy0_oclass = {
+	.handle = NV_ENGINE(COPY0, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_copy0_ctor,
+		.dtor = _nouveau_falcon_dtor,
+		.init = nvc0_copy_init,
+		.fini = _nouveau_falcon_fini,
+		.rd32 = _nouveau_falcon_rd32,
+		.wr32 = _nouveau_falcon_wr32,
+	},
+};
+
+struct nouveau_oclass
+nvc0_copy1_oclass = {
+	.handle = NV_ENGINE(COPY1, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_copy1_ctor,
+		.dtor = _nouveau_falcon_dtor,
+		.init = nvc0_copy_init,
+		.fini = _nouveau_falcon_fini,
+		.rd32 = _nouveau_falcon_rd32,
+		.wr32 = _nouveau_falcon_wr32,
+	},
+};
diff --git a/nvkm/engine/copy/nve0.c b/nvkm/engine/copy/nve0.c
new file mode 100644
index 000000000..dbbe9e899
--- /dev/null
+++ b/nvkm/engine/copy/nve0.c
@@ -0,0 +1,136 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/enum.h>
+#include <core/class.h>
+#include <core/engctx.h>
+
+#include <engine/copy.h>
+
+struct nve0_copy_priv {
+	struct nouveau_engine base;
+};
+
+/*******************************************************************************
+ * Copy object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nve0_copy_sclass[] = {
+	{ 0xa0b5, &nouveau_object_ofuncs },
+	{},
+};
+
+/*******************************************************************************
+ * PCOPY context
+ ******************************************************************************/
+
+static struct nouveau_ofuncs
+nve0_copy_context_ofuncs = {
+	.ctor = _nouveau_engctx_ctor,
+	.dtor = _nouveau_engctx_dtor,
+	.init = _nouveau_engctx_init,
+	.fini = _nouveau_engctx_fini,
+	.rd32 = _nouveau_engctx_rd32,
+	.wr32 = _nouveau_engctx_wr32,
+};
+
+static struct nouveau_oclass
+nve0_copy_cclass = {
+	.handle = NV_ENGCTX(COPY0, 0xc0),
+	.ofuncs = &nve0_copy_context_ofuncs,
+};
+
+/*******************************************************************************
+ * PCOPY engine/subdev functions
+ ******************************************************************************/
+
+static int
+nve0_copy0_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nve0_copy_priv *priv;
+	int ret;
+
+	if (nv_rd32(parent, 0x022500) & 0x00000100)
+		return -ENODEV;
+
+	ret = nouveau_engine_create(parent, engine, oclass, true,
+				    "PCE0", "copy0", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000040;
+	nv_engine(priv)->cclass = &nve0_copy_cclass;
+	nv_engine(priv)->sclass = nve0_copy_sclass;
+	return 0;
+}
+
+static int
+nve0_copy1_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nve0_copy_priv *priv;
+	int ret;
+
+	if (nv_rd32(parent, 0x022500) & 0x00000200)
+		return -ENODEV;
+
+	ret = nouveau_engine_create(parent, engine, oclass, true,
+				    "PCE1", "copy1", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000080;
+	nv_engine(priv)->cclass = &nve0_copy_cclass;
+	nv_engine(priv)->sclass = nve0_copy_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nve0_copy0_oclass = {
+	.handle = NV_ENGINE(COPY0, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nve0_copy0_ctor,
+		.dtor = _nouveau_engine_dtor,
+		.init = _nouveau_engine_init,
+		.fini = _nouveau_engine_fini,
+	},
+};
+
+struct nouveau_oclass
+nve0_copy1_oclass = {
+	.handle = NV_ENGINE(COPY1, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nve0_copy1_ctor,
+		.dtor = _nouveau_engine_dtor,
+		.init = _nouveau_engine_init,
+		.fini = _nouveau_engine_fini,
+	},
+};
diff --git a/nvkm/engine/crypt/Makefile.am b/nvkm/engine/crypt/Makefile.am
new file mode 100644
index 000000000..9d279a365
--- /dev/null
+++ b/nvkm/engine/crypt/Makefile.am
@@ -0,0 +1,13 @@
+noinst_LTLIBRARIES = libcrypt.la
+
+BUILT_SOURCES = fuc/nv98.fuc.h
+CLEANFILES = ${BUILT_SOURCES}
+
+libcrypt_la_SOURCES = ${BUILT_SOURCES} \
+		      nv84.c \
+		      nv98.c
+
+fuc/nv98.fuc.h: fuc/nv98.fuc
+	m4 fuc/nv98.fuc | envyas -a -w -m fuc -V fuc0 -F crypt -o fuc/nv98.fuc.h
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/crypt/fuc/nv98.fuc b/nvkm/engine/crypt/fuc/nv98.fuc
new file mode 100644
index 000000000..629da02dc
--- /dev/null
+++ b/nvkm/engine/crypt/fuc/nv98.fuc
@@ -0,0 +1,698 @@
+/*
+ *  fuc microcode for nv98 pcrypt engine
+ *  Copyright (C) 2010  Marcin Kościelnicki
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ */
+
+.section #nv98_pcrypt_data
+
+ctx_dma:
+ctx_dma_query:		.b32 0
+ctx_dma_src:		.b32 0
+ctx_dma_dst:		.b32 0
+.equ #dma_count 3
+ctx_query_address_high:	.b32 0
+ctx_query_address_low:	.b32 0
+ctx_query_counter:	.b32 0
+ctx_cond_address_high:	.b32 0
+ctx_cond_address_low:	.b32 0
+ctx_cond_off:		.b32 0
+ctx_src_address_high:	.b32 0
+ctx_src_address_low:	.b32 0
+ctx_dst_address_high:	.b32 0
+ctx_dst_address_low:	.b32 0
+ctx_mode:		.b32 0
+.align 16
+ctx_key:		.skip 16
+ctx_iv:			.skip 16
+
+.align 0x80
+swap:
+.skip 32
+
+.align 8
+common_cmd_dtable:
+.b32 #ctx_query_address_high + 0x20000 ~0xff
+.b32 #ctx_query_address_low + 0x20000 ~0xfffffff0
+.b32 #ctx_query_counter + 0x20000 ~0xffffffff
+.b32 #cmd_query_get + 0x00000 ~1
+.b32 #ctx_cond_address_high + 0x20000 ~0xff
+.b32 #ctx_cond_address_low + 0x20000 ~0xfffffff0
+.b32 #cmd_cond_mode + 0x00000 ~7
+.b32 #cmd_wrcache_flush + 0x00000 ~0
+.equ #common_cmd_max 0x88
+
+
+.align 8
+engine_cmd_dtable:
+.b32 #ctx_key + 0x0 + 0x20000 ~0xffffffff
+.b32 #ctx_key + 0x4 + 0x20000 ~0xffffffff
+.b32 #ctx_key + 0x8 + 0x20000 ~0xffffffff
+.b32 #ctx_key + 0xc + 0x20000 ~0xffffffff
+.b32 #ctx_iv + 0x0 + 0x20000 ~0xffffffff
+.b32 #ctx_iv + 0x4 + 0x20000 ~0xffffffff
+.b32 #ctx_iv + 0x8 + 0x20000 ~0xffffffff
+.b32 #ctx_iv + 0xc + 0x20000 ~0xffffffff
+.b32 #ctx_src_address_high + 0x20000 ~0xff
+.b32 #ctx_src_address_low + 0x20000 ~0xfffffff0
+.b32 #ctx_dst_address_high + 0x20000 ~0xff
+.b32 #ctx_dst_address_low + 0x20000 ~0xfffffff0
+.b32 #crypt_cmd_mode + 0x00000 ~0xf
+.b32 #crypt_cmd_length + 0x10000 ~0x0ffffff0
+.equ #engine_cmd_max 0xce
+
+.align 4
+crypt_dtable:
+.b16 #crypt_copy_prep #crypt_do_inout
+.b16 #crypt_store_prep #crypt_do_out
+.b16 #crypt_ecb_e_prep #crypt_do_inout
+.b16 #crypt_ecb_d_prep #crypt_do_inout
+.b16 #crypt_cbc_e_prep #crypt_do_inout
+.b16 #crypt_cbc_d_prep #crypt_do_inout
+.b16 #crypt_pcbc_e_prep #crypt_do_inout
+.b16 #crypt_pcbc_d_prep #crypt_do_inout
+.b16 #crypt_cfb_e_prep #crypt_do_inout
+.b16 #crypt_cfb_d_prep #crypt_do_inout
+.b16 #crypt_ofb_prep #crypt_do_inout
+.b16 #crypt_ctr_prep #crypt_do_inout
+.b16 #crypt_cbc_mac_prep #crypt_do_in
+.b16 #crypt_cmac_finish_complete_prep #crypt_do_in
+.b16 #crypt_cmac_finish_partial_prep #crypt_do_in
+
+.align 0x100
+
+.section #nv98_pcrypt_code
+
+	// $r0 is always set to 0 in our code - this allows some space savings.
+	clear b32 $r0
+
+	// set up the interrupt handler
+	mov $r1 #ih
+	mov $iv0 $r1
+
+	// init stack pointer
+	mov $sp $r0
+
+	// set interrupt dispatch - route timer, fifo, ctxswitch to i0, others to host
+	movw $r1 0xfff0
+	sethi $r1 0
+	mov $r2 0x400
+	iowr I[$r2 + 0x300] $r1
+
+	// enable the interrupts
+	or $r1 0xc
+	iowr I[$r2] $r1
+
+	// enable fifo access and context switching
+	mov $r1 3
+	mov $r2 0x1200
+	iowr I[$r2] $r1
+
+	// enable i0 delivery
+	bset $flags ie0
+
+	// sleep forver, waking only for interrupts.
+	bset $flags $p0
+	spin:
+	sleep $p0
+	bra #spin
+
+// i0 handler
+ih:
+	// see which interrupts we got
+	iord $r1 I[$r0 + 0x200]
+
+	and $r2 $r1 0x8
+	cmpu b32 $r2 0
+	bra e #noctx
+
+		// context switch... prepare the regs for xfer
+		mov $r2 0x7700
+		mov $xtargets $r2
+		mov $xdbase $r0
+		// 128-byte context.
+		mov $r2 0
+		sethi $r2 0x50000
+
+		// read current channel
+		mov $r3 0x1400
+		iord $r4 I[$r3]
+		// if bit 30 set, it's active, so we have to unload it first.
+		shl b32 $r5 $r4 1
+		cmps b32 $r5 0
+		bra nc #ctxload
+
+			// unload the current channel - save the context
+			xdst $r0 $r2
+			xdwait
+			// and clear bit 30, then write back
+			bclr $r4 0x1e
+			iowr I[$r3] $r4
+			// tell PFIFO we unloaded
+			mov $r4 1
+			iowr I[$r3 + 0x200] $r4
+
+		bra #noctx
+
+		ctxload:
+			// no channel loaded - perhaps we're requested to load one
+			iord $r4 I[$r3 + 0x100]
+			shl b32 $r15 $r4 1
+			cmps b32 $r15 0
+			// if bit 30 of next channel not set, probably PFIFO is just
+			// killing a context. do a faux load, without the active bit.
+			bra nc #dummyload
+
+				// ok, do a real context load.
+				xdld $r0 $r2
+				xdwait
+				mov $r5 #ctx_dma
+				mov $r6 #dma_count - 1
+				ctxload_dma_loop:
+					ld b32 $r7 D[$r5 + $r6 * 4]
+					add b32 $r8 $r6 0x180
+					shl b32 $r8 8
+					iowr I[$r8] $r7
+					sub b32 $r6 1
+				bra nc #ctxload_dma_loop
+
+			dummyload:
+			// tell PFIFO we're done
+			mov $r5 2
+			iowr I[$r3 + 0x200] $r5
+
+	noctx:
+	and $r2 $r1 0x4
+	cmpu b32 $r2 0
+	bra e #nocmd
+
+		// incoming fifo command.
+		mov $r3 0x1900
+		iord $r2 I[$r3 + 0x100]
+		iord $r3 I[$r3]
+		// extract the method
+		and $r4 $r2 0x7ff
+		// shift the addr to proper position if we need to interrupt later
+		shl b32 $r2 0x10
+
+		// mthd 0 and 0x100 [NAME, NOP]: ignore
+		and $r5 $r4 0x7bf
+		cmpu b32 $r5 0
+		bra e #cmddone
+
+		mov $r5 #engine_cmd_dtable - 0xc0 * 8
+		mov $r6 #engine_cmd_max
+		cmpu b32 $r4 0xc0
+		bra nc #dtable_cmd
+		mov $r5 #common_cmd_dtable - 0x80 * 8
+		mov $r6 #common_cmd_max
+		cmpu b32 $r4 0x80
+		bra nc #dtable_cmd
+		cmpu b32 $r4 0x60
+		bra nc #dma_cmd
+		cmpu b32 $r4 0x50
+		bra ne #illegal_mthd
+
+			// mthd 0x140: PM_TRIGGER
+			mov $r2 0x2200
+			clear b32 $r3
+			sethi $r3 0x20000
+			iowr I[$r2] $r3
+			bra #cmddone
+
+		dma_cmd:
+			// mthd 0x180...: DMA_*
+			cmpu b32 $r4 0x60+#dma_count
+			bra nc #illegal_mthd
+			shl b32 $r5 $r4 2
+			add b32 $r5 ((#ctx_dma - 0x60 * 4) & 0xffff)
+			bset $r3 0x1e
+			st b32 D[$r5] $r3
+			add b32 $r4 0x180 - 0x60
+			shl b32 $r4 8
+			iowr I[$r4] $r3
+			bra #cmddone
+
+		dtable_cmd:
+			cmpu b32 $r4 $r6
+			bra nc #illegal_mthd
+			shl b32 $r4 3
+			add b32 $r4 $r5
+			ld b32 $r5 D[$r4 + 4]
+			and $r5 $r3
+			cmpu b32 $r5 0
+			bra ne #invalid_bitfield
+			ld b16 $r5 D[$r4]
+			ld b16 $r6 D[$r4 + 2]
+			cmpu b32 $r6 2
+			bra e #cmd_setctx
+			ld b32 $r7 D[$r0 + #ctx_cond_off]
+			and $r6 $r7
+			cmpu b32 $r6 1
+			bra e #cmddone
+			call $r5
+			bra $p1 #dispatch_error
+			bra #cmddone
+
+		cmd_setctx:
+			st b32 D[$r5] $r3
+			bra #cmddone
+
+
+		invalid_bitfield:
+			or $r2 1
+		dispatch_error:
+		illegal_mthd:
+			mov $r4 0x1000
+			iowr I[$r4] $r2
+			iowr I[$r4 + 0x100] $r3
+			mov $r4 0x40
+			iowr I[$r0] $r4
+
+			im_loop:
+				iord $r4 I[$r0 + 0x200]
+				and $r4 0x40
+				cmpu b32 $r4 0
+			bra ne #im_loop
+
+		cmddone:
+		// remove the command from FIFO
+		mov $r3 0x1d00
+		mov $r4 1
+		iowr I[$r3] $r4
+
+	nocmd:
+	// ack the processed interrupts
+	and $r1 $r1 0xc
+	iowr I[$r0 + 0x100] $r1
+iret
+
+cmd_query_get:
+	// if bit 0 of param set, trigger interrupt afterwards.
+	setp $p1 $r3
+	or $r2 3
+
+	// read PTIMER, beware of races...
+	mov $r4 0xb00
+	ptimer_retry:
+		iord $r6 I[$r4 + 0x100]
+		iord $r5 I[$r4]
+		iord $r7 I[$r4 + 0x100]
+		cmpu b32 $r6 $r7
+	bra ne #ptimer_retry
+
+	// prepare the query structure
+	ld b32 $r4 D[$r0 + #ctx_query_counter]
+	st b32 D[$r0 + #swap + 0x0] $r4
+	st b32 D[$r0 + #swap + 0x4] $r0
+	st b32 D[$r0 + #swap + 0x8] $r5
+	st b32 D[$r0 + #swap + 0xc] $r6
+
+	// will use target 0, DMA_QUERY.
+	mov $xtargets $r0
+
+	ld b32 $r4 D[$r0 + #ctx_query_address_high]
+	shl b32 $r4 0x18
+	mov $xdbase $r4
+
+	ld b32 $r4 D[$r0 + #ctx_query_address_low]
+	mov $r5 #swap
+	sethi $r5 0x20000
+	xdst $r4 $r5
+	xdwait
+
+	ret
+
+cmd_cond_mode:
+	// if >= 5, INVALID_ENUM
+	bset $flags $p1
+	or $r2 2
+	cmpu b32 $r3 5
+	bra nc #return
+
+	// otherwise, no error.
+	bclr $flags $p1
+
+	// if < 2, no QUERY object is involved
+	cmpu b32 $r3 2
+	bra nc #cmd_cond_mode_queryful
+
+		xor $r3 1
+		st b32 D[$r0 + #ctx_cond_off] $r3
+	return:
+		ret
+
+	cmd_cond_mode_queryful:
+	// ok, will need to pull a QUERY object, prepare offsets
+	ld b32 $r4 D[$r0 + #ctx_cond_address_high]
+	ld b32 $r5 D[$r0 + #ctx_cond_address_low]
+	and $r6 $r5 0xff
+	shr b32 $r5 8
+	shl b32 $r4 0x18
+	or $r4 $r5
+	mov $xdbase $r4
+	mov $xtargets $r0
+
+	// pull the first one
+	mov $r5 #swap
+	sethi $r5 0x20000
+	xdld $r6 $r5
+
+	// if == 2, only a single QUERY is involved...
+	cmpu b32 $r3 2
+	bra ne #cmd_cond_mode_double
+
+		xdwait
+		ld b32 $r4 D[$r0 + #swap + 4]
+		cmpu b32 $r4 0
+		xbit $r4 $flags z
+		st b32 D[$r0 + #ctx_cond_off] $r4
+		ret
+
+	// ok, we'll need to pull second one too
+	cmd_cond_mode_double:
+	add b32 $r6 0x10
+	add b32 $r5 0x10
+	xdld $r6 $r5
+	xdwait
+
+	// compare COUNTERs
+	ld b32 $r5 D[$r0 + #swap + 0x00]
+	ld b32 $r6 D[$r0 + #swap + 0x10]
+	cmpu b32 $r5 $r6
+	xbit $r4 $flags z
+
+	// compare RESen
+	ld b32 $r5 D[$r0 + #swap + 0x04]
+	ld b32 $r6 D[$r0 + #swap + 0x14]
+	cmpu b32 $r5 $r6
+	xbit $r5 $flags z
+	and $r4 $r5
+
+	// and negate or not, depending on mode
+	cmpu b32 $r3 3
+	xbit $r5 $flags z
+	xor $r4 $r5
+	st b32 D[$r0 + #ctx_cond_off] $r4
+	ret
+
+cmd_wrcache_flush:
+	bclr $flags $p1
+	mov $r2 0x2200
+	clear b32 $r3
+	sethi $r3 0x10000
+	iowr I[$r2] $r3
+	ret
+
+crypt_cmd_mode:
+	// if >= 0xf, INVALID_ENUM
+	bset $flags $p1
+	or $r2 2
+	cmpu b32 $r3 0xf
+	bra nc #crypt_cmd_mode_return
+
+		bclr $flags $p1
+		st b32 D[$r0 + #ctx_mode] $r3
+
+	crypt_cmd_mode_return:
+	ret
+
+crypt_cmd_length:
+	// nop if length == 0
+	cmpu b32 $r3 0
+	bra e #crypt_cmd_mode_return
+
+	// init key, IV
+	cxset 3
+	mov $r4 #ctx_key
+	sethi $r4 0x70000
+	xdst $r0 $r4
+	mov $r4 #ctx_iv
+	sethi $r4 0x60000
+	xdst $r0 $r4
+	xdwait
+	ckeyreg $c7
+
+	// prepare the targets
+	mov $r4 0x2100
+	mov $xtargets $r4
+
+	// prepare src address
+	ld b32 $r4 D[$r0 + #ctx_src_address_high]
+	ld b32 $r5 D[$r0 + #ctx_src_address_low]
+	shr b32 $r8 $r5 8
+	shl b32 $r4 0x18
+	or $r4 $r8
+	and $r5 $r5 0xff
+
+	// prepare dst address
+	ld b32 $r6 D[$r0 + #ctx_dst_address_high]
+	ld b32 $r7 D[$r0 + #ctx_dst_address_low]
+	shr b32 $r8 $r7 8
+	shl b32 $r6 0x18
+	or $r6 $r8
+	and $r7 $r7 0xff
+
+	// find the proper prep & do functions
+	ld b32 $r8 D[$r0 + #ctx_mode]
+	shl b32 $r8 2
+
+	// run prep
+	ld b16 $r9 D[$r8 + #crypt_dtable]
+	call $r9
+
+	// do it
+	ld b16 $r9 D[$r8 + #crypt_dtable + 2]
+	call $r9
+	cxset 1
+	xdwait
+	cxset 0x61
+	xdwait
+	xdwait
+
+	// update src address
+	shr b32 $r8 $r4 0x18
+	shl b32 $r9 $r4 8
+	add b32 $r9 $r5
+	adc b32 $r8 0
+	st b32 D[$r0 + #ctx_src_address_high] $r8
+	st b32 D[$r0 + #ctx_src_address_low] $r9
+
+	// update dst address
+	shr b32 $r8 $r6 0x18
+	shl b32 $r9 $r6 8
+	add b32 $r9 $r7
+	adc b32 $r8 0
+	st b32 D[$r0 + #ctx_dst_address_high] $r8
+	st b32 D[$r0 + #ctx_dst_address_low] $r9
+
+	// pull updated IV
+	cxset 2
+	mov $r4 #ctx_iv
+	sethi $r4 0x60000
+	xdld $r0 $r4
+	xdwait
+
+	ret
+
+
+crypt_copy_prep:
+	cs0begin 2
+		cxsin $c0
+		cxsout $c0
+	ret
+
+crypt_store_prep:
+	cs0begin 1
+		cxsout $c6
+	ret
+
+crypt_ecb_e_prep:
+	cs0begin 3
+		cxsin $c0
+		cenc $c0 $c0
+		cxsout $c0
+	ret
+
+crypt_ecb_d_prep:
+	ckexp $c7 $c7
+	cs0begin 3
+		cxsin $c0
+		cdec $c0 $c0
+		cxsout $c0
+	ret
+
+crypt_cbc_e_prep:
+	cs0begin 4
+		cxsin $c0
+		cxor $c6 $c0
+		cenc $c6 $c6
+		cxsout $c6
+	ret
+
+crypt_cbc_d_prep:
+	ckexp $c7 $c7
+	cs0begin 5
+		cmov $c2 $c6
+		cxsin $c6
+		cdec $c0 $c6
+		cxor $c0 $c2
+		cxsout $c0
+	ret
+
+crypt_pcbc_e_prep:
+	cs0begin 5
+		cxsin $c0
+		cxor $c6 $c0
+		cenc $c6 $c6
+		cxsout $c6
+		cxor $c6 $c0
+	ret
+
+crypt_pcbc_d_prep:
+	ckexp $c7 $c7
+	cs0begin 5
+		cxsin $c0
+		cdec $c1 $c0
+		cxor $c6 $c1
+		cxsout $c6
+		cxor $c6 $c0
+	ret
+
+crypt_cfb_e_prep:
+	cs0begin 4
+		cenc $c6 $c6
+		cxsin $c0
+		cxor $c6 $c0
+		cxsout $c6
+	ret
+
+crypt_cfb_d_prep:
+	cs0begin 4
+		cenc $c0 $c6
+		cxsin $c6
+		cxor $c0 $c6
+		cxsout $c0
+	ret
+
+crypt_ofb_prep:
+	cs0begin 4
+		cenc $c6 $c6
+		cxsin $c0
+		cxor $c0 $c6
+		cxsout $c0
+	ret
+
+crypt_ctr_prep:
+	cs0begin 5
+		cenc $c1 $c6
+		cadd $c6 1
+		cxsin $c0
+		cxor $c0 $c1
+		cxsout $c0
+	ret
+
+crypt_cbc_mac_prep:
+	cs0begin 3
+		cxsin $c0
+		cxor $c6 $c0
+		cenc $c6 $c6
+	ret
+
+crypt_cmac_finish_complete_prep:
+	cs0begin 7
+		cxsin $c0
+		cxor $c6 $c0
+		cxor $c0 $c0
+		cenc $c0 $c0
+		cprecmac $c0 $c0
+		cxor $c6 $c0
+		cenc $c6 $c6
+	ret
+
+crypt_cmac_finish_partial_prep:
+	cs0begin 8
+		cxsin $c0
+		cxor $c6 $c0
+		cxor $c0 $c0
+		cenc $c0 $c0
+		cprecmac $c0 $c0
+		cprecmac $c0 $c0
+		cxor $c6 $c0
+		cenc $c6 $c6
+	ret
+
+// TODO
+crypt_do_in:
+	add b32 $r3 $r5
+	mov $xdbase $r4
+	mov $r9 #swap
+	sethi $r9 0x20000
+	crypt_do_in_loop:
+		xdld $r5 $r9
+		xdwait
+		cxset 0x22
+		xdst $r0 $r9
+		cs0exec 1
+		xdwait
+		add b32 $r5 0x10
+		cmpu b32 $r5 $r3
+	bra ne #crypt_do_in_loop
+	cxset 1
+	xdwait
+	ret
+
+crypt_do_out:
+	add b32 $r3 $r7
+	mov $xdbase $r6
+	mov $r9 #swap
+	sethi $r9 0x20000
+	crypt_do_out_loop:
+		cs0exec 1
+		cxset 0x61
+		xdld $r7 $r9
+		xdst $r7 $r9
+		cxset 1
+		xdwait
+		add b32 $r7 0x10
+		cmpu b32 $r7 $r3
+	bra ne #crypt_do_out_loop
+	ret
+
+crypt_do_inout:
+	add b32 $r3 $r5
+	mov $r9 #swap
+	sethi $r9 0x20000
+	crypt_do_inout_loop:
+		mov $xdbase $r4
+		xdld $r5 $r9
+		xdwait
+		cxset 0x21
+		xdst $r0 $r9
+		cs0exec 1
+		cxset 0x61
+		mov $xdbase $r6
+		xdld $r7 $r9
+		xdst $r7 $r9
+		cxset 1
+		xdwait
+		add b32 $r5 0x10
+		add b32 $r7 0x10
+		cmpu b32 $r5 $r3
+	bra ne #crypt_do_inout_loop
+	ret
+
+.align 0x100
diff --git a/nvkm/engine/crypt/nv84.c b/nvkm/engine/crypt/nv84.c
new file mode 100644
index 000000000..b97490512
--- /dev/null
+++ b/nvkm/engine/crypt/nv84.c
@@ -0,0 +1,195 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/enum.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/gpuobj.h>
+
+#include <subdev/fb.h>
+
+#include <engine/fifo.h>
+#include <engine/crypt.h>
+
+struct nv84_crypt_priv {
+	struct nouveau_engine base;
+};
+
+/*******************************************************************************
+ * Crypt object classes
+ ******************************************************************************/
+
+static int
+nv84_crypt_object_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nouveau_gpuobj *obj;
+	int ret;
+
+	ret = nouveau_gpuobj_create(parent, engine, oclass, 0, parent,
+				    16, 16, 0, &obj);
+	*pobject = nv_object(obj);
+	if (ret)
+		return ret;
+
+	nv_wo32(obj, 0x00, nv_mclass(obj));
+	nv_wo32(obj, 0x04, 0x00000000);
+	nv_wo32(obj, 0x08, 0x00000000);
+	nv_wo32(obj, 0x0c, 0x00000000);
+	return 0;
+}
+
+static struct nouveau_ofuncs
+nv84_crypt_ofuncs = {
+	.ctor = nv84_crypt_object_ctor,
+	.dtor = _nouveau_gpuobj_dtor,
+	.init = _nouveau_gpuobj_init,
+	.fini = _nouveau_gpuobj_fini,
+	.rd32 = _nouveau_gpuobj_rd32,
+	.wr32 = _nouveau_gpuobj_wr32,
+};
+
+static struct nouveau_oclass
+nv84_crypt_sclass[] = {
+	{ 0x74c1, &nv84_crypt_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * PCRYPT context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv84_crypt_cclass = {
+	.handle = NV_ENGCTX(CRYPT, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_engctx_ctor,
+		.dtor = _nouveau_engctx_dtor,
+		.init = _nouveau_engctx_init,
+		.fini = _nouveau_engctx_fini,
+		.rd32 = _nouveau_engctx_rd32,
+		.wr32 = _nouveau_engctx_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PCRYPT engine/subdev functions
+ ******************************************************************************/
+
+static const struct nouveau_bitfield nv84_crypt_intr_mask[] = {
+	{ 0x00000001, "INVALID_STATE" },
+	{ 0x00000002, "ILLEGAL_MTHD" },
+	{ 0x00000004, "ILLEGAL_CLASS" },
+	{ 0x00000080, "QUERY" },
+	{ 0x00000100, "FAULT" },
+	{}
+};
+
+static void
+nv84_crypt_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_fifo *pfifo = nouveau_fifo(subdev);
+	struct nouveau_engine *engine = nv_engine(subdev);
+	struct nouveau_object *engctx;
+	struct nv84_crypt_priv *priv = (void *)subdev;
+	u32 stat = nv_rd32(priv, 0x102130);
+	u32 mthd = nv_rd32(priv, 0x102190);
+	u32 data = nv_rd32(priv, 0x102194);
+	u32 inst = nv_rd32(priv, 0x102188) & 0x7fffffff;
+	int chid;
+
+	engctx = nouveau_engctx_get(engine, inst);
+	chid   = pfifo->chid(pfifo, engctx);
+
+	if (stat) {
+		nv_error(priv, "");
+		nouveau_bitfield_print(nv84_crypt_intr_mask, stat);
+		printk(" ch %d [0x%010llx] mthd 0x%04x data 0x%08x\n",
+		       chid, (u64)inst << 12, mthd, data);
+	}
+
+	nv_wr32(priv, 0x102130, stat);
+	nv_wr32(priv, 0x10200c, 0x10);
+
+	nouveau_engctx_put(engctx);
+}
+
+static int
+nv84_crypt_tlb_flush(struct nouveau_engine *engine)
+{
+	nv50_vm_flush_engine(&engine->base, 0x0a);
+	return 0;
+}
+
+static int
+nv84_crypt_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv84_crypt_priv *priv;
+	int ret;
+
+	ret = nouveau_engine_create(parent, engine, oclass, true,
+				    "PCRYPT", "crypt", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00004000;
+	nv_subdev(priv)->intr = nv84_crypt_intr;
+	nv_engine(priv)->cclass = &nv84_crypt_cclass;
+	nv_engine(priv)->sclass = nv84_crypt_sclass;
+	nv_engine(priv)->tlb_flush = nv84_crypt_tlb_flush;
+	return 0;
+}
+
+static int
+nv84_crypt_init(struct nouveau_object *object)
+{
+	struct nv84_crypt_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_engine_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x102130, 0xffffffff);
+	nv_wr32(priv, 0x102140, 0xffffffbf);
+	nv_wr32(priv, 0x10200c, 0x00000010);
+	return 0;
+}
+
+struct nouveau_oclass
+nv84_crypt_oclass = {
+	.handle = NV_ENGINE(CRYPT, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv84_crypt_ctor,
+		.dtor = _nouveau_engine_dtor,
+		.init = nv84_crypt_init,
+		.fini = _nouveau_engine_fini,
+	},
+};
diff --git a/nvkm/engine/crypt/nv98.c b/nvkm/engine/crypt/nv98.c
new file mode 100644
index 000000000..21986f3bf
--- /dev/null
+++ b/nvkm/engine/crypt/nv98.c
@@ -0,0 +1,163 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/enum.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/falcon.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+
+#include <engine/fifo.h>
+#include <engine/crypt.h>
+
+#include "fuc/nv98.fuc.h"
+
+struct nv98_crypt_priv {
+	struct nouveau_falcon base;
+};
+
+/*******************************************************************************
+ * Crypt object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv98_crypt_sclass[] = {
+	{ 0x88b4, &nouveau_object_ofuncs },
+	{},
+};
+
+/*******************************************************************************
+ * PCRYPT context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv98_crypt_cclass = {
+	.handle = NV_ENGCTX(CRYPT, 0x98),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_falcon_context_ctor,
+		.dtor = _nouveau_falcon_context_dtor,
+		.init = _nouveau_falcon_context_init,
+		.fini = _nouveau_falcon_context_fini,
+		.rd32 = _nouveau_falcon_context_rd32,
+		.wr32 = _nouveau_falcon_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PCRYPT engine/subdev functions
+ ******************************************************************************/
+
+static const struct nouveau_enum nv98_crypt_isr_error_name[] = {
+	{ 0x0000, "ILLEGAL_MTHD" },
+	{ 0x0001, "INVALID_BITFIELD" },
+	{ 0x0002, "INVALID_ENUM" },
+	{ 0x0003, "QUERY" },
+	{}
+};
+
+static void
+nv98_crypt_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_fifo *pfifo = nouveau_fifo(subdev);
+	struct nouveau_engine *engine = nv_engine(subdev);
+	struct nouveau_object *engctx;
+	struct nv98_crypt_priv *priv = (void *)subdev;
+	u32 disp = nv_rd32(priv, 0x08701c);
+	u32 stat = nv_rd32(priv, 0x087008) & disp & ~(disp >> 16);
+	u32 inst = nv_rd32(priv, 0x087050) & 0x3fffffff;
+	u32 ssta = nv_rd32(priv, 0x087040) & 0x0000ffff;
+	u32 addr = nv_rd32(priv, 0x087040) >> 16;
+	u32 mthd = (addr & 0x07ff) << 2;
+	u32 subc = (addr & 0x3800) >> 11;
+	u32 data = nv_rd32(priv, 0x087044);
+	int chid;
+
+	engctx = nouveau_engctx_get(engine, inst);
+	chid   = pfifo->chid(pfifo, engctx);
+
+	if (stat & 0x00000040) {
+		nv_error(priv, "DISPATCH_ERROR [");
+		nouveau_enum_print(nv98_crypt_isr_error_name, ssta);
+		printk("] ch %d [0x%010llx] subc %d mthd 0x%04x data 0x%08x\n",
+		       chid, (u64)inst << 12, subc, mthd, data);
+		nv_wr32(priv, 0x087004, 0x00000040);
+		stat &= ~0x00000040;
+	}
+
+	if (stat) {
+		nv_error(priv, "unhandled intr 0x%08x\n", stat);
+		nv_wr32(priv, 0x087004, stat);
+	}
+
+	nouveau_engctx_put(engctx);
+}
+
+static int
+nv98_crypt_tlb_flush(struct nouveau_engine *engine)
+{
+	nv50_vm_flush_engine(&engine->base, 0x0a);
+	return 0;
+}
+
+static int
+nv98_crypt_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv98_crypt_priv *priv;
+	int ret;
+
+	ret = nouveau_falcon_create(parent, engine, oclass, 0x087000, true,
+				    "PCRYPT", "crypt", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00004000;
+	nv_subdev(priv)->intr = nv98_crypt_intr;
+	nv_engine(priv)->cclass = &nv98_crypt_cclass;
+	nv_engine(priv)->sclass = nv98_crypt_sclass;
+	nv_engine(priv)->tlb_flush = nv98_crypt_tlb_flush;
+	nv_falcon(priv)->code.data = nv98_pcrypt_code;
+	nv_falcon(priv)->code.size = sizeof(nv98_pcrypt_code);
+	nv_falcon(priv)->data.data = nv98_pcrypt_data;
+	nv_falcon(priv)->data.size = sizeof(nv98_pcrypt_data);
+	return 0;
+}
+
+struct nouveau_oclass
+nv98_crypt_oclass = {
+	.handle = NV_ENGINE(CRYPT, 0x98),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv98_crypt_ctor,
+		.dtor = _nouveau_falcon_dtor,
+		.init = _nouveau_falcon_init,
+		.fini = _nouveau_falcon_fini,
+		.rd32 = _nouveau_falcon_rd32,
+		.wr32 = _nouveau_falcon_wr32,
+	},
+};
diff --git a/nvkm/engine/disp/Makefile.am b/nvkm/engine/disp/Makefile.am
new file mode 100644
index 000000000..7e3d55053
--- /dev/null
+++ b/nvkm/engine/disp/Makefile.am
@@ -0,0 +1,22 @@
+noinst_LTLIBRARIES = libdisp.la
+
+libdisp_la_SOURCES = nv04.c \
+		     nv50.c \
+		     nv84.c \
+		     nv94.c \
+		     nva0.c \
+		     nva3.c \
+		     nvd0.c \
+		     nve0.c \
+		     dacnv50.c \
+		     hdanva3.c \
+		     hdanvd0.c \
+		     hdminv84.c \
+		     hdminva3.c \
+		     hdminvd0.c \
+		     sornv50.c \
+		     sornv94.c \
+		     sornvd0.c \
+		     vga.c
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/disp/dacnv50.c b/nvkm/engine/disp/dacnv50.c
new file mode 100644
index 000000000..d0817d944
--- /dev/null
+++ b/nvkm/engine/disp/dacnv50.c
@@ -0,0 +1,88 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/timer.h>
+
+#include "nv50.h"
+
+int
+nv50_dac_power(struct nv50_disp_priv *priv, int or, u32 data)
+{
+	const u32 stat = (data & NV50_DISP_DAC_PWR_HSYNC) |
+		         (data & NV50_DISP_DAC_PWR_VSYNC) |
+		         (data & NV50_DISP_DAC_PWR_DATA) |
+		         (data & NV50_DISP_DAC_PWR_STATE);
+	const u32 doff = (or * 0x800);
+	nv_wait(priv, 0x61a004 + doff, 0x80000000, 0x00000000);
+	nv_mask(priv, 0x61a004 + doff, 0xc000007f, 0x80000000 | stat);
+	nv_wait(priv, 0x61a004 + doff, 0x80000000, 0x00000000);
+	return 0;
+}
+
+int
+nv50_dac_sense(struct nv50_disp_priv *priv, int or, u32 loadval)
+{
+	const u32 doff = (or * 0x800);
+	int load = -EINVAL;
+	nv_wr32(priv, 0x61a00c + doff, 0x00100000 | loadval);
+	udelay(9500);
+	nv_wr32(priv, 0x61a00c + doff, 0x80000000);
+	load = (nv_rd32(priv, 0x61a00c + doff) & 0x38000000) >> 27;
+	nv_wr32(priv, 0x61a00c + doff, 0x00000000);
+	return load;
+}
+
+int
+nv50_dac_mthd(struct nouveau_object *object, u32 mthd, void *args, u32 size)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	const u8 or = (mthd & NV50_DISP_DAC_MTHD_OR);
+	u32 *data = args;
+	int ret;
+
+	if (size < sizeof(u32))
+		return -EINVAL;
+
+	switch (mthd & ~0x3f) {
+	case NV50_DISP_DAC_PWR:
+		ret = priv->dac.power(priv, or, data[0]);
+		break;
+	case NV50_DISP_DAC_LOAD:
+		ret = priv->dac.sense(priv, or, data[0]);
+		if (ret >= 0) {
+			data[0] = ret;
+			ret = 0;
+		}
+		break;
+	default:
+		BUG_ON(1);
+	}
+
+	return ret;
+}
diff --git a/nvkm/engine/disp/hdanva3.c b/nvkm/engine/disp/hdanva3.c
new file mode 100644
index 000000000..373dbcc52
--- /dev/null
+++ b/nvkm/engine/disp/hdanva3.c
@@ -0,0 +1,48 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+
+#include "nv50.h"
+
+int
+nva3_hda_eld(struct nv50_disp_priv *priv, int or, u8 *data, u32 size)
+{
+	const u32 soff = (or * 0x800);
+	int i;
+
+	if (data && data[0]) {
+		for (i = 0; i < size; i++)
+			nv_wr32(priv, 0x61c440 + soff, (i << 8) | data[i]);
+		nv_mask(priv, 0x61c448 + soff, 0x80000003, 0x80000003);
+	} else
+	if (data) {
+		nv_mask(priv, 0x61c448 + soff, 0x80000003, 0x80000001);
+	} else {
+		nv_mask(priv, 0x61c448 + soff, 0x80000003, 0x80000000);
+	}
+
+	return 0;
+}
diff --git a/nvkm/engine/disp/hdanvd0.c b/nvkm/engine/disp/hdanvd0.c
new file mode 100644
index 000000000..dc57e24fc
--- /dev/null
+++ b/nvkm/engine/disp/hdanvd0.c
@@ -0,0 +1,53 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/dp.h>
+#include <subdev/bios/init.h>
+
+#include "nv50.h"
+
+int
+nvd0_hda_eld(struct nv50_disp_priv *priv, int or, u8 *data, u32 size)
+{
+	const u32 soff = (or * 0x030);
+	int i;
+
+	if (data && data[0]) {
+		for (i = 0; i < size; i++)
+			nv_wr32(priv, 0x10ec00 + soff, (i << 8) | data[i]);
+		nv_mask(priv, 0x10ec10 + soff, 0x80000003, 0x80000003);
+	} else
+	if (data) {
+		nv_mask(priv, 0x10ec10 + soff, 0x80000003, 0x80000001);
+	} else {
+		nv_mask(priv, 0x10ec10 + soff, 0x80000003, 0x80000000);
+	}
+
+	return 0;
+}
diff --git a/nvkm/engine/disp/hdminv84.c b/nvkm/engine/disp/hdminv84.c
new file mode 100644
index 000000000..0d36bdc51
--- /dev/null
+++ b/nvkm/engine/disp/hdminv84.c
@@ -0,0 +1,66 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+
+#include "nv50.h"
+
+int
+nv84_hdmi_ctrl(struct nv50_disp_priv *priv, int head, int or, u32 data)
+{
+	const u32 hoff = (head * 0x800);
+
+	if (!(data & NV84_DISP_SOR_HDMI_PWR_STATE_ON)) {
+		nv_mask(priv, 0x6165a4 + hoff, 0x40000000, 0x00000000);
+		nv_mask(priv, 0x616520 + hoff, 0x00000001, 0x00000000);
+		nv_mask(priv, 0x616500 + hoff, 0x00000001, 0x00000000);
+		return 0;
+	}
+
+	/* AVI InfoFrame */
+	nv_mask(priv, 0x616520 + hoff, 0x00000001, 0x00000000);
+	nv_wr32(priv, 0x616528 + hoff, 0x000d0282);
+	nv_wr32(priv, 0x61652c + hoff, 0x0000006f);
+	nv_wr32(priv, 0x616530 + hoff, 0x00000000);
+	nv_wr32(priv, 0x616534 + hoff, 0x00000000);
+	nv_wr32(priv, 0x616538 + hoff, 0x00000000);
+	nv_mask(priv, 0x616520 + hoff, 0x00000001, 0x00000001);
+
+	/* Audio InfoFrame */
+	nv_mask(priv, 0x616500 + hoff, 0x00000001, 0x00000000);
+	nv_wr32(priv, 0x616508 + hoff, 0x000a0184);
+	nv_wr32(priv, 0x61650c + hoff, 0x00000071);
+	nv_wr32(priv, 0x616510 + hoff, 0x00000000);
+	nv_mask(priv, 0x616500 + hoff, 0x00000001, 0x00000001);
+
+	/* ??? */
+	nv_mask(priv, 0x61733c, 0x00100000, 0x00100000); /* RESETF */
+	nv_mask(priv, 0x61733c, 0x10000000, 0x10000000); /* LOOKUP_EN */
+	nv_mask(priv, 0x61733c, 0x00100000, 0x00000000); /* !RESETF */
+
+	/* HDMI_CTRL */
+	nv_mask(priv, 0x6165a4 + hoff, 0x5f1f007f, data | 0x1f000000 /* ??? */);
+	return 0;
+}
diff --git a/nvkm/engine/disp/hdminva3.c b/nvkm/engine/disp/hdminva3.c
new file mode 100644
index 000000000..f065fc248
--- /dev/null
+++ b/nvkm/engine/disp/hdminva3.c
@@ -0,0 +1,66 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+
+#include "nv50.h"
+
+int
+nva3_hdmi_ctrl(struct nv50_disp_priv *priv, int head, int or, u32 data)
+{
+	const u32 soff = (or * 0x800);
+
+	if (!(data & NV84_DISP_SOR_HDMI_PWR_STATE_ON)) {
+		nv_mask(priv, 0x61c5a4 + soff, 0x40000000, 0x00000000);
+		nv_mask(priv, 0x61c520 + soff, 0x00000001, 0x00000000);
+		nv_mask(priv, 0x61c500 + soff, 0x00000001, 0x00000000);
+		return 0;
+	}
+
+	/* AVI InfoFrame */
+	nv_mask(priv, 0x61c520 + soff, 0x00000001, 0x00000000);
+	nv_wr32(priv, 0x61c528 + soff, 0x000d0282);
+	nv_wr32(priv, 0x61c52c + soff, 0x0000006f);
+	nv_wr32(priv, 0x61c530 + soff, 0x00000000);
+	nv_wr32(priv, 0x61c534 + soff, 0x00000000);
+	nv_wr32(priv, 0x61c538 + soff, 0x00000000);
+	nv_mask(priv, 0x61c520 + soff, 0x00000001, 0x00000001);
+
+	/* Audio InfoFrame */
+	nv_mask(priv, 0x61c500 + soff, 0x00000001, 0x00000000);
+	nv_wr32(priv, 0x61c508 + soff, 0x000a0184);
+	nv_wr32(priv, 0x61c50c + soff, 0x00000071);
+	nv_wr32(priv, 0x61c510 + soff, 0x00000000);
+	nv_mask(priv, 0x61c500 + soff, 0x00000001, 0x00000001);
+
+	/* ??? */
+	nv_mask(priv, 0x61733c, 0x00100000, 0x00100000); /* RESETF */
+	nv_mask(priv, 0x61733c, 0x10000000, 0x10000000); /* LOOKUP_EN */
+	nv_mask(priv, 0x61733c, 0x00100000, 0x00000000); /* !RESETF */
+
+	/* HDMI_CTRL */
+	nv_mask(priv, 0x61c5a4 + soff, 0x5f1f007f, data | 0x1f000000 /* ??? */);
+	return 0;
+}
diff --git a/nvkm/engine/disp/hdminvd0.c b/nvkm/engine/disp/hdminvd0.c
new file mode 100644
index 000000000..5151bb261
--- /dev/null
+++ b/nvkm/engine/disp/hdminvd0.c
@@ -0,0 +1,62 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+
+#include "nv50.h"
+
+int
+nvd0_hdmi_ctrl(struct nv50_disp_priv *priv, int head, int or, u32 data)
+{
+	const u32 hoff = (head * 0x800);
+
+	if (!(data & NV84_DISP_SOR_HDMI_PWR_STATE_ON)) {
+		nv_mask(priv, 0x616798 + hoff, 0x40000000, 0x00000000);
+		nv_mask(priv, 0x6167a4 + hoff, 0x00000001, 0x00000000);
+		nv_mask(priv, 0x616714 + hoff, 0x00000001, 0x00000000);
+		return 0;
+	}
+
+	/* AVI InfoFrame */
+	nv_mask(priv, 0x616714 + hoff, 0x00000001, 0x00000000);
+	nv_wr32(priv, 0x61671c + hoff, 0x000d0282);
+	nv_wr32(priv, 0x616720 + hoff, 0x0000006f);
+	nv_wr32(priv, 0x616724 + hoff, 0x00000000);
+	nv_wr32(priv, 0x616728 + hoff, 0x00000000);
+	nv_wr32(priv, 0x61672c + hoff, 0x00000000);
+	nv_mask(priv, 0x616714 + hoff, 0x00000001, 0x00000001);
+
+	/* ??? InfoFrame? */
+	nv_mask(priv, 0x6167a4 + hoff, 0x00000001, 0x00000000);
+	nv_wr32(priv, 0x6167ac + hoff, 0x00000010);
+	nv_mask(priv, 0x6167a4 + hoff, 0x00000001, 0x00000001);
+
+	/* HDMI_CTRL */
+	nv_mask(priv, 0x616798 + hoff, 0x401f007f, data);
+
+	/* NFI, audio doesn't work without it though.. */
+	nv_mask(priv, 0x616548 + hoff, 0x00000070, 0x00000000);
+	return 0;
+}
diff --git a/nvkm/engine/disp/nv04.c b/nvkm/engine/disp/nv04.c
new file mode 100644
index 000000000..1c919f2af
--- /dev/null
+++ b/nvkm/engine/disp/nv04.c
@@ -0,0 +1,90 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <engine/disp.h>
+
+struct nv04_disp_priv {
+	struct nouveau_disp base;
+};
+
+static struct nouveau_oclass
+nv04_disp_sclass[] = {
+	{},
+};
+
+static void
+nv04_disp_intr_vblank(struct nv04_disp_priv *priv, int crtc)
+{
+	struct nouveau_disp *disp = &priv->base;
+	if (disp->vblank.notify)
+		disp->vblank.notify(disp->vblank.data, crtc);
+}
+
+static void
+nv04_disp_intr(struct nouveau_subdev *subdev)
+{
+	struct nv04_disp_priv *priv = (void *)subdev;
+	u32 crtc0 = nv_rd32(priv, 0x600100);
+	u32 crtc1 = nv_rd32(priv, 0x602100);
+
+	if (crtc0 & 0x00000001) {
+		nv04_disp_intr_vblank(priv, 0);
+		nv_wr32(priv, 0x600100, 0x00000001);
+	}
+
+	if (crtc1 & 0x00000001) {
+		nv04_disp_intr_vblank(priv, 1);
+		nv_wr32(priv, 0x602100, 0x00000001);
+	}
+}
+
+static int
+nv04_disp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv04_disp_priv *priv;
+	int ret;
+
+	ret = nouveau_disp_create(parent, engine, oclass, "DISPLAY",
+				  "display", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nv04_disp_sclass;
+	nv_subdev(priv)->intr = nv04_disp_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nv04_disp_oclass = {
+	.handle = NV_ENGINE(DISP, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_disp_ctor,
+		.dtor = _nouveau_disp_dtor,
+		.init = _nouveau_disp_init,
+		.fini = _nouveau_disp_fini,
+	},
+};
diff --git a/nvkm/engine/disp/nv50.c b/nvkm/engine/disp/nv50.c
new file mode 100644
index 000000000..0f09af135
--- /dev/null
+++ b/nvkm/engine/disp/nv50.c
@@ -0,0 +1,1254 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/parent.h>
+#include <core/handle.h>
+#include <core/class.h>
+
+#include <engine/software.h>
+#include <engine/disp.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/disp.h>
+#include <subdev/bios/init.h>
+#include <subdev/bios/pll.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/bar.h>
+#include <subdev/clock.h>
+
+#include "nv50.h"
+
+/*******************************************************************************
+ * EVO channel base class
+ ******************************************************************************/
+
+int
+nv50_disp_chan_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, int chid,
+		       int length, void **pobject)
+{
+	struct nv50_disp_base *base = (void *)parent;
+	struct nv50_disp_chan *chan;
+	int ret;
+
+	if (base->chan & (1 << chid))
+		return -EBUSY;
+	base->chan |= (1 << chid);
+
+	ret = nouveau_namedb_create_(parent, engine, oclass, 0, NULL,
+				     (1ULL << NVDEV_ENGINE_DMAOBJ),
+				     length, pobject);
+	chan = *pobject;
+	if (ret)
+		return ret;
+
+	chan->chid = chid;
+	return 0;
+}
+
+void
+nv50_disp_chan_destroy(struct nv50_disp_chan *chan)
+{
+	struct nv50_disp_base *base = (void *)nv_object(chan)->parent;
+	base->chan &= ~(1 << chan->chid);
+	nouveau_namedb_destroy(&chan->base);
+}
+
+u32
+nv50_disp_chan_rd32(struct nouveau_object *object, u64 addr)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_chan *chan = (void *)object;
+	return nv_rd32(priv, 0x640000 + (chan->chid * 0x1000) + addr);
+}
+
+void
+nv50_disp_chan_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_chan *chan = (void *)object;
+	nv_wr32(priv, 0x640000 + (chan->chid * 0x1000) + addr, data);
+}
+
+/*******************************************************************************
+ * EVO DMA channel base class
+ ******************************************************************************/
+
+static int
+nv50_disp_dmac_object_attach(struct nouveau_object *parent,
+			     struct nouveau_object *object, u32 name)
+{
+	struct nv50_disp_base *base = (void *)parent->parent;
+	struct nv50_disp_chan *chan = (void *)parent;
+	u32 addr = nv_gpuobj(object)->node->offset;
+	u32 chid = chan->chid;
+	u32 data = (chid << 28) | (addr << 10) | chid;
+	return nouveau_ramht_insert(base->ramht, chid, name, data);
+}
+
+static void
+nv50_disp_dmac_object_detach(struct nouveau_object *parent, int cookie)
+{
+	struct nv50_disp_base *base = (void *)parent->parent;
+	nouveau_ramht_remove(base->ramht, cookie);
+}
+
+int
+nv50_disp_dmac_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, u32 pushbuf, int chid,
+		       int length, void **pobject)
+{
+	struct nv50_disp_dmac *dmac;
+	int ret;
+
+	ret = nv50_disp_chan_create_(parent, engine, oclass, chid,
+				     length, pobject);
+	dmac = *pobject;
+	if (ret)
+		return ret;
+
+	dmac->pushdma = (void *)nouveau_handle_ref(parent, pushbuf);
+	if (!dmac->pushdma)
+		return -ENOENT;
+
+	switch (nv_mclass(dmac->pushdma)) {
+	case 0x0002:
+	case 0x003d:
+		if (dmac->pushdma->limit - dmac->pushdma->start != 0xfff)
+			return -EINVAL;
+
+		switch (dmac->pushdma->target) {
+		case NV_MEM_TARGET_VRAM:
+			dmac->push = 0x00000000 | dmac->pushdma->start >> 8;
+			break;
+		case NV_MEM_TARGET_PCI_NOSNOOP:
+			dmac->push = 0x00000003 | dmac->pushdma->start >> 8;
+			break;
+		default:
+			return -EINVAL;
+		}
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+void
+nv50_disp_dmac_dtor(struct nouveau_object *object)
+{
+	struct nv50_disp_dmac *dmac = (void *)object;
+	nouveau_object_ref(NULL, (struct nouveau_object **)&dmac->pushdma);
+	nv50_disp_chan_destroy(&dmac->base);
+}
+
+static int
+nv50_disp_dmac_init(struct nouveau_object *object)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_dmac *dmac = (void *)object;
+	int chid = dmac->base.chid;
+	int ret;
+
+	ret = nv50_disp_chan_init(&dmac->base);
+	if (ret)
+		return ret;
+
+	/* enable error reporting */
+	nv_mask(priv, 0x610028, 0x00010001 << chid, 0x00010001 << chid);
+
+	/* initialise channel for dma command submission */
+	nv_wr32(priv, 0x610204 + (chid * 0x0010), dmac->push);
+	nv_wr32(priv, 0x610208 + (chid * 0x0010), 0x00010000);
+	nv_wr32(priv, 0x61020c + (chid * 0x0010), chid);
+	nv_mask(priv, 0x610200 + (chid * 0x0010), 0x00000010, 0x00000010);
+	nv_wr32(priv, 0x640000 + (chid * 0x1000), 0x00000000);
+	nv_wr32(priv, 0x610200 + (chid * 0x0010), 0x00000013);
+
+	/* wait for it to go inactive */
+	if (!nv_wait(priv, 0x610200 + (chid * 0x10), 0x80000000, 0x00000000)) {
+		nv_error(dmac, "init timeout, 0x%08x\n",
+			 nv_rd32(priv, 0x610200 + (chid * 0x10)));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+static int
+nv50_disp_dmac_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_dmac *dmac = (void *)object;
+	int chid = dmac->base.chid;
+
+	/* deactivate channel */
+	nv_mask(priv, 0x610200 + (chid * 0x0010), 0x00001010, 0x00001000);
+	nv_mask(priv, 0x610200 + (chid * 0x0010), 0x00000003, 0x00000000);
+	if (!nv_wait(priv, 0x610200 + (chid * 0x10), 0x001e0000, 0x00000000)) {
+		nv_error(dmac, "fini timeout, 0x%08x\n",
+			 nv_rd32(priv, 0x610200 + (chid * 0x10)));
+		if (suspend)
+			return -EBUSY;
+	}
+
+	/* disable error reporting */
+	nv_mask(priv, 0x610028, 0x00010001 << chid, 0x00000000 << chid);
+
+	return nv50_disp_chan_fini(&dmac->base, suspend);
+}
+
+/*******************************************************************************
+ * EVO master channel object
+ ******************************************************************************/
+
+static int
+nv50_disp_mast_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_mast_class *args = data;
+	struct nv50_disp_dmac *mast;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nv50_disp_dmac_create_(parent, engine, oclass, args->pushbuf,
+				     0, sizeof(*mast), (void **)&mast);
+	*pobject = nv_object(mast);
+	if (ret)
+		return ret;
+
+	nv_parent(mast)->object_attach = nv50_disp_dmac_object_attach;
+	nv_parent(mast)->object_detach = nv50_disp_dmac_object_detach;
+	return 0;
+}
+
+static int
+nv50_disp_mast_init(struct nouveau_object *object)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_dmac *mast = (void *)object;
+	int ret;
+
+	ret = nv50_disp_chan_init(&mast->base);
+	if (ret)
+		return ret;
+
+	/* enable error reporting */
+	nv_mask(priv, 0x610028, 0x00010001, 0x00010001);
+
+	/* attempt to unstick channel from some unknown state */
+	if ((nv_rd32(priv, 0x610200) & 0x009f0000) == 0x00020000)
+		nv_mask(priv, 0x610200, 0x00800000, 0x00800000);
+	if ((nv_rd32(priv, 0x610200) & 0x003f0000) == 0x00030000)
+		nv_mask(priv, 0x610200, 0x00600000, 0x00600000);
+
+	/* initialise channel for dma command submission */
+	nv_wr32(priv, 0x610204, mast->push);
+	nv_wr32(priv, 0x610208, 0x00010000);
+	nv_wr32(priv, 0x61020c, 0x00000000);
+	nv_mask(priv, 0x610200, 0x00000010, 0x00000010);
+	nv_wr32(priv, 0x640000, 0x00000000);
+	nv_wr32(priv, 0x610200, 0x01000013);
+
+	/* wait for it to go inactive */
+	if (!nv_wait(priv, 0x610200, 0x80000000, 0x00000000)) {
+		nv_error(mast, "init: 0x%08x\n", nv_rd32(priv, 0x610200));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+static int
+nv50_disp_mast_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_dmac *mast = (void *)object;
+
+	/* deactivate channel */
+	nv_mask(priv, 0x610200, 0x00000010, 0x00000000);
+	nv_mask(priv, 0x610200, 0x00000003, 0x00000000);
+	if (!nv_wait(priv, 0x610200, 0x001e0000, 0x00000000)) {
+		nv_error(mast, "fini: 0x%08x\n", nv_rd32(priv, 0x610200));
+		if (suspend)
+			return -EBUSY;
+	}
+
+	/* disable error reporting */
+	nv_mask(priv, 0x610028, 0x00010001, 0x00000000);
+
+	return nv50_disp_chan_fini(&mast->base, suspend);
+}
+
+struct nouveau_ofuncs
+nv50_disp_mast_ofuncs = {
+	.ctor = nv50_disp_mast_ctor,
+	.dtor = nv50_disp_dmac_dtor,
+	.init = nv50_disp_mast_init,
+	.fini = nv50_disp_mast_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * EVO sync channel objects
+ ******************************************************************************/
+
+static int
+nv50_disp_sync_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_sync_class *args = data;
+	struct nv50_disp_dmac *dmac;
+	int ret;
+
+	if (size < sizeof(*data) || args->head > 1)
+		return -EINVAL;
+
+	ret = nv50_disp_dmac_create_(parent, engine, oclass, args->pushbuf,
+				     1 + args->head, sizeof(*dmac),
+				     (void **)&dmac);
+	*pobject = nv_object(dmac);
+	if (ret)
+		return ret;
+
+	nv_parent(dmac)->object_attach = nv50_disp_dmac_object_attach;
+	nv_parent(dmac)->object_detach = nv50_disp_dmac_object_detach;
+	return 0;
+}
+
+struct nouveau_ofuncs
+nv50_disp_sync_ofuncs = {
+	.ctor = nv50_disp_sync_ctor,
+	.dtor = nv50_disp_dmac_dtor,
+	.init = nv50_disp_dmac_init,
+	.fini = nv50_disp_dmac_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * EVO overlay channel objects
+ ******************************************************************************/
+
+static int
+nv50_disp_ovly_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_ovly_class *args = data;
+	struct nv50_disp_dmac *dmac;
+	int ret;
+
+	if (size < sizeof(*data) || args->head > 1)
+		return -EINVAL;
+
+	ret = nv50_disp_dmac_create_(parent, engine, oclass, args->pushbuf,
+				     3 + args->head, sizeof(*dmac),
+				     (void **)&dmac);
+	*pobject = nv_object(dmac);
+	if (ret)
+		return ret;
+
+	nv_parent(dmac)->object_attach = nv50_disp_dmac_object_attach;
+	nv_parent(dmac)->object_detach = nv50_disp_dmac_object_detach;
+	return 0;
+}
+
+struct nouveau_ofuncs
+nv50_disp_ovly_ofuncs = {
+	.ctor = nv50_disp_ovly_ctor,
+	.dtor = nv50_disp_dmac_dtor,
+	.init = nv50_disp_dmac_init,
+	.fini = nv50_disp_dmac_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * EVO PIO channel base class
+ ******************************************************************************/
+
+static int
+nv50_disp_pioc_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, int chid,
+		       int length, void **pobject)
+{
+	return nv50_disp_chan_create_(parent, engine, oclass, chid,
+				      length, pobject);
+}
+
+static void
+nv50_disp_pioc_dtor(struct nouveau_object *object)
+{
+	struct nv50_disp_pioc *pioc = (void *)object;
+	nv50_disp_chan_destroy(&pioc->base);
+}
+
+static int
+nv50_disp_pioc_init(struct nouveau_object *object)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_pioc *pioc = (void *)object;
+	int chid = pioc->base.chid;
+	int ret;
+
+	ret = nv50_disp_chan_init(&pioc->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x610200 + (chid * 0x10), 0x00002000);
+	if (!nv_wait(priv, 0x610200 + (chid * 0x10), 0x00000000, 0x00000000)) {
+		nv_error(pioc, "timeout0: 0x%08x\n",
+			 nv_rd32(priv, 0x610200 + (chid * 0x10)));
+		return -EBUSY;
+	}
+
+	nv_wr32(priv, 0x610200 + (chid * 0x10), 0x00000001);
+	if (!nv_wait(priv, 0x610200 + (chid * 0x10), 0x00030000, 0x00010000)) {
+		nv_error(pioc, "timeout1: 0x%08x\n",
+			 nv_rd32(priv, 0x610200 + (chid * 0x10)));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+static int
+nv50_disp_pioc_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_pioc *pioc = (void *)object;
+	int chid = pioc->base.chid;
+
+	nv_mask(priv, 0x610200 + (chid * 0x10), 0x00000001, 0x00000000);
+	if (!nv_wait(priv, 0x610200 + (chid * 0x10), 0x00030000, 0x00000000)) {
+		nv_error(pioc, "timeout: 0x%08x\n",
+			 nv_rd32(priv, 0x610200 + (chid * 0x10)));
+		if (suspend)
+			return -EBUSY;
+	}
+
+	return nv50_disp_chan_fini(&pioc->base, suspend);
+}
+
+/*******************************************************************************
+ * EVO immediate overlay channel objects
+ ******************************************************************************/
+
+static int
+nv50_disp_oimm_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_oimm_class *args = data;
+	struct nv50_disp_pioc *pioc;
+	int ret;
+
+	if (size < sizeof(*args) || args->head > 1)
+		return -EINVAL;
+
+	ret = nv50_disp_pioc_create_(parent, engine, oclass, 5 + args->head,
+				     sizeof(*pioc), (void **)&pioc);
+	*pobject = nv_object(pioc);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+struct nouveau_ofuncs
+nv50_disp_oimm_ofuncs = {
+	.ctor = nv50_disp_oimm_ctor,
+	.dtor = nv50_disp_pioc_dtor,
+	.init = nv50_disp_pioc_init,
+	.fini = nv50_disp_pioc_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * EVO cursor channel objects
+ ******************************************************************************/
+
+static int
+nv50_disp_curs_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_curs_class *args = data;
+	struct nv50_disp_pioc *pioc;
+	int ret;
+
+	if (size < sizeof(*args) || args->head > 1)
+		return -EINVAL;
+
+	ret = nv50_disp_pioc_create_(parent, engine, oclass, 7 + args->head,
+				     sizeof(*pioc), (void **)&pioc);
+	*pobject = nv_object(pioc);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+struct nouveau_ofuncs
+nv50_disp_curs_ofuncs = {
+	.ctor = nv50_disp_curs_ctor,
+	.dtor = nv50_disp_pioc_dtor,
+	.init = nv50_disp_pioc_init,
+	.fini = nv50_disp_pioc_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * Base display object
+ ******************************************************************************/
+
+static int
+nv50_disp_base_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv = (void *)engine;
+	struct nv50_disp_base *base;
+	int ret;
+
+	ret = nouveau_parent_create(parent, engine, oclass, 0,
+				    priv->sclass, 0, &base);
+	*pobject = nv_object(base);
+	if (ret)
+		return ret;
+
+	return nouveau_ramht_new(parent, parent, 0x1000, 0, &base->ramht);
+}
+
+static void
+nv50_disp_base_dtor(struct nouveau_object *object)
+{
+	struct nv50_disp_base *base = (void *)object;
+	nouveau_ramht_ref(NULL, &base->ramht);
+	nouveau_parent_destroy(&base->base);
+}
+
+static int
+nv50_disp_base_init(struct nouveau_object *object)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_base *base = (void *)object;
+	int ret, i;
+	u32 tmp;
+
+	ret = nouveau_parent_init(&base->base);
+	if (ret)
+		return ret;
+
+	/* The below segments of code copying values from one register to
+	 * another appear to inform EVO of the display capabilities or
+	 * something similar.  NFI what the 0x614004 caps are for..
+	 */
+	tmp = nv_rd32(priv, 0x614004);
+	nv_wr32(priv, 0x610184, tmp);
+
+	/* ... CRTC caps */
+	for (i = 0; i < priv->head.nr; i++) {
+		tmp = nv_rd32(priv, 0x616100 + (i * 0x800));
+		nv_wr32(priv, 0x610190 + (i * 0x10), tmp);
+		tmp = nv_rd32(priv, 0x616104 + (i * 0x800));
+		nv_wr32(priv, 0x610194 + (i * 0x10), tmp);
+		tmp = nv_rd32(priv, 0x616108 + (i * 0x800));
+		nv_wr32(priv, 0x610198 + (i * 0x10), tmp);
+		tmp = nv_rd32(priv, 0x61610c + (i * 0x800));
+		nv_wr32(priv, 0x61019c + (i * 0x10), tmp);
+	}
+
+	/* ... DAC caps */
+	for (i = 0; i < priv->dac.nr; i++) {
+		tmp = nv_rd32(priv, 0x61a000 + (i * 0x800));
+		nv_wr32(priv, 0x6101d0 + (i * 0x04), tmp);
+	}
+
+	/* ... SOR caps */
+	for (i = 0; i < priv->sor.nr; i++) {
+		tmp = nv_rd32(priv, 0x61c000 + (i * 0x800));
+		nv_wr32(priv, 0x6101e0 + (i * 0x04), tmp);
+	}
+
+	/* ... EXT caps */
+	for (i = 0; i < 3; i++) {
+		tmp = nv_rd32(priv, 0x61e000 + (i * 0x800));
+		nv_wr32(priv, 0x6101f0 + (i * 0x04), tmp);
+	}
+
+	/* steal display away from vbios, or something like that */
+	if (nv_rd32(priv, 0x610024) & 0x00000100) {
+		nv_wr32(priv, 0x610024, 0x00000100);
+		nv_mask(priv, 0x6194e8, 0x00000001, 0x00000000);
+		if (!nv_wait(priv, 0x6194e8, 0x00000002, 0x00000000)) {
+			nv_error(priv, "timeout acquiring display\n");
+			return -EBUSY;
+		}
+	}
+
+	/* point at display engine memory area (hash table, objects) */
+	nv_wr32(priv, 0x610010, (nv_gpuobj(base->ramht)->addr >> 8) | 9);
+
+	/* enable supervisor interrupts, disable everything else */
+	nv_wr32(priv, 0x61002c, 0x00000370);
+	nv_wr32(priv, 0x610028, 0x00000000);
+	return 0;
+}
+
+static int
+nv50_disp_base_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_base *base = (void *)object;
+
+	/* disable all interrupts */
+	nv_wr32(priv, 0x610024, 0x00000000);
+	nv_wr32(priv, 0x610020, 0x00000000);
+
+	return nouveau_parent_fini(&base->base, suspend);
+}
+
+struct nouveau_ofuncs
+nv50_disp_base_ofuncs = {
+	.ctor = nv50_disp_base_ctor,
+	.dtor = nv50_disp_base_dtor,
+	.init = nv50_disp_base_init,
+	.fini = nv50_disp_base_fini,
+};
+
+static struct nouveau_omthds
+nv50_disp_base_omthds[] = {
+	{ SOR_MTHD(NV50_DISP_SOR_PWR)         , nv50_sor_mthd },
+	{ SOR_MTHD(NV50_DISP_SOR_LVDS_SCRIPT) , nv50_sor_mthd },
+	{ DAC_MTHD(NV50_DISP_DAC_PWR)         , nv50_dac_mthd },
+	{ DAC_MTHD(NV50_DISP_DAC_LOAD)        , nv50_dac_mthd },
+	{},
+};
+
+static struct nouveau_oclass
+nv50_disp_base_oclass[] = {
+	{ NV50_DISP_CLASS, &nv50_disp_base_ofuncs, nv50_disp_base_omthds },
+	{}
+};
+
+static struct nouveau_oclass
+nv50_disp_sclass[] = {
+	{ NV50_DISP_MAST_CLASS, &nv50_disp_mast_ofuncs },
+	{ NV50_DISP_SYNC_CLASS, &nv50_disp_sync_ofuncs },
+	{ NV50_DISP_OVLY_CLASS, &nv50_disp_ovly_ofuncs },
+	{ NV50_DISP_OIMM_CLASS, &nv50_disp_oimm_ofuncs },
+	{ NV50_DISP_CURS_CLASS, &nv50_disp_curs_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * Display context, tracks instmem allocation and prevents more than one
+ * client using the display hardware at any time.
+ ******************************************************************************/
+
+static int
+nv50_disp_data_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv = (void *)engine;
+	struct nouveau_engctx *ectx;
+	int ret = -EBUSY;
+
+	/* no context needed for channel objects... */
+	if (nv_mclass(parent) != NV_DEVICE_CLASS) {
+		atomic_inc(&parent->refcount);
+		*pobject = parent;
+		return 0;
+	}
+
+	/* allocate display hardware to client */
+	mutex_lock(&nv_subdev(priv)->mutex);
+	if (list_empty(&nv_engine(priv)->contexts)) {
+		ret = nouveau_engctx_create(parent, engine, oclass, NULL,
+					    0x10000, 0x10000,
+					    NVOBJ_FLAG_HEAP, &ectx);
+		*pobject = nv_object(ectx);
+	}
+	mutex_unlock(&nv_subdev(priv)->mutex);
+	return ret;
+}
+
+struct nouveau_oclass
+nv50_disp_cclass = {
+	.handle = NV_ENGCTX(DISP, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_disp_data_ctor,
+		.dtor = _nouveau_engctx_dtor,
+		.init = _nouveau_engctx_init,
+		.fini = _nouveau_engctx_fini,
+		.rd32 = _nouveau_engctx_rd32,
+		.wr32 = _nouveau_engctx_wr32,
+	},
+};
+
+/*******************************************************************************
+ * Display engine implementation
+ ******************************************************************************/
+
+static void
+nv50_disp_intr_error(struct nv50_disp_priv *priv)
+{
+	u32 channels = (nv_rd32(priv, 0x610020) & 0x001f0000) >> 16;
+	u32 addr, data;
+	int chid;
+
+	for (chid = 0; chid < 5; chid++) {
+		if (!(channels & (1 << chid)))
+			continue;
+
+		nv_wr32(priv, 0x610020, 0x00010000 << chid);
+		addr = nv_rd32(priv, 0x610080 + (chid * 0x08));
+		data = nv_rd32(priv, 0x610084 + (chid * 0x08));
+		nv_wr32(priv, 0x610080 + (chid * 0x08), 0x90000000);
+
+		nv_error(priv, "chid %d mthd 0x%04x data 0x%08x 0x%08x\n",
+			 chid, addr & 0xffc, data, addr);
+	}
+}
+
+static void
+nv50_disp_intr_vblank(struct nv50_disp_priv *priv, int crtc)
+{
+	struct nouveau_bar *bar = nouveau_bar(priv);
+	struct nouveau_disp *disp = &priv->base;
+	struct nouveau_software_chan *chan, *temp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&disp->vblank.lock, flags);
+	list_for_each_entry_safe(chan, temp, &disp->vblank.list, vblank.head) {
+		if (chan->vblank.crtc != crtc)
+			continue;
+
+		if (nv_device(priv)->chipset >= 0xc0) {
+			nv_wr32(priv, 0x001718, 0x80000000 | chan->vblank.channel);
+			bar->flush(bar);
+			nv_wr32(priv, 0x06000c,
+				upper_32_bits(chan->vblank.offset));
+			nv_wr32(priv, 0x060010,
+				lower_32_bits(chan->vblank.offset));
+			nv_wr32(priv, 0x060014, chan->vblank.value);
+		} else {
+			nv_wr32(priv, 0x001704, chan->vblank.channel);
+			nv_wr32(priv, 0x001710, 0x80000000 | chan->vblank.ctxdma);
+			bar->flush(bar);
+			if (nv_device(priv)->chipset == 0x50) {
+				nv_wr32(priv, 0x001570, chan->vblank.offset);
+				nv_wr32(priv, 0x001574, chan->vblank.value);
+			} else {
+				nv_wr32(priv, 0x060010, chan->vblank.offset);
+				nv_wr32(priv, 0x060014, chan->vblank.value);
+			}
+		}
+
+		list_del(&chan->vblank.head);
+		if (disp->vblank.put)
+			disp->vblank.put(disp->vblank.data, crtc);
+	}
+	spin_unlock_irqrestore(&disp->vblank.lock, flags);
+
+	if (disp->vblank.notify)
+		disp->vblank.notify(disp->vblank.data, crtc);
+}
+
+static u16
+exec_lookup(struct nv50_disp_priv *priv, int head, int outp, u32 ctrl,
+	    struct dcb_output *dcb, u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+	    struct nvbios_outp *info)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	u16 mask, type, data;
+
+	if (outp < 4) {
+		type = DCB_OUTPUT_ANALOG;
+		mask = 0;
+	} else {
+		outp -= 4;
+		switch (ctrl & 0x00000f00) {
+		case 0x00000000: type = DCB_OUTPUT_LVDS; mask = 1; break;
+		case 0x00000100: type = DCB_OUTPUT_TMDS; mask = 1; break;
+		case 0x00000200: type = DCB_OUTPUT_TMDS; mask = 2; break;
+		case 0x00000500: type = DCB_OUTPUT_TMDS; mask = 3; break;
+		case 0x00000800: type = DCB_OUTPUT_DP; mask = 1; break;
+		case 0x00000900: type = DCB_OUTPUT_DP; mask = 2; break;
+		default:
+			nv_error(priv, "unknown SOR mc 0x%08x\n", ctrl);
+			return 0x0000;
+		}
+	}
+
+	mask  = 0x00c0 & (mask << 6);
+	mask |= 0x0001 << outp;
+	mask |= 0x0100 << head;
+
+	data = dcb_outp_match(bios, type, mask, ver, hdr, dcb);
+	if (!data)
+		return 0x0000;
+
+	return nvbios_outp_match(bios, type, mask, ver, hdr, cnt, len, info);
+}
+
+static bool
+exec_script(struct nv50_disp_priv *priv, int head, int id)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct nvbios_outp info;
+	struct dcb_output dcb;
+	u8  ver, hdr, cnt, len;
+	u16 data;
+	u32 ctrl = 0x00000000;
+	int i;
+
+	for (i = 0; !(ctrl & (1 << head)) && i < 3; i++)
+		ctrl = nv_rd32(priv, 0x610b5c + (i * 8));
+
+	if (nv_device(priv)->chipset  < 0x90 ||
+	    nv_device(priv)->chipset == 0x92 ||
+	    nv_device(priv)->chipset == 0xa0) {
+		for (i = 0; !(ctrl & (1 << head)) && i < 2; i++)
+			ctrl = nv_rd32(priv, 0x610b74 + (i * 8));
+		i += 3;
+	} else {
+		for (i = 0; !(ctrl & (1 << head)) && i < 4; i++)
+			ctrl = nv_rd32(priv, 0x610798 + (i * 8));
+		i += 3;
+	}
+
+	if (!(ctrl & (1 << head)))
+		return false;
+
+	data = exec_lookup(priv, head, i, ctrl, &dcb, &ver, &hdr, &cnt, &len, &info);
+	if (data) {
+		struct nvbios_init init = {
+			.subdev = nv_subdev(priv),
+			.bios = bios,
+			.offset = info.script[id],
+			.outp = &dcb,
+			.crtc = head,
+			.execute = 1,
+		};
+
+		return nvbios_exec(&init) == 0;
+	}
+
+	return false;
+}
+
+static u32
+exec_clkcmp(struct nv50_disp_priv *priv, int head, int id, u32 pclk,
+	    struct dcb_output *outp)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct nvbios_outp info1;
+	struct nvbios_ocfg info2;
+	u8  ver, hdr, cnt, len;
+	u16 data, conf;
+	u32 ctrl = 0x00000000;
+	int i;
+
+	for (i = 0; !(ctrl & (1 << head)) && i < 3; i++)
+		ctrl = nv_rd32(priv, 0x610b58 + (i * 8));
+
+	if (nv_device(priv)->chipset  < 0x90 ||
+	    nv_device(priv)->chipset == 0x92 ||
+	    nv_device(priv)->chipset == 0xa0) {
+		for (i = 0; !(ctrl & (1 << head)) && i < 2; i++)
+			ctrl = nv_rd32(priv, 0x610b70 + (i * 8));
+		i += 3;
+	} else {
+		for (i = 0; !(ctrl & (1 << head)) && i < 4; i++)
+			ctrl = nv_rd32(priv, 0x610794 + (i * 8));
+		i += 3;
+	}
+
+	if (!(ctrl & (1 << head)))
+		return 0x0000;
+
+	data = exec_lookup(priv, head, i, ctrl, outp, &ver, &hdr, &cnt, &len, &info1);
+	if (!data)
+		return 0x0000;
+
+	switch (outp->type) {
+	case DCB_OUTPUT_TMDS:
+		conf = (ctrl & 0x00000f00) >> 8;
+		if (pclk >= 165000)
+			conf |= 0x0100;
+		break;
+	case DCB_OUTPUT_LVDS:
+		conf = priv->sor.lvdsconf;
+		break;
+	case DCB_OUTPUT_DP:
+		conf = (ctrl & 0x00000f00) >> 8;
+		break;
+	case DCB_OUTPUT_ANALOG:
+	default:
+		conf = 0x00ff;
+		break;
+	}
+
+	data = nvbios_ocfg_match(bios, data, conf, &ver, &hdr, &cnt, &len, &info2);
+	if (data) {
+		data = nvbios_oclk_match(bios, info2.clkcmp[id], pclk);
+		if (data) {
+			struct nvbios_init init = {
+				.subdev = nv_subdev(priv),
+				.bios = bios,
+				.offset = data,
+				.outp = outp,
+				.crtc = head,
+				.execute = 1,
+			};
+
+			if (nvbios_exec(&init))
+				return 0x0000;
+			return conf;
+		}
+	}
+
+	return 0x0000;
+}
+
+static void
+nv50_disp_intr_unk10(struct nv50_disp_priv *priv, u32 super)
+{
+	int head = ffs((super & 0x00000060) >> 5) - 1;
+	if (head >= 0) {
+		head = ffs((super & 0x00000180) >> 7) - 1;
+		if (head >= 0)
+			exec_script(priv, head, 1);
+	}
+
+	nv_wr32(priv, 0x610024, 0x00000010);
+	nv_wr32(priv, 0x610030, 0x80000000);
+}
+
+static void
+nv50_disp_intr_unk20_dp(struct nv50_disp_priv *priv,
+		        struct dcb_output *outp, u32 pclk)
+{
+	const int link = !(outp->sorconf.link & 1);
+	const int   or = ffs(outp->or) - 1;
+	const u32 soff = (  or * 0x800);
+	const u32 loff = (link * 0x080) + soff;
+	const u32 ctrl = nv_rd32(priv, 0x610794 + (or * 8));
+	const u32 symbol = 100000;
+	u32 dpctrl = nv_rd32(priv, 0x61c10c + loff) & 0x0000f0000;
+	u32 clksor = nv_rd32(priv, 0x614300 + soff);
+	int bestTU = 0, bestVTUi = 0, bestVTUf = 0, bestVTUa = 0;
+	int TU, VTUi, VTUf, VTUa;
+	u64 link_data_rate, link_ratio, unk;
+	u32 best_diff = 64 * symbol;
+	u32 link_nr, link_bw, bits, r;
+
+	/* calculate packed data rate for each lane */
+	if      (dpctrl > 0x00030000) link_nr = 4;
+	else if (dpctrl > 0x00010000) link_nr = 2;
+	else			      link_nr = 1;
+
+	if (clksor & 0x000c0000)
+		link_bw = 270000;
+	else
+		link_bw = 162000;
+
+	if      ((ctrl & 0xf0000) == 0x60000) bits = 30;
+	else if ((ctrl & 0xf0000) == 0x50000) bits = 24;
+	else                                  bits = 18;
+
+	link_data_rate = (pclk * bits / 8) / link_nr;
+
+	/* calculate ratio of packed data rate to link symbol rate */
+	link_ratio = link_data_rate * symbol;
+	r = do_div(link_ratio, link_bw);
+
+	for (TU = 64; TU >= 32; TU--) {
+		/* calculate average number of valid symbols in each TU */
+		u32 tu_valid = link_ratio * TU;
+		u32 calc, diff;
+
+		/* find a hw representation for the fraction.. */
+		VTUi = tu_valid / symbol;
+		calc = VTUi * symbol;
+		diff = tu_valid - calc;
+		if (diff) {
+			if (diff >= (symbol / 2)) {
+				VTUf = symbol / (symbol - diff);
+				if (symbol - (VTUf * diff))
+					VTUf++;
+
+				if (VTUf <= 15) {
+					VTUa  = 1;
+					calc += symbol - (symbol / VTUf);
+				} else {
+					VTUa  = 0;
+					VTUf  = 1;
+					calc += symbol;
+				}
+			} else {
+				VTUa  = 0;
+				VTUf  = min((int)(symbol / diff), 15);
+				calc += symbol / VTUf;
+			}
+
+			diff = calc - tu_valid;
+		} else {
+			/* no remainder, but the hw doesn't like the fractional
+			 * part to be zero.  decrement the integer part and
+			 * have the fraction add a whole symbol back
+			 */
+			VTUa = 0;
+			VTUf = 1;
+			VTUi--;
+		}
+
+		if (diff < best_diff) {
+			best_diff = diff;
+			bestTU = TU;
+			bestVTUa = VTUa;
+			bestVTUf = VTUf;
+			bestVTUi = VTUi;
+			if (diff == 0)
+				break;
+		}
+	}
+
+	if (!bestTU) {
+		nv_error(priv, "unable to find suitable dp config\n");
+		return;
+	}
+
+	/* XXX close to vbios numbers, but not right */
+	unk  = (symbol - link_ratio) * bestTU;
+	unk *= link_ratio;
+	r = do_div(unk, symbol);
+	r = do_div(unk, symbol);
+	unk += 6;
+
+	nv_mask(priv, 0x61c10c + loff, 0x000001fc, bestTU << 2);
+	nv_mask(priv, 0x61c128 + loff, 0x010f7f3f, bestVTUa << 24 |
+						   bestVTUf << 16 |
+						   bestVTUi << 8 | unk);
+}
+
+static void
+nv50_disp_intr_unk20(struct nv50_disp_priv *priv, u32 super)
+{
+	struct dcb_output outp;
+	u32 addr, mask, data;
+	int head;
+
+	/* finish detaching encoder? */
+	head = ffs((super & 0x00000180) >> 7) - 1;
+	if (head >= 0)
+		exec_script(priv, head, 2);
+
+	/* check whether a vpll change is required */
+	head = ffs((super & 0x00000600) >> 9) - 1;
+	if (head >= 0) {
+		u32 pclk = nv_rd32(priv, 0x610ad0 + (head * 0x540)) & 0x3fffff;
+		if (pclk) {
+			struct nouveau_clock *clk = nouveau_clock(priv);
+			clk->pll_set(clk, PLL_VPLL0 + head, pclk);
+		}
+
+		nv_mask(priv, 0x614200 + head * 0x800, 0x0000000f, 0x00000000);
+	}
+
+	/* (re)attach the relevant OR to the head */
+	head = ffs((super & 0x00000180) >> 7) - 1;
+	if (head >= 0) {
+		u32 pclk = nv_rd32(priv, 0x610ad0 + (head * 0x540)) & 0x3fffff;
+		u32 conf = exec_clkcmp(priv, head, 0, pclk, &outp);
+		if (conf) {
+			if (outp.type == DCB_OUTPUT_ANALOG) {
+				addr = 0x614280 + (ffs(outp.or) - 1) * 0x800;
+				mask = 0xffffffff;
+				data = 0x00000000;
+			} else {
+				if (outp.type == DCB_OUTPUT_DP)
+					nv50_disp_intr_unk20_dp(priv, &outp, pclk);
+				addr = 0x614300 + (ffs(outp.or) - 1) * 0x800;
+				mask = 0x00000707;
+				data = (conf & 0x0100) ? 0x0101 : 0x0000;
+			}
+
+			nv_mask(priv, addr, mask, data);
+		}
+	}
+
+	nv_wr32(priv, 0x610024, 0x00000020);
+	nv_wr32(priv, 0x610030, 0x80000000);
+}
+
+/* If programming a TMDS output on a SOR that can also be configured for
+ * DisplayPort, make sure NV50_SOR_DP_CTRL_ENABLE is forced off.
+ *
+ * It looks like the VBIOS TMDS scripts make an attempt at this, however,
+ * the VBIOS scripts on at least one board I have only switch it off on
+ * link 0, causing a blank display if the output has previously been
+ * programmed for DisplayPort.
+ */
+static void
+nv50_disp_intr_unk40_tmds(struct nv50_disp_priv *priv, struct dcb_output *outp)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	const int link = !(outp->sorconf.link & 1);
+	const int   or = ffs(outp->or) - 1;
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	const u16 mask = (outp->sorconf.link << 6) | outp->or;
+	u8  ver, hdr;
+
+	if (dcb_outp_match(bios, DCB_OUTPUT_DP, mask, &ver, &hdr, outp))
+		nv_mask(priv, 0x61c10c + loff, 0x00000001, 0x00000000);
+}
+
+static void
+nv50_disp_intr_unk40(struct nv50_disp_priv *priv, u32 super)
+{
+	int head = ffs((super & 0x00000180) >> 7) - 1;
+	if (head >= 0) {
+		struct dcb_output outp;
+		u32 pclk = nv_rd32(priv, 0x610ad0 + (head * 0x540)) & 0x3fffff;
+		if (pclk && exec_clkcmp(priv, head, 1, pclk, &outp)) {
+			if (outp.type == DCB_OUTPUT_TMDS)
+				nv50_disp_intr_unk40_tmds(priv, &outp);
+		}
+	}
+
+	nv_wr32(priv, 0x610024, 0x00000040);
+	nv_wr32(priv, 0x610030, 0x80000000);
+}
+
+static void
+nv50_disp_intr_super(struct nv50_disp_priv *priv, u32 intr1)
+{
+	u32 super = nv_rd32(priv, 0x610030);
+
+	nv_debug(priv, "supervisor 0x%08x 0x%08x\n", intr1, super);
+
+	if (intr1 & 0x00000010)
+		nv50_disp_intr_unk10(priv, super);
+	if (intr1 & 0x00000020)
+		nv50_disp_intr_unk20(priv, super);
+	if (intr1 & 0x00000040)
+		nv50_disp_intr_unk40(priv, super);
+}
+
+void
+nv50_disp_intr(struct nouveau_subdev *subdev)
+{
+	struct nv50_disp_priv *priv = (void *)subdev;
+	u32 intr0 = nv_rd32(priv, 0x610020);
+	u32 intr1 = nv_rd32(priv, 0x610024);
+
+	if (intr0 & 0x001f0000) {
+		nv50_disp_intr_error(priv);
+		intr0 &= ~0x001f0000;
+	}
+
+	if (intr1 & 0x00000004) {
+		nv50_disp_intr_vblank(priv, 0);
+		nv_wr32(priv, 0x610024, 0x00000004);
+		intr1 &= ~0x00000004;
+	}
+
+	if (intr1 & 0x00000008) {
+		nv50_disp_intr_vblank(priv, 1);
+		nv_wr32(priv, 0x610024, 0x00000008);
+		intr1 &= ~0x00000008;
+	}
+
+	if (intr1 & 0x00000070) {
+		nv50_disp_intr_super(priv, intr1);
+		intr1 &= ~0x00000070;
+	}
+}
+
+static int
+nv50_disp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv;
+	int ret;
+
+	ret = nouveau_disp_create(parent, engine, oclass, "PDISP",
+				  "display", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nv50_disp_base_oclass;
+	nv_engine(priv)->cclass = &nv50_disp_cclass;
+	nv_subdev(priv)->intr = nv50_disp_intr;
+	priv->sclass = nv50_disp_sclass;
+	priv->head.nr = 2;
+	priv->dac.nr = 3;
+	priv->sor.nr = 2;
+	priv->dac.power = nv50_dac_power;
+	priv->dac.sense = nv50_dac_sense;
+	priv->sor.power = nv50_sor_power;
+
+	INIT_LIST_HEAD(&priv->base.vblank.list);
+	spin_lock_init(&priv->base.vblank.lock);
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_disp_oclass = {
+	.handle = NV_ENGINE(DISP, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_disp_ctor,
+		.dtor = _nouveau_disp_dtor,
+		.init = _nouveau_disp_init,
+		.fini = _nouveau_disp_fini,
+	},
+};
diff --git a/nvkm/engine/disp/nv50.h b/nvkm/engine/disp/nv50.h
new file mode 100644
index 000000000..a6bb93145
--- /dev/null
+++ b/nvkm/engine/disp/nv50.h
@@ -0,0 +1,142 @@
+#ifndef __NV50_DISP_H__
+#define __NV50_DISP_H__
+
+#include <core/parent.h>
+#include <core/namedb.h>
+#include <core/ramht.h>
+
+#include <engine/dmaobj.h>
+#include <engine/disp.h>
+
+struct dcb_output;
+
+struct nv50_disp_priv {
+	struct nouveau_disp base;
+	struct nouveau_oclass *sclass;
+	struct {
+		int nr;
+	} head;
+	struct {
+		int nr;
+		int (*power)(struct nv50_disp_priv *, int dac, u32 data);
+		int (*sense)(struct nv50_disp_priv *, int dac, u32 load);
+	} dac;
+	struct {
+		int nr;
+		int (*power)(struct nv50_disp_priv *, int sor, u32 data);
+		int (*hda_eld)(struct nv50_disp_priv *, int sor, u8 *, u32);
+		int (*hdmi)(struct nv50_disp_priv *, int head, int sor, u32);
+		int (*dp_train_init)(struct nv50_disp_priv *, int sor, int link,
+				     int head, u16 type, u16 mask, u32 data,
+				     struct dcb_output *);
+		int (*dp_train_fini)(struct nv50_disp_priv *, int sor, int link,
+				     int head, u16 type, u16 mask, u32 data,
+				     struct dcb_output *);
+		int (*dp_train)(struct nv50_disp_priv *, int sor, int link,
+				u16 type, u16 mask, u32 data,
+				struct dcb_output *);
+		int (*dp_lnkctl)(struct nv50_disp_priv *, int sor, int link,
+				 int head, u16 type, u16 mask, u32 data,
+				 struct dcb_output *);
+		int (*dp_drvctl)(struct nv50_disp_priv *, int sor, int link,
+				 int lane, u16 type, u16 mask, u32 data,
+				 struct dcb_output *);
+		u32 lvdsconf;
+	} sor;
+};
+
+#define DAC_MTHD(n) (n), (n) + 0x03
+
+int nv50_dac_mthd(struct nouveau_object *, u32, void *, u32);
+int nv50_dac_power(struct nv50_disp_priv *, int, u32);
+int nv50_dac_sense(struct nv50_disp_priv *, int, u32);
+
+#define SOR_MTHD(n) (n), (n) + 0x3f
+
+int nva3_hda_eld(struct nv50_disp_priv *, int, u8 *, u32);
+int nvd0_hda_eld(struct nv50_disp_priv *, int, u8 *, u32);
+
+int nv84_hdmi_ctrl(struct nv50_disp_priv *, int, int, u32);
+int nva3_hdmi_ctrl(struct nv50_disp_priv *, int, int, u32);
+int nvd0_hdmi_ctrl(struct nv50_disp_priv *, int, int, u32);
+
+int nv50_sor_mthd(struct nouveau_object *, u32, void *, u32);
+int nv50_sor_power(struct nv50_disp_priv *, int, u32);
+
+int nv94_sor_dp_train_init(struct nv50_disp_priv *, int, int, int, u16, u16,
+		           u32, struct dcb_output *);
+int nv94_sor_dp_train_fini(struct nv50_disp_priv *, int, int, int, u16, u16,
+		           u32, struct dcb_output *);
+int nv94_sor_dp_train(struct nv50_disp_priv *, int, int, u16, u16, u32,
+		      struct dcb_output *);
+int nv94_sor_dp_lnkctl(struct nv50_disp_priv *, int, int, int, u16, u16, u32,
+		       struct dcb_output *);
+int nv94_sor_dp_drvctl(struct nv50_disp_priv *, int, int, int, u16, u16, u32,
+		       struct dcb_output *);
+
+int nvd0_sor_dp_train(struct nv50_disp_priv *, int, int, u16, u16, u32,
+		      struct dcb_output *);
+int nvd0_sor_dp_lnkctl(struct nv50_disp_priv *, int, int, int, u16, u16, u32,
+		       struct dcb_output *);
+int nvd0_sor_dp_drvctl(struct nv50_disp_priv *, int, int, int, u16, u16, u32,
+		       struct dcb_output *);
+
+struct nv50_disp_base {
+	struct nouveau_parent base;
+	struct nouveau_ramht *ramht;
+	u32 chan;
+};
+
+struct nv50_disp_chan {
+	struct nouveau_namedb base;
+	int chid;
+};
+
+int  nv50_disp_chan_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, int, int, void **);
+void nv50_disp_chan_destroy(struct nv50_disp_chan *);
+u32  nv50_disp_chan_rd32(struct nouveau_object *, u64);
+void nv50_disp_chan_wr32(struct nouveau_object *, u64, u32);
+
+#define nv50_disp_chan_init(a)                                                 \
+	nouveau_namedb_init(&(a)->base)
+#define nv50_disp_chan_fini(a,b)                                               \
+	nouveau_namedb_fini(&(a)->base, (b))
+
+int  nv50_disp_dmac_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, u32, int, int, void **);
+void nv50_disp_dmac_dtor(struct nouveau_object *);
+
+struct nv50_disp_dmac {
+	struct nv50_disp_chan base;
+	struct nouveau_dmaobj *pushdma;
+	u32 push;
+};
+
+struct nv50_disp_pioc {
+	struct nv50_disp_chan base;
+};
+
+extern struct nouveau_ofuncs nv50_disp_mast_ofuncs;
+extern struct nouveau_ofuncs nv50_disp_sync_ofuncs;
+extern struct nouveau_ofuncs nv50_disp_ovly_ofuncs;
+extern struct nouveau_ofuncs nv50_disp_oimm_ofuncs;
+extern struct nouveau_ofuncs nv50_disp_curs_ofuncs;
+extern struct nouveau_ofuncs nv50_disp_base_ofuncs;
+extern struct nouveau_oclass nv50_disp_cclass;
+void nv50_disp_intr(struct nouveau_subdev *);
+
+extern struct nouveau_omthds nv84_disp_base_omthds[];
+
+extern struct nouveau_omthds nva3_disp_base_omthds[];
+
+extern struct nouveau_ofuncs nvd0_disp_mast_ofuncs;
+extern struct nouveau_ofuncs nvd0_disp_sync_ofuncs;
+extern struct nouveau_ofuncs nvd0_disp_ovly_ofuncs;
+extern struct nouveau_ofuncs nvd0_disp_oimm_ofuncs;
+extern struct nouveau_ofuncs nvd0_disp_curs_ofuncs;
+extern struct nouveau_ofuncs nvd0_disp_base_ofuncs;
+extern struct nouveau_oclass nvd0_disp_cclass;
+void nvd0_disp_intr(struct nouveau_subdev *);
+
+#endif
diff --git a/nvkm/engine/disp/nv84.c b/nvkm/engine/disp/nv84.c
new file mode 100644
index 000000000..fc84eacdf
--- /dev/null
+++ b/nvkm/engine/disp/nv84.c
@@ -0,0 +1,98 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <engine/software.h>
+#include <engine/disp.h>
+
+#include <core/class.h>
+
+#include "nv50.h"
+
+static struct nouveau_oclass
+nv84_disp_sclass[] = {
+	{ NV84_DISP_MAST_CLASS, &nv50_disp_mast_ofuncs },
+	{ NV84_DISP_SYNC_CLASS, &nv50_disp_sync_ofuncs },
+	{ NV84_DISP_OVLY_CLASS, &nv50_disp_ovly_ofuncs },
+	{ NV84_DISP_OIMM_CLASS, &nv50_disp_oimm_ofuncs },
+	{ NV84_DISP_CURS_CLASS, &nv50_disp_curs_ofuncs },
+	{}
+};
+
+struct nouveau_omthds
+nv84_disp_base_omthds[] = {
+	{ SOR_MTHD(NV50_DISP_SOR_PWR)         , nv50_sor_mthd },
+	{ SOR_MTHD(NV84_DISP_SOR_HDMI_PWR)    , nv50_sor_mthd },
+	{ SOR_MTHD(NV50_DISP_SOR_LVDS_SCRIPT) , nv50_sor_mthd },
+	{ DAC_MTHD(NV50_DISP_DAC_PWR)         , nv50_dac_mthd },
+	{ DAC_MTHD(NV50_DISP_DAC_LOAD)        , nv50_dac_mthd },
+	{},
+};
+
+static struct nouveau_oclass
+nv84_disp_base_oclass[] = {
+	{ NV84_DISP_CLASS, &nv50_disp_base_ofuncs, nv84_disp_base_omthds },
+	{}
+};
+
+static int
+nv84_disp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv;
+	int ret;
+
+	ret = nouveau_disp_create(parent, engine, oclass, "PDISP",
+				  "display", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nv84_disp_base_oclass;
+	nv_engine(priv)->cclass = &nv50_disp_cclass;
+	nv_subdev(priv)->intr = nv50_disp_intr;
+	priv->sclass = nv84_disp_sclass;
+	priv->head.nr = 2;
+	priv->dac.nr = 3;
+	priv->sor.nr = 2;
+	priv->dac.power = nv50_dac_power;
+	priv->dac.sense = nv50_dac_sense;
+	priv->sor.power = nv50_sor_power;
+	priv->sor.hdmi = nv84_hdmi_ctrl;
+
+	INIT_LIST_HEAD(&priv->base.vblank.list);
+	spin_lock_init(&priv->base.vblank.lock);
+	return 0;
+}
+
+struct nouveau_oclass
+nv84_disp_oclass = {
+	.handle = NV_ENGINE(DISP, 0x82),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv84_disp_ctor,
+		.dtor = _nouveau_disp_dtor,
+		.init = _nouveau_disp_init,
+		.fini = _nouveau_disp_fini,
+	},
+};
diff --git a/nvkm/engine/disp/nv94.c b/nvkm/engine/disp/nv94.c
new file mode 100644
index 000000000..ba9dfd466
--- /dev/null
+++ b/nvkm/engine/disp/nv94.c
@@ -0,0 +1,109 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <engine/software.h>
+#include <engine/disp.h>
+
+#include <core/class.h>
+
+#include "nv50.h"
+
+static struct nouveau_oclass
+nv94_disp_sclass[] = {
+	{ NV94_DISP_MAST_CLASS, &nv50_disp_mast_ofuncs },
+	{ NV94_DISP_SYNC_CLASS, &nv50_disp_sync_ofuncs },
+	{ NV94_DISP_OVLY_CLASS, &nv50_disp_ovly_ofuncs },
+	{ NV94_DISP_OIMM_CLASS, &nv50_disp_oimm_ofuncs },
+	{ NV94_DISP_CURS_CLASS, &nv50_disp_curs_ofuncs },
+	{}
+};
+
+static struct nouveau_omthds
+nv94_disp_base_omthds[] = {
+	{ SOR_MTHD(NV50_DISP_SOR_PWR)         , nv50_sor_mthd },
+	{ SOR_MTHD(NV84_DISP_SOR_HDMI_PWR)    , nv50_sor_mthd },
+	{ SOR_MTHD(NV50_DISP_SOR_LVDS_SCRIPT) , nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_TRAIN)    , nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_LNKCTL)   , nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_DRVCTL(0)), nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_DRVCTL(1)), nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_DRVCTL(2)), nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_DRVCTL(3)), nv50_sor_mthd },
+	{ DAC_MTHD(NV50_DISP_DAC_PWR)         , nv50_dac_mthd },
+	{ DAC_MTHD(NV50_DISP_DAC_LOAD)        , nv50_dac_mthd },
+	{},
+};
+
+static struct nouveau_oclass
+nv94_disp_base_oclass[] = {
+	{ NV94_DISP_CLASS, &nv50_disp_base_ofuncs, nv94_disp_base_omthds },
+	{}
+};
+
+static int
+nv94_disp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv;
+	int ret;
+
+	ret = nouveau_disp_create(parent, engine, oclass, "PDISP",
+				  "display", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nv94_disp_base_oclass;
+	nv_engine(priv)->cclass = &nv50_disp_cclass;
+	nv_subdev(priv)->intr = nv50_disp_intr;
+	priv->sclass = nv94_disp_sclass;
+	priv->head.nr = 2;
+	priv->dac.nr = 3;
+	priv->sor.nr = 4;
+	priv->dac.power = nv50_dac_power;
+	priv->dac.sense = nv50_dac_sense;
+	priv->sor.power = nv50_sor_power;
+	priv->sor.hdmi = nv84_hdmi_ctrl;
+	priv->sor.dp_train = nv94_sor_dp_train;
+	priv->sor.dp_train_init = nv94_sor_dp_train_init;
+	priv->sor.dp_train_fini = nv94_sor_dp_train_fini;
+	priv->sor.dp_lnkctl = nv94_sor_dp_lnkctl;
+	priv->sor.dp_drvctl = nv94_sor_dp_drvctl;
+
+	INIT_LIST_HEAD(&priv->base.vblank.list);
+	spin_lock_init(&priv->base.vblank.lock);
+	return 0;
+}
+
+struct nouveau_oclass
+nv94_disp_oclass = {
+	.handle = NV_ENGINE(DISP, 0x88),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv94_disp_ctor,
+		.dtor = _nouveau_disp_dtor,
+		.init = _nouveau_disp_init,
+		.fini = _nouveau_disp_fini,
+	},
+};
diff --git a/nvkm/engine/disp/nva0.c b/nvkm/engine/disp/nva0.c
new file mode 100644
index 000000000..5d63902cd
--- /dev/null
+++ b/nvkm/engine/disp/nva0.c
@@ -0,0 +1,88 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <engine/software.h>
+#include <engine/disp.h>
+
+#include <core/class.h>
+
+#include "nv50.h"
+
+static struct nouveau_oclass
+nva0_disp_sclass[] = {
+	{ NVA0_DISP_MAST_CLASS, &nv50_disp_mast_ofuncs },
+	{ NVA0_DISP_SYNC_CLASS, &nv50_disp_sync_ofuncs },
+	{ NVA0_DISP_OVLY_CLASS, &nv50_disp_ovly_ofuncs },
+	{ NVA0_DISP_OIMM_CLASS, &nv50_disp_oimm_ofuncs },
+	{ NVA0_DISP_CURS_CLASS, &nv50_disp_curs_ofuncs },
+	{}
+};
+
+static struct nouveau_oclass
+nva0_disp_base_oclass[] = {
+	{ NVA0_DISP_CLASS, &nv50_disp_base_ofuncs, nv84_disp_base_omthds },
+	{}
+};
+
+static int
+nva0_disp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv;
+	int ret;
+
+	ret = nouveau_disp_create(parent, engine, oclass, "PDISP",
+				  "display", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nva0_disp_base_oclass;
+	nv_engine(priv)->cclass = &nv50_disp_cclass;
+	nv_subdev(priv)->intr = nv50_disp_intr;
+	priv->sclass = nva0_disp_sclass;
+	priv->head.nr = 2;
+	priv->dac.nr = 3;
+	priv->sor.nr = 2;
+	priv->dac.power = nv50_dac_power;
+	priv->dac.sense = nv50_dac_sense;
+	priv->sor.power = nv50_sor_power;
+	priv->sor.hdmi = nv84_hdmi_ctrl;
+
+	INIT_LIST_HEAD(&priv->base.vblank.list);
+	spin_lock_init(&priv->base.vblank.lock);
+	return 0;
+}
+
+struct nouveau_oclass
+nva0_disp_oclass = {
+	.handle = NV_ENGINE(DISP, 0x83),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nva0_disp_ctor,
+		.dtor = _nouveau_disp_dtor,
+		.init = _nouveau_disp_init,
+		.fini = _nouveau_disp_fini,
+	},
+};
diff --git a/nvkm/engine/disp/nva3.c b/nvkm/engine/disp/nva3.c
new file mode 100644
index 000000000..e9192ca38
--- /dev/null
+++ b/nvkm/engine/disp/nva3.c
@@ -0,0 +1,111 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <engine/software.h>
+#include <engine/disp.h>
+
+#include <core/class.h>
+
+#include "nv50.h"
+
+static struct nouveau_oclass
+nva3_disp_sclass[] = {
+	{ NVA3_DISP_MAST_CLASS, &nv50_disp_mast_ofuncs },
+	{ NVA3_DISP_SYNC_CLASS, &nv50_disp_sync_ofuncs },
+	{ NVA3_DISP_OVLY_CLASS, &nv50_disp_ovly_ofuncs },
+	{ NVA3_DISP_OIMM_CLASS, &nv50_disp_oimm_ofuncs },
+	{ NVA3_DISP_CURS_CLASS, &nv50_disp_curs_ofuncs },
+	{}
+};
+
+struct nouveau_omthds
+nva3_disp_base_omthds[] = {
+	{ SOR_MTHD(NV50_DISP_SOR_PWR)         , nv50_sor_mthd },
+	{ SOR_MTHD(NVA3_DISP_SOR_HDA_ELD)     , nv50_sor_mthd },
+	{ SOR_MTHD(NV84_DISP_SOR_HDMI_PWR)    , nv50_sor_mthd },
+	{ SOR_MTHD(NV50_DISP_SOR_LVDS_SCRIPT) , nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_TRAIN)    , nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_LNKCTL)   , nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_DRVCTL(0)), nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_DRVCTL(1)), nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_DRVCTL(2)), nv50_sor_mthd },
+	{ SOR_MTHD(NV94_DISP_SOR_DP_DRVCTL(3)), nv50_sor_mthd },
+	{ DAC_MTHD(NV50_DISP_DAC_PWR)         , nv50_dac_mthd },
+	{ DAC_MTHD(NV50_DISP_DAC_LOAD)        , nv50_dac_mthd },
+	{},
+};
+
+static struct nouveau_oclass
+nva3_disp_base_oclass[] = {
+	{ NVA3_DISP_CLASS, &nv50_disp_base_ofuncs, nva3_disp_base_omthds },
+	{}
+};
+
+static int
+nva3_disp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv;
+	int ret;
+
+	ret = nouveau_disp_create(parent, engine, oclass, "PDISP",
+				  "display", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nva3_disp_base_oclass;
+	nv_engine(priv)->cclass = &nv50_disp_cclass;
+	nv_subdev(priv)->intr = nv50_disp_intr;
+	priv->sclass = nva3_disp_sclass;
+	priv->head.nr = 2;
+	priv->dac.nr = 3;
+	priv->sor.nr = 4;
+	priv->dac.power = nv50_dac_power;
+	priv->dac.sense = nv50_dac_sense;
+	priv->sor.power = nv50_sor_power;
+	priv->sor.hda_eld = nva3_hda_eld;
+	priv->sor.hdmi = nva3_hdmi_ctrl;
+	priv->sor.dp_train = nv94_sor_dp_train;
+	priv->sor.dp_train_init = nv94_sor_dp_train_init;
+	priv->sor.dp_train_fini = nv94_sor_dp_train_fini;
+	priv->sor.dp_lnkctl = nv94_sor_dp_lnkctl;
+	priv->sor.dp_drvctl = nv94_sor_dp_drvctl;
+
+	INIT_LIST_HEAD(&priv->base.vblank.list);
+	spin_lock_init(&priv->base.vblank.lock);
+	return 0;
+}
+
+struct nouveau_oclass
+nva3_disp_oclass = {
+	.handle = NV_ENGINE(DISP, 0x85),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nva3_disp_ctor,
+		.dtor = _nouveau_disp_dtor,
+		.init = _nouveau_disp_init,
+		.fini = _nouveau_disp_fini,
+	},
+};
diff --git a/nvkm/engine/disp/nvd0.c b/nvkm/engine/disp/nvd0.c
new file mode 100644
index 000000000..9e38ebff5
--- /dev/null
+++ b/nvkm/engine/disp/nvd0.c
@@ -0,0 +1,976 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/parent.h>
+#include <core/handle.h>
+#include <core/class.h>
+
+#include <engine/software.h>
+#include <engine/disp.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/bar.h>
+#include <subdev/clock.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/disp.h>
+#include <subdev/bios/init.h>
+#include <subdev/bios/pll.h>
+
+#include "nv50.h"
+
+/*******************************************************************************
+ * EVO DMA channel base class
+ ******************************************************************************/
+
+static int
+nvd0_disp_dmac_object_attach(struct nouveau_object *parent,
+			     struct nouveau_object *object, u32 name)
+{
+	struct nv50_disp_base *base = (void *)parent->parent;
+	struct nv50_disp_chan *chan = (void *)parent;
+	u32 addr = nv_gpuobj(object)->node->offset;
+	u32 data = (chan->chid << 27) | (addr << 9) | 0x00000001;
+	return nouveau_ramht_insert(base->ramht, chan->chid, name, data);
+}
+
+static void
+nvd0_disp_dmac_object_detach(struct nouveau_object *parent, int cookie)
+{
+	struct nv50_disp_base *base = (void *)parent->parent;
+	nouveau_ramht_remove(base->ramht, cookie);
+}
+
+static int
+nvd0_disp_dmac_init(struct nouveau_object *object)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_dmac *dmac = (void *)object;
+	int chid = dmac->base.chid;
+	int ret;
+
+	ret = nv50_disp_chan_init(&dmac->base);
+	if (ret)
+		return ret;
+
+	/* enable error reporting */
+	nv_mask(priv, 0x610090, 0x00000001 << chid, 0x00000001 << chid);
+	nv_mask(priv, 0x6100a0, 0x00000001 << chid, 0x00000001 << chid);
+
+	/* initialise channel for dma command submission */
+	nv_wr32(priv, 0x610494 + (chid * 0x0010), dmac->push);
+	nv_wr32(priv, 0x610498 + (chid * 0x0010), 0x00010000);
+	nv_wr32(priv, 0x61049c + (chid * 0x0010), 0x00000001);
+	nv_mask(priv, 0x610490 + (chid * 0x0010), 0x00000010, 0x00000010);
+	nv_wr32(priv, 0x640000 + (chid * 0x1000), 0x00000000);
+	nv_wr32(priv, 0x610490 + (chid * 0x0010), 0x00000013);
+
+	/* wait for it to go inactive */
+	if (!nv_wait(priv, 0x610490 + (chid * 0x10), 0x80000000, 0x00000000)) {
+		nv_error(dmac, "init: 0x%08x\n",
+			 nv_rd32(priv, 0x610490 + (chid * 0x10)));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+static int
+nvd0_disp_dmac_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_dmac *dmac = (void *)object;
+	int chid = dmac->base.chid;
+
+	/* deactivate channel */
+	nv_mask(priv, 0x610490 + (chid * 0x0010), 0x00001010, 0x00001000);
+	nv_mask(priv, 0x610490 + (chid * 0x0010), 0x00000003, 0x00000000);
+	if (!nv_wait(priv, 0x610490 + (chid * 0x10), 0x001e0000, 0x00000000)) {
+		nv_error(dmac, "fini: 0x%08x\n",
+			 nv_rd32(priv, 0x610490 + (chid * 0x10)));
+		if (suspend)
+			return -EBUSY;
+	}
+
+	/* disable error reporting */
+	nv_mask(priv, 0x610090, 0x00000001 << chid, 0x00000000);
+	nv_mask(priv, 0x6100a0, 0x00000001 << chid, 0x00000000);
+
+	return nv50_disp_chan_fini(&dmac->base, suspend);
+}
+
+/*******************************************************************************
+ * EVO master channel object
+ ******************************************************************************/
+
+static int
+nvd0_disp_mast_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_mast_class *args = data;
+	struct nv50_disp_dmac *mast;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nv50_disp_dmac_create_(parent, engine, oclass, args->pushbuf,
+				     0, sizeof(*mast), (void **)&mast);
+	*pobject = nv_object(mast);
+	if (ret)
+		return ret;
+
+	nv_parent(mast)->object_attach = nvd0_disp_dmac_object_attach;
+	nv_parent(mast)->object_detach = nvd0_disp_dmac_object_detach;
+	return 0;
+}
+
+static int
+nvd0_disp_mast_init(struct nouveau_object *object)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_dmac *mast = (void *)object;
+	int ret;
+
+	ret = nv50_disp_chan_init(&mast->base);
+	if (ret)
+		return ret;
+
+	/* enable error reporting */
+	nv_mask(priv, 0x610090, 0x00000001, 0x00000001);
+	nv_mask(priv, 0x6100a0, 0x00000001, 0x00000001);
+
+	/* initialise channel for dma command submission */
+	nv_wr32(priv, 0x610494, mast->push);
+	nv_wr32(priv, 0x610498, 0x00010000);
+	nv_wr32(priv, 0x61049c, 0x00000001);
+	nv_mask(priv, 0x610490, 0x00000010, 0x00000010);
+	nv_wr32(priv, 0x640000, 0x00000000);
+	nv_wr32(priv, 0x610490, 0x01000013);
+
+	/* wait for it to go inactive */
+	if (!nv_wait(priv, 0x610490, 0x80000000, 0x00000000)) {
+		nv_error(mast, "init: 0x%08x\n", nv_rd32(priv, 0x610490));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+static int
+nvd0_disp_mast_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_dmac *mast = (void *)object;
+
+	/* deactivate channel */
+	nv_mask(priv, 0x610490, 0x00000010, 0x00000000);
+	nv_mask(priv, 0x610490, 0x00000003, 0x00000000);
+	if (!nv_wait(priv, 0x610490, 0x001e0000, 0x00000000)) {
+		nv_error(mast, "fini: 0x%08x\n", nv_rd32(priv, 0x610490));
+		if (suspend)
+			return -EBUSY;
+	}
+
+	/* disable error reporting */
+	nv_mask(priv, 0x610090, 0x00000001, 0x00000000);
+	nv_mask(priv, 0x6100a0, 0x00000001, 0x00000000);
+
+	return nv50_disp_chan_fini(&mast->base, suspend);
+}
+
+struct nouveau_ofuncs
+nvd0_disp_mast_ofuncs = {
+	.ctor = nvd0_disp_mast_ctor,
+	.dtor = nv50_disp_dmac_dtor,
+	.init = nvd0_disp_mast_init,
+	.fini = nvd0_disp_mast_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * EVO sync channel objects
+ ******************************************************************************/
+
+static int
+nvd0_disp_sync_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_sync_class *args = data;
+	struct nv50_disp_priv *priv = (void *)engine;
+	struct nv50_disp_dmac *dmac;
+	int ret;
+
+	if (size < sizeof(*data) || args->head >= priv->head.nr)
+		return -EINVAL;
+
+	ret = nv50_disp_dmac_create_(parent, engine, oclass, args->pushbuf,
+				     1 + args->head, sizeof(*dmac),
+				     (void **)&dmac);
+	*pobject = nv_object(dmac);
+	if (ret)
+		return ret;
+
+	nv_parent(dmac)->object_attach = nvd0_disp_dmac_object_attach;
+	nv_parent(dmac)->object_detach = nvd0_disp_dmac_object_detach;
+	return 0;
+}
+
+struct nouveau_ofuncs
+nvd0_disp_sync_ofuncs = {
+	.ctor = nvd0_disp_sync_ctor,
+	.dtor = nv50_disp_dmac_dtor,
+	.init = nvd0_disp_dmac_init,
+	.fini = nvd0_disp_dmac_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * EVO overlay channel objects
+ ******************************************************************************/
+
+static int
+nvd0_disp_ovly_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_ovly_class *args = data;
+	struct nv50_disp_priv *priv = (void *)engine;
+	struct nv50_disp_dmac *dmac;
+	int ret;
+
+	if (size < sizeof(*data) || args->head >= priv->head.nr)
+		return -EINVAL;
+
+	ret = nv50_disp_dmac_create_(parent, engine, oclass, args->pushbuf,
+				     5 + args->head, sizeof(*dmac),
+				     (void **)&dmac);
+	*pobject = nv_object(dmac);
+	if (ret)
+		return ret;
+
+	nv_parent(dmac)->object_attach = nvd0_disp_dmac_object_attach;
+	nv_parent(dmac)->object_detach = nvd0_disp_dmac_object_detach;
+	return 0;
+}
+
+struct nouveau_ofuncs
+nvd0_disp_ovly_ofuncs = {
+	.ctor = nvd0_disp_ovly_ctor,
+	.dtor = nv50_disp_dmac_dtor,
+	.init = nvd0_disp_dmac_init,
+	.fini = nvd0_disp_dmac_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * EVO PIO channel base class
+ ******************************************************************************/
+
+static int
+nvd0_disp_pioc_create_(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, int chid,
+		       int length, void **pobject)
+{
+	return nv50_disp_chan_create_(parent, engine, oclass, chid,
+				      length, pobject);
+}
+
+static void
+nvd0_disp_pioc_dtor(struct nouveau_object *object)
+{
+	struct nv50_disp_pioc *pioc = (void *)object;
+	nv50_disp_chan_destroy(&pioc->base);
+}
+
+static int
+nvd0_disp_pioc_init(struct nouveau_object *object)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_pioc *pioc = (void *)object;
+	int chid = pioc->base.chid;
+	int ret;
+
+	ret = nv50_disp_chan_init(&pioc->base);
+	if (ret)
+		return ret;
+
+	/* enable error reporting */
+	nv_mask(priv, 0x610090, 0x00000001 << chid, 0x00000001 << chid);
+	nv_mask(priv, 0x6100a0, 0x00000001 << chid, 0x00000001 << chid);
+
+	/* activate channel */
+	nv_wr32(priv, 0x610490 + (chid * 0x10), 0x00000001);
+	if (!nv_wait(priv, 0x610490 + (chid * 0x10), 0x00030000, 0x00010000)) {
+		nv_error(pioc, "init: 0x%08x\n",
+			 nv_rd32(priv, 0x610490 + (chid * 0x10)));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+static int
+nvd0_disp_pioc_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_pioc *pioc = (void *)object;
+	int chid = pioc->base.chid;
+
+	nv_mask(priv, 0x610490 + (chid * 0x10), 0x00000001, 0x00000000);
+	if (!nv_wait(priv, 0x610490 + (chid * 0x10), 0x00030000, 0x00000000)) {
+		nv_error(pioc, "timeout: 0x%08x\n",
+			 nv_rd32(priv, 0x610490 + (chid * 0x10)));
+		if (suspend)
+			return -EBUSY;
+	}
+
+	/* disable error reporting */
+	nv_mask(priv, 0x610090, 0x00000001 << chid, 0x00000000);
+	nv_mask(priv, 0x6100a0, 0x00000001 << chid, 0x00000000);
+
+	return nv50_disp_chan_fini(&pioc->base, suspend);
+}
+
+/*******************************************************************************
+ * EVO immediate overlay channel objects
+ ******************************************************************************/
+
+static int
+nvd0_disp_oimm_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_oimm_class *args = data;
+	struct nv50_disp_priv *priv = (void *)engine;
+	struct nv50_disp_pioc *pioc;
+	int ret;
+
+	if (size < sizeof(*args) || args->head >= priv->head.nr)
+		return -EINVAL;
+
+	ret = nvd0_disp_pioc_create_(parent, engine, oclass, 9 + args->head,
+				     sizeof(*pioc), (void **)&pioc);
+	*pobject = nv_object(pioc);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+struct nouveau_ofuncs
+nvd0_disp_oimm_ofuncs = {
+	.ctor = nvd0_disp_oimm_ctor,
+	.dtor = nvd0_disp_pioc_dtor,
+	.init = nvd0_disp_pioc_init,
+	.fini = nvd0_disp_pioc_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * EVO cursor channel objects
+ ******************************************************************************/
+
+static int
+nvd0_disp_curs_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_display_curs_class *args = data;
+	struct nv50_disp_priv *priv = (void *)engine;
+	struct nv50_disp_pioc *pioc;
+	int ret;
+
+	if (size < sizeof(*args) || args->head >= priv->head.nr)
+		return -EINVAL;
+
+	ret = nvd0_disp_pioc_create_(parent, engine, oclass, 13 + args->head,
+				     sizeof(*pioc), (void **)&pioc);
+	*pobject = nv_object(pioc);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+struct nouveau_ofuncs
+nvd0_disp_curs_ofuncs = {
+	.ctor = nvd0_disp_curs_ctor,
+	.dtor = nvd0_disp_pioc_dtor,
+	.init = nvd0_disp_pioc_init,
+	.fini = nvd0_disp_pioc_fini,
+	.rd32 = nv50_disp_chan_rd32,
+	.wr32 = nv50_disp_chan_wr32,
+};
+
+/*******************************************************************************
+ * Base display object
+ ******************************************************************************/
+
+static int
+nvd0_disp_base_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv = (void *)engine;
+	struct nv50_disp_base *base;
+	int ret;
+
+	ret = nouveau_parent_create(parent, engine, oclass, 0,
+				    priv->sclass, 0, &base);
+	*pobject = nv_object(base);
+	if (ret)
+		return ret;
+
+	return nouveau_ramht_new(parent, parent, 0x1000, 0, &base->ramht);
+}
+
+static void
+nvd0_disp_base_dtor(struct nouveau_object *object)
+{
+	struct nv50_disp_base *base = (void *)object;
+	nouveau_ramht_ref(NULL, &base->ramht);
+	nouveau_parent_destroy(&base->base);
+}
+
+static int
+nvd0_disp_base_init(struct nouveau_object *object)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_base *base = (void *)object;
+	int ret, i;
+	u32 tmp;
+
+	ret = nouveau_parent_init(&base->base);
+	if (ret)
+		return ret;
+
+	/* The below segments of code copying values from one register to
+	 * another appear to inform EVO of the display capabilities or
+	 * something similar.
+	 */
+
+	/* ... CRTC caps */
+	for (i = 0; i < priv->head.nr; i++) {
+		tmp = nv_rd32(priv, 0x616104 + (i * 0x800));
+		nv_wr32(priv, 0x6101b4 + (i * 0x800), tmp);
+		tmp = nv_rd32(priv, 0x616108 + (i * 0x800));
+		nv_wr32(priv, 0x6101b8 + (i * 0x800), tmp);
+		tmp = nv_rd32(priv, 0x61610c + (i * 0x800));
+		nv_wr32(priv, 0x6101bc + (i * 0x800), tmp);
+	}
+
+	/* ... DAC caps */
+	for (i = 0; i < priv->dac.nr; i++) {
+		tmp = nv_rd32(priv, 0x61a000 + (i * 0x800));
+		nv_wr32(priv, 0x6101c0 + (i * 0x800), tmp);
+	}
+
+	/* ... SOR caps */
+	for (i = 0; i < priv->sor.nr; i++) {
+		tmp = nv_rd32(priv, 0x61c000 + (i * 0x800));
+		nv_wr32(priv, 0x6301c4 + (i * 0x800), tmp);
+	}
+
+	/* steal display away from vbios, or something like that */
+	if (nv_rd32(priv, 0x6100ac) & 0x00000100) {
+		nv_wr32(priv, 0x6100ac, 0x00000100);
+		nv_mask(priv, 0x6194e8, 0x00000001, 0x00000000);
+		if (!nv_wait(priv, 0x6194e8, 0x00000002, 0x00000000)) {
+			nv_error(priv, "timeout acquiring display\n");
+			return -EBUSY;
+		}
+	}
+
+	/* point at display engine memory area (hash table, objects) */
+	nv_wr32(priv, 0x610010, (nv_gpuobj(object->parent)->addr >> 8) | 9);
+
+	/* enable supervisor interrupts, disable everything else */
+	nv_wr32(priv, 0x610090, 0x00000000);
+	nv_wr32(priv, 0x6100a0, 0x00000000);
+	nv_wr32(priv, 0x6100b0, 0x00000307);
+
+	return 0;
+}
+
+static int
+nvd0_disp_base_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nv50_disp_base *base = (void *)object;
+
+	/* disable all interrupts */
+	nv_wr32(priv, 0x6100b0, 0x00000000);
+
+	return nouveau_parent_fini(&base->base, suspend);
+}
+
+struct nouveau_ofuncs
+nvd0_disp_base_ofuncs = {
+	.ctor = nvd0_disp_base_ctor,
+	.dtor = nvd0_disp_base_dtor,
+	.init = nvd0_disp_base_init,
+	.fini = nvd0_disp_base_fini,
+};
+
+static struct nouveau_oclass
+nvd0_disp_base_oclass[] = {
+	{ NVD0_DISP_CLASS, &nvd0_disp_base_ofuncs, nva3_disp_base_omthds },
+	{}
+};
+
+static struct nouveau_oclass
+nvd0_disp_sclass[] = {
+	{ NVD0_DISP_MAST_CLASS, &nvd0_disp_mast_ofuncs },
+	{ NVD0_DISP_SYNC_CLASS, &nvd0_disp_sync_ofuncs },
+	{ NVD0_DISP_OVLY_CLASS, &nvd0_disp_ovly_ofuncs },
+	{ NVD0_DISP_OIMM_CLASS, &nvd0_disp_oimm_ofuncs },
+	{ NVD0_DISP_CURS_CLASS, &nvd0_disp_curs_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * Display engine implementation
+ ******************************************************************************/
+
+static u16
+exec_lookup(struct nv50_disp_priv *priv, int head, int outp, u32 ctrl,
+	    struct dcb_output *dcb, u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+	    struct nvbios_outp *info)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	u16 mask, type, data;
+
+	if (outp < 4) {
+		type = DCB_OUTPUT_ANALOG;
+		mask = 0;
+	} else {
+		outp -= 4;
+		switch (ctrl & 0x00000f00) {
+		case 0x00000000: type = DCB_OUTPUT_LVDS; mask = 1; break;
+		case 0x00000100: type = DCB_OUTPUT_TMDS; mask = 1; break;
+		case 0x00000200: type = DCB_OUTPUT_TMDS; mask = 2; break;
+		case 0x00000500: type = DCB_OUTPUT_TMDS; mask = 3; break;
+		case 0x00000800: type = DCB_OUTPUT_DP; mask = 1; break;
+		case 0x00000900: type = DCB_OUTPUT_DP; mask = 2; break;
+		default:
+			nv_error(priv, "unknown SOR mc 0x%08x\n", ctrl);
+			return 0x0000;
+		}
+		dcb->sorconf.link = mask;
+	}
+
+	mask  = 0x00c0 & (mask << 6);
+	mask |= 0x0001 << outp;
+	mask |= 0x0100 << head;
+
+	data = dcb_outp_match(bios, type, mask, ver, hdr, dcb);
+	if (!data)
+		return 0x0000;
+
+	return nvbios_outp_match(bios, type, mask, ver, hdr, cnt, len, info);
+}
+
+static bool
+exec_script(struct nv50_disp_priv *priv, int head, int outp, u32 ctrl, int id)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct nvbios_outp info;
+	struct dcb_output dcb;
+	u8  ver, hdr, cnt, len;
+	u16 data;
+
+	data = exec_lookup(priv, head, outp, ctrl, &dcb, &ver, &hdr, &cnt, &len, &info);
+	if (data) {
+		struct nvbios_init init = {
+			.subdev = nv_subdev(priv),
+			.bios = bios,
+			.offset = info.script[id],
+			.outp = &dcb,
+			.crtc = head,
+			.execute = 1,
+		};
+
+		return nvbios_exec(&init) == 0;
+	}
+
+	return false;
+}
+
+static u32
+exec_clkcmp(struct nv50_disp_priv *priv, int head, int outp,
+	    u32 ctrl, int id, u32 pclk)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct nvbios_outp info1;
+	struct nvbios_ocfg info2;
+	struct dcb_output dcb;
+	u8  ver, hdr, cnt, len;
+	u16 data, conf;
+
+	data = exec_lookup(priv, head, outp, ctrl, &dcb, &ver, &hdr, &cnt, &len, &info1);
+	if (data == 0x0000)
+		return false;
+
+	switch (dcb.type) {
+	case DCB_OUTPUT_TMDS:
+		conf = (ctrl & 0x00000f00) >> 8;
+		if (pclk >= 165000)
+			conf |= 0x0100;
+		break;
+	case DCB_OUTPUT_LVDS:
+		conf = priv->sor.lvdsconf;
+		break;
+	case DCB_OUTPUT_DP:
+		conf = (ctrl & 0x00000f00) >> 8;
+		break;
+	case DCB_OUTPUT_ANALOG:
+	default:
+		conf = 0x00ff;
+		break;
+	}
+
+	data = nvbios_ocfg_match(bios, data, conf, &ver, &hdr, &cnt, &len, &info2);
+	if (data) {
+		data = nvbios_oclk_match(bios, info2.clkcmp[id], pclk);
+		if (data) {
+			struct nvbios_init init = {
+				.subdev = nv_subdev(priv),
+				.bios = bios,
+				.offset = data,
+				.outp = &dcb,
+				.crtc = head,
+				.execute = 1,
+			};
+
+			if (nvbios_exec(&init))
+				return 0x0000;
+			return conf;
+		}
+	}
+
+	return 0x0000;
+}
+
+static void
+nvd0_display_unk1_handler(struct nv50_disp_priv *priv, u32 head, u32 mask)
+{
+	int i;
+
+	for (i = 0; mask && i < 8; i++) {
+		u32 mcc = nv_rd32(priv, 0x640180 + (i * 0x20));
+		if (mcc & (1 << head))
+			exec_script(priv, head, i, mcc, 1);
+	}
+
+	nv_wr32(priv, 0x6101d4, 0x00000000);
+	nv_wr32(priv, 0x6109d4, 0x00000000);
+	nv_wr32(priv, 0x6101d0, 0x80000000);
+}
+
+static void
+nvd0_display_unk2_calc_tu(struct nv50_disp_priv *priv, int head, int or)
+{
+	const u32 ctrl = nv_rd32(priv, 0x660200 + (or   * 0x020));
+	const u32 conf = nv_rd32(priv, 0x660404 + (head * 0x300));
+	const u32 pclk = nv_rd32(priv, 0x660450 + (head * 0x300)) / 1000;
+	const u32 link = ((ctrl & 0xf00) == 0x800) ? 0 : 1;
+	const u32 hoff = (head * 0x800);
+	const u32 soff = (  or * 0x800);
+	const u32 loff = (link * 0x080) + soff;
+	const u32 symbol = 100000;
+	const u32 TU = 64;
+	u32 dpctrl = nv_rd32(priv, 0x61c10c + loff) & 0x000f0000;
+	u32 clksor = nv_rd32(priv, 0x612300 + soff);
+	u32 datarate, link_nr, link_bw, bits;
+	u64 ratio, value;
+
+	if      ((conf & 0x3c0) == 0x180) bits = 30;
+	else if ((conf & 0x3c0) == 0x140) bits = 24;
+	else                              bits = 18;
+	datarate = (pclk * bits) / 8;
+
+	if      (dpctrl > 0x00030000) link_nr = 4;
+	else if (dpctrl > 0x00010000) link_nr = 2;
+	else			      link_nr = 1;
+
+	link_bw  = (clksor & 0x007c0000) >> 18;
+	link_bw *= 27000;
+
+	ratio  = datarate;
+	ratio *= symbol;
+	do_div(ratio, link_nr * link_bw);
+
+	value  = (symbol - ratio) * TU;
+	value *= ratio;
+	do_div(value, symbol);
+	do_div(value, symbol);
+
+	value += 5;
+	value |= 0x08000000;
+
+	nv_wr32(priv, 0x616610 + hoff, value);
+}
+
+static void
+nvd0_display_unk2_handler(struct nv50_disp_priv *priv, u32 head, u32 mask)
+{
+	u32 pclk;
+	int i;
+
+	for (i = 0; mask && i < 8; i++) {
+		u32 mcc = nv_rd32(priv, 0x640180 + (i * 0x20));
+		if (mcc & (1 << head))
+			exec_script(priv, head, i, mcc, 2);
+	}
+
+	pclk = nv_rd32(priv, 0x660450 + (head * 0x300)) / 1000;
+	nv_debug(priv, "head %d pclk %d mask 0x%08x\n", head, pclk, mask);
+	if (pclk && (mask & 0x00010000)) {
+		struct nouveau_clock *clk = nouveau_clock(priv);
+		clk->pll_set(clk, PLL_VPLL0 + head, pclk);
+	}
+
+	nv_wr32(priv, 0x612200 + (head * 0x800), 0x00000000);
+
+	for (i = 0; mask && i < 8; i++) {
+		u32 mcp = nv_rd32(priv, 0x660180 + (i * 0x20)), cfg;
+		if (mcp & (1 << head)) {
+			if ((cfg = exec_clkcmp(priv, head, i, mcp, 0, pclk))) {
+				u32 addr, mask, data = 0x00000000;
+				if (i < 4) {
+					addr = 0x612280 + ((i - 0) * 0x800);
+					mask = 0xffffffff;
+				} else {
+					switch (mcp & 0x00000f00) {
+					case 0x00000800:
+					case 0x00000900:
+						nvd0_display_unk2_calc_tu(priv, head, i - 4);
+						break;
+					default:
+						break;
+					}
+
+					addr = 0x612300 + ((i - 4) * 0x800);
+					mask = 0x00000707;
+					if (cfg & 0x00000100)
+						data = 0x00000101;
+				}
+				nv_mask(priv, addr, mask, data);
+			}
+			break;
+		}
+	}
+
+	nv_wr32(priv, 0x6101d4, 0x00000000);
+	nv_wr32(priv, 0x6109d4, 0x00000000);
+	nv_wr32(priv, 0x6101d0, 0x80000000);
+}
+
+static void
+nvd0_display_unk4_handler(struct nv50_disp_priv *priv, u32 head, u32 mask)
+{
+	int pclk, i;
+
+	pclk = nv_rd32(priv, 0x660450 + (head * 0x300)) / 1000;
+
+	for (i = 0; mask && i < 8; i++) {
+		u32 mcp = nv_rd32(priv, 0x660180 + (i * 0x20));
+		if (mcp & (1 << head))
+			exec_clkcmp(priv, head, i, mcp, 1, pclk);
+	}
+
+	nv_wr32(priv, 0x6101d4, 0x00000000);
+	nv_wr32(priv, 0x6109d4, 0x00000000);
+	nv_wr32(priv, 0x6101d0, 0x80000000);
+}
+
+static void
+nvd0_disp_intr_vblank(struct nv50_disp_priv *priv, int crtc)
+{
+	struct nouveau_bar *bar = nouveau_bar(priv);
+	struct nouveau_disp *disp = &priv->base;
+	struct nouveau_software_chan *chan, *temp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&disp->vblank.lock, flags);
+	list_for_each_entry_safe(chan, temp, &disp->vblank.list, vblank.head) {
+		if (chan->vblank.crtc != crtc)
+			continue;
+
+		nv_wr32(priv, 0x001718, 0x80000000 | chan->vblank.channel);
+		bar->flush(bar);
+		nv_wr32(priv, 0x06000c, upper_32_bits(chan->vblank.offset));
+		nv_wr32(priv, 0x060010, lower_32_bits(chan->vblank.offset));
+		nv_wr32(priv, 0x060014, chan->vblank.value);
+
+		list_del(&chan->vblank.head);
+		if (disp->vblank.put)
+			disp->vblank.put(disp->vblank.data, crtc);
+	}
+	spin_unlock_irqrestore(&disp->vblank.lock, flags);
+
+	if (disp->vblank.notify)
+		disp->vblank.notify(disp->vblank.data, crtc);
+}
+
+void
+nvd0_disp_intr(struct nouveau_subdev *subdev)
+{
+	struct nv50_disp_priv *priv = (void *)subdev;
+	u32 intr = nv_rd32(priv, 0x610088);
+	int i;
+
+	if (intr & 0x00000001) {
+		u32 stat = nv_rd32(priv, 0x61008c);
+		nv_wr32(priv, 0x61008c, stat);
+		intr &= ~0x00000001;
+	}
+
+	if (intr & 0x00000002) {
+		u32 stat = nv_rd32(priv, 0x61009c);
+		int chid = ffs(stat) - 1;
+		if (chid >= 0) {
+			u32 mthd = nv_rd32(priv, 0x6101f0 + (chid * 12));
+			u32 data = nv_rd32(priv, 0x6101f4 + (chid * 12));
+			u32 unkn = nv_rd32(priv, 0x6101f8 + (chid * 12));
+
+			nv_error(priv, "chid %d mthd 0x%04x data 0x%08x "
+				       "0x%08x 0x%08x\n",
+				 chid, (mthd & 0x0000ffc), data, mthd, unkn);
+			nv_wr32(priv, 0x61009c, (1 << chid));
+			nv_wr32(priv, 0x6101f0 + (chid * 12), 0x90000000);
+		}
+
+		intr &= ~0x00000002;
+	}
+
+	if (intr & 0x00100000) {
+		u32 stat = nv_rd32(priv, 0x6100ac);
+		u32 mask = 0, crtc = ~0;
+
+		while (!mask && ++crtc < priv->head.nr)
+			mask = nv_rd32(priv, 0x6101d4 + (crtc * 0x800));
+
+		if (stat & 0x00000001) {
+			nv_wr32(priv, 0x6100ac, 0x00000001);
+			nvd0_display_unk1_handler(priv, crtc, mask);
+			stat &= ~0x00000001;
+		}
+
+		if (stat & 0x00000002) {
+			nv_wr32(priv, 0x6100ac, 0x00000002);
+			nvd0_display_unk2_handler(priv, crtc, mask);
+			stat &= ~0x00000002;
+		}
+
+		if (stat & 0x00000004) {
+			nv_wr32(priv, 0x6100ac, 0x00000004);
+			nvd0_display_unk4_handler(priv, crtc, mask);
+			stat &= ~0x00000004;
+		}
+
+		if (stat) {
+			nv_info(priv, "unknown intr24 0x%08x\n", stat);
+			nv_wr32(priv, 0x6100ac, stat);
+		}
+
+		intr &= ~0x00100000;
+	}
+
+	for (i = 0; i < priv->head.nr; i++) {
+		u32 mask = 0x01000000 << i;
+		if (mask & intr) {
+			u32 stat = nv_rd32(priv, 0x6100bc + (i * 0x800));
+			if (stat & 0x00000001)
+				nvd0_disp_intr_vblank(priv, i);
+			nv_mask(priv, 0x6100bc + (i * 0x800), 0, 0);
+			nv_rd32(priv, 0x6100c0 + (i * 0x800));
+		}
+	}
+}
+
+static int
+nvd0_disp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv;
+	int ret;
+
+	ret = nouveau_disp_create(parent, engine, oclass, "PDISP",
+				  "display", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nvd0_disp_base_oclass;
+	nv_engine(priv)->cclass = &nv50_disp_cclass;
+	nv_subdev(priv)->intr = nvd0_disp_intr;
+	priv->sclass = nvd0_disp_sclass;
+	priv->head.nr = nv_rd32(priv, 0x022448);
+	priv->dac.nr = 3;
+	priv->sor.nr = 4;
+	priv->dac.power = nv50_dac_power;
+	priv->dac.sense = nv50_dac_sense;
+	priv->sor.power = nv50_sor_power;
+	priv->sor.hda_eld = nvd0_hda_eld;
+	priv->sor.hdmi = nvd0_hdmi_ctrl;
+	priv->sor.dp_train = nvd0_sor_dp_train;
+	priv->sor.dp_train_init = nv94_sor_dp_train_init;
+	priv->sor.dp_train_fini = nv94_sor_dp_train_fini;
+	priv->sor.dp_lnkctl = nvd0_sor_dp_lnkctl;
+	priv->sor.dp_drvctl = nvd0_sor_dp_drvctl;
+
+	INIT_LIST_HEAD(&priv->base.vblank.list);
+	spin_lock_init(&priv->base.vblank.lock);
+	return 0;
+}
+
+struct nouveau_oclass
+nvd0_disp_oclass = {
+	.handle = NV_ENGINE(DISP, 0x90),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvd0_disp_ctor,
+		.dtor = _nouveau_disp_dtor,
+		.init = _nouveau_disp_init,
+		.fini = _nouveau_disp_fini,
+	},
+};
diff --git a/nvkm/engine/disp/nve0.c b/nvkm/engine/disp/nve0.c
new file mode 100644
index 000000000..259537c45
--- /dev/null
+++ b/nvkm/engine/disp/nve0.c
@@ -0,0 +1,94 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <engine/software.h>
+#include <engine/disp.h>
+
+#include <core/class.h>
+
+#include "nv50.h"
+
+static struct nouveau_oclass
+nve0_disp_sclass[] = {
+	{ NVE0_DISP_MAST_CLASS, &nvd0_disp_mast_ofuncs },
+	{ NVE0_DISP_SYNC_CLASS, &nvd0_disp_sync_ofuncs },
+	{ NVE0_DISP_OVLY_CLASS, &nvd0_disp_ovly_ofuncs },
+	{ NVE0_DISP_OIMM_CLASS, &nvd0_disp_oimm_ofuncs },
+	{ NVE0_DISP_CURS_CLASS, &nvd0_disp_curs_ofuncs },
+	{}
+};
+
+static struct nouveau_oclass
+nve0_disp_base_oclass[] = {
+	{ NVE0_DISP_CLASS, &nvd0_disp_base_ofuncs, nva3_disp_base_omthds },
+	{}
+};
+
+static int
+nve0_disp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_disp_priv *priv;
+	int ret;
+
+	ret = nouveau_disp_create(parent, engine, oclass, "PDISP",
+				  "display", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nve0_disp_base_oclass;
+	nv_engine(priv)->cclass = &nv50_disp_cclass;
+	nv_subdev(priv)->intr = nvd0_disp_intr;
+	priv->sclass = nve0_disp_sclass;
+	priv->head.nr = nv_rd32(priv, 0x022448);
+	priv->dac.nr = 3;
+	priv->sor.nr = 4;
+	priv->dac.power = nv50_dac_power;
+	priv->dac.sense = nv50_dac_sense;
+	priv->sor.power = nv50_sor_power;
+	priv->sor.hda_eld = nvd0_hda_eld;
+	priv->sor.hdmi = nvd0_hdmi_ctrl;
+	priv->sor.dp_train = nvd0_sor_dp_train;
+	priv->sor.dp_train_init = nv94_sor_dp_train_init;
+	priv->sor.dp_train_fini = nv94_sor_dp_train_fini;
+	priv->sor.dp_lnkctl = nvd0_sor_dp_lnkctl;
+	priv->sor.dp_drvctl = nvd0_sor_dp_drvctl;
+
+	INIT_LIST_HEAD(&priv->base.vblank.list);
+	spin_lock_init(&priv->base.vblank.lock);
+	return 0;
+}
+
+struct nouveau_oclass
+nve0_disp_oclass = {
+	.handle = NV_ENGINE(DISP, 0x91),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nve0_disp_ctor,
+		.dtor = _nouveau_disp_dtor,
+		.init = _nouveau_disp_init,
+		.fini = _nouveau_disp_fini,
+	},
+};
diff --git a/nvkm/engine/disp/sornv50.c b/nvkm/engine/disp/sornv50.c
new file mode 100644
index 000000000..39b6b6773
--- /dev/null
+++ b/nvkm/engine/disp/sornv50.c
@@ -0,0 +1,112 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/timer.h>
+
+#include "nv50.h"
+
+int
+nv50_sor_power(struct nv50_disp_priv *priv, int or, u32 data)
+{
+	const u32 stat = data & NV50_DISP_SOR_PWR_STATE;
+	const u32 soff = (or * 0x800);
+	nv_wait(priv, 0x61c004 + soff, 0x80000000, 0x00000000);
+	nv_mask(priv, 0x61c004 + soff, 0x80000001, 0x80000000 | stat);
+	nv_wait(priv, 0x61c004 + soff, 0x80000000, 0x00000000);
+	nv_wait(priv, 0x61c030 + soff, 0x10000000, 0x00000000);
+	return 0;
+}
+
+int
+nv50_sor_mthd(struct nouveau_object *object, u32 mthd, void *args, u32 size)
+{
+	struct nv50_disp_priv *priv = (void *)object->engine;
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	const u16 type = (mthd & NV50_DISP_SOR_MTHD_TYPE) >> 12;
+	const u8  head = (mthd & NV50_DISP_SOR_MTHD_HEAD) >> 3;
+	const u8  link = (mthd & NV50_DISP_SOR_MTHD_LINK) >> 2;
+	const u8    or = (mthd & NV50_DISP_SOR_MTHD_OR);
+	const u16 mask = (0x0100 << head) | (0x0040 << link) | (0x0001 << or);
+	struct dcb_output outp;
+	u8  ver, hdr;
+	u32 data;
+	int ret = -EINVAL;
+
+	if (size < sizeof(u32))
+		return -EINVAL;
+	data = *(u32 *)args;
+
+	if (type && !dcb_outp_match(bios, type, mask, &ver, &hdr, &outp))
+		return -ENODEV;
+
+	switch (mthd & ~0x3f) {
+	case NV50_DISP_SOR_PWR:
+		ret = priv->sor.power(priv, or, data);
+		break;
+	case NVA3_DISP_SOR_HDA_ELD:
+		ret = priv->sor.hda_eld(priv, or, args, size);
+		break;
+	case NV84_DISP_SOR_HDMI_PWR:
+		ret = priv->sor.hdmi(priv, head, or, data);
+		break;
+	case NV50_DISP_SOR_LVDS_SCRIPT:
+		priv->sor.lvdsconf = data & NV50_DISP_SOR_LVDS_SCRIPT_ID;
+		ret = 0;
+		break;
+	case NV94_DISP_SOR_DP_TRAIN:
+		switch (data & NV94_DISP_SOR_DP_TRAIN_OP) {
+		case NV94_DISP_SOR_DP_TRAIN_OP_PATTERN:
+			ret = priv->sor.dp_train(priv, or, link, type, mask, data, &outp);
+			break;
+		case NV94_DISP_SOR_DP_TRAIN_OP_INIT:
+			ret = priv->sor.dp_train_init(priv, or, link, head, type, mask, data, &outp);
+			break;
+		case NV94_DISP_SOR_DP_TRAIN_OP_FINI:
+			ret = priv->sor.dp_train_fini(priv, or, link, head, type, mask, data, &outp);
+			break;
+		default:
+			break;
+		}
+		break;
+	case NV94_DISP_SOR_DP_LNKCTL:
+		ret = priv->sor.dp_lnkctl(priv, or, link, head, type, mask, data, &outp);
+		break;
+	case NV94_DISP_SOR_DP_DRVCTL(0):
+	case NV94_DISP_SOR_DP_DRVCTL(1):
+	case NV94_DISP_SOR_DP_DRVCTL(2):
+	case NV94_DISP_SOR_DP_DRVCTL(3):
+		ret = priv->sor.dp_drvctl(priv, or, link, (mthd & 0xc0) >> 6,
+				          type, mask, data, &outp);
+		break;
+	default:
+		BUG_ON(1);
+	}
+
+	return ret;
+}
diff --git a/nvkm/engine/disp/sornv94.c b/nvkm/engine/disp/sornv94.c
new file mode 100644
index 000000000..f6edd0097
--- /dev/null
+++ b/nvkm/engine/disp/sornv94.c
@@ -0,0 +1,190 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/dp.h>
+#include <subdev/bios/init.h>
+
+#include "nv50.h"
+
+static inline u32
+nv94_sor_dp_lane_map(struct nv50_disp_priv *priv, u8 lane)
+{
+	static const u8 nvaf[] = { 24, 16, 8, 0 }; /* thanks, apple.. */
+	static const u8 nv94[] = { 16, 8, 0, 24 };
+	if (nv_device(priv)->chipset == 0xaf)
+		return nvaf[lane];
+	return nv94[lane];
+}
+
+int
+nv94_sor_dp_train_init(struct nv50_disp_priv *priv, int or, int link, int head,
+		       u16 type, u16 mask, u32 data, struct dcb_output *dcbo)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct nvbios_dpout info;
+	u8  ver, hdr, cnt, len;
+	u16 outp;
+
+	outp = nvbios_dpout_match(bios, type, mask, &ver, &hdr, &cnt, &len, &info);
+	if (outp) {
+		struct nvbios_init init = {
+			.subdev = nv_subdev(priv),
+			.bios = bios,
+			.outp = dcbo,
+			.crtc = head,
+			.execute = 1,
+		};
+
+		if (data & NV94_DISP_SOR_DP_TRAIN_INIT_SPREAD_ON)
+			init.offset = info.script[2];
+		else
+			init.offset = info.script[3];
+		nvbios_exec(&init);
+
+		init.offset = info.script[0];
+		nvbios_exec(&init);
+	}
+
+	return 0;
+}
+
+int
+nv94_sor_dp_train_fini(struct nv50_disp_priv *priv, int or, int link, int head,
+		       u16 type, u16 mask, u32 data, struct dcb_output *dcbo)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct nvbios_dpout info;
+	u8  ver, hdr, cnt, len;
+	u16 outp;
+
+	outp = nvbios_dpout_match(bios, type, mask, &ver, &hdr, &cnt, &len, &info);
+	if (outp) {
+		struct nvbios_init init = {
+			.subdev = nv_subdev(priv),
+			.bios = bios,
+			.offset = info.script[1],
+			.outp = dcbo,
+			.crtc = head,
+			.execute = 1,
+		};
+
+		nvbios_exec(&init);
+	}
+
+	return 0;
+}
+
+int
+nv94_sor_dp_train(struct nv50_disp_priv *priv, int or, int link,
+		  u16 type, u16 mask, u32 data, struct dcb_output *info)
+{
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	const u32 patt = (data & NV94_DISP_SOR_DP_TRAIN_PATTERN);
+	nv_mask(priv, 0x61c10c + loff, 0x0f000000, patt << 24);
+	return 0;
+}
+
+int
+nv94_sor_dp_lnkctl(struct nv50_disp_priv *priv, int or, int link, int head,
+		   u16 type, u16 mask, u32 data, struct dcb_output *dcbo)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	const u32 soff = (or * 0x800);
+	u16 link_bw = (data & NV94_DISP_SOR_DP_LNKCTL_WIDTH) >> 8;
+	u8  link_nr = (data & NV94_DISP_SOR_DP_LNKCTL_COUNT);
+	u32 dpctrl = 0x00000000;
+	u32 clksor = 0x00000000;
+	u32 outp, lane = 0;
+	u8  ver, hdr, cnt, len;
+	struct nvbios_dpout info;
+	int i;
+
+	/* -> 10Khz units */
+	link_bw *= 2700;
+
+	outp = nvbios_dpout_match(bios, type, mask, &ver, &hdr, &cnt, &len, &info);
+	if (outp && info.lnkcmp) {
+		struct nvbios_init init = {
+			.subdev = nv_subdev(priv),
+			.bios = bios,
+			.offset = 0x0000,
+			.outp = dcbo,
+			.crtc = head,
+			.execute = 1,
+		};
+
+		while (link_bw < nv_ro16(bios, info.lnkcmp))
+			info.lnkcmp += 4;
+		init.offset = nv_ro16(bios, info.lnkcmp + 2);
+
+		nvbios_exec(&init);
+	}
+
+	dpctrl |= ((1 << link_nr) - 1) << 16;
+	if (data & NV94_DISP_SOR_DP_LNKCTL_FRAME_ENH)
+		dpctrl |= 0x00004000;
+	if (link_bw > 16200)
+		clksor |= 0x00040000;
+
+	for (i = 0; i < link_nr; i++)
+		lane |= 1 << (nv94_sor_dp_lane_map(priv, i) >> 3);
+
+	nv_mask(priv, 0x614300 + soff, 0x000c0000, clksor);
+	nv_mask(priv, 0x61c10c + loff, 0x001f4000, dpctrl);
+	nv_mask(priv, 0x61c130 + loff, 0x0000000f, lane);
+	return 0;
+}
+
+int
+nv94_sor_dp_drvctl(struct nv50_disp_priv *priv, int or, int link, int lane,
+		   u16 type, u16 mask, u32 data, struct dcb_output *dcbo)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	const u8 swing = (data & NV94_DISP_SOR_DP_DRVCTL_VS) >> 8;
+	const u8 preem = (data & NV94_DISP_SOR_DP_DRVCTL_PE);
+	u32 addr, shift = nv94_sor_dp_lane_map(priv, lane);
+	u8  ver, hdr, cnt, len;
+	struct nvbios_dpout outp;
+	struct nvbios_dpcfg ocfg;
+
+	addr = nvbios_dpout_match(bios, type, mask, &ver, &hdr, &cnt, &len, &outp);
+	if (!addr)
+		return -ENODEV;
+
+	addr = nvbios_dpcfg_match(bios, addr, 0, swing, preem, &ver, &hdr, &cnt, &len, &ocfg);
+	if (!addr)
+		return -EINVAL;
+
+	nv_mask(priv, 0x61c118 + loff, 0x000000ff << shift, ocfg.drv << shift);
+	nv_mask(priv, 0x61c120 + loff, 0x000000ff << shift, ocfg.pre << shift);
+	nv_mask(priv, 0x61c130 + loff, 0x0000ff00, ocfg.unk << 8);
+	return 0;
+}
diff --git a/nvkm/engine/disp/sornvd0.c b/nvkm/engine/disp/sornvd0.c
new file mode 100644
index 000000000..c37ce7e29
--- /dev/null
+++ b/nvkm/engine/disp/sornvd0.c
@@ -0,0 +1,126 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/dp.h>
+#include <subdev/bios/init.h>
+
+#include "nv50.h"
+
+static inline u32
+nvd0_sor_dp_lane_map(struct nv50_disp_priv *priv, u8 lane)
+{
+	static const u8 nvd0[] = { 16, 8, 0, 24 };
+	return nvd0[lane];
+}
+
+int
+nvd0_sor_dp_train(struct nv50_disp_priv *priv, int or, int link,
+		  u16 type, u16 mask, u32 data, struct dcb_output *info)
+{
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	const u32 patt = (data & NV94_DISP_SOR_DP_TRAIN_PATTERN);
+	nv_mask(priv, 0x61c110 + loff, 0x0f0f0f0f, 0x01010101 * patt);
+	return 0;
+}
+
+int
+nvd0_sor_dp_lnkctl(struct nv50_disp_priv *priv, int or, int link, int head,
+		   u16 type, u16 mask, u32 data, struct dcb_output *dcbo)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	const u32 soff = (or * 0x800);
+	const u8  link_bw = (data & NV94_DISP_SOR_DP_LNKCTL_WIDTH) >> 8;
+	const u8  link_nr = (data & NV94_DISP_SOR_DP_LNKCTL_COUNT);
+	u32 dpctrl = 0x00000000;
+	u32 clksor = 0x00000000;
+	u32 outp, lane = 0;
+	u8  ver, hdr, cnt, len;
+	struct nvbios_dpout info;
+	int i;
+
+	outp = nvbios_dpout_match(bios, type, mask, &ver, &hdr, &cnt, &len, &info);
+	if (outp && info.lnkcmp) {
+		struct nvbios_init init = {
+			.subdev = nv_subdev(priv),
+			.bios = bios,
+			.offset = 0x0000,
+			.outp = dcbo,
+			.crtc = head,
+			.execute = 1,
+		};
+
+		while (nv_ro08(bios, info.lnkcmp) < link_bw)
+			info.lnkcmp += 3;
+		init.offset = nv_ro16(bios, info.lnkcmp + 1);
+
+		nvbios_exec(&init);
+	}
+
+	clksor |= link_bw << 18;
+	dpctrl |= ((1 << link_nr) - 1) << 16;
+	if (data & NV94_DISP_SOR_DP_LNKCTL_FRAME_ENH)
+		dpctrl |= 0x00004000;
+
+	for (i = 0; i < link_nr; i++)
+		lane |= 1 << (nvd0_sor_dp_lane_map(priv, i) >> 3);
+
+	nv_mask(priv, 0x612300 + soff, 0x007c0000, clksor);
+	nv_mask(priv, 0x61c10c + loff, 0x001f4000, dpctrl);
+	nv_mask(priv, 0x61c130 + loff, 0x0000000f, lane);
+	return 0;
+}
+
+int
+nvd0_sor_dp_drvctl(struct nv50_disp_priv *priv, int or, int link, int lane,
+		   u16 type, u16 mask, u32 data, struct dcb_output *dcbo)
+{
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	const u32 loff = (or * 0x800) + (link * 0x80);
+	const u8 swing = (data & NV94_DISP_SOR_DP_DRVCTL_VS) >> 8;
+	const u8 preem = (data & NV94_DISP_SOR_DP_DRVCTL_PE);
+	u32 addr, shift = nvd0_sor_dp_lane_map(priv, lane);
+	u8  ver, hdr, cnt, len;
+	struct nvbios_dpout outp;
+	struct nvbios_dpcfg ocfg;
+
+	addr = nvbios_dpout_match(bios, type, mask, &ver, &hdr, &cnt, &len, &outp);
+	if (!addr)
+		return -ENODEV;
+
+	addr = nvbios_dpcfg_match(bios, addr, 0, swing, preem, &ver, &hdr, &cnt, &len, &ocfg);
+	if (!addr)
+		return -EINVAL;
+
+	nv_mask(priv, 0x61c118 + loff, 0x000000ff << shift, ocfg.drv << shift);
+	nv_mask(priv, 0x61c120 + loff, 0x000000ff << shift, ocfg.pre << shift);
+	nv_mask(priv, 0x61c130 + loff, 0x0000ff00, ocfg.unk << 8);
+	nv_mask(priv, 0x61c13c + loff, 0x00000000, 0x00000000);
+	return 0;
+}
diff --git a/nvkm/engine/disp/vga.c b/nvkm/engine/disp/vga.c
new file mode 100644
index 000000000..5a1c68474
--- /dev/null
+++ b/nvkm/engine/disp/vga.c
@@ -0,0 +1,215 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/subdev.h>
+#include <core/device.h>
+#include <subdev/vga.h>
+
+u8
+nv_rdport(void *obj, int head, u16 port)
+{
+	struct nouveau_device *device = nv_device(obj);
+
+	if (device->card_type >= NV_50)
+		return nv_rd08(obj, 0x601000 + port);
+
+	if (port == 0x03c0 || port == 0x03c1 ||	/* AR */
+	    port == 0x03c2 || port == 0x03da ||	/* INP0 */
+	    port == 0x03d4 || port == 0x03d5)	/* CR */
+		return nv_rd08(obj, 0x601000 + (head * 0x2000) + port);
+
+	if (port == 0x03c2 || port == 0x03cc ||	/* MISC */
+	    port == 0x03c4 || port == 0x03c5 ||	/* SR */
+	    port == 0x03ce || port == 0x03cf) {	/* GR */
+		if (device->card_type < NV_40)
+			head = 0; /* CR44 selects head */
+		return nv_rd08(obj, 0x0c0000 + (head * 0x2000) + port);
+	}
+
+	nv_error(obj, "unknown vga port 0x%04x\n", port);
+	return 0x00;
+}
+
+void
+nv_wrport(void *obj, int head, u16 port, u8 data)
+{
+	struct nouveau_device *device = nv_device(obj);
+
+	if (device->card_type >= NV_50)
+		nv_wr08(obj, 0x601000 + port, data);
+	else
+	if (port == 0x03c0 || port == 0x03c1 ||	/* AR */
+	    port == 0x03c2 || port == 0x03da ||	/* INP0 */
+	    port == 0x03d4 || port == 0x03d5)	/* CR */
+		nv_wr08(obj, 0x601000 + (head * 0x2000) + port, data);
+	else
+	if (port == 0x03c2 || port == 0x03cc ||	/* MISC */
+	    port == 0x03c4 || port == 0x03c5 ||	/* SR */
+	    port == 0x03ce || port == 0x03cf) {	/* GR */
+		if (device->card_type < NV_40)
+			head = 0; /* CR44 selects head */
+		nv_wr08(obj, 0x0c0000 + (head * 0x2000) + port, data);
+	} else
+		nv_error(obj, "unknown vga port 0x%04x\n", port);
+}
+
+u8
+nv_rdvgas(void *obj, int head, u8 index)
+{
+	nv_wrport(obj, head, 0x03c4, index);
+	return nv_rdport(obj, head, 0x03c5);
+}
+
+void
+nv_wrvgas(void *obj, int head, u8 index, u8 value)
+{
+	nv_wrport(obj, head, 0x03c4, index);
+	nv_wrport(obj, head, 0x03c5, value);
+}
+
+u8
+nv_rdvgag(void *obj, int head, u8 index)
+{
+	nv_wrport(obj, head, 0x03ce, index);
+	return nv_rdport(obj, head, 0x03cf);
+}
+
+void
+nv_wrvgag(void *obj, int head, u8 index, u8 value)
+{
+	nv_wrport(obj, head, 0x03ce, index);
+	nv_wrport(obj, head, 0x03cf, value);
+}
+
+u8
+nv_rdvgac(void *obj, int head, u8 index)
+{
+	nv_wrport(obj, head, 0x03d4, index);
+	return nv_rdport(obj, head, 0x03d5);
+}
+
+void
+nv_wrvgac(void *obj, int head, u8 index, u8 value)
+{
+	nv_wrport(obj, head, 0x03d4, index);
+	nv_wrport(obj, head, 0x03d5, value);
+}
+
+u8
+nv_rdvgai(void *obj, int head, u16 port, u8 index)
+{
+	if (port == 0x03c4) return nv_rdvgas(obj, head, index);
+	if (port == 0x03ce) return nv_rdvgag(obj, head, index);
+	if (port == 0x03d4) return nv_rdvgac(obj, head, index);
+	nv_error(obj, "unknown indexed vga port 0x%04x\n", port);
+	return 0x00;
+}
+
+void
+nv_wrvgai(void *obj, int head, u16 port, u8 index, u8 value)
+{
+	if      (port == 0x03c4) nv_wrvgas(obj, head, index, value);
+	else if (port == 0x03ce) nv_wrvgag(obj, head, index, value);
+	else if (port == 0x03d4) nv_wrvgac(obj, head, index, value);
+	else nv_error(obj, "unknown indexed vga port 0x%04x\n", port);
+}
+
+bool
+nv_lockvgac(void *obj, bool lock)
+{
+	bool locked = !nv_rdvgac(obj, 0, 0x1f);
+	u8 data = lock ? 0x99 : 0x57;
+	nv_wrvgac(obj, 0, 0x1f, data);
+	if (nv_device(obj)->chipset == 0x11) {
+		if (!(nv_rd32(obj, 0x001084) & 0x10000000))
+			nv_wrvgac(obj, 1, 0x1f, data);
+	}
+	return locked;
+}
+
+/* CR44 takes values 0 (head A), 3 (head B) and 4 (heads tied)
+ * it affects only the 8 bit vga io regs, which we access using mmio at
+ * 0xc{0,2}3c*, 0x60{1,3}3*, and 0x68{1,3}3d*
+ * in general, the set value of cr44 does not matter: reg access works as
+ * expected and values can be set for the appropriate head by using a 0x2000
+ * offset as required
+ * however:
+ * a) pre nv40, the head B range of PRMVIO regs at 0xc23c* was not exposed and
+ *    cr44 must be set to 0 or 3 for accessing values on the correct head
+ *    through the common 0xc03c* addresses
+ * b) in tied mode (4) head B is programmed to the values set on head A, and
+ *    access using the head B addresses can have strange results, ergo we leave
+ *    tied mode in init once we know to what cr44 should be restored on exit
+ *
+ * the owner parameter is slightly abused:
+ * 0 and 1 are treated as head values and so the set value is (owner * 3)
+ * other values are treated as literal values to set
+ */
+u8
+nv_rdvgaowner(void *obj)
+{
+	if (nv_device(obj)->card_type < NV_50) {
+		if (nv_device(obj)->chipset == 0x11) {
+			u32 tied = nv_rd32(obj, 0x001084) & 0x10000000;
+			if (tied == 0) {
+				u8 slA = nv_rdvgac(obj, 0, 0x28) & 0x80;
+				u8 tvA = nv_rdvgac(obj, 0, 0x33) & 0x01;
+				u8 slB = nv_rdvgac(obj, 1, 0x28) & 0x80;
+				u8 tvB = nv_rdvgac(obj, 1, 0x33) & 0x01;
+				if (slA && !tvA) return 0x00;
+				if (slB && !tvB) return 0x03;
+				if (slA) return 0x00;
+				if (slB) return 0x03;
+				return 0x00;
+			}
+			return 0x04;
+		}
+
+		return nv_rdvgac(obj, 0, 0x44);
+	}
+
+	nv_error(obj, "rdvgaowner after nv4x\n");
+	return 0x00;
+}
+
+void
+nv_wrvgaowner(void *obj, u8 select)
+{
+	if (nv_device(obj)->card_type < NV_50) {
+		u8 owner = (select == 1) ? 3 : select;
+		if (nv_device(obj)->chipset == 0x11) {
+			/* workaround hw lockup bug */
+			nv_rdvgac(obj, 0, 0x1f);
+			nv_rdvgac(obj, 1, 0x1f);
+		}
+
+		nv_wrvgac(obj, 0, 0x44, owner);
+
+		if (nv_device(obj)->chipset == 0x11) {
+			nv_wrvgac(obj, 0, 0x2e, owner);
+			nv_wrvgac(obj, 0, 0x2e, owner);
+		}
+	} else
+		nv_error(obj, "wrvgaowner after nv4x\n");
+}
diff --git a/nvkm/engine/dmaobj/Makefile.am b/nvkm/engine/dmaobj/Makefile.am
new file mode 100644
index 000000000..d2633c46c
--- /dev/null
+++ b/nvkm/engine/dmaobj/Makefile.am
@@ -0,0 +1,9 @@
+noinst_LTLIBRARIES = libdmaobj.la
+
+libdmaobj_la_SOURCES = base.c \
+		       nv04.c \
+		       nv50.c \
+		       nvc0.c \
+		       nvd0.c
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/dmaobj/base.c b/nvkm/engine/dmaobj/base.c
new file mode 100644
index 000000000..5103e88d1
--- /dev/null
+++ b/nvkm/engine/dmaobj/base.c
@@ -0,0 +1,120 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/class.h>
+
+#include <subdev/fb.h>
+#include <engine/dmaobj.h>
+
+static int
+nouveau_dmaobj_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nouveau_dmaeng *dmaeng = (void *)engine;
+	struct nouveau_dmaobj *dmaobj;
+	struct nouveau_gpuobj *gpuobj;
+	struct nv_dma_class *args = data;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_object_create(parent, engine, oclass, 0, &dmaobj);
+	*pobject = nv_object(dmaobj);
+	if (ret)
+		return ret;
+
+	switch (args->flags & NV_DMA_TARGET_MASK) {
+	case NV_DMA_TARGET_VM:
+		dmaobj->target = NV_MEM_TARGET_VM;
+		break;
+	case NV_DMA_TARGET_VRAM:
+		dmaobj->target = NV_MEM_TARGET_VRAM;
+		break;
+	case NV_DMA_TARGET_PCI:
+		dmaobj->target = NV_MEM_TARGET_PCI;
+		break;
+	case NV_DMA_TARGET_PCI_US:
+	case NV_DMA_TARGET_AGP:
+		dmaobj->target = NV_MEM_TARGET_PCI_NOSNOOP;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	switch (args->flags & NV_DMA_ACCESS_MASK) {
+	case NV_DMA_ACCESS_VM:
+		dmaobj->access = NV_MEM_ACCESS_VM;
+		break;
+	case NV_DMA_ACCESS_RD:
+		dmaobj->access = NV_MEM_ACCESS_RO;
+		break;
+	case NV_DMA_ACCESS_WR:
+		dmaobj->access = NV_MEM_ACCESS_WO;
+		break;
+	case NV_DMA_ACCESS_RDWR:
+		dmaobj->access = NV_MEM_ACCESS_RW;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	dmaobj->start = args->start;
+	dmaobj->limit = args->limit;
+	dmaobj->conf0 = args->conf0;
+
+	switch (nv_mclass(parent)) {
+	case NV_DEVICE_CLASS:
+		/* delayed, or no, binding */
+		break;
+	default:
+		ret = dmaeng->bind(dmaeng, *pobject, dmaobj, &gpuobj);
+		if (ret == 0) {
+			nouveau_object_ref(NULL, pobject);
+			*pobject = nv_object(gpuobj);
+		}
+		break;
+	}
+
+	return ret;
+}
+
+static struct nouveau_ofuncs
+nouveau_dmaobj_ofuncs = {
+	.ctor = nouveau_dmaobj_ctor,
+	.dtor = nouveau_object_destroy,
+	.init = nouveau_object_init,
+	.fini = nouveau_object_fini,
+};
+
+struct nouveau_oclass
+nouveau_dmaobj_sclass[] = {
+	{ NV_DMA_FROM_MEMORY_CLASS, &nouveau_dmaobj_ofuncs },
+	{ NV_DMA_TO_MEMORY_CLASS, &nouveau_dmaobj_ofuncs },
+	{ NV_DMA_IN_MEMORY_CLASS, &nouveau_dmaobj_ofuncs },
+	{}
+};
diff --git a/nvkm/engine/dmaobj/nv04.c b/nvkm/engine/dmaobj/nv04.c
new file mode 100644
index 000000000..027d8217c
--- /dev/null
+++ b/nvkm/engine/dmaobj/nv04.c
@@ -0,0 +1,143 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/gpuobj.h>
+#include <core/class.h>
+
+#include <subdev/fb.h>
+#include <subdev/vm/nv04.h>
+
+#include <engine/dmaobj.h>
+
+struct nv04_dmaeng_priv {
+	struct nouveau_dmaeng base;
+};
+
+static int
+nv04_dmaobj_bind(struct nouveau_dmaeng *dmaeng,
+		 struct nouveau_object *parent,
+		 struct nouveau_dmaobj *dmaobj,
+		 struct nouveau_gpuobj **pgpuobj)
+{
+	struct nv04_vmmgr_priv *vmm = nv04_vmmgr(dmaeng);
+	struct nouveau_gpuobj *gpuobj;
+	u32 flags0 = nv_mclass(dmaobj);
+	u32 flags2 = 0x00000000;
+	u64 offset = dmaobj->start & 0xfffff000;
+	u64 adjust = dmaobj->start & 0x00000fff;
+	u32 length = dmaobj->limit - dmaobj->start;
+	int ret;
+
+	if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
+		switch (nv_mclass(parent->parent)) {
+		case NV03_CHANNEL_DMA_CLASS:
+		case NV10_CHANNEL_DMA_CLASS:
+		case NV17_CHANNEL_DMA_CLASS:
+		case NV40_CHANNEL_DMA_CLASS:
+			break;
+		default:
+			return -EINVAL;
+		}
+	}
+
+	if (dmaobj->target == NV_MEM_TARGET_VM) {
+		if (nv_object(vmm)->oclass == &nv04_vmmgr_oclass) {
+			struct nouveau_gpuobj *pgt = vmm->vm->pgt[0].obj[0];
+			if (!dmaobj->start)
+				return nouveau_gpuobj_dup(parent, pgt, pgpuobj);
+			offset  = nv_ro32(pgt, 8 + (offset >> 10));
+			offset &= 0xfffff000;
+		}
+
+		dmaobj->target = NV_MEM_TARGET_PCI;
+		dmaobj->access = NV_MEM_ACCESS_RW;
+	}
+
+	switch (dmaobj->target) {
+	case NV_MEM_TARGET_VRAM:
+		flags0 |= 0x00003000;
+		break;
+	case NV_MEM_TARGET_PCI:
+		flags0 |= 0x00023000;
+		break;
+	case NV_MEM_TARGET_PCI_NOSNOOP:
+		flags0 |= 0x00033000;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	switch (dmaobj->access) {
+	case NV_MEM_ACCESS_RO:
+		flags0 |= 0x00004000;
+		break;
+	case NV_MEM_ACCESS_WO:
+		flags0 |= 0x00008000;
+	case NV_MEM_ACCESS_RW:
+		flags2 |= 0x00000002;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	ret = nouveau_gpuobj_new(parent, parent, 16, 16, 0, &gpuobj);
+	*pgpuobj = gpuobj;
+	if (ret == 0) {
+		nv_wo32(*pgpuobj, 0x00, flags0 | (adjust << 20));
+		nv_wo32(*pgpuobj, 0x04, length);
+		nv_wo32(*pgpuobj, 0x08, flags2 | offset);
+		nv_wo32(*pgpuobj, 0x0c, flags2 | offset);
+	}
+
+	return ret;
+}
+
+static int
+nv04_dmaeng_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		 struct nouveau_oclass *oclass, void *data, u32 size,
+		 struct nouveau_object **pobject)
+{
+	struct nv04_dmaeng_priv *priv;
+	int ret;
+
+	ret = nouveau_dmaeng_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nouveau_dmaobj_sclass;
+	priv->base.bind = nv04_dmaobj_bind;
+	return 0;
+}
+
+struct nouveau_oclass
+nv04_dmaeng_oclass = {
+	.handle = NV_ENGINE(DMAOBJ, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_dmaeng_ctor,
+		.dtor = _nouveau_dmaeng_dtor,
+		.init = _nouveau_dmaeng_init,
+		.fini = _nouveau_dmaeng_fini,
+	},
+};
diff --git a/nvkm/engine/dmaobj/nv50.c b/nvkm/engine/dmaobj/nv50.c
new file mode 100644
index 000000000..750183f7c
--- /dev/null
+++ b/nvkm/engine/dmaobj/nv50.c
@@ -0,0 +1,161 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/gpuobj.h>
+#include <core/class.h>
+
+#include <subdev/fb.h>
+#include <engine/dmaobj.h>
+
+struct nv50_dmaeng_priv {
+	struct nouveau_dmaeng base;
+};
+
+static int
+nv50_dmaobj_bind(struct nouveau_dmaeng *dmaeng,
+		 struct nouveau_object *parent,
+		 struct nouveau_dmaobj *dmaobj,
+		 struct nouveau_gpuobj **pgpuobj)
+{
+	u32 flags0 = nv_mclass(dmaobj);
+	u32 flags5 = 0x00000000;
+	int ret;
+
+	if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
+		switch (nv_mclass(parent->parent)) {
+		case NV50_CHANNEL_DMA_CLASS:
+		case NV84_CHANNEL_DMA_CLASS:
+		case NV50_CHANNEL_IND_CLASS:
+		case NV84_CHANNEL_IND_CLASS:
+		case NV50_DISP_MAST_CLASS:
+		case NV84_DISP_MAST_CLASS:
+		case NV94_DISP_MAST_CLASS:
+		case NVA0_DISP_MAST_CLASS:
+		case NVA3_DISP_MAST_CLASS:
+		case NV50_DISP_SYNC_CLASS:
+		case NV84_DISP_SYNC_CLASS:
+		case NV94_DISP_SYNC_CLASS:
+		case NVA0_DISP_SYNC_CLASS:
+		case NVA3_DISP_SYNC_CLASS:
+		case NV50_DISP_OVLY_CLASS:
+		case NV84_DISP_OVLY_CLASS:
+		case NV94_DISP_OVLY_CLASS:
+		case NVA0_DISP_OVLY_CLASS:
+		case NVA3_DISP_OVLY_CLASS:
+			break;
+		default:
+			return -EINVAL;
+		}
+	}
+
+	if (!(dmaobj->conf0 & NV50_DMA_CONF0_ENABLE)) {
+		if (dmaobj->target == NV_MEM_TARGET_VM) {
+			dmaobj->conf0  = NV50_DMA_CONF0_PRIV_VM;
+			dmaobj->conf0 |= NV50_DMA_CONF0_PART_VM;
+			dmaobj->conf0 |= NV50_DMA_CONF0_COMP_VM;
+			dmaobj->conf0 |= NV50_DMA_CONF0_TYPE_VM;
+		} else {
+			dmaobj->conf0  = NV50_DMA_CONF0_PRIV_US;
+			dmaobj->conf0 |= NV50_DMA_CONF0_PART_256;
+			dmaobj->conf0 |= NV50_DMA_CONF0_COMP_NONE;
+			dmaobj->conf0 |= NV50_DMA_CONF0_TYPE_LINEAR;
+		}
+	}
+
+	flags0 |= (dmaobj->conf0 & NV50_DMA_CONF0_COMP) << 22;
+	flags0 |= (dmaobj->conf0 & NV50_DMA_CONF0_TYPE) << 22;
+	flags0 |= (dmaobj->conf0 & NV50_DMA_CONF0_PRIV);
+	flags5 |= (dmaobj->conf0 & NV50_DMA_CONF0_PART);
+
+	switch (dmaobj->target) {
+	case NV_MEM_TARGET_VM:
+		flags0 |= 0x00000000;
+		break;
+	case NV_MEM_TARGET_VRAM:
+		flags0 |= 0x00010000;
+		break;
+	case NV_MEM_TARGET_PCI:
+		flags0 |= 0x00020000;
+		break;
+	case NV_MEM_TARGET_PCI_NOSNOOP:
+		flags0 |= 0x00030000;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	switch (dmaobj->access) {
+	case NV_MEM_ACCESS_VM:
+		break;
+	case NV_MEM_ACCESS_RO:
+		flags0 |= 0x00040000;
+		break;
+	case NV_MEM_ACCESS_WO:
+	case NV_MEM_ACCESS_RW:
+		flags0 |= 0x00080000;
+		break;
+	}
+
+	ret = nouveau_gpuobj_new(parent, parent, 24, 32, 0, pgpuobj);
+	if (ret == 0) {
+		nv_wo32(*pgpuobj, 0x00, flags0);
+		nv_wo32(*pgpuobj, 0x04, lower_32_bits(dmaobj->limit));
+		nv_wo32(*pgpuobj, 0x08, lower_32_bits(dmaobj->start));
+		nv_wo32(*pgpuobj, 0x0c, upper_32_bits(dmaobj->limit) << 24 |
+					upper_32_bits(dmaobj->start));
+		nv_wo32(*pgpuobj, 0x10, 0x00000000);
+		nv_wo32(*pgpuobj, 0x14, flags5);
+	}
+
+	return ret;
+}
+
+static int
+nv50_dmaeng_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		 struct nouveau_oclass *oclass, void *data, u32 size,
+		 struct nouveau_object **pobject)
+{
+	struct nv50_dmaeng_priv *priv;
+	int ret;
+
+	ret = nouveau_dmaeng_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nouveau_dmaobj_sclass;
+	priv->base.bind = nv50_dmaobj_bind;
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_dmaeng_oclass = {
+	.handle = NV_ENGINE(DMAOBJ, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_dmaeng_ctor,
+		.dtor = _nouveau_dmaeng_dtor,
+		.init = _nouveau_dmaeng_init,
+		.fini = _nouveau_dmaeng_fini,
+	},
+};
diff --git a/nvkm/engine/dmaobj/nvc0.c b/nvkm/engine/dmaobj/nvc0.c
new file mode 100644
index 000000000..cd3970d03
--- /dev/null
+++ b/nvkm/engine/dmaobj/nvc0.c
@@ -0,0 +1,143 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/device.h>
+#include <core/gpuobj.h>
+#include <core/class.h>
+
+#include <subdev/fb.h>
+#include <engine/dmaobj.h>
+
+struct nvc0_dmaeng_priv {
+	struct nouveau_dmaeng base;
+};
+
+static int
+nvc0_dmaobj_bind(struct nouveau_dmaeng *dmaeng,
+		 struct nouveau_object *parent,
+		 struct nouveau_dmaobj *dmaobj,
+		 struct nouveau_gpuobj **pgpuobj)
+{
+	u32 flags0 = nv_mclass(dmaobj);
+	u32 flags5 = 0x00000000;
+	int ret;
+
+	if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
+		switch (nv_mclass(parent->parent)) {
+		case NVA3_DISP_MAST_CLASS:
+		case NVA3_DISP_SYNC_CLASS:
+		case NVA3_DISP_OVLY_CLASS:
+			break;
+		default:
+			return -EINVAL;
+		}
+	} else
+		return 0;
+
+	if (!(dmaobj->conf0 & NVC0_DMA_CONF0_ENABLE)) {
+		if (dmaobj->target == NV_MEM_TARGET_VM) {
+			dmaobj->conf0  = NVC0_DMA_CONF0_PRIV_VM;
+			dmaobj->conf0 |= NVC0_DMA_CONF0_TYPE_VM;
+		} else {
+			dmaobj->conf0  = NVC0_DMA_CONF0_PRIV_US;
+			dmaobj->conf0 |= NVC0_DMA_CONF0_TYPE_LINEAR;
+			dmaobj->conf0 |= 0x00020000;
+		}
+	}
+
+	flags0 |= (dmaobj->conf0 & NVC0_DMA_CONF0_TYPE) << 22;
+	flags0 |= (dmaobj->conf0 & NVC0_DMA_CONF0_PRIV);
+	flags5 |= (dmaobj->conf0 & NVC0_DMA_CONF0_UNKN);
+
+	switch (dmaobj->target) {
+	case NV_MEM_TARGET_VM:
+		flags0 |= 0x00000000;
+		break;
+	case NV_MEM_TARGET_VRAM:
+		flags0 |= 0x00010000;
+		break;
+	case NV_MEM_TARGET_PCI:
+		flags0 |= 0x00020000;
+		break;
+	case NV_MEM_TARGET_PCI_NOSNOOP:
+		flags0 |= 0x00030000;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	switch (dmaobj->access) {
+	case NV_MEM_ACCESS_VM:
+		break;
+	case NV_MEM_ACCESS_RO:
+		flags0 |= 0x00040000;
+		break;
+	case NV_MEM_ACCESS_WO:
+	case NV_MEM_ACCESS_RW:
+		flags0 |= 0x00080000;
+		break;
+	}
+
+	ret = nouveau_gpuobj_new(parent, parent, 24, 32, 0, pgpuobj);
+	if (ret == 0) {
+		nv_wo32(*pgpuobj, 0x00, flags0);
+		nv_wo32(*pgpuobj, 0x04, lower_32_bits(dmaobj->limit));
+		nv_wo32(*pgpuobj, 0x08, lower_32_bits(dmaobj->start));
+		nv_wo32(*pgpuobj, 0x0c, upper_32_bits(dmaobj->limit) << 24 |
+					upper_32_bits(dmaobj->start));
+		nv_wo32(*pgpuobj, 0x10, 0x00000000);
+		nv_wo32(*pgpuobj, 0x14, flags5);
+	}
+
+	return ret;
+}
+
+static int
+nvc0_dmaeng_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		 struct nouveau_oclass *oclass, void *data, u32 size,
+		 struct nouveau_object **pobject)
+{
+	struct nvc0_dmaeng_priv *priv;
+	int ret;
+
+	ret = nouveau_dmaeng_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nouveau_dmaobj_sclass;
+	priv->base.bind = nvc0_dmaobj_bind;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_dmaeng_oclass = {
+	.handle = NV_ENGINE(DMAOBJ, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_dmaeng_ctor,
+		.dtor = _nouveau_dmaeng_dtor,
+		.init = _nouveau_dmaeng_init,
+		.fini = _nouveau_dmaeng_fini,
+	},
+};
diff --git a/nvkm/engine/dmaobj/nvd0.c b/nvkm/engine/dmaobj/nvd0.c
new file mode 100644
index 000000000..d15287529
--- /dev/null
+++ b/nvkm/engine/dmaobj/nvd0.c
@@ -0,0 +1,122 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/device.h>
+#include <core/gpuobj.h>
+#include <core/class.h>
+
+#include <subdev/fb.h>
+#include <engine/dmaobj.h>
+
+struct nvd0_dmaeng_priv {
+	struct nouveau_dmaeng base;
+};
+
+static int
+nvd0_dmaobj_bind(struct nouveau_dmaeng *dmaeng,
+		 struct nouveau_object *parent,
+		 struct nouveau_dmaobj *dmaobj,
+		 struct nouveau_gpuobj **pgpuobj)
+{
+	u32 flags0 = 0x00000000;
+	int ret;
+
+	if (!nv_iclass(parent, NV_ENGCTX_CLASS)) {
+		switch (nv_mclass(parent->parent)) {
+		case NVD0_DISP_MAST_CLASS:
+		case NVD0_DISP_SYNC_CLASS:
+		case NVD0_DISP_OVLY_CLASS:
+		case NVE0_DISP_MAST_CLASS:
+		case NVE0_DISP_SYNC_CLASS:
+		case NVE0_DISP_OVLY_CLASS:
+			break;
+		default:
+			return -EINVAL;
+		}
+	} else
+		return 0;
+
+	if (!(dmaobj->conf0 & NVD0_DMA_CONF0_ENABLE)) {
+		if (dmaobj->target == NV_MEM_TARGET_VM) {
+			dmaobj->conf0 |= NVD0_DMA_CONF0_TYPE_VM;
+			dmaobj->conf0 |= NVD0_DMA_CONF0_PAGE_LP;
+		} else {
+			dmaobj->conf0 |= NVD0_DMA_CONF0_TYPE_LINEAR;
+			dmaobj->conf0 |= NVD0_DMA_CONF0_PAGE_SP;
+		}
+	}
+
+	flags0 |= (dmaobj->conf0 & NVD0_DMA_CONF0_TYPE) << 20;
+	flags0 |= (dmaobj->conf0 & NVD0_DMA_CONF0_PAGE) >> 4;
+
+	switch (dmaobj->target) {
+	case NV_MEM_TARGET_VRAM:
+		flags0 |= 0x00000009;
+		break;
+	default:
+		return -EINVAL;
+		break;
+	}
+
+	ret = nouveau_gpuobj_new(parent, parent, 24, 32, 0, pgpuobj);
+	if (ret == 0) {
+		nv_wo32(*pgpuobj, 0x00, flags0);
+		nv_wo32(*pgpuobj, 0x04, dmaobj->start >> 8);
+		nv_wo32(*pgpuobj, 0x08, dmaobj->limit >> 8);
+		nv_wo32(*pgpuobj, 0x0c, 0x00000000);
+		nv_wo32(*pgpuobj, 0x10, 0x00000000);
+		nv_wo32(*pgpuobj, 0x14, 0x00000000);
+	}
+
+	return ret;
+}
+
+static int
+nvd0_dmaeng_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		 struct nouveau_oclass *oclass, void *data, u32 size,
+		 struct nouveau_object **pobject)
+{
+	struct nvd0_dmaeng_priv *priv;
+	int ret;
+
+	ret = nouveau_dmaeng_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->sclass = nouveau_dmaobj_sclass;
+	priv->base.bind = nvd0_dmaobj_bind;
+	return 0;
+}
+
+struct nouveau_oclass
+nvd0_dmaeng_oclass = {
+	.handle = NV_ENGINE(DMAOBJ, 0xd0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvd0_dmaeng_ctor,
+		.dtor = _nouveau_dmaeng_dtor,
+		.init = _nouveau_dmaeng_init,
+		.fini = _nouveau_dmaeng_fini,
+	},
+};
diff --git a/nvkm/engine/fifo/Makefile.am b/nvkm/engine/fifo/Makefile.am
new file mode 100644
index 000000000..6c2764f60
--- /dev/null
+++ b/nvkm/engine/fifo/Makefile.am
@@ -0,0 +1,13 @@
+noinst_LTLIBRARIES = libfifo.la
+
+libfifo_la_SOURCES = base.c \
+		     nv04.c \
+		     nv10.c \
+		     nv17.c \
+		     nv40.c \
+		     nv50.c \
+		     nv84.c \
+		     nvc0.c \
+		     nve0.c
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/fifo/base.c b/nvkm/engine/fifo/base.c
new file mode 100644
index 000000000..c2b9db335
--- /dev/null
+++ b/nvkm/engine/fifo/base.c
@@ -0,0 +1,180 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/handle.h>
+#include <core/class.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+
+int
+nouveau_fifo_channel_create_(struct nouveau_object *parent,
+			     struct nouveau_object *engine,
+			     struct nouveau_oclass *oclass,
+			     int bar, u32 addr, u32 size, u32 pushbuf,
+			     u64 engmask, int len, void **ptr)
+{
+	struct nouveau_device *device = nv_device(engine);
+	struct nouveau_fifo *priv = (void *)engine;
+	struct nouveau_fifo_chan *chan;
+	struct nouveau_dmaeng *dmaeng;
+	unsigned long flags;
+	int ret;
+
+	/* create base object class */
+	ret = nouveau_namedb_create_(parent, engine, oclass, 0, NULL,
+				     engmask, len, ptr);
+	chan = *ptr;
+	if (ret)
+		return ret;
+
+	/* validate dma object representing push buffer */
+	chan->pushdma = (void *)nouveau_handle_ref(parent, pushbuf);
+	if (!chan->pushdma)
+		return -ENOENT;
+
+	dmaeng = (void *)chan->pushdma->base.engine;
+	switch (chan->pushdma->base.oclass->handle) {
+	case NV_DMA_FROM_MEMORY_CLASS:
+	case NV_DMA_IN_MEMORY_CLASS:
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	ret = dmaeng->bind(dmaeng, parent, chan->pushdma, &chan->pushgpu);
+	if (ret)
+		return ret;
+
+	/* find a free fifo channel */
+	spin_lock_irqsave(&priv->lock, flags);
+	for (chan->chid = priv->min; chan->chid < priv->max; chan->chid++) {
+		if (!priv->channel[chan->chid]) {
+			priv->channel[chan->chid] = nv_object(chan);
+			break;
+		}
+	}
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	if (chan->chid == priv->max) {
+		nv_error(priv, "no free channels\n");
+		return -ENOSPC;
+	}
+
+	/* map fifo control registers */
+	chan->user = ioremap(pci_resource_start(device->pdev, bar) + addr +
+			     (chan->chid * size), size);
+	if (!chan->user)
+		return -EFAULT;
+
+	chan->size = size;
+	return 0;
+}
+
+void
+nouveau_fifo_channel_destroy(struct nouveau_fifo_chan *chan)
+{
+	struct nouveau_fifo *priv = (void *)nv_object(chan)->engine;
+	unsigned long flags;
+
+	iounmap(chan->user);
+
+	spin_lock_irqsave(&priv->lock, flags);
+	priv->channel[chan->chid] = NULL;
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	nouveau_gpuobj_ref(NULL, &chan->pushgpu);
+	nouveau_object_ref(NULL, (struct nouveau_object **)&chan->pushdma);
+	nouveau_namedb_destroy(&chan->base);
+}
+
+void
+_nouveau_fifo_channel_dtor(struct nouveau_object *object)
+{
+	struct nouveau_fifo_chan *chan = (void *)object;
+	nouveau_fifo_channel_destroy(chan);
+}
+
+u32
+_nouveau_fifo_channel_rd32(struct nouveau_object *object, u64 addr)
+{
+	struct nouveau_fifo_chan *chan = (void *)object;
+	return ioread32_native(chan->user + addr);
+}
+
+void
+_nouveau_fifo_channel_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	struct nouveau_fifo_chan *chan = (void *)object;
+	iowrite32_native(data, chan->user + addr);
+}
+
+static int
+nouveau_fifo_chid(struct nouveau_fifo *priv, struct nouveau_object *object)
+{
+	int engidx = nv_hclass(priv) & 0xff;
+
+	while (object && object->parent) {
+		if ( nv_iclass(object->parent, NV_ENGCTX_CLASS) &&
+		    (nv_hclass(object->parent) & 0xff) == engidx)
+			return nouveau_fifo_chan(object)->chid;
+		object = object->parent;
+	}
+
+	return -1;
+}
+
+void
+nouveau_fifo_destroy(struct nouveau_fifo *priv)
+{
+	kfree(priv->channel);
+	nouveau_engine_destroy(&priv->base);
+}
+
+int
+nouveau_fifo_create_(struct nouveau_object *parent,
+		     struct nouveau_object *engine,
+		     struct nouveau_oclass *oclass,
+		     int min, int max, int length, void **pobject)
+{
+	struct nouveau_fifo *priv;
+	int ret;
+
+	ret = nouveau_engine_create_(parent, engine, oclass, true, "PFIFO",
+				     "fifo", length, pobject);
+	priv = *pobject;
+	if (ret)
+		return ret;
+
+	priv->min = min;
+	priv->max = max;
+	priv->channel = kzalloc(sizeof(*priv->channel) * (max + 1), GFP_KERNEL);
+	if (!priv->channel)
+		return -ENOMEM;
+
+	priv->chid = nouveau_fifo_chid;
+	spin_lock_init(&priv->lock);
+	return 0;
+}
diff --git a/nvkm/engine/fifo/nv04.c b/nvkm/engine/fifo/nv04.c
new file mode 100644
index 000000000..a47a8548f
--- /dev/null
+++ b/nvkm/engine/fifo/nv04.c
@@ -0,0 +1,629 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/namedb.h>
+#include <core/handle.h>
+#include <core/ramht.h>
+
+#include <subdev/instmem.h>
+#include <subdev/instmem/nv04.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+
+#include <engine/fifo.h>
+
+#include "nv04.h"
+
+static struct ramfc_desc
+nv04_ramfc[] = {
+	{ 32,  0, 0x00,  0, NV04_PFIFO_CACHE1_DMA_PUT },
+	{ 32,  0, 0x04,  0, NV04_PFIFO_CACHE1_DMA_GET },
+	{ 16,  0, 0x08,  0, NV04_PFIFO_CACHE1_DMA_INSTANCE },
+	{ 16, 16, 0x08,  0, NV04_PFIFO_CACHE1_DMA_DCOUNT },
+	{ 32,  0, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_STATE },
+	{ 32,  0, 0x10,  0, NV04_PFIFO_CACHE1_DMA_FETCH },
+	{ 32,  0, 0x14,  0, NV04_PFIFO_CACHE1_ENGINE },
+	{ 32,  0, 0x18,  0, NV04_PFIFO_CACHE1_PULL1 },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO channel objects
+ ******************************************************************************/
+
+int
+nv04_fifo_object_attach(struct nouveau_object *parent,
+			struct nouveau_object *object, u32 handle)
+{
+	struct nv04_fifo_priv *priv = (void *)parent->engine;
+	struct nv04_fifo_chan *chan = (void *)parent;
+	u32 context, chid = chan->base.chid;
+	int ret;
+
+	if (nv_iclass(object, NV_GPUOBJ_CLASS))
+		context = nv_gpuobj(object)->addr >> 4;
+	else
+		context = 0x00000004; /* just non-zero */
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_DMAOBJ:
+	case NVDEV_ENGINE_SW:
+		context |= 0x00000000;
+		break;
+	case NVDEV_ENGINE_GR:
+		context |= 0x00010000;
+		break;
+	case NVDEV_ENGINE_MPEG:
+		context |= 0x00020000;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	context |= 0x80000000; /* valid */
+	context |= chid << 24;
+
+	mutex_lock(&nv_subdev(priv)->mutex);
+	ret = nouveau_ramht_insert(priv->ramht, chid, handle, context);
+	mutex_unlock(&nv_subdev(priv)->mutex);
+	return ret;
+}
+
+void
+nv04_fifo_object_detach(struct nouveau_object *parent, int cookie)
+{
+	struct nv04_fifo_priv *priv = (void *)parent->engine;
+	mutex_lock(&nv_subdev(priv)->mutex);
+	nouveau_ramht_remove(priv->ramht, cookie);
+	mutex_unlock(&nv_subdev(priv)->mutex);
+}
+
+int
+nv04_fifo_context_attach(struct nouveau_object *parent,
+			 struct nouveau_object *object)
+{
+	nv_engctx(object)->addr = nouveau_fifo_chan(parent)->chid;
+	return 0;
+}
+
+static int
+nv04_fifo_chan_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv04_fifo_priv *priv = (void *)engine;
+	struct nv04_fifo_chan *chan;
+	struct nv03_channel_dma_class *args = data;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 0, 0x800000,
+					  0x10000, args->pushbuf,
+					  (1ULL << NVDEV_ENGINE_DMAOBJ) |
+					  (1ULL << NVDEV_ENGINE_SW) |
+					  (1ULL << NVDEV_ENGINE_GR), &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->object_attach = nv04_fifo_object_attach;
+	nv_parent(chan)->object_detach = nv04_fifo_object_detach;
+	nv_parent(chan)->context_attach = nv04_fifo_context_attach;
+	chan->ramfc = chan->base.chid * 32;
+
+	nv_wo32(priv->ramfc, chan->ramfc + 0x00, args->offset);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x04, args->offset);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x08, chan->base.pushgpu->addr >> 4);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x10,
+			     NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+			     NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+#ifdef __BIG_ENDIAN
+			     NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+			     NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	return 0;
+}
+
+void
+nv04_fifo_chan_dtor(struct nouveau_object *object)
+{
+	struct nv04_fifo_priv *priv = (void *)object->engine;
+	struct nv04_fifo_chan *chan = (void *)object;
+	struct ramfc_desc *c = priv->ramfc_desc;
+
+	do {
+		nv_wo32(priv->ramfc, chan->ramfc + c->ctxp, 0x00000000);
+	} while ((++c)->bits);
+
+	nouveau_fifo_channel_destroy(&chan->base);
+}
+
+int
+nv04_fifo_chan_init(struct nouveau_object *object)
+{
+	struct nv04_fifo_priv *priv = (void *)object->engine;
+	struct nv04_fifo_chan *chan = (void *)object;
+	u32 mask = 1 << chan->base.chid;
+	unsigned long flags;
+	int ret;
+
+	ret = nouveau_fifo_channel_init(&chan->base);
+	if (ret)
+		return ret;
+
+	spin_lock_irqsave(&priv->base.lock, flags);
+	nv_mask(priv, NV04_PFIFO_MODE, mask, mask);
+	spin_unlock_irqrestore(&priv->base.lock, flags);
+	return 0;
+}
+
+int
+nv04_fifo_chan_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv04_fifo_priv *priv = (void *)object->engine;
+	struct nv04_fifo_chan *chan = (void *)object;
+	struct nouveau_gpuobj *fctx = priv->ramfc;
+	struct ramfc_desc *c;
+	unsigned long flags;
+	u32 data = chan->ramfc;
+	u32 chid;
+
+	/* prevent fifo context switches */
+	spin_lock_irqsave(&priv->base.lock, flags);
+	nv_wr32(priv, NV03_PFIFO_CACHES, 0);
+
+	/* if this channel is active, replace it with a null context */
+	chid = nv_rd32(priv, NV03_PFIFO_CACHE1_PUSH1) & priv->base.max;
+	if (chid == chan->base.chid) {
+		nv_mask(priv, NV04_PFIFO_CACHE1_DMA_PUSH, 0x00000001, 0);
+		nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH0, 0);
+		nv_mask(priv, NV04_PFIFO_CACHE1_PULL0, 0x00000001, 0);
+
+		c = priv->ramfc_desc;
+		do {
+			u32 rm = ((1ULL << c->bits) - 1) << c->regs;
+			u32 cm = ((1ULL << c->bits) - 1) << c->ctxs;
+			u32 rv = (nv_rd32(priv, c->regp) &  rm) >> c->regs;
+			u32 cv = (nv_ro32(fctx, c->ctxp + data) & ~cm);
+			nv_wo32(fctx, c->ctxp + data, cv | (rv << c->ctxs));
+		} while ((++c)->bits);
+
+		c = priv->ramfc_desc;
+		do {
+			nv_wr32(priv, c->regp, 0x00000000);
+		} while ((++c)->bits);
+
+		nv_wr32(priv, NV03_PFIFO_CACHE1_GET, 0);
+		nv_wr32(priv, NV03_PFIFO_CACHE1_PUT, 0);
+		nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH1, priv->base.max);
+		nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH0, 1);
+		nv_wr32(priv, NV04_PFIFO_CACHE1_PULL0, 1);
+	}
+
+	/* restore normal operation, after disabling dma mode */
+	nv_mask(priv, NV04_PFIFO_MODE, 1 << chan->base.chid, 0);
+	nv_wr32(priv, NV03_PFIFO_CACHES, 1);
+	spin_unlock_irqrestore(&priv->base.lock, flags);
+
+	return nouveau_fifo_channel_fini(&chan->base, suspend);
+}
+
+static struct nouveau_ofuncs
+nv04_fifo_ofuncs = {
+	.ctor = nv04_fifo_chan_ctor,
+	.dtor = nv04_fifo_chan_dtor,
+	.init = nv04_fifo_chan_init,
+	.fini = nv04_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_oclass
+nv04_fifo_sclass[] = {
+	{ NV03_CHANNEL_DMA_CLASS, &nv04_fifo_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO context - basically just the instmem reserved for the channel
+ ******************************************************************************/
+
+int
+nv04_fifo_context_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nv04_fifo_base *base;
+	int ret;
+
+	ret = nouveau_fifo_context_create(parent, engine, oclass, NULL, 0x1000,
+				          0x1000, NVOBJ_FLAG_HEAP, &base);
+	*pobject = nv_object(base);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static struct nouveau_oclass
+nv04_fifo_cclass = {
+	.handle = NV_ENGCTX(FIFO, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_fifo_context_ctor,
+		.dtor = _nouveau_fifo_context_dtor,
+		.init = _nouveau_fifo_context_init,
+		.fini = _nouveau_fifo_context_fini,
+		.rd32 = _nouveau_fifo_context_rd32,
+		.wr32 = _nouveau_fifo_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PFIFO engine
+ ******************************************************************************/
+
+void
+nv04_fifo_pause(struct nouveau_fifo *pfifo, unsigned long *pflags)
+__acquires(priv->base.lock)
+{
+	struct nv04_fifo_priv *priv = (void *)pfifo;
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->base.lock, flags);
+	*pflags = flags;
+
+	nv_wr32(priv, NV03_PFIFO_CACHES, 0x00000000);
+	nv_mask(priv, NV04_PFIFO_CACHE1_PULL0, 0x00000001, 0x00000000);
+
+	/* in some cases the puller may be left in an inconsistent state
+	 * if you try to stop it while it's busy translating handles.
+	 * sometimes you get a CACHE_ERROR, sometimes it just fails
+	 * silently; sending incorrect instance offsets to PGRAPH after
+	 * it's started up again.
+	 *
+	 * to avoid this, we invalidate the most recently calculated
+	 * instance.
+	 */
+	if (!nv_wait(priv, NV04_PFIFO_CACHE1_PULL0,
+			   NV04_PFIFO_CACHE1_PULL0_HASH_BUSY, 0x00000000))
+		nv_warn(priv, "timeout idling puller\n");
+
+	if (nv_rd32(priv, NV04_PFIFO_CACHE1_PULL0) &
+			  NV04_PFIFO_CACHE1_PULL0_HASH_FAILED)
+		nv_wr32(priv, NV03_PFIFO_INTR_0, NV_PFIFO_INTR_CACHE_ERROR);
+
+	nv_wr32(priv, NV04_PFIFO_CACHE1_HASH, 0x00000000);
+}
+
+void
+nv04_fifo_start(struct nouveau_fifo *pfifo, unsigned long *pflags)
+__releases(priv->base.lock)
+{
+	struct nv04_fifo_priv *priv = (void *)pfifo;
+	unsigned long flags = *pflags;
+
+	nv_mask(priv, NV04_PFIFO_CACHE1_PULL0, 0x00000001, 0x00000001);
+	nv_wr32(priv, NV03_PFIFO_CACHES, 0x00000001);
+
+	spin_unlock_irqrestore(&priv->base.lock, flags);
+}
+
+static const char *
+nv_dma_state_err(u32 state)
+{
+	static const char * const desc[] = {
+		"NONE", "CALL_SUBR_ACTIVE", "INVALID_MTHD", "RET_SUBR_INACTIVE",
+		"INVALID_CMD", "IB_EMPTY"/* NV50+ */, "MEM_FAULT", "UNK"
+	};
+	return desc[(state >> 29) & 0x7];
+}
+
+static bool
+nv04_fifo_swmthd(struct nv04_fifo_priv *priv, u32 chid, u32 addr, u32 data)
+{
+	struct nv04_fifo_chan *chan = NULL;
+	struct nouveau_handle *bind;
+	const int subc = (addr >> 13) & 0x7;
+	const int mthd = addr & 0x1ffc;
+	bool handled = false;
+	unsigned long flags;
+	u32 engine;
+
+	spin_lock_irqsave(&priv->base.lock, flags);
+	if (likely(chid >= priv->base.min && chid <= priv->base.max))
+		chan = (void *)priv->base.channel[chid];
+	if (unlikely(!chan))
+		goto out;
+
+	switch (mthd) {
+	case 0x0000:
+		bind = nouveau_namedb_get(nv_namedb(chan), data);
+		if (unlikely(!bind))
+			break;
+
+		if (nv_engidx(bind->object->engine) == NVDEV_ENGINE_SW) {
+			engine = 0x0000000f << (subc * 4);
+			chan->subc[subc] = data;
+			handled = true;
+
+			nv_mask(priv, NV04_PFIFO_CACHE1_ENGINE, engine, 0);
+		}
+
+		nouveau_namedb_put(bind);
+		break;
+	default:
+		engine = nv_rd32(priv, NV04_PFIFO_CACHE1_ENGINE);
+		if (unlikely(((engine >> (subc * 4)) & 0xf) != 0))
+			break;
+
+		bind = nouveau_namedb_get(nv_namedb(chan), chan->subc[subc]);
+		if (likely(bind)) {
+			if (!nv_call(bind->object, mthd, data))
+				handled = true;
+			nouveau_namedb_put(bind);
+		}
+		break;
+	}
+
+out:
+	spin_unlock_irqrestore(&priv->base.lock, flags);
+	return handled;
+}
+
+void
+nv04_fifo_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_device *device = nv_device(subdev);
+	struct nv04_fifo_priv *priv = (void *)subdev;
+	uint32_t status, reassign;
+	int cnt = 0;
+
+	reassign = nv_rd32(priv, NV03_PFIFO_CACHES) & 1;
+	while ((status = nv_rd32(priv, NV03_PFIFO_INTR_0)) && (cnt++ < 100)) {
+		uint32_t chid, get;
+
+		nv_wr32(priv, NV03_PFIFO_CACHES, 0);
+
+		chid = nv_rd32(priv, NV03_PFIFO_CACHE1_PUSH1) & priv->base.max;
+		get  = nv_rd32(priv, NV03_PFIFO_CACHE1_GET);
+
+		if (status & NV_PFIFO_INTR_CACHE_ERROR) {
+			uint32_t mthd, data;
+			int ptr;
+
+			/* NV_PFIFO_CACHE1_GET actually goes to 0xffc before
+			 * wrapping on my G80 chips, but CACHE1 isn't big
+			 * enough for this much data.. Tests show that it
+			 * wraps around to the start at GET=0x800.. No clue
+			 * as to why..
+			 */
+			ptr = (get & 0x7ff) >> 2;
+
+			if (device->card_type < NV_40) {
+				mthd = nv_rd32(priv,
+					NV04_PFIFO_CACHE1_METHOD(ptr));
+				data = nv_rd32(priv,
+					NV04_PFIFO_CACHE1_DATA(ptr));
+			} else {
+				mthd = nv_rd32(priv,
+					NV40_PFIFO_CACHE1_METHOD(ptr));
+				data = nv_rd32(priv,
+					NV40_PFIFO_CACHE1_DATA(ptr));
+			}
+
+			if (!nv04_fifo_swmthd(priv, chid, mthd, data)) {
+				nv_error(priv, "CACHE_ERROR - Ch %d/%d "
+					      "Mthd 0x%04x Data 0x%08x\n",
+					chid, (mthd >> 13) & 7, mthd & 0x1ffc,
+					data);
+			}
+
+			nv_wr32(priv, NV04_PFIFO_CACHE1_DMA_PUSH, 0);
+			nv_wr32(priv, NV03_PFIFO_INTR_0,
+						NV_PFIFO_INTR_CACHE_ERROR);
+
+			nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH0,
+				nv_rd32(priv, NV03_PFIFO_CACHE1_PUSH0) & ~1);
+			nv_wr32(priv, NV03_PFIFO_CACHE1_GET, get + 4);
+			nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH0,
+				nv_rd32(priv, NV03_PFIFO_CACHE1_PUSH0) | 1);
+			nv_wr32(priv, NV04_PFIFO_CACHE1_HASH, 0);
+
+			nv_wr32(priv, NV04_PFIFO_CACHE1_DMA_PUSH,
+				nv_rd32(priv, NV04_PFIFO_CACHE1_DMA_PUSH) | 1);
+			nv_wr32(priv, NV04_PFIFO_CACHE1_PULL0, 1);
+
+			status &= ~NV_PFIFO_INTR_CACHE_ERROR;
+		}
+
+		if (status & NV_PFIFO_INTR_DMA_PUSHER) {
+			u32 dma_get = nv_rd32(priv, 0x003244);
+			u32 dma_put = nv_rd32(priv, 0x003240);
+			u32 push = nv_rd32(priv, 0x003220);
+			u32 state = nv_rd32(priv, 0x003228);
+
+			if (device->card_type == NV_50) {
+				u32 ho_get = nv_rd32(priv, 0x003328);
+				u32 ho_put = nv_rd32(priv, 0x003320);
+				u32 ib_get = nv_rd32(priv, 0x003334);
+				u32 ib_put = nv_rd32(priv, 0x003330);
+
+				nv_error(priv, "DMA_PUSHER - Ch %d Get 0x%02x%08x "
+				     "Put 0x%02x%08x IbGet 0x%08x IbPut 0x%08x "
+				     "State 0x%08x (err: %s) Push 0x%08x\n",
+					chid, ho_get, dma_get, ho_put,
+					dma_put, ib_get, ib_put, state,
+					nv_dma_state_err(state),
+					push);
+
+				/* METHOD_COUNT, in DMA_STATE on earlier chipsets */
+				nv_wr32(priv, 0x003364, 0x00000000);
+				if (dma_get != dma_put || ho_get != ho_put) {
+					nv_wr32(priv, 0x003244, dma_put);
+					nv_wr32(priv, 0x003328, ho_put);
+				} else
+				if (ib_get != ib_put) {
+					nv_wr32(priv, 0x003334, ib_put);
+				}
+			} else {
+				nv_error(priv, "DMA_PUSHER - Ch %d Get 0x%08x "
+					     "Put 0x%08x State 0x%08x (err: %s) Push 0x%08x\n",
+					chid, dma_get, dma_put, state,
+					nv_dma_state_err(state), push);
+
+				if (dma_get != dma_put)
+					nv_wr32(priv, 0x003244, dma_put);
+			}
+
+			nv_wr32(priv, 0x003228, 0x00000000);
+			nv_wr32(priv, 0x003220, 0x00000001);
+			nv_wr32(priv, 0x002100, NV_PFIFO_INTR_DMA_PUSHER);
+			status &= ~NV_PFIFO_INTR_DMA_PUSHER;
+		}
+
+		if (status & NV_PFIFO_INTR_SEMAPHORE) {
+			uint32_t sem;
+
+			status &= ~NV_PFIFO_INTR_SEMAPHORE;
+			nv_wr32(priv, NV03_PFIFO_INTR_0,
+				NV_PFIFO_INTR_SEMAPHORE);
+
+			sem = nv_rd32(priv, NV10_PFIFO_CACHE1_SEMAPHORE);
+			nv_wr32(priv, NV10_PFIFO_CACHE1_SEMAPHORE, sem | 0x1);
+
+			nv_wr32(priv, NV03_PFIFO_CACHE1_GET, get + 4);
+			nv_wr32(priv, NV04_PFIFO_CACHE1_PULL0, 1);
+		}
+
+		if (device->card_type == NV_50) {
+			if (status & 0x00000010) {
+				status &= ~0x00000010;
+				nv_wr32(priv, 0x002100, 0x00000010);
+			}
+		}
+
+		if (status) {
+			nv_warn(priv, "unknown intr 0x%08x, ch %d\n",
+				status, chid);
+			nv_wr32(priv, NV03_PFIFO_INTR_0, status);
+			status = 0;
+		}
+
+		nv_wr32(priv, NV03_PFIFO_CACHES, reassign);
+	}
+
+	if (status) {
+		nv_error(priv, "still angry after %d spins, halt\n", cnt);
+		nv_wr32(priv, 0x002140, 0);
+		nv_wr32(priv, 0x000140, 0);
+	}
+
+	nv_wr32(priv, 0x000100, 0x00000100);
+}
+
+static int
+nv04_fifo_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv04_instmem_priv *imem = nv04_instmem(parent);
+	struct nv04_fifo_priv *priv;
+	int ret;
+
+	ret = nouveau_fifo_create(parent, engine, oclass, 0, 15, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nouveau_ramht_ref(imem->ramht, &priv->ramht);
+	nouveau_gpuobj_ref(imem->ramro, &priv->ramro);
+	nouveau_gpuobj_ref(imem->ramfc, &priv->ramfc);
+
+	nv_subdev(priv)->unit = 0x00000100;
+	nv_subdev(priv)->intr = nv04_fifo_intr;
+	nv_engine(priv)->cclass = &nv04_fifo_cclass;
+	nv_engine(priv)->sclass = nv04_fifo_sclass;
+	priv->base.pause = nv04_fifo_pause;
+	priv->base.start = nv04_fifo_start;
+	priv->ramfc_desc = nv04_ramfc;
+	return 0;
+}
+
+void
+nv04_fifo_dtor(struct nouveau_object *object)
+{
+	struct nv04_fifo_priv *priv = (void *)object;
+	nouveau_gpuobj_ref(NULL, &priv->ramfc);
+	nouveau_gpuobj_ref(NULL, &priv->ramro);
+	nouveau_ramht_ref(NULL, &priv->ramht);
+	nouveau_fifo_destroy(&priv->base);
+}
+
+int
+nv04_fifo_init(struct nouveau_object *object)
+{
+	struct nv04_fifo_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_fifo_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, NV04_PFIFO_DELAY_0, 0x000000ff);
+	nv_wr32(priv, NV04_PFIFO_DMA_TIMESLICE, 0x0101ffff);
+
+	nv_wr32(priv, NV03_PFIFO_RAMHT, (0x03 << 24) /* search 128 */ |
+				       ((priv->ramht->bits - 9) << 16) |
+				        (priv->ramht->base.addr >> 8));
+	nv_wr32(priv, NV03_PFIFO_RAMRO, priv->ramro->addr >> 8);
+	nv_wr32(priv, NV03_PFIFO_RAMFC, priv->ramfc->addr >> 8);
+
+	nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH1, priv->base.max);
+
+	nv_wr32(priv, NV03_PFIFO_INTR_0, 0xffffffff);
+	nv_wr32(priv, NV03_PFIFO_INTR_EN_0, 0xffffffff);
+
+	nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH0, 1);
+	nv_wr32(priv, NV04_PFIFO_CACHE1_PULL0, 1);
+	nv_wr32(priv, NV03_PFIFO_CACHES, 1);
+	return 0;
+}
+
+struct nouveau_oclass
+nv04_fifo_oclass = {
+	.handle = NV_ENGINE(FIFO, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_fifo_ctor,
+		.dtor = nv04_fifo_dtor,
+		.init = nv04_fifo_init,
+		.fini = _nouveau_fifo_fini,
+	},
+};
diff --git a/nvkm/engine/fifo/nv04.h b/nvkm/engine/fifo/nv04.h
new file mode 100644
index 000000000..496a4b4fd
--- /dev/null
+++ b/nvkm/engine/fifo/nv04.h
@@ -0,0 +1,178 @@
+#ifndef __NV04_FIFO_H__
+#define __NV04_FIFO_H__
+
+#include <engine/fifo.h>
+
+#define NV04_PFIFO_DELAY_0                                 0x00002040
+#define NV04_PFIFO_DMA_TIMESLICE                           0x00002044
+#define NV04_PFIFO_NEXT_CHANNEL                            0x00002050
+#define NV03_PFIFO_INTR_0                                  0x00002100
+#define NV03_PFIFO_INTR_EN_0                               0x00002140
+#    define NV_PFIFO_INTR_CACHE_ERROR                          (1<<0)
+#    define NV_PFIFO_INTR_RUNOUT                               (1<<4)
+#    define NV_PFIFO_INTR_RUNOUT_OVERFLOW                      (1<<8)
+#    define NV_PFIFO_INTR_DMA_PUSHER                          (1<<12)
+#    define NV_PFIFO_INTR_DMA_PT                              (1<<16)
+#    define NV_PFIFO_INTR_SEMAPHORE                           (1<<20)
+#    define NV_PFIFO_INTR_ACQUIRE_TIMEOUT                     (1<<24)
+#define NV03_PFIFO_RAMHT                                   0x00002210
+#define NV03_PFIFO_RAMFC                                   0x00002214
+#define NV03_PFIFO_RAMRO                                   0x00002218
+#define NV40_PFIFO_RAMFC                                   0x00002220
+#define NV03_PFIFO_CACHES                                  0x00002500
+#define NV04_PFIFO_MODE                                    0x00002504
+#define NV04_PFIFO_DMA                                     0x00002508
+#define NV04_PFIFO_SIZE                                    0x0000250c
+#define NV50_PFIFO_CTX_TABLE(c)                        (0x2600+(c)*4)
+#define NV50_PFIFO_CTX_TABLE__SIZE                                128
+#define NV50_PFIFO_CTX_TABLE_CHANNEL_ENABLED                  (1<<31)
+#define NV50_PFIFO_CTX_TABLE_UNK30_BAD                        (1<<30)
+#define NV50_PFIFO_CTX_TABLE_INSTANCE_MASK_G80             0x0FFFFFFF
+#define NV50_PFIFO_CTX_TABLE_INSTANCE_MASK_G84             0x00FFFFFF
+#define NV03_PFIFO_CACHE0_PUSH0                            0x00003000
+#define NV03_PFIFO_CACHE0_PULL0                            0x00003040
+#define NV04_PFIFO_CACHE0_PULL0                            0x00003050
+#define NV04_PFIFO_CACHE0_PULL1                            0x00003054
+#define NV03_PFIFO_CACHE1_PUSH0                            0x00003200
+#define NV03_PFIFO_CACHE1_PUSH1                            0x00003204
+#define NV03_PFIFO_CACHE1_PUSH1_DMA                            (1<<8)
+#define NV40_PFIFO_CACHE1_PUSH1_DMA                           (1<<16)
+#define NV03_PFIFO_CACHE1_PUSH1_CHID_MASK                  0x0000000f
+#define NV10_PFIFO_CACHE1_PUSH1_CHID_MASK                  0x0000001f
+#define NV50_PFIFO_CACHE1_PUSH1_CHID_MASK                  0x0000007f
+#define NV03_PFIFO_CACHE1_PUT                              0x00003210
+#define NV04_PFIFO_CACHE1_DMA_PUSH                         0x00003220
+#define NV04_PFIFO_CACHE1_DMA_FETCH                        0x00003224
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_8_BYTES         0x00000000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_16_BYTES        0x00000008
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_24_BYTES        0x00000010
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_32_BYTES        0x00000018
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_40_BYTES        0x00000020
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_48_BYTES        0x00000028
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_56_BYTES        0x00000030
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_64_BYTES        0x00000038
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_72_BYTES        0x00000040
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_80_BYTES        0x00000048
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_88_BYTES        0x00000050
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_96_BYTES        0x00000058
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_104_BYTES       0x00000060
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_112_BYTES       0x00000068
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_120_BYTES       0x00000070
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES       0x00000078
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_136_BYTES       0x00000080
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_144_BYTES       0x00000088
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_152_BYTES       0x00000090
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_160_BYTES       0x00000098
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_168_BYTES       0x000000A0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_176_BYTES       0x000000A8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_184_BYTES       0x000000B0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_192_BYTES       0x000000B8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_200_BYTES       0x000000C0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_208_BYTES       0x000000C8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_216_BYTES       0x000000D0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_224_BYTES       0x000000D8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_232_BYTES       0x000000E0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_240_BYTES       0x000000E8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_248_BYTES       0x000000F0
+#    define NV_PFIFO_CACHE1_DMA_FETCH_TRIG_256_BYTES       0x000000F8
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE                 0x0000E000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_32_BYTES        0x00000000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_64_BYTES        0x00002000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_96_BYTES        0x00004000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES       0x00006000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_160_BYTES       0x00008000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_192_BYTES       0x0000A000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_224_BYTES       0x0000C000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_SIZE_256_BYTES       0x0000E000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS             0x001F0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_0           0x00000000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_1           0x00010000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_2           0x00020000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_3           0x00030000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_4           0x00040000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_5           0x00050000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_6           0x00060000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_7           0x00070000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8           0x00080000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_9           0x00090000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_10          0x000A0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_11          0x000B0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_12          0x000C0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_13          0x000D0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_14          0x000E0000
+#    define NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_15          0x000F0000
+#    define NV_PFIFO_CACHE1_ENDIAN                         0x80000000
+#    define NV_PFIFO_CACHE1_LITTLE_ENDIAN                  0x7FFFFFFF
+#    define NV_PFIFO_CACHE1_BIG_ENDIAN                     0x80000000
+#define NV04_PFIFO_CACHE1_DMA_STATE                        0x00003228
+#define NV04_PFIFO_CACHE1_DMA_INSTANCE                     0x0000322c
+#define NV04_PFIFO_CACHE1_DMA_CTL                          0x00003230
+#define NV04_PFIFO_CACHE1_DMA_PUT                          0x00003240
+#define NV04_PFIFO_CACHE1_DMA_GET                          0x00003244
+#define NV10_PFIFO_CACHE1_REF_CNT                          0x00003248
+#define NV10_PFIFO_CACHE1_DMA_SUBROUTINE                   0x0000324C
+#define NV03_PFIFO_CACHE1_PULL0                            0x00003240
+#define NV04_PFIFO_CACHE1_PULL0                            0x00003250
+#    define NV04_PFIFO_CACHE1_PULL0_HASH_FAILED            0x00000010
+#    define NV04_PFIFO_CACHE1_PULL0_HASH_BUSY              0x00001000
+#define NV03_PFIFO_CACHE1_PULL1                            0x00003250
+#define NV04_PFIFO_CACHE1_PULL1                            0x00003254
+#define NV04_PFIFO_CACHE1_HASH                             0x00003258
+#define NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT                  0x00003260
+#define NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP                0x00003264
+#define NV10_PFIFO_CACHE1_ACQUIRE_VALUE                    0x00003268
+#define NV10_PFIFO_CACHE1_SEMAPHORE                        0x0000326C
+#define NV03_PFIFO_CACHE1_GET                              0x00003270
+#define NV04_PFIFO_CACHE1_ENGINE                           0x00003280
+#define NV04_PFIFO_CACHE1_DMA_DCOUNT                       0x000032A0
+#define NV40_PFIFO_GRCTX_INSTANCE                          0x000032E0
+#define NV40_PFIFO_UNK32E4                                 0x000032E4
+#define NV04_PFIFO_CACHE1_METHOD(i)                (0x00003800+(i*8))
+#define NV04_PFIFO_CACHE1_DATA(i)                  (0x00003804+(i*8))
+#define NV40_PFIFO_CACHE1_METHOD(i)                (0x00090000+(i*8))
+#define NV40_PFIFO_CACHE1_DATA(i)                  (0x00090004+(i*8))
+
+struct ramfc_desc {
+	unsigned bits:6;
+	unsigned ctxs:5;
+	unsigned ctxp:8;
+	unsigned regs:5;
+	unsigned regp;
+};
+
+struct nv04_fifo_priv {
+	struct nouveau_fifo base;
+	struct ramfc_desc *ramfc_desc;
+	struct nouveau_ramht  *ramht;
+	struct nouveau_gpuobj *ramro;
+	struct nouveau_gpuobj *ramfc;
+};
+
+struct nv04_fifo_base {
+	struct nouveau_fifo_base base;
+};
+
+struct nv04_fifo_chan {
+	struct nouveau_fifo_chan base;
+	u32 subc[8];
+	u32 ramfc;
+};
+
+int  nv04_fifo_object_attach(struct nouveau_object *,
+			     struct nouveau_object *, u32);
+void nv04_fifo_object_detach(struct nouveau_object *, int);
+
+void nv04_fifo_chan_dtor(struct nouveau_object *);
+int  nv04_fifo_chan_init(struct nouveau_object *);
+int  nv04_fifo_chan_fini(struct nouveau_object *, bool suspend);
+
+int  nv04_fifo_context_ctor(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, void *, u32,
+			    struct nouveau_object **);
+
+void nv04_fifo_dtor(struct nouveau_object *);
+int  nv04_fifo_init(struct nouveau_object *);
+void nv04_fifo_pause(struct nouveau_fifo *, unsigned long *);
+void nv04_fifo_start(struct nouveau_fifo *, unsigned long *);
+
+#endif
diff --git a/nvkm/engine/fifo/nv10.c b/nvkm/engine/fifo/nv10.c
new file mode 100644
index 000000000..2c927c1d1
--- /dev/null
+++ b/nvkm/engine/fifo/nv10.c
@@ -0,0 +1,171 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/ramht.h>
+
+#include <subdev/instmem.h>
+#include <subdev/instmem/nv04.h>
+#include <subdev/fb.h>
+
+#include <engine/fifo.h>
+
+#include "nv04.h"
+
+static struct ramfc_desc
+nv10_ramfc[] = {
+	{ 32,  0, 0x00,  0, NV04_PFIFO_CACHE1_DMA_PUT },
+	{ 32,  0, 0x04,  0, NV04_PFIFO_CACHE1_DMA_GET },
+	{ 32,  0, 0x08,  0, NV10_PFIFO_CACHE1_REF_CNT },
+	{ 16,  0, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_INSTANCE },
+	{ 16, 16, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_DCOUNT },
+	{ 32,  0, 0x10,  0, NV04_PFIFO_CACHE1_DMA_STATE },
+	{ 32,  0, 0x14,  0, NV04_PFIFO_CACHE1_DMA_FETCH },
+	{ 32,  0, 0x18,  0, NV04_PFIFO_CACHE1_ENGINE },
+	{ 32,  0, 0x1c,  0, NV04_PFIFO_CACHE1_PULL1 },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO channel objects
+ ******************************************************************************/
+
+static int
+nv10_fifo_chan_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv04_fifo_priv *priv = (void *)engine;
+	struct nv04_fifo_chan *chan;
+	struct nv03_channel_dma_class *args = data;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 0, 0x800000,
+					  0x10000, args->pushbuf,
+					  (1ULL << NVDEV_ENGINE_DMAOBJ) |
+					  (1ULL << NVDEV_ENGINE_SW) |
+					  (1ULL << NVDEV_ENGINE_GR), &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->object_attach = nv04_fifo_object_attach;
+	nv_parent(chan)->object_detach = nv04_fifo_object_detach;
+	nv_parent(chan)->context_attach = nv04_fifo_context_attach;
+	chan->ramfc = chan->base.chid * 32;
+
+	nv_wo32(priv->ramfc, chan->ramfc + 0x00, args->offset);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x04, args->offset);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x0c, chan->base.pushgpu->addr >> 4);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x14,
+			     NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+			     NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+#ifdef __BIG_ENDIAN
+			     NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+			     NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	return 0;
+}
+
+static struct nouveau_ofuncs
+nv10_fifo_ofuncs = {
+	.ctor = nv10_fifo_chan_ctor,
+	.dtor = nv04_fifo_chan_dtor,
+	.init = nv04_fifo_chan_init,
+	.fini = nv04_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_oclass
+nv10_fifo_sclass[] = {
+	{ NV10_CHANNEL_DMA_CLASS, &nv10_fifo_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO context - basically just the instmem reserved for the channel
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv10_fifo_cclass = {
+	.handle = NV_ENGCTX(FIFO, 0x10),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_fifo_context_ctor,
+		.dtor = _nouveau_fifo_context_dtor,
+		.init = _nouveau_fifo_context_init,
+		.fini = _nouveau_fifo_context_fini,
+		.rd32 = _nouveau_fifo_context_rd32,
+		.wr32 = _nouveau_fifo_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PFIFO engine
+ ******************************************************************************/
+
+static int
+nv10_fifo_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv04_instmem_priv *imem = nv04_instmem(parent);
+	struct nv04_fifo_priv *priv;
+	int ret;
+
+	ret = nouveau_fifo_create(parent, engine, oclass, 0, 31, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nouveau_ramht_ref(imem->ramht, &priv->ramht);
+	nouveau_gpuobj_ref(imem->ramro, &priv->ramro);
+	nouveau_gpuobj_ref(imem->ramfc, &priv->ramfc);
+
+	nv_subdev(priv)->unit = 0x00000100;
+	nv_subdev(priv)->intr = nv04_fifo_intr;
+	nv_engine(priv)->cclass = &nv10_fifo_cclass;
+	nv_engine(priv)->sclass = nv10_fifo_sclass;
+	priv->base.pause = nv04_fifo_pause;
+	priv->base.start = nv04_fifo_start;
+	priv->ramfc_desc = nv10_ramfc;
+	return 0;
+}
+
+struct nouveau_oclass
+nv10_fifo_oclass = {
+	.handle = NV_ENGINE(FIFO, 0x10),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv10_fifo_ctor,
+		.dtor = nv04_fifo_dtor,
+		.init = nv04_fifo_init,
+		.fini = _nouveau_fifo_fini,
+	},
+};
diff --git a/nvkm/engine/fifo/nv17.c b/nvkm/engine/fifo/nv17.c
new file mode 100644
index 000000000..a9cb51d38
--- /dev/null
+++ b/nvkm/engine/fifo/nv17.c
@@ -0,0 +1,208 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/ramht.h>
+
+#include <subdev/instmem.h>
+#include <subdev/instmem/nv04.h>
+#include <subdev/fb.h>
+
+#include <engine/fifo.h>
+
+#include "nv04.h"
+
+static struct ramfc_desc
+nv17_ramfc[] = {
+	{ 32,  0, 0x00,  0, NV04_PFIFO_CACHE1_DMA_PUT },
+	{ 32,  0, 0x04,  0, NV04_PFIFO_CACHE1_DMA_GET },
+	{ 32,  0, 0x08,  0, NV10_PFIFO_CACHE1_REF_CNT },
+	{ 16,  0, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_INSTANCE },
+	{ 16, 16, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_DCOUNT },
+	{ 32,  0, 0x10,  0, NV04_PFIFO_CACHE1_DMA_STATE },
+	{ 32,  0, 0x14,  0, NV04_PFIFO_CACHE1_DMA_FETCH },
+	{ 32,  0, 0x18,  0, NV04_PFIFO_CACHE1_ENGINE },
+	{ 32,  0, 0x1c,  0, NV04_PFIFO_CACHE1_PULL1 },
+	{ 32,  0, 0x20,  0, NV10_PFIFO_CACHE1_ACQUIRE_VALUE },
+	{ 32,  0, 0x24,  0, NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP },
+	{ 32,  0, 0x28,  0, NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT },
+	{ 32,  0, 0x2c,  0, NV10_PFIFO_CACHE1_SEMAPHORE },
+	{ 32,  0, 0x30,  0, NV10_PFIFO_CACHE1_DMA_SUBROUTINE },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO channel objects
+ ******************************************************************************/
+
+static int
+nv17_fifo_chan_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv04_fifo_priv *priv = (void *)engine;
+	struct nv04_fifo_chan *chan;
+	struct nv03_channel_dma_class *args = data;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 0, 0x800000,
+					  0x10000, args->pushbuf,
+					  (1ULL << NVDEV_ENGINE_DMAOBJ) |
+					  (1ULL << NVDEV_ENGINE_SW) |
+					  (1ULL << NVDEV_ENGINE_GR) |
+					  (1ULL << NVDEV_ENGINE_MPEG), /* NV31- */
+					  &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->object_attach = nv04_fifo_object_attach;
+	nv_parent(chan)->object_detach = nv04_fifo_object_detach;
+	nv_parent(chan)->context_attach = nv04_fifo_context_attach;
+	chan->ramfc = chan->base.chid * 64;
+
+	nv_wo32(priv->ramfc, chan->ramfc + 0x00, args->offset);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x04, args->offset);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x0c, chan->base.pushgpu->addr >> 4);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x14,
+			     NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+			     NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+#ifdef __BIG_ENDIAN
+			     NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+			     NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	return 0;
+}
+
+static struct nouveau_ofuncs
+nv17_fifo_ofuncs = {
+	.ctor = nv17_fifo_chan_ctor,
+	.dtor = nv04_fifo_chan_dtor,
+	.init = nv04_fifo_chan_init,
+	.fini = nv04_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_oclass
+nv17_fifo_sclass[] = {
+	{ NV17_CHANNEL_DMA_CLASS, &nv17_fifo_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO context - basically just the instmem reserved for the channel
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv17_fifo_cclass = {
+	.handle = NV_ENGCTX(FIFO, 0x17),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_fifo_context_ctor,
+		.dtor = _nouveau_fifo_context_dtor,
+		.init = _nouveau_fifo_context_init,
+		.fini = _nouveau_fifo_context_fini,
+		.rd32 = _nouveau_fifo_context_rd32,
+		.wr32 = _nouveau_fifo_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PFIFO engine
+ ******************************************************************************/
+
+static int
+nv17_fifo_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv04_instmem_priv *imem = nv04_instmem(parent);
+	struct nv04_fifo_priv *priv;
+	int ret;
+
+	ret = nouveau_fifo_create(parent, engine, oclass, 0, 31, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nouveau_ramht_ref(imem->ramht, &priv->ramht);
+	nouveau_gpuobj_ref(imem->ramro, &priv->ramro);
+	nouveau_gpuobj_ref(imem->ramfc, &priv->ramfc);
+
+	nv_subdev(priv)->unit = 0x00000100;
+	nv_subdev(priv)->intr = nv04_fifo_intr;
+	nv_engine(priv)->cclass = &nv17_fifo_cclass;
+	nv_engine(priv)->sclass = nv17_fifo_sclass;
+	priv->base.pause = nv04_fifo_pause;
+	priv->base.start = nv04_fifo_start;
+	priv->ramfc_desc = nv17_ramfc;
+	return 0;
+}
+
+static int
+nv17_fifo_init(struct nouveau_object *object)
+{
+	struct nv04_fifo_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_fifo_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, NV04_PFIFO_DELAY_0, 0x000000ff);
+	nv_wr32(priv, NV04_PFIFO_DMA_TIMESLICE, 0x0101ffff);
+
+	nv_wr32(priv, NV03_PFIFO_RAMHT, (0x03 << 24) /* search 128 */ |
+				       ((priv->ramht->bits - 9) << 16) |
+				        (priv->ramht->base.addr >> 8));
+	nv_wr32(priv, NV03_PFIFO_RAMRO, priv->ramro->addr >> 8);
+	nv_wr32(priv, NV03_PFIFO_RAMFC, priv->ramfc->addr >> 8 | 0x00010000);
+
+	nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH1, priv->base.max);
+
+	nv_wr32(priv, NV03_PFIFO_INTR_0, 0xffffffff);
+	nv_wr32(priv, NV03_PFIFO_INTR_EN_0, 0xffffffff);
+
+	nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH0, 1);
+	nv_wr32(priv, NV04_PFIFO_CACHE1_PULL0, 1);
+	nv_wr32(priv, NV03_PFIFO_CACHES, 1);
+	return 0;
+}
+
+struct nouveau_oclass
+nv17_fifo_oclass = {
+	.handle = NV_ENGINE(FIFO, 0x17),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv17_fifo_ctor,
+		.dtor = nv04_fifo_dtor,
+		.init = nv17_fifo_init,
+		.fini = _nouveau_fifo_fini,
+	},
+};
diff --git a/nvkm/engine/fifo/nv40.c b/nvkm/engine/fifo/nv40.c
new file mode 100644
index 000000000..2b1f91721
--- /dev/null
+++ b/nvkm/engine/fifo/nv40.c
@@ -0,0 +1,349 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/ramht.h>
+
+#include <subdev/instmem.h>
+#include <subdev/instmem/nv04.h>
+#include <subdev/fb.h>
+
+#include <engine/fifo.h>
+
+#include "nv04.h"
+
+static struct ramfc_desc
+nv40_ramfc[] = {
+	{ 32,  0, 0x00,  0, NV04_PFIFO_CACHE1_DMA_PUT },
+	{ 32,  0, 0x04,  0, NV04_PFIFO_CACHE1_DMA_GET },
+	{ 32,  0, 0x08,  0, NV10_PFIFO_CACHE1_REF_CNT },
+	{ 32,  0, 0x0c,  0, NV04_PFIFO_CACHE1_DMA_INSTANCE },
+	{ 32,  0, 0x10,  0, NV04_PFIFO_CACHE1_DMA_DCOUNT },
+	{ 32,  0, 0x14,  0, NV04_PFIFO_CACHE1_DMA_STATE },
+	{ 28,  0, 0x18,  0, NV04_PFIFO_CACHE1_DMA_FETCH },
+	{  2, 28, 0x18, 28, 0x002058 },
+	{ 32,  0, 0x1c,  0, NV04_PFIFO_CACHE1_ENGINE },
+	{ 32,  0, 0x20,  0, NV04_PFIFO_CACHE1_PULL1 },
+	{ 32,  0, 0x24,  0, NV10_PFIFO_CACHE1_ACQUIRE_VALUE },
+	{ 32,  0, 0x28,  0, NV10_PFIFO_CACHE1_ACQUIRE_TIMESTAMP },
+	{ 32,  0, 0x2c,  0, NV10_PFIFO_CACHE1_ACQUIRE_TIMEOUT },
+	{ 32,  0, 0x30,  0, NV10_PFIFO_CACHE1_SEMAPHORE },
+	{ 32,  0, 0x34,  0, NV10_PFIFO_CACHE1_DMA_SUBROUTINE },
+	{ 32,  0, 0x38,  0, NV40_PFIFO_GRCTX_INSTANCE },
+	{ 17,  0, 0x3c,  0, NV04_PFIFO_DMA_TIMESLICE },
+	{ 32,  0, 0x40,  0, 0x0032e4 },
+	{ 32,  0, 0x44,  0, 0x0032e8 },
+	{ 32,  0, 0x4c,  0, 0x002088 },
+	{ 32,  0, 0x50,  0, 0x003300 },
+	{ 32,  0, 0x54,  0, 0x00330c },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO channel objects
+ ******************************************************************************/
+
+static int
+nv40_fifo_object_attach(struct nouveau_object *parent,
+			struct nouveau_object *object, u32 handle)
+{
+	struct nv04_fifo_priv *priv = (void *)parent->engine;
+	struct nv04_fifo_chan *chan = (void *)parent;
+	u32 context, chid = chan->base.chid;
+	int ret;
+
+	if (nv_iclass(object, NV_GPUOBJ_CLASS))
+		context = nv_gpuobj(object)->addr >> 4;
+	else
+		context = 0x00000004; /* just non-zero */
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_DMAOBJ:
+	case NVDEV_ENGINE_SW:
+		context |= 0x00000000;
+		break;
+	case NVDEV_ENGINE_GR:
+		context |= 0x00100000;
+		break;
+	case NVDEV_ENGINE_MPEG:
+		context |= 0x00200000;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	context |= chid << 23;
+
+	mutex_lock(&nv_subdev(priv)->mutex);
+	ret = nouveau_ramht_insert(priv->ramht, chid, handle, context);
+	mutex_unlock(&nv_subdev(priv)->mutex);
+	return ret;
+}
+
+static int
+nv40_fifo_context_attach(struct nouveau_object *parent,
+			 struct nouveau_object *engctx)
+{
+	struct nv04_fifo_priv *priv = (void *)parent->engine;
+	struct nv04_fifo_chan *chan = (void *)parent;
+	unsigned long flags;
+	u32 reg, ctx;
+
+	switch (nv_engidx(engctx->engine)) {
+	case NVDEV_ENGINE_SW:
+		return 0;
+	case NVDEV_ENGINE_GR:
+		reg = 0x32e0;
+		ctx = 0x38;
+		break;
+	case NVDEV_ENGINE_MPEG:
+		reg = 0x330c;
+		ctx = 0x54;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	spin_lock_irqsave(&priv->base.lock, flags);
+	nv_engctx(engctx)->addr = nv_gpuobj(engctx)->addr >> 4;
+	nv_mask(priv, 0x002500, 0x00000001, 0x00000000);
+
+	if ((nv_rd32(priv, 0x003204) & priv->base.max) == chan->base.chid)
+		nv_wr32(priv, reg, nv_engctx(engctx)->addr);
+	nv_wo32(priv->ramfc, chan->ramfc + ctx, nv_engctx(engctx)->addr);
+
+	nv_mask(priv, 0x002500, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&priv->base.lock, flags);
+	return 0;
+}
+
+static int
+nv40_fifo_context_detach(struct nouveau_object *parent, bool suspend,
+			 struct nouveau_object *engctx)
+{
+	struct nv04_fifo_priv *priv = (void *)parent->engine;
+	struct nv04_fifo_chan *chan = (void *)parent;
+	unsigned long flags;
+	u32 reg, ctx;
+
+	switch (nv_engidx(engctx->engine)) {
+	case NVDEV_ENGINE_SW:
+		return 0;
+	case NVDEV_ENGINE_GR:
+		reg = 0x32e0;
+		ctx = 0x38;
+		break;
+	case NVDEV_ENGINE_MPEG:
+		reg = 0x330c;
+		ctx = 0x54;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	spin_lock_irqsave(&priv->base.lock, flags);
+	nv_mask(priv, 0x002500, 0x00000001, 0x00000000);
+
+	if ((nv_rd32(priv, 0x003204) & priv->base.max) == chan->base.chid)
+		nv_wr32(priv, reg, 0x00000000);
+	nv_wo32(priv->ramfc, chan->ramfc + ctx, 0x00000000);
+
+	nv_mask(priv, 0x002500, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&priv->base.lock, flags);
+	return 0;
+}
+
+static int
+nv40_fifo_chan_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nv04_fifo_priv *priv = (void *)engine;
+	struct nv04_fifo_chan *chan;
+	struct nv03_channel_dma_class *args = data;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 0, 0xc00000,
+					  0x1000, args->pushbuf,
+					  (1ULL << NVDEV_ENGINE_DMAOBJ) |
+					  (1ULL << NVDEV_ENGINE_SW) |
+					  (1ULL << NVDEV_ENGINE_GR) |
+					  (1ULL << NVDEV_ENGINE_MPEG), &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->context_attach = nv40_fifo_context_attach;
+	nv_parent(chan)->context_detach = nv40_fifo_context_detach;
+	nv_parent(chan)->object_attach = nv40_fifo_object_attach;
+	nv_parent(chan)->object_detach = nv04_fifo_object_detach;
+	chan->ramfc = chan->base.chid * 128;
+
+	nv_wo32(priv->ramfc, chan->ramfc + 0x00, args->offset);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x04, args->offset);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x0c, chan->base.pushgpu->addr >> 4);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x18, 0x30000000 |
+			     NV_PFIFO_CACHE1_DMA_FETCH_TRIG_128_BYTES |
+			     NV_PFIFO_CACHE1_DMA_FETCH_SIZE_128_BYTES |
+#ifdef __BIG_ENDIAN
+			     NV_PFIFO_CACHE1_BIG_ENDIAN |
+#endif
+			     NV_PFIFO_CACHE1_DMA_FETCH_MAX_REQS_8);
+	nv_wo32(priv->ramfc, chan->ramfc + 0x3c, 0x0001ffff);
+	return 0;
+}
+
+static struct nouveau_ofuncs
+nv40_fifo_ofuncs = {
+	.ctor = nv40_fifo_chan_ctor,
+	.dtor = nv04_fifo_chan_dtor,
+	.init = nv04_fifo_chan_init,
+	.fini = nv04_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_oclass
+nv40_fifo_sclass[] = {
+	{ NV40_CHANNEL_DMA_CLASS, &nv40_fifo_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO context - basically just the instmem reserved for the channel
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv40_fifo_cclass = {
+	.handle = NV_ENGCTX(FIFO, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_fifo_context_ctor,
+		.dtor = _nouveau_fifo_context_dtor,
+		.init = _nouveau_fifo_context_init,
+		.fini = _nouveau_fifo_context_fini,
+		.rd32 = _nouveau_fifo_context_rd32,
+		.wr32 = _nouveau_fifo_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PFIFO engine
+ ******************************************************************************/
+
+static int
+nv40_fifo_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv04_instmem_priv *imem = nv04_instmem(parent);
+	struct nv04_fifo_priv *priv;
+	int ret;
+
+	ret = nouveau_fifo_create(parent, engine, oclass, 0, 31, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nouveau_ramht_ref(imem->ramht, &priv->ramht);
+	nouveau_gpuobj_ref(imem->ramro, &priv->ramro);
+	nouveau_gpuobj_ref(imem->ramfc, &priv->ramfc);
+
+	nv_subdev(priv)->unit = 0x00000100;
+	nv_subdev(priv)->intr = nv04_fifo_intr;
+	nv_engine(priv)->cclass = &nv40_fifo_cclass;
+	nv_engine(priv)->sclass = nv40_fifo_sclass;
+	priv->base.pause = nv04_fifo_pause;
+	priv->base.start = nv04_fifo_start;
+	priv->ramfc_desc = nv40_ramfc;
+	return 0;
+}
+
+static int
+nv40_fifo_init(struct nouveau_object *object)
+{
+	struct nv04_fifo_priv *priv = (void *)object;
+	struct nouveau_fb *pfb = nouveau_fb(object);
+	int ret;
+
+	ret = nouveau_fifo_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x002040, 0x000000ff);
+	nv_wr32(priv, 0x002044, 0x2101ffff);
+	nv_wr32(priv, 0x002058, 0x00000001);
+
+	nv_wr32(priv, NV03_PFIFO_RAMHT, (0x03 << 24) /* search 128 */ |
+				       ((priv->ramht->bits - 9) << 16) |
+				        (priv->ramht->base.addr >> 8));
+	nv_wr32(priv, NV03_PFIFO_RAMRO, priv->ramro->addr >> 8);
+
+	switch (nv_device(priv)->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		nv_wr32(priv, 0x002230, 0x00000001);
+	case 0x40:
+	case 0x41:
+	case 0x42:
+	case 0x43:
+	case 0x45:
+	case 0x48:
+		nv_wr32(priv, 0x002220, 0x00030002);
+		break;
+	default:
+		nv_wr32(priv, 0x002230, 0x00000000);
+		nv_wr32(priv, 0x002220, ((pfb->ram.size - 512 * 1024 +
+					 priv->ramfc->addr) >> 16) |
+					0x00030000);
+		break;
+	}
+
+	nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH1, priv->base.max);
+
+	nv_wr32(priv, NV03_PFIFO_INTR_0, 0xffffffff);
+	nv_wr32(priv, NV03_PFIFO_INTR_EN_0, 0xffffffff);
+
+	nv_wr32(priv, NV03_PFIFO_CACHE1_PUSH0, 1);
+	nv_wr32(priv, NV04_PFIFO_CACHE1_PULL0, 1);
+	nv_wr32(priv, NV03_PFIFO_CACHES, 1);
+	return 0;
+}
+
+struct nouveau_oclass
+nv40_fifo_oclass = {
+	.handle = NV_ENGINE(FIFO, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv40_fifo_ctor,
+		.dtor = nv04_fifo_dtor,
+		.init = nv40_fifo_init,
+		.fini = _nouveau_fifo_fini,
+	},
+};
diff --git a/nvkm/engine/fifo/nv50.c b/nvkm/engine/fifo/nv50.c
new file mode 100644
index 000000000..bd096364f
--- /dev/null
+++ b/nvkm/engine/fifo/nv50.c
@@ -0,0 +1,504 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/client.h>
+#include <core/engctx.h>
+#include <core/ramht.h>
+#include <core/class.h>
+#include <core/math.h>
+
+#include <subdev/timer.h>
+#include <subdev/bar.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+
+#include "nv50.h"
+
+/*******************************************************************************
+ * FIFO channel objects
+ ******************************************************************************/
+
+void
+nv50_fifo_playlist_update(struct nv50_fifo_priv *priv)
+{
+	struct nouveau_bar *bar = nouveau_bar(priv);
+	struct nouveau_gpuobj *cur;
+	int i, p;
+
+	cur = priv->playlist[priv->cur_playlist];
+	priv->cur_playlist = !priv->cur_playlist;
+
+	for (i = priv->base.min, p = 0; i < priv->base.max; i++) {
+		if (nv_rd32(priv, 0x002600 + (i * 4)) & 0x80000000)
+			nv_wo32(cur, p++ * 4, i);
+	}
+
+	bar->flush(bar);
+
+	nv_wr32(priv, 0x0032f4, cur->addr >> 12);
+	nv_wr32(priv, 0x0032ec, p);
+	nv_wr32(priv, 0x002500, 0x00000101);
+}
+
+static int
+nv50_fifo_context_attach(struct nouveau_object *parent,
+			 struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nv50_fifo_base *base = (void *)parent->parent;
+	struct nouveau_gpuobj *ectx = (void *)object;
+	u64 limit = ectx->addr + ectx->size - 1;
+	u64 start = ectx->addr;
+	u32 addr;
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_SW   : return 0;
+	case NVDEV_ENGINE_GR   : addr = 0x0000; break;
+	case NVDEV_ENGINE_MPEG : addr = 0x0060; break;
+	default:
+		return -EINVAL;
+	}
+
+	nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
+	nv_wo32(base->eng, addr + 0x00, 0x00190000);
+	nv_wo32(base->eng, addr + 0x04, lower_32_bits(limit));
+	nv_wo32(base->eng, addr + 0x08, lower_32_bits(start));
+	nv_wo32(base->eng, addr + 0x0c, upper_32_bits(limit) << 24 |
+					upper_32_bits(start));
+	nv_wo32(base->eng, addr + 0x10, 0x00000000);
+	nv_wo32(base->eng, addr + 0x14, 0x00000000);
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nv50_fifo_context_detach(struct nouveau_object *parent, bool suspend,
+			 struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nv50_fifo_priv *priv = (void *)parent->engine;
+	struct nv50_fifo_base *base = (void *)parent->parent;
+	struct nv50_fifo_chan *chan = (void *)parent;
+	u32 addr, me;
+	int ret = 0;
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_SW   : return 0;
+	case NVDEV_ENGINE_GR   : addr = 0x0000; break;
+	case NVDEV_ENGINE_MPEG : addr = 0x0060; break;
+	default:
+		return -EINVAL;
+	}
+
+	/* HW bug workaround:
+	 *
+	 * PFIFO will hang forever if the connected engines don't report
+	 * that they've processed the context switch request.
+	 *
+	 * In order for the kickoff to work, we need to ensure all the
+	 * connected engines are in a state where they can answer.
+	 *
+	 * Newer chipsets don't seem to suffer from this issue, and well,
+	 * there's also a "ignore these engines" bitmask reg we can use
+	 * if we hit the issue there..
+	 */
+	me = nv_mask(priv, 0x00b860, 0x00000001, 0x00000001);
+
+	/* do the kickoff... */
+	nv_wr32(priv, 0x0032fc, nv_gpuobj(base)->addr >> 12);
+	if (!nv_wait_ne(priv, 0x0032fc, 0xffffffff, 0xffffffff)) {
+		nv_error(priv, "channel %d unload timeout\n", chan->base.chid);
+		if (suspend)
+			ret = -EBUSY;
+	}
+	nv_wr32(priv, 0x00b860, me);
+
+	if (ret == 0) {
+		nv_wo32(base->eng, addr + 0x00, 0x00000000);
+		nv_wo32(base->eng, addr + 0x04, 0x00000000);
+		nv_wo32(base->eng, addr + 0x08, 0x00000000);
+		nv_wo32(base->eng, addr + 0x0c, 0x00000000);
+		nv_wo32(base->eng, addr + 0x10, 0x00000000);
+		nv_wo32(base->eng, addr + 0x14, 0x00000000);
+		bar->flush(bar);
+	}
+
+	return ret;
+}
+
+static int
+nv50_fifo_object_attach(struct nouveau_object *parent,
+			struct nouveau_object *object, u32 handle)
+{
+	struct nv50_fifo_chan *chan = (void *)parent;
+	u32 context;
+
+	if (nv_iclass(object, NV_GPUOBJ_CLASS))
+		context = nv_gpuobj(object)->node->offset >> 4;
+	else
+		context = 0x00000004; /* just non-zero */
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_DMAOBJ:
+	case NVDEV_ENGINE_SW    : context |= 0x00000000; break;
+	case NVDEV_ENGINE_GR    : context |= 0x00100000; break;
+	case NVDEV_ENGINE_MPEG  : context |= 0x00200000; break;
+	default:
+		return -EINVAL;
+	}
+
+	return nouveau_ramht_insert(chan->ramht, 0, handle, context);
+}
+
+void
+nv50_fifo_object_detach(struct nouveau_object *parent, int cookie)
+{
+	struct nv50_fifo_chan *chan = (void *)parent;
+	nouveau_ramht_remove(chan->ramht, cookie);
+}
+
+static int
+nv50_fifo_chan_ctor_dma(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nv50_fifo_base *base = (void *)parent;
+	struct nv50_fifo_chan *chan;
+	struct nv03_channel_dma_class *args = data;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 0, 0xc00000,
+					  0x2000, args->pushbuf,
+					  (1ULL << NVDEV_ENGINE_DMAOBJ) |
+					  (1ULL << NVDEV_ENGINE_SW) |
+					  (1ULL << NVDEV_ENGINE_GR) |
+					  (1ULL << NVDEV_ENGINE_MPEG), &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->context_attach = nv50_fifo_context_attach;
+	nv_parent(chan)->context_detach = nv50_fifo_context_detach;
+	nv_parent(chan)->object_attach = nv50_fifo_object_attach;
+	nv_parent(chan)->object_detach = nv50_fifo_object_detach;
+
+	ret = nouveau_ramht_new(parent, parent, 0x8000, 16, &chan->ramht);
+	if (ret)
+		return ret;
+
+	nv_wo32(base->ramfc, 0x08, lower_32_bits(args->offset));
+	nv_wo32(base->ramfc, 0x0c, upper_32_bits(args->offset));
+	nv_wo32(base->ramfc, 0x10, lower_32_bits(args->offset));
+	nv_wo32(base->ramfc, 0x14, upper_32_bits(args->offset));
+	nv_wo32(base->ramfc, 0x3c, 0x003f6078);
+	nv_wo32(base->ramfc, 0x44, 0x01003fff);
+	nv_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
+	nv_wo32(base->ramfc, 0x4c, 0xffffffff);
+	nv_wo32(base->ramfc, 0x60, 0x7fffffff);
+	nv_wo32(base->ramfc, 0x78, 0x00000000);
+	nv_wo32(base->ramfc, 0x7c, 0x30000001);
+	nv_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				   (4 << 24) /* SEARCH_FULL */ |
+				   (chan->ramht->base.node->offset >> 4));
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nv50_fifo_chan_ctor_ind(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nv50_channel_ind_class *args = data;
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nv50_fifo_base *base = (void *)parent;
+	struct nv50_fifo_chan *chan;
+	u64 ioffset, ilength;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 0, 0xc00000,
+					  0x2000, args->pushbuf,
+					  (1ULL << NVDEV_ENGINE_DMAOBJ) |
+					  (1ULL << NVDEV_ENGINE_SW) |
+					  (1ULL << NVDEV_ENGINE_GR) |
+					  (1ULL << NVDEV_ENGINE_MPEG), &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->context_attach = nv50_fifo_context_attach;
+	nv_parent(chan)->context_detach = nv50_fifo_context_detach;
+	nv_parent(chan)->object_attach = nv50_fifo_object_attach;
+	nv_parent(chan)->object_detach = nv50_fifo_object_detach;
+
+	ret = nouveau_ramht_new(parent, parent, 0x8000, 16, &chan->ramht);
+	if (ret)
+		return ret;
+
+	ioffset = args->ioffset;
+	ilength = log2i(args->ilength / 8);
+
+	nv_wo32(base->ramfc, 0x3c, 0x403f6078);
+	nv_wo32(base->ramfc, 0x44, 0x01003fff);
+	nv_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
+	nv_wo32(base->ramfc, 0x50, lower_32_bits(ioffset));
+	nv_wo32(base->ramfc, 0x54, upper_32_bits(ioffset) | (ilength << 16));
+	nv_wo32(base->ramfc, 0x60, 0x7fffffff);
+	nv_wo32(base->ramfc, 0x78, 0x00000000);
+	nv_wo32(base->ramfc, 0x7c, 0x30000001);
+	nv_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				   (4 << 24) /* SEARCH_FULL */ |
+				   (chan->ramht->base.node->offset >> 4));
+	bar->flush(bar);
+	return 0;
+}
+
+void
+nv50_fifo_chan_dtor(struct nouveau_object *object)
+{
+	struct nv50_fifo_chan *chan = (void *)object;
+	nouveau_ramht_ref(NULL, &chan->ramht);
+	nouveau_fifo_channel_destroy(&chan->base);
+}
+
+static int
+nv50_fifo_chan_init(struct nouveau_object *object)
+{
+	struct nv50_fifo_priv *priv = (void *)object->engine;
+	struct nv50_fifo_base *base = (void *)object->parent;
+	struct nv50_fifo_chan *chan = (void *)object;
+	struct nouveau_gpuobj *ramfc = base->ramfc;
+	u32 chid = chan->base.chid;
+	int ret;
+
+	ret = nouveau_fifo_channel_init(&chan->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x002600 + (chid * 4), 0x80000000 | ramfc->addr >> 12);
+	nv50_fifo_playlist_update(priv);
+	return 0;
+}
+
+int
+nv50_fifo_chan_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_fifo_priv *priv = (void *)object->engine;
+	struct nv50_fifo_chan *chan = (void *)object;
+	u32 chid = chan->base.chid;
+
+	/* remove channel from playlist, fifo will unload context */
+	nv_mask(priv, 0x002600 + (chid * 4), 0x80000000, 0x00000000);
+	nv50_fifo_playlist_update(priv);
+	nv_wr32(priv, 0x002600 + (chid * 4), 0x00000000);
+
+	return nouveau_fifo_channel_fini(&chan->base, suspend);
+}
+
+static struct nouveau_ofuncs
+nv50_fifo_ofuncs_dma = {
+	.ctor = nv50_fifo_chan_ctor_dma,
+	.dtor = nv50_fifo_chan_dtor,
+	.init = nv50_fifo_chan_init,
+	.fini = nv50_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_ofuncs
+nv50_fifo_ofuncs_ind = {
+	.ctor = nv50_fifo_chan_ctor_ind,
+	.dtor = nv50_fifo_chan_dtor,
+	.init = nv50_fifo_chan_init,
+	.fini = nv50_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_oclass
+nv50_fifo_sclass[] = {
+	{ NV50_CHANNEL_DMA_CLASS, &nv50_fifo_ofuncs_dma },
+	{ NV50_CHANNEL_IND_CLASS, &nv50_fifo_ofuncs_ind },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO context - basically just the instmem reserved for the channel
+ ******************************************************************************/
+
+static int
+nv50_fifo_context_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nv50_fifo_base *base;
+	int ret;
+
+	ret = nouveau_fifo_context_create(parent, engine, oclass, NULL, 0x10000,
+				          0x1000, NVOBJ_FLAG_HEAP, &base);
+	*pobject = nv_object(base);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, nv_object(base), 0x0200, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC, &base->ramfc);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, nv_object(base), 0x1200, 0,
+				 NVOBJ_FLAG_ZERO_ALLOC, &base->eng);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, nv_object(base), 0x4000, 0, 0,
+				&base->pgd);
+	if (ret)
+		return ret;
+
+	ret = nouveau_vm_ref(nouveau_client(parent)->vm, &base->vm, base->pgd);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+void
+nv50_fifo_context_dtor(struct nouveau_object *object)
+{
+	struct nv50_fifo_base *base = (void *)object;
+	nouveau_vm_ref(NULL, &base->vm, base->pgd);
+	nouveau_gpuobj_ref(NULL, &base->pgd);
+	nouveau_gpuobj_ref(NULL, &base->eng);
+	nouveau_gpuobj_ref(NULL, &base->ramfc);
+	nouveau_gpuobj_ref(NULL, &base->cache);
+	nouveau_fifo_context_destroy(&base->base);
+}
+
+static struct nouveau_oclass
+nv50_fifo_cclass = {
+	.handle = NV_ENGCTX(FIFO, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_fifo_context_ctor,
+		.dtor = nv50_fifo_context_dtor,
+		.init = _nouveau_fifo_context_init,
+		.fini = _nouveau_fifo_context_fini,
+		.rd32 = _nouveau_fifo_context_rd32,
+		.wr32 = _nouveau_fifo_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PFIFO engine
+ ******************************************************************************/
+
+static int
+nv50_fifo_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_fifo_priv *priv;
+	int ret;
+
+	ret = nouveau_fifo_create(parent, engine, oclass, 1, 127, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 128 * 4, 0x1000, 0,
+				&priv->playlist[0]);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 128 * 4, 0x1000, 0,
+				&priv->playlist[1]);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000100;
+	nv_subdev(priv)->intr = nv04_fifo_intr;
+	nv_engine(priv)->cclass = &nv50_fifo_cclass;
+	nv_engine(priv)->sclass = nv50_fifo_sclass;
+	return 0;
+}
+
+void
+nv50_fifo_dtor(struct nouveau_object *object)
+{
+	struct nv50_fifo_priv *priv = (void *)object;
+
+	nouveau_gpuobj_ref(NULL, &priv->playlist[1]);
+	nouveau_gpuobj_ref(NULL, &priv->playlist[0]);
+
+	nouveau_fifo_destroy(&priv->base);
+}
+
+int
+nv50_fifo_init(struct nouveau_object *object)
+{
+	struct nv50_fifo_priv *priv = (void *)object;
+	int ret, i;
+
+	ret = nouveau_fifo_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_mask(priv, 0x000200, 0x00000100, 0x00000000);
+	nv_mask(priv, 0x000200, 0x00000100, 0x00000100);
+	nv_wr32(priv, 0x00250c, 0x6f3cfc34);
+	nv_wr32(priv, 0x002044, 0x01003fff);
+
+	nv_wr32(priv, 0x002100, 0xffffffff);
+	nv_wr32(priv, 0x002140, 0xffffffff);
+
+	for (i = 0; i < 128; i++)
+		nv_wr32(priv, 0x002600 + (i * 4), 0x00000000);
+	nv50_fifo_playlist_update(priv);
+
+	nv_wr32(priv, 0x003200, 0x00000001);
+	nv_wr32(priv, 0x003250, 0x00000001);
+	nv_wr32(priv, 0x002500, 0x00000001);
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_fifo_oclass = {
+	.handle = NV_ENGINE(FIFO, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_fifo_ctor,
+		.dtor = nv50_fifo_dtor,
+		.init = nv50_fifo_init,
+		.fini = _nouveau_fifo_fini,
+	},
+};
diff --git a/nvkm/engine/fifo/nv50.h b/nvkm/engine/fifo/nv50.h
new file mode 100644
index 000000000..3a9ceb315
--- /dev/null
+++ b/nvkm/engine/fifo/nv50.h
@@ -0,0 +1,36 @@
+#ifndef __NV50_FIFO_H__
+#define __NV50_FIFO_H__
+
+struct nv50_fifo_priv {
+	struct nouveau_fifo base;
+	struct nouveau_gpuobj *playlist[2];
+	int cur_playlist;
+};
+
+struct nv50_fifo_base {
+	struct nouveau_fifo_base base;
+	struct nouveau_gpuobj *ramfc;
+	struct nouveau_gpuobj *cache;
+	struct nouveau_gpuobj *eng;
+	struct nouveau_gpuobj *pgd;
+	struct nouveau_vm *vm;
+};
+
+struct nv50_fifo_chan {
+	struct nouveau_fifo_chan base;
+	u32 subc[8];
+	struct nouveau_ramht *ramht;
+};
+
+void nv50_fifo_playlist_update(struct nv50_fifo_priv *);
+
+void nv50_fifo_object_detach(struct nouveau_object *, int);
+void nv50_fifo_chan_dtor(struct nouveau_object *);
+int  nv50_fifo_chan_fini(struct nouveau_object *, bool);
+
+void nv50_fifo_context_dtor(struct nouveau_object *);
+
+void nv50_fifo_dtor(struct nouveau_object *);
+int  nv50_fifo_init(struct nouveau_object *);
+
+#endif
diff --git a/nvkm/engine/fifo/nv84.c b/nvkm/engine/fifo/nv84.c
new file mode 100644
index 000000000..1eb1c512f
--- /dev/null
+++ b/nvkm/engine/fifo/nv84.c
@@ -0,0 +1,420 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/client.h>
+#include <core/engctx.h>
+#include <core/ramht.h>
+#include <core/class.h>
+#include <core/math.h>
+
+#include <subdev/timer.h>
+#include <subdev/bar.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+
+#include "nv50.h"
+
+/*******************************************************************************
+ * FIFO channel objects
+ ******************************************************************************/
+
+static int
+nv84_fifo_context_attach(struct nouveau_object *parent,
+			 struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nv50_fifo_base *base = (void *)parent->parent;
+	struct nouveau_gpuobj *ectx = (void *)object;
+	u64 limit = ectx->addr + ectx->size - 1;
+	u64 start = ectx->addr;
+	u32 addr;
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_SW   : return 0;
+	case NVDEV_ENGINE_GR   : addr = 0x0020; break;
+	case NVDEV_ENGINE_MPEG : addr = 0x0060; break;
+	case NVDEV_ENGINE_CRYPT: addr = 0x00a0; break;
+	case NVDEV_ENGINE_COPY0: addr = 0x00c0; break;
+	default:
+		return -EINVAL;
+	}
+
+	nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
+	nv_wo32(base->eng, addr + 0x00, 0x00190000);
+	nv_wo32(base->eng, addr + 0x04, lower_32_bits(limit));
+	nv_wo32(base->eng, addr + 0x08, lower_32_bits(start));
+	nv_wo32(base->eng, addr + 0x0c, upper_32_bits(limit) << 24 |
+					upper_32_bits(start));
+	nv_wo32(base->eng, addr + 0x10, 0x00000000);
+	nv_wo32(base->eng, addr + 0x14, 0x00000000);
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nv84_fifo_context_detach(struct nouveau_object *parent, bool suspend,
+			 struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nv50_fifo_priv *priv = (void *)parent->engine;
+	struct nv50_fifo_base *base = (void *)parent->parent;
+	struct nv50_fifo_chan *chan = (void *)parent;
+	u32 addr, save, engn;
+	bool done;
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_SW   : return 0;
+	case NVDEV_ENGINE_GR   : engn = 0; addr = 0x0020; break;
+	case NVDEV_ENGINE_MPEG : engn = 1; addr = 0x0060; break;
+	case NVDEV_ENGINE_CRYPT: engn = 4; addr = 0x00a0; break;
+	case NVDEV_ENGINE_COPY0: engn = 2; addr = 0x00c0; break;
+	default:
+		return -EINVAL;
+	}
+
+	save = nv_mask(priv, 0x002520, 0x0000003f, 1 << engn);
+	nv_wr32(priv, 0x0032fc, nv_gpuobj(base)->addr >> 12);
+	done = nv_wait_ne(priv, 0x0032fc, 0xffffffff, 0xffffffff);
+	nv_wr32(priv, 0x002520, save);
+	if (!done) {
+		nv_error(priv, "channel %d unload timeout\n", chan->base.chid);
+		if (suspend)
+			return -EBUSY;
+	}
+
+	nv_wo32(base->eng, addr + 0x00, 0x00000000);
+	nv_wo32(base->eng, addr + 0x04, 0x00000000);
+	nv_wo32(base->eng, addr + 0x08, 0x00000000);
+	nv_wo32(base->eng, addr + 0x0c, 0x00000000);
+	nv_wo32(base->eng, addr + 0x10, 0x00000000);
+	nv_wo32(base->eng, addr + 0x14, 0x00000000);
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nv84_fifo_object_attach(struct nouveau_object *parent,
+			struct nouveau_object *object, u32 handle)
+{
+	struct nv50_fifo_chan *chan = (void *)parent;
+	u32 context;
+
+	if (nv_iclass(object, NV_GPUOBJ_CLASS))
+		context = nv_gpuobj(object)->node->offset >> 4;
+	else
+		context = 0x00000004; /* just non-zero */
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_DMAOBJ:
+	case NVDEV_ENGINE_SW    : context |= 0x00000000; break;
+	case NVDEV_ENGINE_GR    : context |= 0x00100000; break;
+	case NVDEV_ENGINE_MPEG  :
+	case NVDEV_ENGINE_PPP   : context |= 0x00200000; break;
+	case NVDEV_ENGINE_ME    :
+	case NVDEV_ENGINE_COPY0 : context |= 0x00300000; break;
+	case NVDEV_ENGINE_VP    : context |= 0x00400000; break;
+	case NVDEV_ENGINE_CRYPT :
+	case NVDEV_ENGINE_UNK1C1: context |= 0x00500000; break;
+	case NVDEV_ENGINE_BSP   : context |= 0x00600000; break;
+	default:
+		return -EINVAL;
+	}
+
+	return nouveau_ramht_insert(chan->ramht, 0, handle, context);
+}
+
+static int
+nv84_fifo_chan_ctor_dma(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nv50_fifo_base *base = (void *)parent;
+	struct nv50_fifo_chan *chan;
+	struct nv03_channel_dma_class *args = data;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 0, 0xc00000,
+					  0x2000, args->pushbuf,
+					  (1ULL << NVDEV_ENGINE_DMAOBJ) |
+					  (1ULL << NVDEV_ENGINE_SW) |
+					  (1ULL << NVDEV_ENGINE_GR) |
+					  (1ULL << NVDEV_ENGINE_MPEG) |
+					  (1ULL << NVDEV_ENGINE_ME) |
+					  (1ULL << NVDEV_ENGINE_VP) |
+					  (1ULL << NVDEV_ENGINE_CRYPT) |
+					  (1ULL << NVDEV_ENGINE_BSP) |
+					  (1ULL << NVDEV_ENGINE_PPP) |
+					  (1ULL << NVDEV_ENGINE_COPY0) |
+					  (1ULL << NVDEV_ENGINE_UNK1C1), &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	ret = nouveau_ramht_new(parent, parent, 0x8000, 16, &chan->ramht);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->context_attach = nv84_fifo_context_attach;
+	nv_parent(chan)->context_detach = nv84_fifo_context_detach;
+	nv_parent(chan)->object_attach = nv84_fifo_object_attach;
+	nv_parent(chan)->object_detach = nv50_fifo_object_detach;
+
+	nv_wo32(base->ramfc, 0x08, lower_32_bits(args->offset));
+	nv_wo32(base->ramfc, 0x0c, upper_32_bits(args->offset));
+	nv_wo32(base->ramfc, 0x10, lower_32_bits(args->offset));
+	nv_wo32(base->ramfc, 0x14, upper_32_bits(args->offset));
+	nv_wo32(base->ramfc, 0x3c, 0x003f6078);
+	nv_wo32(base->ramfc, 0x44, 0x01003fff);
+	nv_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
+	nv_wo32(base->ramfc, 0x4c, 0xffffffff);
+	nv_wo32(base->ramfc, 0x60, 0x7fffffff);
+	nv_wo32(base->ramfc, 0x78, 0x00000000);
+	nv_wo32(base->ramfc, 0x7c, 0x30000001);
+	nv_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				   (4 << 24) /* SEARCH_FULL */ |
+				   (chan->ramht->base.node->offset >> 4));
+	nv_wo32(base->ramfc, 0x88, base->cache->addr >> 10);
+	nv_wo32(base->ramfc, 0x98, nv_gpuobj(base)->addr >> 12);
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nv84_fifo_chan_ctor_ind(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nv50_fifo_base *base = (void *)parent;
+	struct nv50_fifo_chan *chan;
+	struct nv50_channel_ind_class *args = data;
+	u64 ioffset, ilength;
+	int ret;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 0, 0xc00000,
+					  0x2000, args->pushbuf,
+					  (1ULL << NVDEV_ENGINE_DMAOBJ) |
+					  (1ULL << NVDEV_ENGINE_SW) |
+					  (1ULL << NVDEV_ENGINE_GR) |
+					  (1ULL << NVDEV_ENGINE_MPEG) |
+					  (1ULL << NVDEV_ENGINE_ME) |
+					  (1ULL << NVDEV_ENGINE_VP) |
+					  (1ULL << NVDEV_ENGINE_CRYPT) |
+					  (1ULL << NVDEV_ENGINE_BSP) |
+					  (1ULL << NVDEV_ENGINE_PPP) |
+					  (1ULL << NVDEV_ENGINE_COPY0) |
+					  (1ULL << NVDEV_ENGINE_UNK1C1), &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	ret = nouveau_ramht_new(parent, parent, 0x8000, 16, &chan->ramht);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->context_attach = nv84_fifo_context_attach;
+	nv_parent(chan)->context_detach = nv84_fifo_context_detach;
+	nv_parent(chan)->object_attach = nv84_fifo_object_attach;
+	nv_parent(chan)->object_detach = nv50_fifo_object_detach;
+
+	ioffset = args->ioffset;
+	ilength = log2i(args->ilength / 8);
+
+	nv_wo32(base->ramfc, 0x3c, 0x403f6078);
+	nv_wo32(base->ramfc, 0x44, 0x01003fff);
+	nv_wo32(base->ramfc, 0x48, chan->base.pushgpu->node->offset >> 4);
+	nv_wo32(base->ramfc, 0x50, lower_32_bits(ioffset));
+	nv_wo32(base->ramfc, 0x54, upper_32_bits(ioffset) | (ilength << 16));
+	nv_wo32(base->ramfc, 0x60, 0x7fffffff);
+	nv_wo32(base->ramfc, 0x78, 0x00000000);
+	nv_wo32(base->ramfc, 0x7c, 0x30000001);
+	nv_wo32(base->ramfc, 0x80, ((chan->ramht->bits - 9) << 27) |
+				   (4 << 24) /* SEARCH_FULL */ |
+				   (chan->ramht->base.node->offset >> 4));
+	nv_wo32(base->ramfc, 0x88, base->cache->addr >> 10);
+	nv_wo32(base->ramfc, 0x98, nv_gpuobj(base)->addr >> 12);
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nv84_fifo_chan_init(struct nouveau_object *object)
+{
+	struct nv50_fifo_priv *priv = (void *)object->engine;
+	struct nv50_fifo_base *base = (void *)object->parent;
+	struct nv50_fifo_chan *chan = (void *)object;
+	struct nouveau_gpuobj *ramfc = base->ramfc;
+	u32 chid = chan->base.chid;
+	int ret;
+
+	ret = nouveau_fifo_channel_init(&chan->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x002600 + (chid * 4), 0x80000000 | ramfc->addr >> 8);
+	nv50_fifo_playlist_update(priv);
+	return 0;
+}
+
+static struct nouveau_ofuncs
+nv84_fifo_ofuncs_dma = {
+	.ctor = nv84_fifo_chan_ctor_dma,
+	.dtor = nv50_fifo_chan_dtor,
+	.init = nv84_fifo_chan_init,
+	.fini = nv50_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_ofuncs
+nv84_fifo_ofuncs_ind = {
+	.ctor = nv84_fifo_chan_ctor_ind,
+	.dtor = nv50_fifo_chan_dtor,
+	.init = nv84_fifo_chan_init,
+	.fini = nv50_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_oclass
+nv84_fifo_sclass[] = {
+	{ NV84_CHANNEL_DMA_CLASS, &nv84_fifo_ofuncs_dma },
+	{ NV84_CHANNEL_IND_CLASS, &nv84_fifo_ofuncs_ind },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO context - basically just the instmem reserved for the channel
+ ******************************************************************************/
+
+static int
+nv84_fifo_context_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nv50_fifo_base *base;
+	int ret;
+
+	ret = nouveau_fifo_context_create(parent, engine, oclass, NULL, 0x10000,
+				          0x1000, NVOBJ_FLAG_HEAP, &base);
+	*pobject = nv_object(base);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, nv_object(base), 0x0200, 0,
+				 NVOBJ_FLAG_ZERO_ALLOC, &base->eng);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, nv_object(base), 0x4000, 0,
+				 0, &base->pgd);
+	if (ret)
+		return ret;
+
+	ret = nouveau_vm_ref(nouveau_client(parent)->vm, &base->vm, base->pgd);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, nv_object(base), 0x1000, 0x400,
+				 NVOBJ_FLAG_ZERO_ALLOC, &base->cache);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, nv_object(base), 0x0100, 0x100,
+				 NVOBJ_FLAG_ZERO_ALLOC, &base->ramfc);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static struct nouveau_oclass
+nv84_fifo_cclass = {
+	.handle = NV_ENGCTX(FIFO, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv84_fifo_context_ctor,
+		.dtor = nv50_fifo_context_dtor,
+		.init = _nouveau_fifo_context_init,
+		.fini = _nouveau_fifo_context_fini,
+		.rd32 = _nouveau_fifo_context_rd32,
+		.wr32 = _nouveau_fifo_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PFIFO engine
+ ******************************************************************************/
+
+static int
+nv84_fifo_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_fifo_priv *priv;
+	int ret;
+
+	ret = nouveau_fifo_create(parent, engine, oclass, 1, 127, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 128 * 4, 0x1000, 0,
+				&priv->playlist[0]);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 128 * 4, 0x1000, 0,
+				&priv->playlist[1]);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000100;
+	nv_subdev(priv)->intr = nv04_fifo_intr;
+	nv_engine(priv)->cclass = &nv84_fifo_cclass;
+	nv_engine(priv)->sclass = nv84_fifo_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nv84_fifo_oclass = {
+	.handle = NV_ENGINE(FIFO, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv84_fifo_ctor,
+		.dtor = nv50_fifo_dtor,
+		.init = nv50_fifo_init,
+		.fini = _nouveau_fifo_fini,
+	},
+};
diff --git a/nvkm/engine/fifo/nvc0.c b/nvkm/engine/fifo/nvc0.c
new file mode 100644
index 000000000..b4365dde1
--- /dev/null
+++ b/nvkm/engine/fifo/nvc0.c
@@ -0,0 +1,655 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/client.h>
+#include <core/handle.h>
+#include <core/namedb.h>
+#include <core/gpuobj.h>
+#include <core/engctx.h>
+#include <core/class.h>
+#include <core/math.h>
+#include <core/enum.h>
+
+#include <subdev/timer.h>
+#include <subdev/bar.h>
+#include <subdev/vm.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+
+struct nvc0_fifo_priv {
+	struct nouveau_fifo base;
+	struct nouveau_gpuobj *playlist[2];
+	int cur_playlist;
+	struct {
+		struct nouveau_gpuobj *mem;
+		struct nouveau_vma bar;
+	} user;
+	int spoon_nr;
+};
+
+struct nvc0_fifo_base {
+	struct nouveau_fifo_base base;
+	struct nouveau_gpuobj *pgd;
+	struct nouveau_vm *vm;
+};
+
+struct nvc0_fifo_chan {
+	struct nouveau_fifo_chan base;
+};
+
+/*******************************************************************************
+ * FIFO channel objects
+ ******************************************************************************/
+
+static void
+nvc0_fifo_playlist_update(struct nvc0_fifo_priv *priv)
+{
+	struct nouveau_bar *bar = nouveau_bar(priv);
+	struct nouveau_gpuobj *cur;
+	int i, p;
+
+	cur = priv->playlist[priv->cur_playlist];
+	priv->cur_playlist = !priv->cur_playlist;
+
+	for (i = 0, p = 0; i < 128; i++) {
+		if (!(nv_rd32(priv, 0x003004 + (i * 8)) & 1))
+			continue;
+		nv_wo32(cur, p + 0, i);
+		nv_wo32(cur, p + 4, 0x00000004);
+		p += 8;
+	}
+	bar->flush(bar);
+
+	nv_wr32(priv, 0x002270, cur->addr >> 12);
+	nv_wr32(priv, 0x002274, 0x01f00000 | (p >> 3));
+	if (!nv_wait(priv, 0x00227c, 0x00100000, 0x00000000))
+		nv_error(priv, "playlist update failed\n");
+}
+
+static int
+nvc0_fifo_context_attach(struct nouveau_object *parent,
+			 struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nvc0_fifo_base *base = (void *)parent->parent;
+	struct nouveau_engctx *ectx = (void *)object;
+	u32 addr;
+	int ret;
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_SW   : return 0;
+	case NVDEV_ENGINE_GR   : addr = 0x0210; break;
+	case NVDEV_ENGINE_COPY0: addr = 0x0230; break;
+	case NVDEV_ENGINE_COPY1: addr = 0x0240; break;
+	case NVDEV_ENGINE_BSP  : addr = 0x0270; break;
+	case NVDEV_ENGINE_VP   : addr = 0x0250; break;
+	case NVDEV_ENGINE_PPP  : addr = 0x0260; break;
+	default:
+		return -EINVAL;
+	}
+
+	if (!ectx->vma.node) {
+		ret = nouveau_gpuobj_map_vm(nv_gpuobj(ectx), base->vm,
+					    NV_MEM_ACCESS_RW, &ectx->vma);
+		if (ret)
+			return ret;
+
+		nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
+	}
+
+	nv_wo32(base, addr + 0x00, lower_32_bits(ectx->vma.offset) | 4);
+	nv_wo32(base, addr + 0x04, upper_32_bits(ectx->vma.offset));
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nvc0_fifo_context_detach(struct nouveau_object *parent, bool suspend,
+			 struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nvc0_fifo_priv *priv = (void *)parent->engine;
+	struct nvc0_fifo_base *base = (void *)parent->parent;
+	struct nvc0_fifo_chan *chan = (void *)parent;
+	u32 addr;
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_SW   : return 0;
+	case NVDEV_ENGINE_GR   : addr = 0x0210; break;
+	case NVDEV_ENGINE_COPY0: addr = 0x0230; break;
+	case NVDEV_ENGINE_COPY1: addr = 0x0240; break;
+	case NVDEV_ENGINE_BSP  : addr = 0x0270; break;
+	case NVDEV_ENGINE_VP   : addr = 0x0250; break;
+	case NVDEV_ENGINE_PPP  : addr = 0x0260; break;
+	default:
+		return -EINVAL;
+	}
+
+	nv_wr32(priv, 0x002634, chan->base.chid);
+	if (!nv_wait(priv, 0x002634, 0xffffffff, chan->base.chid)) {
+		nv_error(priv, "channel %d kick timeout\n", chan->base.chid);
+		if (suspend)
+			return -EBUSY;
+	}
+
+	nv_wo32(base, addr + 0x00, 0x00000000);
+	nv_wo32(base, addr + 0x04, 0x00000000);
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nvc0_fifo_chan_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nvc0_fifo_priv *priv = (void *)engine;
+	struct nvc0_fifo_base *base = (void *)parent;
+	struct nvc0_fifo_chan *chan;
+	struct nv50_channel_ind_class *args = data;
+	u64 usermem, ioffset, ilength;
+	int ret, i;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 1,
+					  priv->user.bar.offset, 0x1000,
+					  args->pushbuf,
+					  (1ULL << NVDEV_ENGINE_SW) |
+					  (1ULL << NVDEV_ENGINE_GR) |
+					  (1ULL << NVDEV_ENGINE_COPY0) |
+					  (1ULL << NVDEV_ENGINE_COPY1) |
+					  (1ULL << NVDEV_ENGINE_BSP) |
+					  (1ULL << NVDEV_ENGINE_VP) |
+					  (1ULL << NVDEV_ENGINE_PPP), &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->context_attach = nvc0_fifo_context_attach;
+	nv_parent(chan)->context_detach = nvc0_fifo_context_detach;
+
+	usermem = chan->base.chid * 0x1000;
+	ioffset = args->ioffset;
+	ilength = log2i(args->ilength / 8);
+
+	for (i = 0; i < 0x1000; i += 4)
+		nv_wo32(priv->user.mem, usermem + i, 0x00000000);
+
+	nv_wo32(base, 0x08, lower_32_bits(priv->user.mem->addr + usermem));
+	nv_wo32(base, 0x0c, upper_32_bits(priv->user.mem->addr + usermem));
+	nv_wo32(base, 0x10, 0x0000face);
+	nv_wo32(base, 0x30, 0xfffff902);
+	nv_wo32(base, 0x48, lower_32_bits(ioffset));
+	nv_wo32(base, 0x4c, upper_32_bits(ioffset) | (ilength << 16));
+	nv_wo32(base, 0x54, 0x00000002);
+	nv_wo32(base, 0x84, 0x20400000);
+	nv_wo32(base, 0x94, 0x30000001);
+	nv_wo32(base, 0x9c, 0x00000100);
+	nv_wo32(base, 0xa4, 0x1f1f1f1f);
+	nv_wo32(base, 0xa8, 0x1f1f1f1f);
+	nv_wo32(base, 0xac, 0x0000001f);
+	nv_wo32(base, 0xb8, 0xf8000000);
+	nv_wo32(base, 0xf8, 0x10003080); /* 0x002310 */
+	nv_wo32(base, 0xfc, 0x10000010); /* 0x002350 */
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nvc0_fifo_chan_init(struct nouveau_object *object)
+{
+	struct nouveau_gpuobj *base = nv_gpuobj(object->parent);
+	struct nvc0_fifo_priv *priv = (void *)object->engine;
+	struct nvc0_fifo_chan *chan = (void *)object;
+	u32 chid = chan->base.chid;
+	int ret;
+
+	ret = nouveau_fifo_channel_init(&chan->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x003000 + (chid * 8), 0xc0000000 | base->addr >> 12);
+	nv_wr32(priv, 0x003004 + (chid * 8), 0x001f0001);
+	nvc0_fifo_playlist_update(priv);
+	return 0;
+}
+
+static int
+nvc0_fifo_chan_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nvc0_fifo_priv *priv = (void *)object->engine;
+	struct nvc0_fifo_chan *chan = (void *)object;
+	u32 chid = chan->base.chid;
+
+	nv_mask(priv, 0x003004 + (chid * 8), 0x00000001, 0x00000000);
+	nvc0_fifo_playlist_update(priv);
+	nv_wr32(priv, 0x003000 + (chid * 8), 0x00000000);
+
+	return nouveau_fifo_channel_fini(&chan->base, suspend);
+}
+
+static struct nouveau_ofuncs
+nvc0_fifo_ofuncs = {
+	.ctor = nvc0_fifo_chan_ctor,
+	.dtor = _nouveau_fifo_channel_dtor,
+	.init = nvc0_fifo_chan_init,
+	.fini = nvc0_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_oclass
+nvc0_fifo_sclass[] = {
+	{ NVC0_CHANNEL_IND_CLASS, &nvc0_fifo_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO context - instmem heap and vm setup
+ ******************************************************************************/
+
+static int
+nvc0_fifo_context_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nvc0_fifo_base *base;
+	int ret;
+
+	ret = nouveau_fifo_context_create(parent, engine, oclass, NULL, 0x1000,
+				          0x1000, NVOBJ_FLAG_ZERO_ALLOC |
+					  NVOBJ_FLAG_HEAP, &base);
+	*pobject = nv_object(base);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x10000, 0x1000, 0, &base->pgd);
+	if (ret)
+		return ret;
+
+	nv_wo32(base, 0x0200, lower_32_bits(base->pgd->addr));
+	nv_wo32(base, 0x0204, upper_32_bits(base->pgd->addr));
+	nv_wo32(base, 0x0208, 0xffffffff);
+	nv_wo32(base, 0x020c, 0x000000ff);
+
+	ret = nouveau_vm_ref(nouveau_client(parent)->vm, &base->vm, base->pgd);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static void
+nvc0_fifo_context_dtor(struct nouveau_object *object)
+{
+	struct nvc0_fifo_base *base = (void *)object;
+	nouveau_vm_ref(NULL, &base->vm, base->pgd);
+	nouveau_gpuobj_ref(NULL, &base->pgd);
+	nouveau_fifo_context_destroy(&base->base);
+}
+
+static struct nouveau_oclass
+nvc0_fifo_cclass = {
+	.handle = NV_ENGCTX(FIFO, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_fifo_context_ctor,
+		.dtor = nvc0_fifo_context_dtor,
+		.init = _nouveau_fifo_context_init,
+		.fini = _nouveau_fifo_context_fini,
+		.rd32 = _nouveau_fifo_context_rd32,
+		.wr32 = _nouveau_fifo_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PFIFO engine
+ ******************************************************************************/
+
+static const struct nouveau_enum nvc0_fifo_fault_unit[] = {
+	{ 0x00, "PGRAPH" },
+	{ 0x03, "PEEPHOLE" },
+	{ 0x04, "BAR1" },
+	{ 0x05, "BAR3" },
+	{ 0x07, "PFIFO" },
+	{ 0x10, "PBSP" },
+	{ 0x11, "PPPP" },
+	{ 0x13, "PCOUNTER" },
+	{ 0x14, "PVP" },
+	{ 0x15, "PCOPY0" },
+	{ 0x16, "PCOPY1" },
+	{ 0x17, "PDAEMON" },
+	{}
+};
+
+static const struct nouveau_enum nvc0_fifo_fault_reason[] = {
+	{ 0x00, "PT_NOT_PRESENT" },
+	{ 0x01, "PT_TOO_SHORT" },
+	{ 0x02, "PAGE_NOT_PRESENT" },
+	{ 0x03, "VM_LIMIT_EXCEEDED" },
+	{ 0x04, "NO_CHANNEL" },
+	{ 0x05, "PAGE_SYSTEM_ONLY" },
+	{ 0x06, "PAGE_READ_ONLY" },
+	{ 0x0a, "COMPRESSED_SYSRAM" },
+	{ 0x0c, "INVALID_STORAGE_TYPE" },
+	{}
+};
+
+static const struct nouveau_enum nvc0_fifo_fault_hubclient[] = {
+	{ 0x01, "PCOPY0" },
+	{ 0x02, "PCOPY1" },
+	{ 0x04, "DISPATCH" },
+	{ 0x05, "CTXCTL" },
+	{ 0x06, "PFIFO" },
+	{ 0x07, "BAR_READ" },
+	{ 0x08, "BAR_WRITE" },
+	{ 0x0b, "PVP" },
+	{ 0x0c, "PPPP" },
+	{ 0x0d, "PBSP" },
+	{ 0x11, "PCOUNTER" },
+	{ 0x12, "PDAEMON" },
+	{ 0x14, "CCACHE" },
+	{ 0x15, "CCACHE_POST" },
+	{}
+};
+
+static const struct nouveau_enum nvc0_fifo_fault_gpcclient[] = {
+	{ 0x01, "TEX" },
+	{ 0x0c, "ESETUP" },
+	{ 0x0e, "CTXCTL" },
+	{ 0x0f, "PROP" },
+	{}
+};
+
+static const struct nouveau_bitfield nvc0_fifo_subfifo_intr[] = {
+/*	{ 0x00008000, "" }	seen with null ib push */
+	{ 0x00200000, "ILLEGAL_MTHD" },
+	{ 0x00800000, "EMPTY_SUBC" },
+	{}
+};
+
+static void
+nvc0_fifo_isr_vm_fault(struct nvc0_fifo_priv *priv, int unit)
+{
+	u32 inst = nv_rd32(priv, 0x002800 + (unit * 0x10));
+	u32 valo = nv_rd32(priv, 0x002804 + (unit * 0x10));
+	u32 vahi = nv_rd32(priv, 0x002808 + (unit * 0x10));
+	u32 stat = nv_rd32(priv, 0x00280c + (unit * 0x10));
+	u32 client = (stat & 0x00001f00) >> 8;
+
+	switch (unit) {
+	case 3: /* PEEPHOLE */
+		nv_mask(priv, 0x001718, 0x00000000, 0x00000000);
+		break;
+	case 4: /* BAR1 */
+		nv_mask(priv, 0x001704, 0x00000000, 0x00000000);
+		break;
+	case 5: /* BAR3 */
+		nv_mask(priv, 0x001714, 0x00000000, 0x00000000);
+		break;
+	default:
+		break;
+	}
+
+	nv_error(priv, "%s fault at 0x%010llx [", (stat & 0x00000080) ?
+		 "write" : "read", (u64)vahi << 32 | valo);
+	nouveau_enum_print(nvc0_fifo_fault_reason, stat & 0x0000000f);
+	printk("] from ");
+	nouveau_enum_print(nvc0_fifo_fault_unit, unit);
+	if (stat & 0x00000040) {
+		printk("/");
+		nouveau_enum_print(nvc0_fifo_fault_hubclient, client);
+	} else {
+		printk("/GPC%d/", (stat & 0x1f000000) >> 24);
+		nouveau_enum_print(nvc0_fifo_fault_gpcclient, client);
+	}
+	printk(" on channel 0x%010llx\n", (u64)inst << 12);
+}
+
+static int
+nvc0_fifo_swmthd(struct nvc0_fifo_priv *priv, u32 chid, u32 mthd, u32 data)
+{
+	struct nvc0_fifo_chan *chan = NULL;
+	struct nouveau_handle *bind;
+	unsigned long flags;
+	int ret = -EINVAL;
+
+	spin_lock_irqsave(&priv->base.lock, flags);
+	if (likely(chid >= priv->base.min && chid <= priv->base.max))
+		chan = (void *)priv->base.channel[chid];
+	if (unlikely(!chan))
+		goto out;
+
+	bind = nouveau_namedb_get_class(nv_namedb(chan), 0x906e);
+	if (likely(bind)) {
+		if (!mthd || !nv_call(bind->object, mthd, data))
+			ret = 0;
+		nouveau_namedb_put(bind);
+	}
+
+out:
+	spin_unlock_irqrestore(&priv->base.lock, flags);
+	return ret;
+}
+
+static void
+nvc0_fifo_isr_subfifo_intr(struct nvc0_fifo_priv *priv, int unit)
+{
+	u32 stat = nv_rd32(priv, 0x040108 + (unit * 0x2000));
+	u32 addr = nv_rd32(priv, 0x0400c0 + (unit * 0x2000));
+	u32 data = nv_rd32(priv, 0x0400c4 + (unit * 0x2000));
+	u32 chid = nv_rd32(priv, 0x040120 + (unit * 0x2000)) & 0x7f;
+	u32 subc = (addr & 0x00070000) >> 16;
+	u32 mthd = (addr & 0x00003ffc);
+	u32 show = stat;
+
+	if (stat & 0x00200000) {
+		if (mthd == 0x0054) {
+			if (!nvc0_fifo_swmthd(priv, chid, 0x0500, 0x00000000))
+				show &= ~0x00200000;
+		}
+	}
+
+	if (stat & 0x00800000) {
+		if (!nvc0_fifo_swmthd(priv, chid, mthd, data))
+			show &= ~0x00800000;
+	}
+
+	if (show) {
+		nv_error(priv, "SUBFIFO%d:", unit);
+		nouveau_bitfield_print(nvc0_fifo_subfifo_intr, show);
+		printk("\n");
+		nv_error(priv, "SUBFIFO%d: ch %d subc %d mthd 0x%04x "
+			       "data 0x%08x\n",
+			 unit, chid, subc, mthd, data);
+	}
+
+	nv_wr32(priv, 0x0400c0 + (unit * 0x2000), 0x80600008);
+	nv_wr32(priv, 0x040108 + (unit * 0x2000), stat);
+}
+
+static void
+nvc0_fifo_intr(struct nouveau_subdev *subdev)
+{
+	struct nvc0_fifo_priv *priv = (void *)subdev;
+	u32 mask = nv_rd32(priv, 0x002140);
+	u32 stat = nv_rd32(priv, 0x002100) & mask;
+
+	if (stat & 0x00000100) {
+		nv_warn(priv, "unknown status 0x00000100\n");
+		nv_wr32(priv, 0x002100, 0x00000100);
+		stat &= ~0x00000100;
+	}
+
+	if (stat & 0x10000000) {
+		u32 units = nv_rd32(priv, 0x00259c);
+		u32 u = units;
+
+		while (u) {
+			int i = ffs(u) - 1;
+			nvc0_fifo_isr_vm_fault(priv, i);
+			u &= ~(1 << i);
+		}
+
+		nv_wr32(priv, 0x00259c, units);
+		stat &= ~0x10000000;
+	}
+
+	if (stat & 0x20000000) {
+		u32 units = nv_rd32(priv, 0x0025a0);
+		u32 u = units;
+
+		while (u) {
+			int i = ffs(u) - 1;
+			nvc0_fifo_isr_subfifo_intr(priv, i);
+			u &= ~(1 << i);
+		}
+
+		nv_wr32(priv, 0x0025a0, units);
+		stat &= ~0x20000000;
+	}
+
+	if (stat & 0x40000000) {
+		nv_warn(priv, "unknown status 0x40000000\n");
+		nv_mask(priv, 0x002a00, 0x00000000, 0x00000000);
+		stat &= ~0x40000000;
+	}
+
+	if (stat) {
+		nv_fatal(priv, "unhandled status 0x%08x\n", stat);
+		nv_wr32(priv, 0x002100, stat);
+		nv_wr32(priv, 0x002140, 0);
+	}
+}
+
+static int
+nvc0_fifo_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nvc0_fifo_priv *priv;
+	int ret;
+
+	ret = nouveau_fifo_create(parent, engine, oclass, 0, 127, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 0x1000, 0,
+				&priv->playlist[0]);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 0x1000, 0,
+				&priv->playlist[1]);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 128 * 0x1000, 0x1000, 0,
+				&priv->user.mem);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_map(priv->user.mem, NV_MEM_ACCESS_RW,
+				&priv->user.bar);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000100;
+	nv_subdev(priv)->intr = nvc0_fifo_intr;
+	nv_engine(priv)->cclass = &nvc0_fifo_cclass;
+	nv_engine(priv)->sclass = nvc0_fifo_sclass;
+	return 0;
+}
+
+static void
+nvc0_fifo_dtor(struct nouveau_object *object)
+{
+	struct nvc0_fifo_priv *priv = (void *)object;
+
+	nouveau_gpuobj_unmap(&priv->user.bar);
+	nouveau_gpuobj_ref(NULL, &priv->user.mem);
+	nouveau_gpuobj_ref(NULL, &priv->playlist[1]);
+	nouveau_gpuobj_ref(NULL, &priv->playlist[0]);
+
+	nouveau_fifo_destroy(&priv->base);
+}
+
+static int
+nvc0_fifo_init(struct nouveau_object *object)
+{
+	struct nvc0_fifo_priv *priv = (void *)object;
+	int ret, i;
+
+	ret = nouveau_fifo_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x000204, 0xffffffff);
+	nv_wr32(priv, 0x002204, 0xffffffff);
+
+	priv->spoon_nr = hweight32(nv_rd32(priv, 0x002204));
+	nv_debug(priv, "%d subfifo(s)\n", priv->spoon_nr);
+
+	/* assign engines to subfifos */
+	if (priv->spoon_nr >= 3) {
+		nv_wr32(priv, 0x002208, ~(1 << 0)); /* PGRAPH */
+		nv_wr32(priv, 0x00220c, ~(1 << 1)); /* PVP */
+		nv_wr32(priv, 0x002210, ~(1 << 1)); /* PPP */
+		nv_wr32(priv, 0x002214, ~(1 << 1)); /* PBSP */
+		nv_wr32(priv, 0x002218, ~(1 << 2)); /* PCE0 */
+		nv_wr32(priv, 0x00221c, ~(1 << 1)); /* PCE1 */
+	}
+
+	/* PSUBFIFO[n] */
+	for (i = 0; i < priv->spoon_nr; i++) {
+		nv_mask(priv, 0x04013c + (i * 0x2000), 0x10000100, 0x00000000);
+		nv_wr32(priv, 0x040108 + (i * 0x2000), 0xffffffff); /* INTR */
+		nv_wr32(priv, 0x04010c + (i * 0x2000), 0xfffffeff); /* INTREN */
+	}
+
+	nv_mask(priv, 0x002200, 0x00000001, 0x00000001);
+	nv_wr32(priv, 0x002254, 0x10000000 | priv->user.bar.offset >> 12);
+
+	nv_wr32(priv, 0x002a00, 0xffffffff); /* clears PFIFO.INTR bit 30 */
+	nv_wr32(priv, 0x002100, 0xffffffff);
+	nv_wr32(priv, 0x002140, 0xbfffffff);
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_fifo_oclass = {
+	.handle = NV_ENGINE(FIFO, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_fifo_ctor,
+		.dtor = nvc0_fifo_dtor,
+		.init = nvc0_fifo_init,
+		.fini = _nouveau_fifo_fini,
+	},
+};
diff --git a/nvkm/engine/fifo/nve0.c b/nvkm/engine/fifo/nve0.c
new file mode 100644
index 000000000..c930da99c
--- /dev/null
+++ b/nvkm/engine/fifo/nve0.c
@@ -0,0 +1,633 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/client.h>
+#include <core/handle.h>
+#include <core/namedb.h>
+#include <core/gpuobj.h>
+#include <core/engctx.h>
+#include <core/class.h>
+#include <core/math.h>
+#include <core/enum.h>
+
+#include <subdev/timer.h>
+#include <subdev/bar.h>
+#include <subdev/vm.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+
+#define _(a,b) { (a), ((1ULL << (a)) | (b)) }
+static const struct {
+	u64 subdev;
+	u64 mask;
+} fifo_engine[] = {
+	_(NVDEV_ENGINE_GR      , (1ULL << NVDEV_ENGINE_SW)),
+	_(NVDEV_ENGINE_VP      , 0),
+	_(NVDEV_ENGINE_PPP     , 0),
+	_(NVDEV_ENGINE_BSP     , 0),
+	_(NVDEV_ENGINE_COPY0   , 0),
+	_(NVDEV_ENGINE_COPY1   , 0),
+	_(NVDEV_ENGINE_VENC    , 0),
+};
+#undef _
+#define FIFO_ENGINE_NR ARRAY_SIZE(fifo_engine)
+
+struct nve0_fifo_engn {
+	struct nouveau_gpuobj *playlist[2];
+	int cur_playlist;
+};
+
+struct nve0_fifo_priv {
+	struct nouveau_fifo base;
+	struct nve0_fifo_engn engine[FIFO_ENGINE_NR];
+	struct {
+		struct nouveau_gpuobj *mem;
+		struct nouveau_vma bar;
+	} user;
+	int spoon_nr;
+};
+
+struct nve0_fifo_base {
+	struct nouveau_fifo_base base;
+	struct nouveau_gpuobj *pgd;
+	struct nouveau_vm *vm;
+};
+
+struct nve0_fifo_chan {
+	struct nouveau_fifo_chan base;
+	u32 engine;
+};
+
+/*******************************************************************************
+ * FIFO channel objects
+ ******************************************************************************/
+
+static void
+nve0_fifo_playlist_update(struct nve0_fifo_priv *priv, u32 engine)
+{
+	struct nouveau_bar *bar = nouveau_bar(priv);
+	struct nve0_fifo_engn *engn = &priv->engine[engine];
+	struct nouveau_gpuobj *cur;
+	u32 match = (engine << 16) | 0x00000001;
+	int i, p;
+
+	cur = engn->playlist[engn->cur_playlist];
+	if (unlikely(cur == NULL)) {
+		int ret = nouveau_gpuobj_new(nv_object(priv)->parent, NULL,
+					     0x8000, 0x1000, 0, &cur);
+		if (ret) {
+			nv_error(priv, "playlist alloc failed\n");
+			return;
+		}
+
+		engn->playlist[engn->cur_playlist] = cur;
+	}
+
+	engn->cur_playlist = !engn->cur_playlist;
+
+	for (i = 0, p = 0; i < priv->base.max; i++) {
+		u32 ctrl = nv_rd32(priv, 0x800004 + (i * 8)) & 0x001f0001;
+		if (ctrl != match)
+			continue;
+		nv_wo32(cur, p + 0, i);
+		nv_wo32(cur, p + 4, 0x00000000);
+		p += 8;
+	}
+	bar->flush(bar);
+
+	nv_wr32(priv, 0x002270, cur->addr >> 12);
+	nv_wr32(priv, 0x002274, (engine << 20) | (p >> 3));
+	if (!nv_wait(priv, 0x002284 + (engine * 4), 0x00100000, 0x00000000))
+		nv_error(priv, "playlist %d update timeout\n", engine);
+}
+
+static int
+nve0_fifo_context_attach(struct nouveau_object *parent,
+			 struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nve0_fifo_base *base = (void *)parent->parent;
+	struct nouveau_engctx *ectx = (void *)object;
+	u32 addr;
+	int ret;
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_SW   : return 0;
+	case NVDEV_ENGINE_GR   :
+	case NVDEV_ENGINE_COPY0:
+	case NVDEV_ENGINE_COPY1: addr = 0x0210; break;
+	case NVDEV_ENGINE_BSP  : addr = 0x0270; break;
+	case NVDEV_ENGINE_VP   : addr = 0x0250; break;
+	case NVDEV_ENGINE_PPP  : addr = 0x0260; break;
+	default:
+		return -EINVAL;
+	}
+
+	if (!ectx->vma.node) {
+		ret = nouveau_gpuobj_map_vm(nv_gpuobj(ectx), base->vm,
+					    NV_MEM_ACCESS_RW, &ectx->vma);
+		if (ret)
+			return ret;
+
+		nv_engctx(ectx)->addr = nv_gpuobj(base)->addr >> 12;
+	}
+
+	nv_wo32(base, addr + 0x00, lower_32_bits(ectx->vma.offset) | 4);
+	nv_wo32(base, addr + 0x04, upper_32_bits(ectx->vma.offset));
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nve0_fifo_context_detach(struct nouveau_object *parent, bool suspend,
+			 struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nve0_fifo_priv *priv = (void *)parent->engine;
+	struct nve0_fifo_base *base = (void *)parent->parent;
+	struct nve0_fifo_chan *chan = (void *)parent;
+	u32 addr;
+
+	switch (nv_engidx(object->engine)) {
+	case NVDEV_ENGINE_SW   : return 0;
+	case NVDEV_ENGINE_GR   :
+	case NVDEV_ENGINE_COPY0:
+	case NVDEV_ENGINE_COPY1: addr = 0x0210; break;
+	case NVDEV_ENGINE_BSP  : addr = 0x0270; break;
+	case NVDEV_ENGINE_VP   : addr = 0x0250; break;
+	case NVDEV_ENGINE_PPP  : addr = 0x0260; break;
+	default:
+		return -EINVAL;
+	}
+
+	nv_wr32(priv, 0x002634, chan->base.chid);
+	if (!nv_wait(priv, 0x002634, 0xffffffff, chan->base.chid)) {
+		nv_error(priv, "channel %d kick timeout\n", chan->base.chid);
+		if (suspend)
+			return -EBUSY;
+	}
+
+	nv_wo32(base, addr + 0x00, 0x00000000);
+	nv_wo32(base, addr + 0x04, 0x00000000);
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nve0_fifo_chan_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nve0_fifo_priv *priv = (void *)engine;
+	struct nve0_fifo_base *base = (void *)parent;
+	struct nve0_fifo_chan *chan;
+	struct nve0_channel_ind_class *args = data;
+	u64 usermem, ioffset, ilength;
+	int ret, i;
+
+	if (size < sizeof(*args))
+		return -EINVAL;
+
+	for (i = 0; i < FIFO_ENGINE_NR; i++) {
+		if (args->engine & (1 << i)) {
+			if (nouveau_engine(parent, fifo_engine[i].subdev)) {
+				args->engine = (1 << i);
+				break;
+			}
+		}
+	}
+
+	if (i == FIFO_ENGINE_NR)
+		return -ENODEV;
+
+	ret = nouveau_fifo_channel_create(parent, engine, oclass, 1,
+					  priv->user.bar.offset, 0x200,
+					  args->pushbuf,
+					  fifo_engine[i].mask, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv_parent(chan)->context_attach = nve0_fifo_context_attach;
+	nv_parent(chan)->context_detach = nve0_fifo_context_detach;
+	chan->engine = i;
+
+	usermem = chan->base.chid * 0x200;
+	ioffset = args->ioffset;
+	ilength = log2i(args->ilength / 8);
+
+	for (i = 0; i < 0x200; i += 4)
+		nv_wo32(priv->user.mem, usermem + i, 0x00000000);
+
+	nv_wo32(base, 0x08, lower_32_bits(priv->user.mem->addr + usermem));
+	nv_wo32(base, 0x0c, upper_32_bits(priv->user.mem->addr + usermem));
+	nv_wo32(base, 0x10, 0x0000face);
+	nv_wo32(base, 0x30, 0xfffff902);
+	nv_wo32(base, 0x48, lower_32_bits(ioffset));
+	nv_wo32(base, 0x4c, upper_32_bits(ioffset) | (ilength << 16));
+	nv_wo32(base, 0x84, 0x20400000);
+	nv_wo32(base, 0x94, 0x30000001);
+	nv_wo32(base, 0x9c, 0x00000100);
+	nv_wo32(base, 0xac, 0x0000001f);
+	nv_wo32(base, 0xe8, chan->base.chid);
+	nv_wo32(base, 0xb8, 0xf8000000);
+	nv_wo32(base, 0xf8, 0x10003080); /* 0x002310 */
+	nv_wo32(base, 0xfc, 0x10000010); /* 0x002350 */
+	bar->flush(bar);
+	return 0;
+}
+
+static int
+nve0_fifo_chan_init(struct nouveau_object *object)
+{
+	struct nouveau_gpuobj *base = nv_gpuobj(object->parent);
+	struct nve0_fifo_priv *priv = (void *)object->engine;
+	struct nve0_fifo_chan *chan = (void *)object;
+	u32 chid = chan->base.chid;
+	int ret;
+
+	ret = nouveau_fifo_channel_init(&chan->base);
+	if (ret)
+		return ret;
+
+	nv_mask(priv, 0x800004 + (chid * 8), 0x000f0000, chan->engine << 16);
+	nv_wr32(priv, 0x800000 + (chid * 8), 0x80000000 | base->addr >> 12);
+	nv_mask(priv, 0x800004 + (chid * 8), 0x00000400, 0x00000400);
+	nve0_fifo_playlist_update(priv, chan->engine);
+	nv_mask(priv, 0x800004 + (chid * 8), 0x00000400, 0x00000400);
+	return 0;
+}
+
+static int
+nve0_fifo_chan_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nve0_fifo_priv *priv = (void *)object->engine;
+	struct nve0_fifo_chan *chan = (void *)object;
+	u32 chid = chan->base.chid;
+
+	nv_mask(priv, 0x800004 + (chid * 8), 0x00000800, 0x00000800);
+	nve0_fifo_playlist_update(priv, chan->engine);
+	nv_wr32(priv, 0x800000 + (chid * 8), 0x00000000);
+
+	return nouveau_fifo_channel_fini(&chan->base, suspend);
+}
+
+static struct nouveau_ofuncs
+nve0_fifo_ofuncs = {
+	.ctor = nve0_fifo_chan_ctor,
+	.dtor = _nouveau_fifo_channel_dtor,
+	.init = nve0_fifo_chan_init,
+	.fini = nve0_fifo_chan_fini,
+	.rd32 = _nouveau_fifo_channel_rd32,
+	.wr32 = _nouveau_fifo_channel_wr32,
+};
+
+static struct nouveau_oclass
+nve0_fifo_sclass[] = {
+	{ NVE0_CHANNEL_IND_CLASS, &nve0_fifo_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * FIFO context - instmem heap and vm setup
+ ******************************************************************************/
+
+static int
+nve0_fifo_context_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nve0_fifo_base *base;
+	int ret;
+
+	ret = nouveau_fifo_context_create(parent, engine, oclass, NULL, 0x1000,
+				          0x1000, NVOBJ_FLAG_ZERO_ALLOC, &base);
+	*pobject = nv_object(base);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x10000, 0x1000, 0, &base->pgd);
+	if (ret)
+		return ret;
+
+	nv_wo32(base, 0x0200, lower_32_bits(base->pgd->addr));
+	nv_wo32(base, 0x0204, upper_32_bits(base->pgd->addr));
+	nv_wo32(base, 0x0208, 0xffffffff);
+	nv_wo32(base, 0x020c, 0x000000ff);
+
+	ret = nouveau_vm_ref(nouveau_client(parent)->vm, &base->vm, base->pgd);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static void
+nve0_fifo_context_dtor(struct nouveau_object *object)
+{
+	struct nve0_fifo_base *base = (void *)object;
+	nouveau_vm_ref(NULL, &base->vm, base->pgd);
+	nouveau_gpuobj_ref(NULL, &base->pgd);
+	nouveau_fifo_context_destroy(&base->base);
+}
+
+static struct nouveau_oclass
+nve0_fifo_cclass = {
+	.handle = NV_ENGCTX(FIFO, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nve0_fifo_context_ctor,
+		.dtor = nve0_fifo_context_dtor,
+		.init = _nouveau_fifo_context_init,
+		.fini = _nouveau_fifo_context_fini,
+		.rd32 = _nouveau_fifo_context_rd32,
+		.wr32 = _nouveau_fifo_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PFIFO engine
+ ******************************************************************************/
+
+static const struct nouveau_enum nve0_fifo_fault_unit[] = {
+	{}
+};
+
+static const struct nouveau_enum nve0_fifo_fault_reason[] = {
+	{ 0x00, "PT_NOT_PRESENT" },
+	{ 0x01, "PT_TOO_SHORT" },
+	{ 0x02, "PAGE_NOT_PRESENT" },
+	{ 0x03, "VM_LIMIT_EXCEEDED" },
+	{ 0x04, "NO_CHANNEL" },
+	{ 0x05, "PAGE_SYSTEM_ONLY" },
+	{ 0x06, "PAGE_READ_ONLY" },
+	{ 0x0a, "COMPRESSED_SYSRAM" },
+	{ 0x0c, "INVALID_STORAGE_TYPE" },
+	{}
+};
+
+static const struct nouveau_enum nve0_fifo_fault_hubclient[] = {
+	{}
+};
+
+static const struct nouveau_enum nve0_fifo_fault_gpcclient[] = {
+	{}
+};
+
+static const struct nouveau_bitfield nve0_fifo_subfifo_intr[] = {
+	{ 0x00200000, "ILLEGAL_MTHD" },
+	{ 0x00800000, "EMPTY_SUBC" },
+	{}
+};
+
+static void
+nve0_fifo_isr_vm_fault(struct nve0_fifo_priv *priv, int unit)
+{
+	u32 inst = nv_rd32(priv, 0x2800 + (unit * 0x10));
+	u32 valo = nv_rd32(priv, 0x2804 + (unit * 0x10));
+	u32 vahi = nv_rd32(priv, 0x2808 + (unit * 0x10));
+	u32 stat = nv_rd32(priv, 0x280c + (unit * 0x10));
+	u32 client = (stat & 0x00001f00) >> 8;
+
+	nv_error(priv, "PFIFO: %s fault at 0x%010llx [", (stat & 0x00000080) ?
+		       "write" : "read", (u64)vahi << 32 | valo);
+	nouveau_enum_print(nve0_fifo_fault_reason, stat & 0x0000000f);
+	printk("] from ");
+	nouveau_enum_print(nve0_fifo_fault_unit, unit);
+	if (stat & 0x00000040) {
+		printk("/");
+		nouveau_enum_print(nve0_fifo_fault_hubclient, client);
+	} else {
+		printk("/GPC%d/", (stat & 0x1f000000) >> 24);
+		nouveau_enum_print(nve0_fifo_fault_gpcclient, client);
+	}
+	printk(" on channel 0x%010llx\n", (u64)inst << 12);
+}
+
+static int
+nve0_fifo_swmthd(struct nve0_fifo_priv *priv, u32 chid, u32 mthd, u32 data)
+{
+	struct nve0_fifo_chan *chan = NULL;
+	struct nouveau_handle *bind;
+	unsigned long flags;
+	int ret = -EINVAL;
+
+	spin_lock_irqsave(&priv->base.lock, flags);
+	if (likely(chid >= priv->base.min && chid <= priv->base.max))
+		chan = (void *)priv->base.channel[chid];
+	if (unlikely(!chan))
+		goto out;
+
+	bind = nouveau_namedb_get_class(nv_namedb(chan), 0x906e);
+	if (likely(bind)) {
+		if (!mthd || !nv_call(bind->object, mthd, data))
+			ret = 0;
+		nouveau_namedb_put(bind);
+	}
+
+out:
+	spin_unlock_irqrestore(&priv->base.lock, flags);
+	return ret;
+}
+
+static void
+nve0_fifo_isr_subfifo_intr(struct nve0_fifo_priv *priv, int unit)
+{
+	u32 stat = nv_rd32(priv, 0x040108 + (unit * 0x2000));
+	u32 addr = nv_rd32(priv, 0x0400c0 + (unit * 0x2000));
+	u32 data = nv_rd32(priv, 0x0400c4 + (unit * 0x2000));
+	u32 chid = nv_rd32(priv, 0x040120 + (unit * 0x2000)) & 0xfff;
+	u32 subc = (addr & 0x00070000) >> 16;
+	u32 mthd = (addr & 0x00003ffc);
+	u32 show = stat;
+
+	if (stat & 0x00200000) {
+		if (mthd == 0x0054) {
+			if (!nve0_fifo_swmthd(priv, chid, 0x0500, 0x00000000))
+				show &= ~0x00200000;
+		}
+	}
+
+	if (stat & 0x00800000) {
+		if (!nve0_fifo_swmthd(priv, chid, mthd, data))
+			show &= ~0x00800000;
+	}
+
+	if (show) {
+		nv_error(priv, "SUBFIFO%d:", unit);
+		nouveau_bitfield_print(nve0_fifo_subfifo_intr, show);
+		printk("\n");
+		nv_error(priv, "SUBFIFO%d: ch %d subc %d mthd 0x%04x "
+			       "data 0x%08x\n",
+			 unit, chid, subc, mthd, data);
+	}
+
+	nv_wr32(priv, 0x0400c0 + (unit * 0x2000), 0x80600008);
+	nv_wr32(priv, 0x040108 + (unit * 0x2000), stat);
+}
+
+static void
+nve0_fifo_intr(struct nouveau_subdev *subdev)
+{
+	struct nve0_fifo_priv *priv = (void *)subdev;
+	u32 mask = nv_rd32(priv, 0x002140);
+	u32 stat = nv_rd32(priv, 0x002100) & mask;
+
+	if (stat & 0x00000100) {
+		nv_warn(priv, "unknown status 0x00000100\n");
+		nv_wr32(priv, 0x002100, 0x00000100);
+		stat &= ~0x00000100;
+	}
+
+	if (stat & 0x10000000) {
+		u32 units = nv_rd32(priv, 0x00259c);
+		u32 u = units;
+
+		while (u) {
+			int i = ffs(u) - 1;
+			nve0_fifo_isr_vm_fault(priv, i);
+			u &= ~(1 << i);
+		}
+
+		nv_wr32(priv, 0x00259c, units);
+		stat &= ~0x10000000;
+	}
+
+	if (stat & 0x20000000) {
+		u32 units = nv_rd32(priv, 0x0025a0);
+		u32 u = units;
+
+		while (u) {
+			int i = ffs(u) - 1;
+			nve0_fifo_isr_subfifo_intr(priv, i);
+			u &= ~(1 << i);
+		}
+
+		nv_wr32(priv, 0x0025a0, units);
+		stat &= ~0x20000000;
+	}
+
+	if (stat & 0x40000000) {
+		nv_warn(priv, "unknown status 0x40000000\n");
+		nv_mask(priv, 0x002a00, 0x00000000, 0x00000000);
+		stat &= ~0x40000000;
+	}
+
+	if (stat) {
+		nv_fatal(priv, "unhandled status 0x%08x\n", stat);
+		nv_wr32(priv, 0x002100, stat);
+		nv_wr32(priv, 0x002140, 0);
+	}
+}
+
+static int
+nve0_fifo_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nve0_fifo_priv *priv;
+	int ret;
+
+	ret = nouveau_fifo_create(parent, engine, oclass, 0, 4095, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 4096 * 0x200, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC, &priv->user.mem);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_map(priv->user.mem, NV_MEM_ACCESS_RW,
+				&priv->user.bar);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000100;
+	nv_subdev(priv)->intr = nve0_fifo_intr;
+	nv_engine(priv)->cclass = &nve0_fifo_cclass;
+	nv_engine(priv)->sclass = nve0_fifo_sclass;
+	return 0;
+}
+
+static void
+nve0_fifo_dtor(struct nouveau_object *object)
+{
+	struct nve0_fifo_priv *priv = (void *)object;
+	int i;
+
+	nouveau_gpuobj_unmap(&priv->user.bar);
+	nouveau_gpuobj_ref(NULL, &priv->user.mem);
+
+	for (i = 0; i < ARRAY_SIZE(priv->engine); i++) {
+		nouveau_gpuobj_ref(NULL, &priv->engine[i].playlist[1]);
+		nouveau_gpuobj_ref(NULL, &priv->engine[i].playlist[0]);
+	}
+
+	nouveau_fifo_destroy(&priv->base);
+}
+
+static int
+nve0_fifo_init(struct nouveau_object *object)
+{
+	struct nve0_fifo_priv *priv = (void *)object;
+	int ret, i;
+
+	ret = nouveau_fifo_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* enable all available PSUBFIFOs */
+	nv_wr32(priv, 0x000204, 0xffffffff);
+	priv->spoon_nr = hweight32(nv_rd32(priv, 0x000204));
+	nv_debug(priv, "%d subfifo(s)\n", priv->spoon_nr);
+
+	/* PSUBFIFO[n] */
+	for (i = 0; i < priv->spoon_nr; i++) {
+		nv_mask(priv, 0x04013c + (i * 0x2000), 0x10000100, 0x00000000);
+		nv_wr32(priv, 0x040108 + (i * 0x2000), 0xffffffff); /* INTR */
+		nv_wr32(priv, 0x04010c + (i * 0x2000), 0xfffffeff); /* INTREN */
+	}
+
+	nv_wr32(priv, 0x002254, 0x10000000 | priv->user.bar.offset >> 12);
+
+	nv_wr32(priv, 0x002a00, 0xffffffff);
+	nv_wr32(priv, 0x002100, 0xffffffff);
+	nv_wr32(priv, 0x002140, 0xbfffffff);
+	return 0;
+}
+
+struct nouveau_oclass
+nve0_fifo_oclass = {
+	.handle = NV_ENGINE(FIFO, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nve0_fifo_ctor,
+		.dtor = nve0_fifo_dtor,
+		.init = nve0_fifo_init,
+		.fini = _nouveau_fifo_fini,
+	},
+};
diff --git a/nvkm/engine/graph/Makefile.am b/nvkm/engine/graph/Makefile.am
new file mode 100644
index 000000000..567d606d2
--- /dev/null
+++ b/nvkm/engine/graph/Makefile.am
@@ -0,0 +1,47 @@
+noinst_LTLIBRARIES = libgraph.la
+
+BUILT_SOURCES = fuc/hubnvc0.fuc.h \
+		fuc/gpcnvc0.fuc.h \
+		fuc/hubnve0.fuc.h \
+		fuc/gpcnve0.fuc.h
+CLEANFILES = ${BUILT_SOURCES}
+
+libgraph_la_SOURCES = ${BUILT_SOURCES} \
+		      ctxnv40.c \
+		      ctxnv50.c \
+		      ctxnvc0.c \
+		      ctxnve0.c \
+		      nv04.c \
+		      nv10.c \
+		      nv20.c \
+		      nv25.c \
+		      nv2a.c \
+		      nv30.c \
+		      nv34.c \
+		      nv35.c \
+		      nv40.c \
+		      nv50.c \
+		      nvc0.c \
+		      nve0.c
+
+fuc/hubnvc0.fuc.h: fuc/nvc0.fuc fuc/hubnvc0.fuc
+	cd fuc && \
+	m4 hubnvc0.fuc | envyas -a -w -m fuc -V fuc3 -o hubnvc0.fuc.h && \
+	cd ..
+
+fuc/gpcnvc0.fuc.h: fuc/nvc0.fuc fuc/gpcnvc0.fuc
+	cd fuc && \
+	m4 gpcnvc0.fuc | envyas -a -w -m fuc -V fuc3 -o gpcnvc0.fuc.h && \
+	cd ..
+
+fuc/hubnve0.fuc.h: fuc/nve0.fuc fuc/hubnve0.fuc
+	cd fuc && \
+	m4 hubnve0.fuc | envyas -a -w -m fuc -V fuc3 -o hubnve0.fuc.h && \
+	cd ..
+
+fuc/gpcnve0.fuc.h: fuc/nve0.fuc fuc/gpcnve0.fuc
+	cd fuc && \
+	m4 gpcnve0.fuc | envyas -a -w -m fuc -V fuc3 -o gpcnve0.fuc.h && \
+	cd ..
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/graph/ctx.h b/nvkm/engine/graph/ctx.h
new file mode 100644
index 000000000..e1947013d
--- /dev/null
+++ b/nvkm/engine/graph/ctx.h
@@ -0,0 +1,129 @@
+#ifndef __NOUVEAU_GRCTX_H__
+#define __NOUVEAU_GRCTX_H__
+
+struct nouveau_grctx {
+	struct nouveau_device *device;
+
+	enum {
+		NOUVEAU_GRCTX_PROG,
+		NOUVEAU_GRCTX_VALS
+	} mode;
+	void *data;
+
+	u32 ctxprog_max;
+	u32 ctxprog_len;
+	u32 ctxprog_reg;
+	int ctxprog_label[32];
+	u32 ctxvals_pos;
+	u32 ctxvals_base;
+};
+
+static inline void
+cp_out(struct nouveau_grctx *ctx, u32 inst)
+{
+	u32 *ctxprog = ctx->data;
+
+	if (ctx->mode != NOUVEAU_GRCTX_PROG)
+		return;
+
+	BUG_ON(ctx->ctxprog_len == ctx->ctxprog_max);
+	ctxprog[ctx->ctxprog_len++] = inst;
+}
+
+static inline void
+cp_lsr(struct nouveau_grctx *ctx, u32 val)
+{
+	cp_out(ctx, CP_LOAD_SR | val);
+}
+
+static inline void
+cp_ctx(struct nouveau_grctx *ctx, u32 reg, u32 length)
+{
+	ctx->ctxprog_reg = (reg - 0x00400000) >> 2;
+
+	ctx->ctxvals_base = ctx->ctxvals_pos;
+	ctx->ctxvals_pos = ctx->ctxvals_base + length;
+
+	if (length > (CP_CTX_COUNT >> CP_CTX_COUNT_SHIFT)) {
+		cp_lsr(ctx, length);
+		length = 0;
+	}
+
+	cp_out(ctx, CP_CTX | (length << CP_CTX_COUNT_SHIFT) | ctx->ctxprog_reg);
+}
+
+static inline void
+cp_name(struct nouveau_grctx *ctx, int name)
+{
+	u32 *ctxprog = ctx->data;
+	int i;
+
+	if (ctx->mode != NOUVEAU_GRCTX_PROG)
+		return;
+
+	ctx->ctxprog_label[name] = ctx->ctxprog_len;
+	for (i = 0; i < ctx->ctxprog_len; i++) {
+		if ((ctxprog[i] & 0xfff00000) != 0xff400000)
+			continue;
+		if ((ctxprog[i] & CP_BRA_IP) != ((name) << CP_BRA_IP_SHIFT))
+			continue;
+		ctxprog[i] = (ctxprog[i] & 0x00ff00ff) |
+			     (ctx->ctxprog_len << CP_BRA_IP_SHIFT);
+	}
+}
+
+static inline void
+_cp_bra(struct nouveau_grctx *ctx, u32 mod, int flag, int state, int name)
+{
+	int ip = 0;
+
+	if (mod != 2) {
+		ip = ctx->ctxprog_label[name] << CP_BRA_IP_SHIFT;
+		if (ip == 0)
+			ip = 0xff000000 | (name << CP_BRA_IP_SHIFT);
+	}
+
+	cp_out(ctx, CP_BRA | (mod << 18) | ip | flag |
+		    (state ? 0 : CP_BRA_IF_CLEAR));
+}
+#define cp_bra(c, f, s, n) _cp_bra((c), 0, CP_FLAG_##f, CP_FLAG_##f##_##s, n)
+#define cp_cal(c, f, s, n) _cp_bra((c), 1, CP_FLAG_##f, CP_FLAG_##f##_##s, n)
+#define cp_ret(c, f, s) _cp_bra((c), 2, CP_FLAG_##f, CP_FLAG_##f##_##s, 0)
+
+static inline void
+_cp_wait(struct nouveau_grctx *ctx, int flag, int state)
+{
+	cp_out(ctx, CP_WAIT | flag | (state ? CP_WAIT_SET : 0));
+}
+#define cp_wait(c, f, s) _cp_wait((c), CP_FLAG_##f, CP_FLAG_##f##_##s)
+
+static inline void
+_cp_set(struct nouveau_grctx *ctx, int flag, int state)
+{
+	cp_out(ctx, CP_SET | flag | (state ? CP_SET_1 : 0));
+}
+#define cp_set(c, f, s) _cp_set((c), CP_FLAG_##f, CP_FLAG_##f##_##s)
+
+static inline void
+cp_pos(struct nouveau_grctx *ctx, int offset)
+{
+	ctx->ctxvals_pos = offset;
+	ctx->ctxvals_base = ctx->ctxvals_pos;
+
+	cp_lsr(ctx, ctx->ctxvals_pos);
+	cp_out(ctx, CP_SET_CONTEXT_POINTER);
+}
+
+static inline void
+gr_def(struct nouveau_grctx *ctx, u32 reg, u32 val)
+{
+	if (ctx->mode != NOUVEAU_GRCTX_VALS)
+		return;
+
+	reg = (reg - 0x00400000) / 4;
+	reg = (reg - ctx->ctxprog_reg) + ctx->ctxvals_base;
+
+	nv_wo32(ctx->data, reg * 4, val);
+}
+
+#endif
diff --git a/nvkm/engine/graph/ctxnv40.c b/nvkm/engine/graph/ctxnv40.c
new file mode 100644
index 000000000..7bbb1e1b7
--- /dev/null
+++ b/nvkm/engine/graph/ctxnv40.c
@@ -0,0 +1,695 @@
+/*
+ * Copyright 2009 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/gpuobj.h>
+
+/* NVIDIA context programs handle a number of other conditions which are
+ * not implemented in our versions.  It's not clear why NVIDIA context
+ * programs have this code, nor whether it's strictly necessary for
+ * correct operation.  We'll implement additional handling if/when we
+ * discover it's necessary.
+ *
+ * - On context save, NVIDIA set 0x400314 bit 0 to 1 if the "3D state"
+ *   flag is set, this gets saved into the context.
+ * - On context save, the context program for all cards load nsource
+ *   into a flag register and check for ILLEGAL_MTHD.  If it's set,
+ *   opcode 0x60000d is called before resuming normal operation.
+ * - Some context programs check more conditions than the above.  NV44
+ *   checks: ((nsource & 0x0857) || (0x400718 & 0x0100) || (intr & 0x0001))
+ *   and calls 0x60000d before resuming normal operation.
+ * - At the very beginning of NVIDIA's context programs, flag 9 is checked
+ *   and if true 0x800001 is called with count=0, pos=0, the flag is cleared
+ *   and then the ctxprog is aborted.  It looks like a complicated NOP,
+ *   its purpose is unknown.
+ * - In the section of code that loads the per-vs state, NVIDIA check
+ *   flag 10.  If it's set, they only transfer the small 0x300 byte block
+ *   of state + the state for a single vs as opposed to the state for
+ *   all vs units.  It doesn't seem likely that it'll occur in normal
+ *   operation, especially seeing as it appears NVIDIA may have screwed
+ *   up the ctxprogs for some cards and have an invalid instruction
+ *   rather than a cp_lsr(ctx, dwords_for_1_vs_unit) instruction.
+ * - There's a number of places where context offset 0 (where we place
+ *   the PRAMIN offset of the context) is loaded into either 0x408000,
+ *   0x408004 or 0x408008.  Not sure what's up there either.
+ * - The ctxprogs for some cards save 0x400a00 again during the cleanup
+ *   path for auto-loadctx.
+ */
+
+#define CP_FLAG_CLEAR                 0
+#define CP_FLAG_SET                   1
+#define CP_FLAG_SWAP_DIRECTION        ((0 * 32) + 0)
+#define CP_FLAG_SWAP_DIRECTION_LOAD   0
+#define CP_FLAG_SWAP_DIRECTION_SAVE   1
+#define CP_FLAG_USER_SAVE             ((0 * 32) + 5)
+#define CP_FLAG_USER_SAVE_NOT_PENDING 0
+#define CP_FLAG_USER_SAVE_PENDING     1
+#define CP_FLAG_USER_LOAD             ((0 * 32) + 6)
+#define CP_FLAG_USER_LOAD_NOT_PENDING 0
+#define CP_FLAG_USER_LOAD_PENDING     1
+#define CP_FLAG_STATUS                ((3 * 32) + 0)
+#define CP_FLAG_STATUS_IDLE           0
+#define CP_FLAG_STATUS_BUSY           1
+#define CP_FLAG_AUTO_SAVE             ((3 * 32) + 4)
+#define CP_FLAG_AUTO_SAVE_NOT_PENDING 0
+#define CP_FLAG_AUTO_SAVE_PENDING     1
+#define CP_FLAG_AUTO_LOAD             ((3 * 32) + 5)
+#define CP_FLAG_AUTO_LOAD_NOT_PENDING 0
+#define CP_FLAG_AUTO_LOAD_PENDING     1
+#define CP_FLAG_UNK54                 ((3 * 32) + 6)
+#define CP_FLAG_UNK54_CLEAR           0
+#define CP_FLAG_UNK54_SET             1
+#define CP_FLAG_ALWAYS                ((3 * 32) + 8)
+#define CP_FLAG_ALWAYS_FALSE          0
+#define CP_FLAG_ALWAYS_TRUE           1
+#define CP_FLAG_UNK57                 ((3 * 32) + 9)
+#define CP_FLAG_UNK57_CLEAR           0
+#define CP_FLAG_UNK57_SET             1
+
+#define CP_CTX                   0x00100000
+#define CP_CTX_COUNT             0x000fc000
+#define CP_CTX_COUNT_SHIFT               14
+#define CP_CTX_REG               0x00003fff
+#define CP_LOAD_SR               0x00200000
+#define CP_LOAD_SR_VALUE         0x000fffff
+#define CP_BRA                   0x00400000
+#define CP_BRA_IP                0x0000ff00
+#define CP_BRA_IP_SHIFT                   8
+#define CP_BRA_IF_CLEAR          0x00000080
+#define CP_BRA_FLAG              0x0000007f
+#define CP_WAIT                  0x00500000
+#define CP_WAIT_SET              0x00000080
+#define CP_WAIT_FLAG             0x0000007f
+#define CP_SET                   0x00700000
+#define CP_SET_1                 0x00000080
+#define CP_SET_FLAG              0x0000007f
+#define CP_NEXT_TO_SWAP          0x00600007
+#define CP_NEXT_TO_CURRENT       0x00600009
+#define CP_SET_CONTEXT_POINTER   0x0060000a
+#define CP_END                   0x0060000e
+#define CP_LOAD_MAGIC_UNK01      0x00800001 /* unknown */
+#define CP_LOAD_MAGIC_NV44TCL    0x00800029 /* per-vs state (0x4497) */
+#define CP_LOAD_MAGIC_NV40TCL    0x00800041 /* per-vs state (0x4097) */
+
+#include "nv40.h"
+#include "ctx.h"
+
+/* TODO:
+ *  - get vs count from 0x1540
+ */
+
+static int
+nv40_graph_vs_count(struct nouveau_device *device)
+{
+
+	switch (device->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		return 8;
+	case 0x40:
+		return 6;
+	case 0x41:
+	case 0x42:
+		return 5;
+	case 0x43:
+	case 0x44:
+	case 0x46:
+	case 0x4a:
+		return 3;
+	case 0x4c:
+	case 0x4e:
+	case 0x67:
+	default:
+		return 1;
+	}
+}
+
+
+enum cp_label {
+	cp_check_load = 1,
+	cp_setup_auto_load,
+	cp_setup_load,
+	cp_setup_save,
+	cp_swap_state,
+	cp_swap_state3d_3_is_save,
+	cp_prepare_exit,
+	cp_exit,
+};
+
+static void
+nv40_graph_construct_general(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int i;
+
+	cp_ctx(ctx, 0x4000a4, 1);
+	gr_def(ctx, 0x4000a4, 0x00000008);
+	cp_ctx(ctx, 0x400144, 58);
+	gr_def(ctx, 0x400144, 0x00000001);
+	cp_ctx(ctx, 0x400314, 1);
+	gr_def(ctx, 0x400314, 0x00000000);
+	cp_ctx(ctx, 0x400400, 10);
+	cp_ctx(ctx, 0x400480, 10);
+	cp_ctx(ctx, 0x400500, 19);
+	gr_def(ctx, 0x400514, 0x00040000);
+	gr_def(ctx, 0x400524, 0x55555555);
+	gr_def(ctx, 0x400528, 0x55555555);
+	gr_def(ctx, 0x40052c, 0x55555555);
+	gr_def(ctx, 0x400530, 0x55555555);
+	cp_ctx(ctx, 0x400560, 6);
+	gr_def(ctx, 0x400568, 0x0000ffff);
+	gr_def(ctx, 0x40056c, 0x0000ffff);
+	cp_ctx(ctx, 0x40057c, 5);
+	cp_ctx(ctx, 0x400710, 3);
+	gr_def(ctx, 0x400710, 0x20010001);
+	gr_def(ctx, 0x400714, 0x0f73ef00);
+	cp_ctx(ctx, 0x400724, 1);
+	gr_def(ctx, 0x400724, 0x02008821);
+	cp_ctx(ctx, 0x400770, 3);
+	if (device->chipset == 0x40) {
+		cp_ctx(ctx, 0x400814, 4);
+		cp_ctx(ctx, 0x400828, 5);
+		cp_ctx(ctx, 0x400840, 5);
+		gr_def(ctx, 0x400850, 0x00000040);
+		cp_ctx(ctx, 0x400858, 4);
+		gr_def(ctx, 0x400858, 0x00000040);
+		gr_def(ctx, 0x40085c, 0x00000040);
+		gr_def(ctx, 0x400864, 0x80000000);
+		cp_ctx(ctx, 0x40086c, 9);
+		gr_def(ctx, 0x40086c, 0x80000000);
+		gr_def(ctx, 0x400870, 0x80000000);
+		gr_def(ctx, 0x400874, 0x80000000);
+		gr_def(ctx, 0x400878, 0x80000000);
+		gr_def(ctx, 0x400888, 0x00000040);
+		gr_def(ctx, 0x40088c, 0x80000000);
+		cp_ctx(ctx, 0x4009c0, 8);
+		gr_def(ctx, 0x4009cc, 0x80000000);
+		gr_def(ctx, 0x4009dc, 0x80000000);
+	} else {
+		cp_ctx(ctx, 0x400840, 20);
+		if (nv44_graph_class(ctx->device)) {
+			for (i = 0; i < 8; i++)
+				gr_def(ctx, 0x400860 + (i * 4), 0x00000001);
+		}
+		gr_def(ctx, 0x400880, 0x00000040);
+		gr_def(ctx, 0x400884, 0x00000040);
+		gr_def(ctx, 0x400888, 0x00000040);
+		cp_ctx(ctx, 0x400894, 11);
+		gr_def(ctx, 0x400894, 0x00000040);
+		if (!nv44_graph_class(ctx->device)) {
+			for (i = 0; i < 8; i++)
+				gr_def(ctx, 0x4008a0 + (i * 4), 0x80000000);
+		}
+		cp_ctx(ctx, 0x4008e0, 2);
+		cp_ctx(ctx, 0x4008f8, 2);
+		if (device->chipset == 0x4c ||
+		    (device->chipset & 0xf0) == 0x60)
+			cp_ctx(ctx, 0x4009f8, 1);
+	}
+	cp_ctx(ctx, 0x400a00, 73);
+	gr_def(ctx, 0x400b0c, 0x0b0b0b0c);
+	cp_ctx(ctx, 0x401000, 4);
+	cp_ctx(ctx, 0x405004, 1);
+	switch (device->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x403448, 1);
+		gr_def(ctx, 0x403448, 0x00001010);
+		break;
+	default:
+		cp_ctx(ctx, 0x403440, 1);
+		switch (device->chipset) {
+		case 0x40:
+			gr_def(ctx, 0x403440, 0x00000010);
+			break;
+		case 0x44:
+		case 0x46:
+		case 0x4a:
+			gr_def(ctx, 0x403440, 0x00003010);
+			break;
+		case 0x41:
+		case 0x42:
+		case 0x43:
+		case 0x4c:
+		case 0x4e:
+		case 0x67:
+		default:
+			gr_def(ctx, 0x403440, 0x00001010);
+			break;
+		}
+		break;
+	}
+}
+
+static void
+nv40_graph_construct_state3d(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int i;
+
+	if (device->chipset == 0x40) {
+		cp_ctx(ctx, 0x401880, 51);
+		gr_def(ctx, 0x401940, 0x00000100);
+	} else
+	if (device->chipset == 0x46 || device->chipset == 0x47 ||
+	    device->chipset == 0x49 || device->chipset == 0x4b) {
+		cp_ctx(ctx, 0x401880, 32);
+		for (i = 0; i < 16; i++)
+			gr_def(ctx, 0x401880 + (i * 4), 0x00000111);
+		if (device->chipset == 0x46)
+			cp_ctx(ctx, 0x401900, 16);
+		cp_ctx(ctx, 0x401940, 3);
+	}
+	cp_ctx(ctx, 0x40194c, 18);
+	gr_def(ctx, 0x401954, 0x00000111);
+	gr_def(ctx, 0x401958, 0x00080060);
+	gr_def(ctx, 0x401974, 0x00000080);
+	gr_def(ctx, 0x401978, 0xffff0000);
+	gr_def(ctx, 0x40197c, 0x00000001);
+	gr_def(ctx, 0x401990, 0x46400000);
+	if (device->chipset == 0x40) {
+		cp_ctx(ctx, 0x4019a0, 2);
+		cp_ctx(ctx, 0x4019ac, 5);
+	} else {
+		cp_ctx(ctx, 0x4019a0, 1);
+		cp_ctx(ctx, 0x4019b4, 3);
+	}
+	gr_def(ctx, 0x4019bc, 0xffff0000);
+	switch (device->chipset) {
+	case 0x46:
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x4019c0, 18);
+		for (i = 0; i < 16; i++)
+			gr_def(ctx, 0x4019c0 + (i * 4), 0x88888888);
+		break;
+	}
+	cp_ctx(ctx, 0x401a08, 8);
+	gr_def(ctx, 0x401a10, 0x0fff0000);
+	gr_def(ctx, 0x401a14, 0x0fff0000);
+	gr_def(ctx, 0x401a1c, 0x00011100);
+	cp_ctx(ctx, 0x401a2c, 4);
+	cp_ctx(ctx, 0x401a44, 26);
+	for (i = 0; i < 16; i++)
+		gr_def(ctx, 0x401a44 + (i * 4), 0x07ff0000);
+	gr_def(ctx, 0x401a8c, 0x4b7fffff);
+	if (device->chipset == 0x40) {
+		cp_ctx(ctx, 0x401ab8, 3);
+	} else {
+		cp_ctx(ctx, 0x401ab8, 1);
+		cp_ctx(ctx, 0x401ac0, 1);
+	}
+	cp_ctx(ctx, 0x401ad0, 8);
+	gr_def(ctx, 0x401ad0, 0x30201000);
+	gr_def(ctx, 0x401ad4, 0x70605040);
+	gr_def(ctx, 0x401ad8, 0xb8a89888);
+	gr_def(ctx, 0x401adc, 0xf8e8d8c8);
+	cp_ctx(ctx, 0x401b10, device->chipset == 0x40 ? 2 : 1);
+	gr_def(ctx, 0x401b10, 0x40100000);
+	cp_ctx(ctx, 0x401b18, device->chipset == 0x40 ? 6 : 5);
+	gr_def(ctx, 0x401b28, device->chipset == 0x40 ?
+			      0x00000004 : 0x00000000);
+	cp_ctx(ctx, 0x401b30, 25);
+	gr_def(ctx, 0x401b34, 0x0000ffff);
+	gr_def(ctx, 0x401b68, 0x435185d6);
+	gr_def(ctx, 0x401b6c, 0x2155b699);
+	gr_def(ctx, 0x401b70, 0xfedcba98);
+	gr_def(ctx, 0x401b74, 0x00000098);
+	gr_def(ctx, 0x401b84, 0xffffffff);
+	gr_def(ctx, 0x401b88, 0x00ff7000);
+	gr_def(ctx, 0x401b8c, 0x0000ffff);
+	if (device->chipset != 0x44 && device->chipset != 0x4a &&
+	    device->chipset != 0x4e)
+		cp_ctx(ctx, 0x401b94, 1);
+	cp_ctx(ctx, 0x401b98, 8);
+	gr_def(ctx, 0x401b9c, 0x00ff0000);
+	cp_ctx(ctx, 0x401bc0, 9);
+	gr_def(ctx, 0x401be0, 0x00ffff00);
+	cp_ctx(ctx, 0x401c00, 192);
+	for (i = 0; i < 16; i++) { /* fragment texture units */
+		gr_def(ctx, 0x401c40 + (i * 4), 0x00018488);
+		gr_def(ctx, 0x401c80 + (i * 4), 0x00028202);
+		gr_def(ctx, 0x401d00 + (i * 4), 0x0000aae4);
+		gr_def(ctx, 0x401d40 + (i * 4), 0x01012000);
+		gr_def(ctx, 0x401d80 + (i * 4), 0x00080008);
+		gr_def(ctx, 0x401e00 + (i * 4), 0x00100008);
+	}
+	for (i = 0; i < 4; i++) { /* vertex texture units */
+		gr_def(ctx, 0x401e90 + (i * 4), 0x0001bc80);
+		gr_def(ctx, 0x401ea0 + (i * 4), 0x00000202);
+		gr_def(ctx, 0x401ec0 + (i * 4), 0x00000008);
+		gr_def(ctx, 0x401ee0 + (i * 4), 0x00080008);
+	}
+	cp_ctx(ctx, 0x400f5c, 3);
+	gr_def(ctx, 0x400f5c, 0x00000002);
+	cp_ctx(ctx, 0x400f84, 1);
+}
+
+static void
+nv40_graph_construct_state3d_2(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int i;
+
+	cp_ctx(ctx, 0x402000, 1);
+	cp_ctx(ctx, 0x402404, device->chipset == 0x40 ? 1 : 2);
+	switch (device->chipset) {
+	case 0x40:
+		gr_def(ctx, 0x402404, 0x00000001);
+		break;
+	case 0x4c:
+	case 0x4e:
+	case 0x67:
+		gr_def(ctx, 0x402404, 0x00000020);
+		break;
+	case 0x46:
+	case 0x49:
+	case 0x4b:
+		gr_def(ctx, 0x402404, 0x00000421);
+		break;
+	default:
+		gr_def(ctx, 0x402404, 0x00000021);
+	}
+	if (device->chipset != 0x40)
+		gr_def(ctx, 0x402408, 0x030c30c3);
+	switch (device->chipset) {
+	case 0x44:
+	case 0x46:
+	case 0x4a:
+	case 0x4c:
+	case 0x4e:
+	case 0x67:
+		cp_ctx(ctx, 0x402440, 1);
+		gr_def(ctx, 0x402440, 0x00011001);
+		break;
+	default:
+		break;
+	}
+	cp_ctx(ctx, 0x402480, device->chipset == 0x40 ? 8 : 9);
+	gr_def(ctx, 0x402488, 0x3e020200);
+	gr_def(ctx, 0x40248c, 0x00ffffff);
+	switch (device->chipset) {
+	case 0x40:
+		gr_def(ctx, 0x402490, 0x60103f00);
+		break;
+	case 0x47:
+		gr_def(ctx, 0x402490, 0x40103f00);
+		break;
+	case 0x41:
+	case 0x42:
+	case 0x49:
+	case 0x4b:
+		gr_def(ctx, 0x402490, 0x20103f00);
+		break;
+	default:
+		gr_def(ctx, 0x402490, 0x0c103f00);
+		break;
+	}
+	gr_def(ctx, 0x40249c, device->chipset <= 0x43 ?
+			      0x00020000 : 0x00040000);
+	cp_ctx(ctx, 0x402500, 31);
+	gr_def(ctx, 0x402530, 0x00008100);
+	if (device->chipset == 0x40)
+		cp_ctx(ctx, 0x40257c, 6);
+	cp_ctx(ctx, 0x402594, 16);
+	cp_ctx(ctx, 0x402800, 17);
+	gr_def(ctx, 0x402800, 0x00000001);
+	switch (device->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x402864, 1);
+		gr_def(ctx, 0x402864, 0x00001001);
+		cp_ctx(ctx, 0x402870, 3);
+		gr_def(ctx, 0x402878, 0x00000003);
+		if (device->chipset != 0x47) { /* belong at end!! */
+			cp_ctx(ctx, 0x402900, 1);
+			cp_ctx(ctx, 0x402940, 1);
+			cp_ctx(ctx, 0x402980, 1);
+			cp_ctx(ctx, 0x4029c0, 1);
+			cp_ctx(ctx, 0x402a00, 1);
+			cp_ctx(ctx, 0x402a40, 1);
+			cp_ctx(ctx, 0x402a80, 1);
+			cp_ctx(ctx, 0x402ac0, 1);
+		}
+		break;
+	case 0x40:
+		cp_ctx(ctx, 0x402844, 1);
+		gr_def(ctx, 0x402844, 0x00000001);
+		cp_ctx(ctx, 0x402850, 1);
+		break;
+	default:
+		cp_ctx(ctx, 0x402844, 1);
+		gr_def(ctx, 0x402844, 0x00001001);
+		cp_ctx(ctx, 0x402850, 2);
+		gr_def(ctx, 0x402854, 0x00000003);
+		break;
+	}
+
+	cp_ctx(ctx, 0x402c00, 4);
+	gr_def(ctx, 0x402c00, device->chipset == 0x40 ?
+			      0x80800001 : 0x00888001);
+	switch (device->chipset) {
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x402c20, 40);
+		for (i = 0; i < 32; i++)
+			gr_def(ctx, 0x402c40 + (i * 4), 0xffffffff);
+		cp_ctx(ctx, 0x4030b8, 13);
+		gr_def(ctx, 0x4030dc, 0x00000005);
+		gr_def(ctx, 0x4030e8, 0x0000ffff);
+		break;
+	default:
+		cp_ctx(ctx, 0x402c10, 4);
+		if (device->chipset == 0x40)
+			cp_ctx(ctx, 0x402c20, 36);
+		else
+		if (device->chipset <= 0x42)
+			cp_ctx(ctx, 0x402c20, 24);
+		else
+		if (device->chipset <= 0x4a)
+			cp_ctx(ctx, 0x402c20, 16);
+		else
+			cp_ctx(ctx, 0x402c20, 8);
+		cp_ctx(ctx, 0x402cb0, device->chipset == 0x40 ? 12 : 13);
+		gr_def(ctx, 0x402cd4, 0x00000005);
+		if (device->chipset != 0x40)
+			gr_def(ctx, 0x402ce0, 0x0000ffff);
+		break;
+	}
+
+	cp_ctx(ctx, 0x403400, device->chipset == 0x40 ? 4 : 3);
+	cp_ctx(ctx, 0x403410, device->chipset == 0x40 ? 4 : 3);
+	cp_ctx(ctx, 0x403420, nv40_graph_vs_count(ctx->device));
+	for (i = 0; i < nv40_graph_vs_count(ctx->device); i++)
+		gr_def(ctx, 0x403420 + (i * 4), 0x00005555);
+
+	if (device->chipset != 0x40) {
+		cp_ctx(ctx, 0x403600, 1);
+		gr_def(ctx, 0x403600, 0x00000001);
+	}
+	cp_ctx(ctx, 0x403800, 1);
+
+	cp_ctx(ctx, 0x403c18, 1);
+	gr_def(ctx, 0x403c18, 0x00000001);
+	switch (device->chipset) {
+	case 0x46:
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+		cp_ctx(ctx, 0x405018, 1);
+		gr_def(ctx, 0x405018, 0x08e00001);
+		cp_ctx(ctx, 0x405c24, 1);
+		gr_def(ctx, 0x405c24, 0x000e3000);
+		break;
+	}
+	if (device->chipset != 0x4e)
+		cp_ctx(ctx, 0x405800, 11);
+	cp_ctx(ctx, 0x407000, 1);
+}
+
+static void
+nv40_graph_construct_state3d_3(struct nouveau_grctx *ctx)
+{
+	int len = nv44_graph_class(ctx->device) ? 0x0084 : 0x0684;
+
+	cp_out (ctx, 0x300000);
+	cp_lsr (ctx, len - 4);
+	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_swap_state3d_3_is_save);
+	cp_lsr (ctx, len);
+	cp_name(ctx, cp_swap_state3d_3_is_save);
+	cp_out (ctx, 0x800001);
+
+	ctx->ctxvals_pos += len;
+}
+
+static void
+nv40_graph_construct_shader(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	struct nouveau_gpuobj *obj = ctx->data;
+	int vs, vs_nr, vs_len, vs_nr_b0, vs_nr_b1, b0_offset, b1_offset;
+	int offset, i;
+
+	vs_nr    = nv40_graph_vs_count(ctx->device);
+	vs_nr_b0 = 363;
+	vs_nr_b1 = device->chipset == 0x40 ? 128 : 64;
+	if (device->chipset == 0x40) {
+		b0_offset = 0x2200/4; /* 33a0 */
+		b1_offset = 0x55a0/4; /* 1500 */
+		vs_len = 0x6aa0/4;
+	} else
+	if (device->chipset == 0x41 || device->chipset == 0x42) {
+		b0_offset = 0x2200/4; /* 2200 */
+		b1_offset = 0x4400/4; /* 0b00 */
+		vs_len = 0x4f00/4;
+	} else {
+		b0_offset = 0x1d40/4; /* 2200 */
+		b1_offset = 0x3f40/4; /* 0b00 : 0a40 */
+		vs_len = nv44_graph_class(device) ? 0x4980/4 : 0x4a40/4;
+	}
+
+	cp_lsr(ctx, vs_len * vs_nr + 0x300/4);
+	cp_out(ctx, nv44_graph_class(device) ? 0x800029 : 0x800041);
+
+	offset = ctx->ctxvals_pos;
+	ctx->ctxvals_pos += (0x0300/4 + (vs_nr * vs_len));
+
+	if (ctx->mode != NOUVEAU_GRCTX_VALS)
+		return;
+
+	offset += 0x0280/4;
+	for (i = 0; i < 16; i++, offset += 2)
+		nv_wo32(obj, offset * 4, 0x3f800000);
+
+	for (vs = 0; vs < vs_nr; vs++, offset += vs_len) {
+		for (i = 0; i < vs_nr_b0 * 6; i += 6)
+			nv_wo32(obj, (offset + b0_offset + i) * 4, 0x00000001);
+		for (i = 0; i < vs_nr_b1 * 4; i += 4)
+			nv_wo32(obj, (offset + b1_offset + i) * 4, 0x3f800000);
+	}
+}
+
+static void
+nv40_grctx_generate(struct nouveau_grctx *ctx)
+{
+	/* decide whether we're loading/unloading the context */
+	cp_bra (ctx, AUTO_SAVE, PENDING, cp_setup_save);
+	cp_bra (ctx, USER_SAVE, PENDING, cp_setup_save);
+
+	cp_name(ctx, cp_check_load);
+	cp_bra (ctx, AUTO_LOAD, PENDING, cp_setup_auto_load);
+	cp_bra (ctx, USER_LOAD, PENDING, cp_setup_load);
+	cp_bra (ctx, ALWAYS, TRUE, cp_exit);
+
+	/* setup for context load */
+	cp_name(ctx, cp_setup_auto_load);
+	cp_wait(ctx, STATUS, IDLE);
+	cp_out (ctx, CP_NEXT_TO_SWAP);
+	cp_name(ctx, cp_setup_load);
+	cp_wait(ctx, STATUS, IDLE);
+	cp_set (ctx, SWAP_DIRECTION, LOAD);
+	cp_out (ctx, 0x00910880); /* ?? */
+	cp_out (ctx, 0x00901ffe); /* ?? */
+	cp_out (ctx, 0x01940000); /* ?? */
+	cp_lsr (ctx, 0x20);
+	cp_out (ctx, 0x0060000b); /* ?? */
+	cp_wait(ctx, UNK57, CLEAR);
+	cp_out (ctx, 0x0060000c); /* ?? */
+	cp_bra (ctx, ALWAYS, TRUE, cp_swap_state);
+
+	/* setup for context save */
+	cp_name(ctx, cp_setup_save);
+	cp_set (ctx, SWAP_DIRECTION, SAVE);
+
+	/* general PGRAPH state */
+	cp_name(ctx, cp_swap_state);
+	cp_pos (ctx, 0x00020/4);
+	nv40_graph_construct_general(ctx);
+	cp_wait(ctx, STATUS, IDLE);
+
+	/* 3D state, block 1 */
+	cp_bra (ctx, UNK54, CLEAR, cp_prepare_exit);
+	nv40_graph_construct_state3d(ctx);
+	cp_wait(ctx, STATUS, IDLE);
+
+	/* 3D state, block 2 */
+	nv40_graph_construct_state3d_2(ctx);
+
+	/* Some other block of "random" state */
+	nv40_graph_construct_state3d_3(ctx);
+
+	/* Per-vertex shader state */
+	cp_pos (ctx, ctx->ctxvals_pos);
+	nv40_graph_construct_shader(ctx);
+
+	/* pre-exit state updates */
+	cp_name(ctx, cp_prepare_exit);
+	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_check_load);
+	cp_bra (ctx, USER_SAVE, PENDING, cp_exit);
+	cp_out (ctx, CP_NEXT_TO_CURRENT);
+
+	cp_name(ctx, cp_exit);
+	cp_set (ctx, USER_SAVE, NOT_PENDING);
+	cp_set (ctx, USER_LOAD, NOT_PENDING);
+	cp_out (ctx, CP_END);
+}
+
+void
+nv40_grctx_fill(struct nouveau_device *device, struct nouveau_gpuobj *mem)
+{
+	nv40_grctx_generate(&(struct nouveau_grctx) {
+			     .device = device,
+			     .mode = NOUVEAU_GRCTX_VALS,
+			     .data = mem,
+			   });
+}
+
+int
+nv40_grctx_init(struct nouveau_device *device, u32 *size)
+{
+	u32 *ctxprog = kmalloc(256 * 4, GFP_KERNEL), i;
+	struct nouveau_grctx ctx = {
+		.device = device,
+		.mode = NOUVEAU_GRCTX_PROG,
+		.data = ctxprog,
+		.ctxprog_max = 256,
+	};
+
+	if (!ctxprog)
+		return -ENOMEM;
+
+	nv40_grctx_generate(&ctx);
+
+	nv_wr32(device, 0x400324, 0);
+	for (i = 0; i < ctx.ctxprog_len; i++)
+		nv_wr32(device, 0x400328, ctxprog[i]);
+	*size = ctx.ctxvals_pos * 4;
+
+	kfree(ctxprog);
+	return 0;
+}
diff --git a/nvkm/engine/graph/ctxnv50.c b/nvkm/engine/graph/ctxnv50.c
new file mode 100644
index 000000000..552fdbd45
--- /dev/null
+++ b/nvkm/engine/graph/ctxnv50.c
@@ -0,0 +1,3341 @@
+/*
+ * Copyright 2009 Marcin Kościelnicki
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <core/gpuobj.h>
+
+#define CP_FLAG_CLEAR                 0
+#define CP_FLAG_SET                   1
+#define CP_FLAG_SWAP_DIRECTION        ((0 * 32) + 0)
+#define CP_FLAG_SWAP_DIRECTION_LOAD   0
+#define CP_FLAG_SWAP_DIRECTION_SAVE   1
+#define CP_FLAG_UNK01                 ((0 * 32) + 1)
+#define CP_FLAG_UNK01_CLEAR           0
+#define CP_FLAG_UNK01_SET             1
+#define CP_FLAG_UNK03                 ((0 * 32) + 3)
+#define CP_FLAG_UNK03_CLEAR           0
+#define CP_FLAG_UNK03_SET             1
+#define CP_FLAG_USER_SAVE             ((0 * 32) + 5)
+#define CP_FLAG_USER_SAVE_NOT_PENDING 0
+#define CP_FLAG_USER_SAVE_PENDING     1
+#define CP_FLAG_USER_LOAD             ((0 * 32) + 6)
+#define CP_FLAG_USER_LOAD_NOT_PENDING 0
+#define CP_FLAG_USER_LOAD_PENDING     1
+#define CP_FLAG_UNK0B                 ((0 * 32) + 0xb)
+#define CP_FLAG_UNK0B_CLEAR           0
+#define CP_FLAG_UNK0B_SET             1
+#define CP_FLAG_XFER_SWITCH           ((0 * 32) + 0xe)
+#define CP_FLAG_XFER_SWITCH_DISABLE   0
+#define CP_FLAG_XFER_SWITCH_ENABLE    1
+#define CP_FLAG_STATE                 ((0 * 32) + 0x1c)
+#define CP_FLAG_STATE_STOPPED         0
+#define CP_FLAG_STATE_RUNNING         1
+#define CP_FLAG_UNK1D                 ((0 * 32) + 0x1d)
+#define CP_FLAG_UNK1D_CLEAR           0
+#define CP_FLAG_UNK1D_SET             1
+#define CP_FLAG_UNK20                 ((1 * 32) + 0)
+#define CP_FLAG_UNK20_CLEAR           0
+#define CP_FLAG_UNK20_SET             1
+#define CP_FLAG_STATUS                ((2 * 32) + 0)
+#define CP_FLAG_STATUS_BUSY           0
+#define CP_FLAG_STATUS_IDLE           1
+#define CP_FLAG_AUTO_SAVE             ((2 * 32) + 4)
+#define CP_FLAG_AUTO_SAVE_NOT_PENDING 0
+#define CP_FLAG_AUTO_SAVE_PENDING     1
+#define CP_FLAG_AUTO_LOAD             ((2 * 32) + 5)
+#define CP_FLAG_AUTO_LOAD_NOT_PENDING 0
+#define CP_FLAG_AUTO_LOAD_PENDING     1
+#define CP_FLAG_NEWCTX                ((2 * 32) + 10)
+#define CP_FLAG_NEWCTX_BUSY           0
+#define CP_FLAG_NEWCTX_DONE           1
+#define CP_FLAG_XFER                  ((2 * 32) + 11)
+#define CP_FLAG_XFER_IDLE             0
+#define CP_FLAG_XFER_BUSY             1
+#define CP_FLAG_ALWAYS                ((2 * 32) + 13)
+#define CP_FLAG_ALWAYS_FALSE          0
+#define CP_FLAG_ALWAYS_TRUE           1
+#define CP_FLAG_INTR                  ((2 * 32) + 15)
+#define CP_FLAG_INTR_NOT_PENDING      0
+#define CP_FLAG_INTR_PENDING          1
+
+#define CP_CTX                   0x00100000
+#define CP_CTX_COUNT             0x000f0000
+#define CP_CTX_COUNT_SHIFT               16
+#define CP_CTX_REG               0x00003fff
+#define CP_LOAD_SR               0x00200000
+#define CP_LOAD_SR_VALUE         0x000fffff
+#define CP_BRA                   0x00400000
+#define CP_BRA_IP                0x0001ff00
+#define CP_BRA_IP_SHIFT                   8
+#define CP_BRA_IF_CLEAR          0x00000080
+#define CP_BRA_FLAG              0x0000007f
+#define CP_WAIT                  0x00500000
+#define CP_WAIT_SET              0x00000080
+#define CP_WAIT_FLAG             0x0000007f
+#define CP_SET                   0x00700000
+#define CP_SET_1                 0x00000080
+#define CP_SET_FLAG              0x0000007f
+#define CP_NEWCTX                0x00600004
+#define CP_NEXT_TO_SWAP          0x00600005
+#define CP_SET_CONTEXT_POINTER   0x00600006
+#define CP_SET_XFER_POINTER      0x00600007
+#define CP_ENABLE                0x00600009
+#define CP_END                   0x0060000c
+#define CP_NEXT_TO_CURRENT       0x0060000d
+#define CP_DISABLE1              0x0090ffff
+#define CP_DISABLE2              0x0091ffff
+#define CP_XFER_1      0x008000ff
+#define CP_XFER_2      0x008800ff
+#define CP_SEEK_1      0x00c000ff
+#define CP_SEEK_2      0x00c800ff
+
+#include "nv50.h"
+#include "ctx.h"
+
+#define IS_NVA3F(x) (((x) > 0xa0 && (x) < 0xaa) || (x) == 0xaf)
+#define IS_NVAAF(x) ((x) >= 0xaa && (x) <= 0xac)
+
+/*
+ * This code deals with PGRAPH contexts on NV50 family cards. Like NV40, it's
+ * the GPU itself that does context-switching, but it needs a special
+ * microcode to do it. And it's the driver's task to supply this microcode,
+ * further known as ctxprog, as well as the initial context values, known
+ * as ctxvals.
+ *
+ * Without ctxprog, you cannot switch contexts. Not even in software, since
+ * the majority of context [xfer strands] isn't accessible directly. You're
+ * stuck with a single channel, and you also suffer all the problems resulting
+ * from missing ctxvals, since you cannot load them.
+ *
+ * Without ctxvals, you're stuck with PGRAPH's default context. It's enough to
+ * run 2d operations, but trying to utilise 3d or CUDA will just lock you up,
+ * since you don't have... some sort of needed setup.
+ *
+ * Nouveau will just disable acceleration if not given ctxprog + ctxvals, since
+ * it's too much hassle to handle no-ctxprog as a special case.
+ */
+
+/*
+ * How ctxprogs work.
+ *
+ * The ctxprog is written in its own kind of microcode, with very small and
+ * crappy set of available commands. You upload it to a small [512 insns]
+ * area of memory on PGRAPH, and it'll be run when PFIFO wants PGRAPH to
+ * switch channel. or when the driver explicitely requests it. Stuff visible
+ * to ctxprog consists of: PGRAPH MMIO registers, PGRAPH context strands,
+ * the per-channel context save area in VRAM [known as ctxvals or grctx],
+ * 4 flags registers, a scratch register, two grctx pointers, plus many
+ * random poorly-understood details.
+ *
+ * When ctxprog runs, it's supposed to check what operations are asked of it,
+ * save old context if requested, optionally reset PGRAPH and switch to the
+ * new channel, and load the new context. Context consists of three major
+ * parts: subset of MMIO registers and two "xfer areas".
+ */
+
+/* TODO:
+ *  - document unimplemented bits compared to nvidia
+ *  - NVAx: make a TP subroutine, use it.
+ *  - use 0x4008fc instead of 0x1540?
+ */
+
+enum cp_label {
+	cp_check_load = 1,
+	cp_setup_auto_load,
+	cp_setup_load,
+	cp_setup_save,
+	cp_swap_state,
+	cp_prepare_exit,
+	cp_exit,
+};
+
+static void nv50_graph_construct_mmio(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_xfer1(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_xfer2(struct nouveau_grctx *ctx);
+
+/* Main function: construct the ctxprog skeleton, call the other functions. */
+
+static int
+nv50_grctx_generate(struct nouveau_grctx *ctx)
+{
+	cp_set (ctx, STATE, RUNNING);
+	cp_set (ctx, XFER_SWITCH, ENABLE);
+	/* decide whether we're loading/unloading the context */
+	cp_bra (ctx, AUTO_SAVE, PENDING, cp_setup_save);
+	cp_bra (ctx, USER_SAVE, PENDING, cp_setup_save);
+
+	cp_name(ctx, cp_check_load);
+	cp_bra (ctx, AUTO_LOAD, PENDING, cp_setup_auto_load);
+	cp_bra (ctx, USER_LOAD, PENDING, cp_setup_load);
+	cp_bra (ctx, ALWAYS, TRUE, cp_prepare_exit);
+
+	/* setup for context load */
+	cp_name(ctx, cp_setup_auto_load);
+	cp_out (ctx, CP_DISABLE1);
+	cp_out (ctx, CP_DISABLE2);
+	cp_out (ctx, CP_ENABLE);
+	cp_out (ctx, CP_NEXT_TO_SWAP);
+	cp_set (ctx, UNK01, SET);
+	cp_name(ctx, cp_setup_load);
+	cp_out (ctx, CP_NEWCTX);
+	cp_wait(ctx, NEWCTX, BUSY);
+	cp_set (ctx, UNK1D, CLEAR);
+	cp_set (ctx, SWAP_DIRECTION, LOAD);
+	cp_bra (ctx, UNK0B, SET, cp_prepare_exit);
+	cp_bra (ctx, ALWAYS, TRUE, cp_swap_state);
+
+	/* setup for context save */
+	cp_name(ctx, cp_setup_save);
+	cp_set (ctx, UNK1D, SET);
+	cp_wait(ctx, STATUS, BUSY);
+	cp_wait(ctx, INTR, PENDING);
+	cp_bra (ctx, STATUS, BUSY, cp_setup_save);
+	cp_set (ctx, UNK01, SET);
+	cp_set (ctx, SWAP_DIRECTION, SAVE);
+
+	/* general PGRAPH state */
+	cp_name(ctx, cp_swap_state);
+	cp_set (ctx, UNK03, SET);
+	cp_pos (ctx, 0x00004/4);
+	cp_ctx (ctx, 0x400828, 1); /* needed. otherwise, flickering happens. */
+	cp_pos (ctx, 0x00100/4);
+	nv50_graph_construct_mmio(ctx);
+	nv50_graph_construct_xfer1(ctx);
+	nv50_graph_construct_xfer2(ctx);
+
+	cp_bra (ctx, SWAP_DIRECTION, SAVE, cp_check_load);
+
+	cp_set (ctx, UNK20, SET);
+	cp_set (ctx, SWAP_DIRECTION, SAVE); /* no idea why this is needed, but fixes at least one lockup. */
+	cp_lsr (ctx, ctx->ctxvals_base);
+	cp_out (ctx, CP_SET_XFER_POINTER);
+	cp_lsr (ctx, 4);
+	cp_out (ctx, CP_SEEK_1);
+	cp_out (ctx, CP_XFER_1);
+	cp_wait(ctx, XFER, BUSY);
+
+	/* pre-exit state updates */
+	cp_name(ctx, cp_prepare_exit);
+	cp_set (ctx, UNK01, CLEAR);
+	cp_set (ctx, UNK03, CLEAR);
+	cp_set (ctx, UNK1D, CLEAR);
+
+	cp_bra (ctx, USER_SAVE, PENDING, cp_exit);
+	cp_out (ctx, CP_NEXT_TO_CURRENT);
+
+	cp_name(ctx, cp_exit);
+	cp_set (ctx, USER_SAVE, NOT_PENDING);
+	cp_set (ctx, USER_LOAD, NOT_PENDING);
+	cp_set (ctx, XFER_SWITCH, DISABLE);
+	cp_set (ctx, STATE, STOPPED);
+	cp_out (ctx, CP_END);
+	ctx->ctxvals_pos += 0x400; /* padding... no idea why you need it */
+
+	return 0;
+}
+
+void
+nv50_grctx_fill(struct nouveau_device *device, struct nouveau_gpuobj *mem)
+{
+	nv50_grctx_generate(&(struct nouveau_grctx) {
+			     .device = device,
+			     .mode = NOUVEAU_GRCTX_VALS,
+			     .data = mem,
+			   });
+}
+
+int
+nv50_grctx_init(struct nouveau_device *device, u32 *size)
+{
+	u32 *ctxprog = kmalloc(512 * 4, GFP_KERNEL), i;
+	struct nouveau_grctx ctx = {
+		.device = device,
+		.mode = NOUVEAU_GRCTX_PROG,
+		.data = ctxprog,
+		.ctxprog_max = 512,
+	};
+
+	if (!ctxprog)
+		return -ENOMEM;
+	nv50_grctx_generate(&ctx);
+
+	nv_wr32(device, 0x400324, 0);
+	for (i = 0; i < ctx.ctxprog_len; i++)
+		nv_wr32(device, 0x400328, ctxprog[i]);
+	*size = ctx.ctxvals_pos * 4;
+	kfree(ctxprog);
+	return 0;
+}
+
+/*
+ * Constructs MMIO part of ctxprog and ctxvals. Just a matter of knowing which
+ * registers to save/restore and the default values for them.
+ */
+
+static void
+nv50_graph_construct_mmio_ddata(struct nouveau_grctx *ctx);
+
+static void
+nv50_graph_construct_mmio(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int i, j;
+	int offset, base;
+	u32 units = nv_rd32 (ctx->device, 0x1540);
+
+	/* 0800: DISPATCH */
+	cp_ctx(ctx, 0x400808, 7);
+	gr_def(ctx, 0x400814, 0x00000030);
+	cp_ctx(ctx, 0x400834, 0x32);
+	if (device->chipset == 0x50) {
+		gr_def(ctx, 0x400834, 0xff400040);
+		gr_def(ctx, 0x400838, 0xfff00080);
+		gr_def(ctx, 0x40083c, 0xfff70090);
+		gr_def(ctx, 0x400840, 0xffe806a8);
+	}
+	gr_def(ctx, 0x400844, 0x00000002);
+	if (IS_NVA3F(device->chipset))
+		gr_def(ctx, 0x400894, 0x00001000);
+	gr_def(ctx, 0x4008e8, 0x00000003);
+	gr_def(ctx, 0x4008ec, 0x00001000);
+	if (device->chipset == 0x50)
+		cp_ctx(ctx, 0x400908, 0xb);
+	else if (device->chipset < 0xa0)
+		cp_ctx(ctx, 0x400908, 0xc);
+	else
+		cp_ctx(ctx, 0x400908, 0xe);
+
+	if (device->chipset >= 0xa0)
+		cp_ctx(ctx, 0x400b00, 0x1);
+	if (IS_NVA3F(device->chipset)) {
+		cp_ctx(ctx, 0x400b10, 0x1);
+		gr_def(ctx, 0x400b10, 0x0001629d);
+		cp_ctx(ctx, 0x400b20, 0x1);
+		gr_def(ctx, 0x400b20, 0x0001629d);
+	}
+
+	nv50_graph_construct_mmio_ddata(ctx);
+
+	/* 0C00: VFETCH */
+	cp_ctx(ctx, 0x400c08, 0x2);
+	gr_def(ctx, 0x400c08, 0x0000fe0c);
+
+	/* 1000 */
+	if (device->chipset < 0xa0) {
+		cp_ctx(ctx, 0x401008, 0x4);
+		gr_def(ctx, 0x401014, 0x00001000);
+	} else if (!IS_NVA3F(device->chipset)) {
+		cp_ctx(ctx, 0x401008, 0x5);
+		gr_def(ctx, 0x401018, 0x00001000);
+	} else {
+		cp_ctx(ctx, 0x401008, 0x5);
+		gr_def(ctx, 0x401018, 0x00004000);
+	}
+
+	/* 1400 */
+	cp_ctx(ctx, 0x401400, 0x8);
+	cp_ctx(ctx, 0x401424, 0x3);
+	if (device->chipset == 0x50)
+		gr_def(ctx, 0x40142c, 0x0001fd87);
+	else
+		gr_def(ctx, 0x40142c, 0x00000187);
+	cp_ctx(ctx, 0x401540, 0x5);
+	gr_def(ctx, 0x401550, 0x00001018);
+
+	/* 1800: STREAMOUT */
+	cp_ctx(ctx, 0x401814, 0x1);
+	gr_def(ctx, 0x401814, 0x000000ff);
+	if (device->chipset == 0x50) {
+		cp_ctx(ctx, 0x40181c, 0xe);
+		gr_def(ctx, 0x401850, 0x00000004);
+	} else if (device->chipset < 0xa0) {
+		cp_ctx(ctx, 0x40181c, 0xf);
+		gr_def(ctx, 0x401854, 0x00000004);
+	} else {
+		cp_ctx(ctx, 0x40181c, 0x13);
+		gr_def(ctx, 0x401864, 0x00000004);
+	}
+
+	/* 1C00 */
+	cp_ctx(ctx, 0x401c00, 0x1);
+	switch (device->chipset) {
+	case 0x50:
+		gr_def(ctx, 0x401c00, 0x0001005f);
+		break;
+	case 0x84:
+	case 0x86:
+	case 0x94:
+		gr_def(ctx, 0x401c00, 0x044d00df);
+		break;
+	case 0x92:
+	case 0x96:
+	case 0x98:
+	case 0xa0:
+	case 0xaa:
+	case 0xac:
+		gr_def(ctx, 0x401c00, 0x042500df);
+		break;
+	case 0xa3:
+	case 0xa5:
+	case 0xa8:
+	case 0xaf:
+		gr_def(ctx, 0x401c00, 0x142500df);
+		break;
+	}
+
+	/* 2000 */
+
+	/* 2400 */
+	cp_ctx(ctx, 0x402400, 0x1);
+	if (device->chipset == 0x50)
+		cp_ctx(ctx, 0x402408, 0x1);
+	else
+		cp_ctx(ctx, 0x402408, 0x2);
+	gr_def(ctx, 0x402408, 0x00000600);
+
+	/* 2800: CSCHED */
+	cp_ctx(ctx, 0x402800, 0x1);
+	if (device->chipset == 0x50)
+		gr_def(ctx, 0x402800, 0x00000006);
+
+	/* 2C00: ZCULL */
+	cp_ctx(ctx, 0x402c08, 0x6);
+	if (device->chipset != 0x50)
+		gr_def(ctx, 0x402c14, 0x01000000);
+	gr_def(ctx, 0x402c18, 0x000000ff);
+	if (device->chipset == 0x50)
+		cp_ctx(ctx, 0x402ca0, 0x1);
+	else
+		cp_ctx(ctx, 0x402ca0, 0x2);
+	if (device->chipset < 0xa0)
+		gr_def(ctx, 0x402ca0, 0x00000400);
+	else if (!IS_NVA3F(device->chipset))
+		gr_def(ctx, 0x402ca0, 0x00000800);
+	else
+		gr_def(ctx, 0x402ca0, 0x00000400);
+	cp_ctx(ctx, 0x402cac, 0x4);
+
+	/* 3000: ENG2D */
+	cp_ctx(ctx, 0x403004, 0x1);
+	gr_def(ctx, 0x403004, 0x00000001);
+
+	/* 3400 */
+	if (device->chipset >= 0xa0) {
+		cp_ctx(ctx, 0x403404, 0x1);
+		gr_def(ctx, 0x403404, 0x00000001);
+	}
+
+	/* 5000: CCACHE */
+	cp_ctx(ctx, 0x405000, 0x1);
+	switch (device->chipset) {
+	case 0x50:
+		gr_def(ctx, 0x405000, 0x00300080);
+		break;
+	case 0x84:
+	case 0xa0:
+	case 0xa3:
+	case 0xa5:
+	case 0xa8:
+	case 0xaa:
+	case 0xac:
+	case 0xaf:
+		gr_def(ctx, 0x405000, 0x000e0080);
+		break;
+	case 0x86:
+	case 0x92:
+	case 0x94:
+	case 0x96:
+	case 0x98:
+		gr_def(ctx, 0x405000, 0x00000080);
+		break;
+	}
+	cp_ctx(ctx, 0x405014, 0x1);
+	gr_def(ctx, 0x405014, 0x00000004);
+	cp_ctx(ctx, 0x40501c, 0x1);
+	cp_ctx(ctx, 0x405024, 0x1);
+	cp_ctx(ctx, 0x40502c, 0x1);
+
+	/* 6000? */
+	if (device->chipset == 0x50)
+		cp_ctx(ctx, 0x4063e0, 0x1);
+
+	/* 6800: M2MF */
+	if (device->chipset < 0x90) {
+		cp_ctx(ctx, 0x406814, 0x2b);
+		gr_def(ctx, 0x406818, 0x00000f80);
+		gr_def(ctx, 0x406860, 0x007f0080);
+		gr_def(ctx, 0x40689c, 0x007f0080);
+	} else {
+		cp_ctx(ctx, 0x406814, 0x4);
+		if (device->chipset == 0x98)
+			gr_def(ctx, 0x406818, 0x00000f80);
+		else
+			gr_def(ctx, 0x406818, 0x00001f80);
+		if (IS_NVA3F(device->chipset))
+			gr_def(ctx, 0x40681c, 0x00000030);
+		cp_ctx(ctx, 0x406830, 0x3);
+	}
+
+	/* 7000: per-ROP group state */
+	for (i = 0; i < 8; i++) {
+		if (units & (1<<(i+16))) {
+			cp_ctx(ctx, 0x407000 + (i<<8), 3);
+			if (device->chipset == 0x50)
+				gr_def(ctx, 0x407000 + (i<<8), 0x1b74f820);
+			else if (device->chipset != 0xa5)
+				gr_def(ctx, 0x407000 + (i<<8), 0x3b74f821);
+			else
+				gr_def(ctx, 0x407000 + (i<<8), 0x7b74f821);
+			gr_def(ctx, 0x407004 + (i<<8), 0x89058001);
+
+			if (device->chipset == 0x50) {
+				cp_ctx(ctx, 0x407010 + (i<<8), 1);
+			} else if (device->chipset < 0xa0) {
+				cp_ctx(ctx, 0x407010 + (i<<8), 2);
+				gr_def(ctx, 0x407010 + (i<<8), 0x00001000);
+				gr_def(ctx, 0x407014 + (i<<8), 0x0000001f);
+			} else {
+				cp_ctx(ctx, 0x407010 + (i<<8), 3);
+				gr_def(ctx, 0x407010 + (i<<8), 0x00001000);
+				if (device->chipset != 0xa5)
+					gr_def(ctx, 0x407014 + (i<<8), 0x000000ff);
+				else
+					gr_def(ctx, 0x407014 + (i<<8), 0x000001ff);
+			}
+
+			cp_ctx(ctx, 0x407080 + (i<<8), 4);
+			if (device->chipset != 0xa5)
+				gr_def(ctx, 0x407080 + (i<<8), 0x027c10fa);
+			else
+				gr_def(ctx, 0x407080 + (i<<8), 0x827c10fa);
+			if (device->chipset == 0x50)
+				gr_def(ctx, 0x407084 + (i<<8), 0x000000c0);
+			else
+				gr_def(ctx, 0x407084 + (i<<8), 0x400000c0);
+			gr_def(ctx, 0x407088 + (i<<8), 0xb7892080);
+
+			if (device->chipset < 0xa0)
+				cp_ctx(ctx, 0x407094 + (i<<8), 1);
+			else if (!IS_NVA3F(device->chipset))
+				cp_ctx(ctx, 0x407094 + (i<<8), 3);
+			else {
+				cp_ctx(ctx, 0x407094 + (i<<8), 4);
+				gr_def(ctx, 0x4070a0 + (i<<8), 1);
+			}
+		}
+	}
+
+	cp_ctx(ctx, 0x407c00, 0x3);
+	if (device->chipset < 0x90)
+		gr_def(ctx, 0x407c00, 0x00010040);
+	else if (device->chipset < 0xa0)
+		gr_def(ctx, 0x407c00, 0x00390040);
+	else
+		gr_def(ctx, 0x407c00, 0x003d0040);
+	gr_def(ctx, 0x407c08, 0x00000022);
+	if (device->chipset >= 0xa0) {
+		cp_ctx(ctx, 0x407c10, 0x3);
+		cp_ctx(ctx, 0x407c20, 0x1);
+		cp_ctx(ctx, 0x407c2c, 0x1);
+	}
+
+	if (device->chipset < 0xa0) {
+		cp_ctx(ctx, 0x407d00, 0x9);
+	} else {
+		cp_ctx(ctx, 0x407d00, 0x15);
+	}
+	if (device->chipset == 0x98)
+		gr_def(ctx, 0x407d08, 0x00380040);
+	else {
+		if (device->chipset < 0x90)
+			gr_def(ctx, 0x407d08, 0x00010040);
+		else if (device->chipset < 0xa0)
+			gr_def(ctx, 0x407d08, 0x00390040);
+		else
+			gr_def(ctx, 0x407d08, 0x003d0040);
+		gr_def(ctx, 0x407d0c, 0x00000022);
+	}
+
+	/* 8000+: per-TP state */
+	for (i = 0; i < 10; i++) {
+		if (units & (1<<i)) {
+			if (device->chipset < 0xa0)
+				base = 0x408000 + (i<<12);
+			else
+				base = 0x408000 + (i<<11);
+			if (device->chipset < 0xa0)
+				offset = base + 0xc00;
+			else
+				offset = base + 0x80;
+			cp_ctx(ctx, offset + 0x00, 1);
+			gr_def(ctx, offset + 0x00, 0x0000ff0a);
+			cp_ctx(ctx, offset + 0x08, 1);
+
+			/* per-MP state */
+			for (j = 0; j < (device->chipset < 0xa0 ? 2 : 4); j++) {
+				if (!(units & (1 << (j+24)))) continue;
+				if (device->chipset < 0xa0)
+					offset = base + 0x200 + (j<<7);
+				else
+					offset = base + 0x100 + (j<<7);
+				cp_ctx(ctx, offset, 0x20);
+				gr_def(ctx, offset + 0x00, 0x01800000);
+				gr_def(ctx, offset + 0x04, 0x00160000);
+				gr_def(ctx, offset + 0x08, 0x01800000);
+				gr_def(ctx, offset + 0x18, 0x0003ffff);
+				switch (device->chipset) {
+				case 0x50:
+					gr_def(ctx, offset + 0x1c, 0x00080000);
+					break;
+				case 0x84:
+					gr_def(ctx, offset + 0x1c, 0x00880000);
+					break;
+				case 0x86:
+					gr_def(ctx, offset + 0x1c, 0x018c0000);
+					break;
+				case 0x92:
+				case 0x96:
+				case 0x98:
+					gr_def(ctx, offset + 0x1c, 0x118c0000);
+					break;
+				case 0x94:
+					gr_def(ctx, offset + 0x1c, 0x10880000);
+					break;
+				case 0xa0:
+				case 0xa5:
+					gr_def(ctx, offset + 0x1c, 0x310c0000);
+					break;
+				case 0xa3:
+				case 0xa8:
+				case 0xaa:
+				case 0xac:
+				case 0xaf:
+					gr_def(ctx, offset + 0x1c, 0x300c0000);
+					break;
+				}
+				gr_def(ctx, offset + 0x40, 0x00010401);
+				if (device->chipset == 0x50)
+					gr_def(ctx, offset + 0x48, 0x00000040);
+				else
+					gr_def(ctx, offset + 0x48, 0x00000078);
+				gr_def(ctx, offset + 0x50, 0x000000bf);
+				gr_def(ctx, offset + 0x58, 0x00001210);
+				if (device->chipset == 0x50)
+					gr_def(ctx, offset + 0x5c, 0x00000080);
+				else
+					gr_def(ctx, offset + 0x5c, 0x08000080);
+				if (device->chipset >= 0xa0)
+					gr_def(ctx, offset + 0x68, 0x0000003e);
+			}
+
+			if (device->chipset < 0xa0)
+				cp_ctx(ctx, base + 0x300, 0x4);
+			else
+				cp_ctx(ctx, base + 0x300, 0x5);
+			if (device->chipset == 0x50)
+				gr_def(ctx, base + 0x304, 0x00007070);
+			else if (device->chipset < 0xa0)
+				gr_def(ctx, base + 0x304, 0x00027070);
+			else if (!IS_NVA3F(device->chipset))
+				gr_def(ctx, base + 0x304, 0x01127070);
+			else
+				gr_def(ctx, base + 0x304, 0x05127070);
+
+			if (device->chipset < 0xa0)
+				cp_ctx(ctx, base + 0x318, 1);
+			else
+				cp_ctx(ctx, base + 0x320, 1);
+			if (device->chipset == 0x50)
+				gr_def(ctx, base + 0x318, 0x0003ffff);
+			else if (device->chipset < 0xa0)
+				gr_def(ctx, base + 0x318, 0x03ffffff);
+			else
+				gr_def(ctx, base + 0x320, 0x07ffffff);
+
+			if (device->chipset < 0xa0)
+				cp_ctx(ctx, base + 0x324, 5);
+			else
+				cp_ctx(ctx, base + 0x328, 4);
+
+			if (device->chipset < 0xa0) {
+				cp_ctx(ctx, base + 0x340, 9);
+				offset = base + 0x340;
+			} else if (!IS_NVA3F(device->chipset)) {
+				cp_ctx(ctx, base + 0x33c, 0xb);
+				offset = base + 0x344;
+			} else {
+				cp_ctx(ctx, base + 0x33c, 0xd);
+				offset = base + 0x344;
+			}
+			gr_def(ctx, offset + 0x0, 0x00120407);
+			gr_def(ctx, offset + 0x4, 0x05091507);
+			if (device->chipset == 0x84)
+				gr_def(ctx, offset + 0x8, 0x05100202);
+			else
+				gr_def(ctx, offset + 0x8, 0x05010202);
+			gr_def(ctx, offset + 0xc, 0x00030201);
+			if (device->chipset == 0xa3)
+				cp_ctx(ctx, base + 0x36c, 1);
+
+			cp_ctx(ctx, base + 0x400, 2);
+			gr_def(ctx, base + 0x404, 0x00000040);
+			cp_ctx(ctx, base + 0x40c, 2);
+			gr_def(ctx, base + 0x40c, 0x0d0c0b0a);
+			gr_def(ctx, base + 0x410, 0x00141210);
+
+			if (device->chipset < 0xa0)
+				offset = base + 0x800;
+			else
+				offset = base + 0x500;
+			cp_ctx(ctx, offset, 6);
+			gr_def(ctx, offset + 0x0, 0x000001f0);
+			gr_def(ctx, offset + 0x4, 0x00000001);
+			gr_def(ctx, offset + 0x8, 0x00000003);
+			if (device->chipset == 0x50 || IS_NVAAF(device->chipset))
+				gr_def(ctx, offset + 0xc, 0x00008000);
+			gr_def(ctx, offset + 0x14, 0x00039e00);
+			cp_ctx(ctx, offset + 0x1c, 2);
+			if (device->chipset == 0x50)
+				gr_def(ctx, offset + 0x1c, 0x00000040);
+			else
+				gr_def(ctx, offset + 0x1c, 0x00000100);
+			gr_def(ctx, offset + 0x20, 0x00003800);
+
+			if (device->chipset >= 0xa0) {
+				cp_ctx(ctx, base + 0x54c, 2);
+				if (!IS_NVA3F(device->chipset))
+					gr_def(ctx, base + 0x54c, 0x003fe006);
+				else
+					gr_def(ctx, base + 0x54c, 0x003fe007);
+				gr_def(ctx, base + 0x550, 0x003fe000);
+			}
+
+			if (device->chipset < 0xa0)
+				offset = base + 0xa00;
+			else
+				offset = base + 0x680;
+			cp_ctx(ctx, offset, 1);
+			gr_def(ctx, offset, 0x00404040);
+
+			if (device->chipset < 0xa0)
+				offset = base + 0xe00;
+			else
+				offset = base + 0x700;
+			cp_ctx(ctx, offset, 2);
+			if (device->chipset < 0xa0)
+				gr_def(ctx, offset, 0x0077f005);
+			else if (device->chipset == 0xa5)
+				gr_def(ctx, offset, 0x6cf7f007);
+			else if (device->chipset == 0xa8)
+				gr_def(ctx, offset, 0x6cfff007);
+			else if (device->chipset == 0xac)
+				gr_def(ctx, offset, 0x0cfff007);
+			else
+				gr_def(ctx, offset, 0x0cf7f007);
+			if (device->chipset == 0x50)
+				gr_def(ctx, offset + 0x4, 0x00007fff);
+			else if (device->chipset < 0xa0)
+				gr_def(ctx, offset + 0x4, 0x003f7fff);
+			else
+				gr_def(ctx, offset + 0x4, 0x02bf7fff);
+			cp_ctx(ctx, offset + 0x2c, 1);
+			if (device->chipset == 0x50) {
+				cp_ctx(ctx, offset + 0x50, 9);
+				gr_def(ctx, offset + 0x54, 0x000003ff);
+				gr_def(ctx, offset + 0x58, 0x00000003);
+				gr_def(ctx, offset + 0x5c, 0x00000003);
+				gr_def(ctx, offset + 0x60, 0x000001ff);
+				gr_def(ctx, offset + 0x64, 0x0000001f);
+				gr_def(ctx, offset + 0x68, 0x0000000f);
+				gr_def(ctx, offset + 0x6c, 0x0000000f);
+			} else if (device->chipset < 0xa0) {
+				cp_ctx(ctx, offset + 0x50, 1);
+				cp_ctx(ctx, offset + 0x70, 1);
+			} else {
+				cp_ctx(ctx, offset + 0x50, 1);
+				cp_ctx(ctx, offset + 0x60, 5);
+			}
+		}
+	}
+}
+
+static void
+dd_emit(struct nouveau_grctx *ctx, int num, u32 val) {
+	int i;
+	if (val && ctx->mode == NOUVEAU_GRCTX_VALS)
+		for (i = 0; i < num; i++)
+			nv_wo32(ctx->data, 4 * (ctx->ctxvals_pos + i), val);
+	ctx->ctxvals_pos += num;
+}
+
+static void
+nv50_graph_construct_mmio_ddata(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int base, num;
+	base = ctx->ctxvals_pos;
+
+	/* tesla state */
+	dd_emit(ctx, 1, 0);	/* 00000001 UNK0F90 */
+	dd_emit(ctx, 1, 0);	/* 00000001 UNK135C */
+
+	/* SRC_TIC state */
+	dd_emit(ctx, 1, 0);	/* 00000007 SRC_TILE_MODE_Z */
+	dd_emit(ctx, 1, 2);	/* 00000007 SRC_TILE_MODE_Y */
+	dd_emit(ctx, 1, 1);	/* 00000001 SRC_LINEAR #1 */
+	dd_emit(ctx, 1, 0);	/* 000000ff SRC_ADDRESS_HIGH */
+	dd_emit(ctx, 1, 0);	/* 00000001 SRC_SRGB */
+	if (device->chipset >= 0x94)
+		dd_emit(ctx, 1, 0);	/* 00000003 eng2d UNK0258 */
+	dd_emit(ctx, 1, 1);	/* 00000fff SRC_DEPTH */
+	dd_emit(ctx, 1, 0x100);	/* 0000ffff SRC_HEIGHT */
+
+	/* turing state */
+	dd_emit(ctx, 1, 0);		/* 0000000f TEXTURES_LOG2 */
+	dd_emit(ctx, 1, 0);		/* 0000000f SAMPLERS_LOG2 */
+	dd_emit(ctx, 1, 0);		/* 000000ff CB_DEF_ADDRESS_HIGH */
+	dd_emit(ctx, 1, 0);		/* ffffffff CB_DEF_ADDRESS_LOW */
+	dd_emit(ctx, 1, 0);		/* ffffffff SHARED_SIZE */
+	dd_emit(ctx, 1, 2);		/* ffffffff REG_MODE */
+	dd_emit(ctx, 1, 1);		/* 0000ffff BLOCK_ALLOC_THREADS */
+	dd_emit(ctx, 1, 1);		/* 00000001 LANES32 */
+	dd_emit(ctx, 1, 0);		/* 000000ff UNK370 */
+	dd_emit(ctx, 1, 0);		/* 000000ff USER_PARAM_UNK */
+	dd_emit(ctx, 1, 0);		/* 000000ff USER_PARAM_COUNT */
+	dd_emit(ctx, 1, 1);		/* 000000ff UNK384 bits 8-15 */
+	dd_emit(ctx, 1, 0x3fffff);	/* 003fffff TIC_LIMIT */
+	dd_emit(ctx, 1, 0x1fff);	/* 000fffff TSC_LIMIT */
+	dd_emit(ctx, 1, 0);		/* 0000ffff CB_ADDR_INDEX */
+	dd_emit(ctx, 1, 1);		/* 000007ff BLOCKDIM_X */
+	dd_emit(ctx, 1, 1);		/* 000007ff BLOCKDIM_XMY */
+	dd_emit(ctx, 1, 0);		/* 00000001 BLOCKDIM_XMY_OVERFLOW */
+	dd_emit(ctx, 1, 1);		/* 0003ffff BLOCKDIM_XMYMZ */
+	dd_emit(ctx, 1, 1);		/* 000007ff BLOCKDIM_Y */
+	dd_emit(ctx, 1, 1);		/* 0000007f BLOCKDIM_Z */
+	dd_emit(ctx, 1, 4);		/* 000000ff CP_REG_ALLOC_TEMP */
+	dd_emit(ctx, 1, 1);		/* 00000001 BLOCKDIM_DIRTY */
+	if (IS_NVA3F(device->chipset))
+		dd_emit(ctx, 1, 0);	/* 00000003 UNK03E8 */
+	dd_emit(ctx, 1, 1);		/* 0000007f BLOCK_ALLOC_HALFWARPS */
+	dd_emit(ctx, 1, 1);		/* 00000007 LOCAL_WARPS_NO_CLAMP */
+	dd_emit(ctx, 1, 7);		/* 00000007 LOCAL_WARPS_LOG_ALLOC */
+	dd_emit(ctx, 1, 1);		/* 00000007 STACK_WARPS_NO_CLAMP */
+	dd_emit(ctx, 1, 7);		/* 00000007 STACK_WARPS_LOG_ALLOC */
+	dd_emit(ctx, 1, 1);		/* 00001fff BLOCK_ALLOC_REGSLOTS_PACKED */
+	dd_emit(ctx, 1, 1);		/* 00001fff BLOCK_ALLOC_REGSLOTS_STRIDED */
+	dd_emit(ctx, 1, 1);		/* 000007ff BLOCK_ALLOC_THREADS */
+
+	/* compat 2d state */
+	if (device->chipset == 0x50) {
+		dd_emit(ctx, 4, 0);		/* 0000ffff clip X, Y, W, H */
+
+		dd_emit(ctx, 1, 1);		/* ffffffff chroma COLOR_FORMAT */
+
+		dd_emit(ctx, 1, 1);		/* ffffffff pattern COLOR_FORMAT */
+		dd_emit(ctx, 1, 0);		/* ffffffff pattern SHAPE */
+		dd_emit(ctx, 1, 1);		/* ffffffff pattern PATTERN_SELECT */
+
+		dd_emit(ctx, 1, 0xa);		/* ffffffff surf2d SRC_FORMAT */
+		dd_emit(ctx, 1, 0);		/* ffffffff surf2d DMA_SRC */
+		dd_emit(ctx, 1, 0);		/* 000000ff surf2d SRC_ADDRESS_HIGH */
+		dd_emit(ctx, 1, 0);		/* ffffffff surf2d SRC_ADDRESS_LOW */
+		dd_emit(ctx, 1, 0x40);		/* 0000ffff surf2d SRC_PITCH */
+		dd_emit(ctx, 1, 0);		/* 0000000f surf2d SRC_TILE_MODE_Z */
+		dd_emit(ctx, 1, 2);		/* 0000000f surf2d SRC_TILE_MODE_Y */
+		dd_emit(ctx, 1, 0x100);		/* ffffffff surf2d SRC_HEIGHT */
+		dd_emit(ctx, 1, 1);		/* 00000001 surf2d SRC_LINEAR */
+		dd_emit(ctx, 1, 0x100);		/* ffffffff surf2d SRC_WIDTH */
+
+		dd_emit(ctx, 1, 0);		/* 0000ffff gdirect CLIP_B_X */
+		dd_emit(ctx, 1, 0);		/* 0000ffff gdirect CLIP_B_Y */
+		dd_emit(ctx, 1, 0);		/* 0000ffff gdirect CLIP_C_X */
+		dd_emit(ctx, 1, 0);		/* 0000ffff gdirect CLIP_C_Y */
+		dd_emit(ctx, 1, 0);		/* 0000ffff gdirect CLIP_D_X */
+		dd_emit(ctx, 1, 0);		/* 0000ffff gdirect CLIP_D_Y */
+		dd_emit(ctx, 1, 1);		/* ffffffff gdirect COLOR_FORMAT */
+		dd_emit(ctx, 1, 0);		/* ffffffff gdirect OPERATION */
+		dd_emit(ctx, 1, 0);		/* 0000ffff gdirect POINT_X */
+		dd_emit(ctx, 1, 0);		/* 0000ffff gdirect POINT_Y */
+
+		dd_emit(ctx, 1, 0);		/* 0000ffff blit SRC_Y */
+		dd_emit(ctx, 1, 0);		/* ffffffff blit OPERATION */
+
+		dd_emit(ctx, 1, 0);		/* ffffffff ifc OPERATION */
+
+		dd_emit(ctx, 1, 0);		/* ffffffff iifc INDEX_FORMAT */
+		dd_emit(ctx, 1, 0);		/* ffffffff iifc LUT_OFFSET */
+		dd_emit(ctx, 1, 4);		/* ffffffff iifc COLOR_FORMAT */
+		dd_emit(ctx, 1, 0);		/* ffffffff iifc OPERATION */
+	}
+
+	/* m2mf state */
+	dd_emit(ctx, 1, 0);		/* ffffffff m2mf LINE_COUNT */
+	dd_emit(ctx, 1, 0);		/* ffffffff m2mf LINE_LENGTH_IN */
+	dd_emit(ctx, 2, 0);		/* ffffffff m2mf OFFSET_IN, OFFSET_OUT */
+	dd_emit(ctx, 1, 1);		/* ffffffff m2mf TILING_DEPTH_OUT */
+	dd_emit(ctx, 1, 0x100);		/* ffffffff m2mf TILING_HEIGHT_OUT */
+	dd_emit(ctx, 1, 0);		/* ffffffff m2mf TILING_POSITION_OUT_Z */
+	dd_emit(ctx, 1, 1);		/* 00000001 m2mf LINEAR_OUT */
+	dd_emit(ctx, 2, 0);		/* 0000ffff m2mf TILING_POSITION_OUT_X, Y */
+	dd_emit(ctx, 1, 0x100);		/* ffffffff m2mf TILING_PITCH_OUT */
+	dd_emit(ctx, 1, 1);		/* ffffffff m2mf TILING_DEPTH_IN */
+	dd_emit(ctx, 1, 0x100);		/* ffffffff m2mf TILING_HEIGHT_IN */
+	dd_emit(ctx, 1, 0);		/* ffffffff m2mf TILING_POSITION_IN_Z */
+	dd_emit(ctx, 1, 1);		/* 00000001 m2mf LINEAR_IN */
+	dd_emit(ctx, 2, 0);		/* 0000ffff m2mf TILING_POSITION_IN_X, Y */
+	dd_emit(ctx, 1, 0x100);		/* ffffffff m2mf TILING_PITCH_IN */
+
+	/* more compat 2d state */
+	if (device->chipset == 0x50) {
+		dd_emit(ctx, 1, 1);		/* ffffffff line COLOR_FORMAT */
+		dd_emit(ctx, 1, 0);		/* ffffffff line OPERATION */
+
+		dd_emit(ctx, 1, 1);		/* ffffffff triangle COLOR_FORMAT */
+		dd_emit(ctx, 1, 0);		/* ffffffff triangle OPERATION */
+
+		dd_emit(ctx, 1, 0);		/* 0000000f sifm TILE_MODE_Z */
+		dd_emit(ctx, 1, 2);		/* 0000000f sifm TILE_MODE_Y */
+		dd_emit(ctx, 1, 0);		/* 000000ff sifm FORMAT_FILTER */
+		dd_emit(ctx, 1, 1);		/* 000000ff sifm FORMAT_ORIGIN */
+		dd_emit(ctx, 1, 0);		/* 0000ffff sifm SRC_PITCH */
+		dd_emit(ctx, 1, 1);		/* 00000001 sifm SRC_LINEAR */
+		dd_emit(ctx, 1, 0);		/* 000000ff sifm SRC_OFFSET_HIGH */
+		dd_emit(ctx, 1, 0);		/* ffffffff sifm SRC_OFFSET */
+		dd_emit(ctx, 1, 0);		/* 0000ffff sifm SRC_HEIGHT */
+		dd_emit(ctx, 1, 0);		/* 0000ffff sifm SRC_WIDTH */
+		dd_emit(ctx, 1, 3);		/* ffffffff sifm COLOR_FORMAT */
+		dd_emit(ctx, 1, 0);		/* ffffffff sifm OPERATION */
+
+		dd_emit(ctx, 1, 0);		/* ffffffff sifc OPERATION */
+	}
+
+	/* tesla state */
+	dd_emit(ctx, 1, 0);		/* 0000000f GP_TEXTURES_LOG2 */
+	dd_emit(ctx, 1, 0);		/* 0000000f GP_SAMPLERS_LOG2 */
+	dd_emit(ctx, 1, 0);		/* 000000ff */
+	dd_emit(ctx, 1, 0);		/* ffffffff */
+	dd_emit(ctx, 1, 4);		/* 000000ff UNK12B0_0 */
+	dd_emit(ctx, 1, 0x70);		/* 000000ff UNK12B0_1 */
+	dd_emit(ctx, 1, 0x80);		/* 000000ff UNK12B0_3 */
+	dd_emit(ctx, 1, 0);		/* 000000ff UNK12B0_2 */
+	dd_emit(ctx, 1, 0);		/* 0000000f FP_TEXTURES_LOG2 */
+	dd_emit(ctx, 1, 0);		/* 0000000f FP_SAMPLERS_LOG2 */
+	if (IS_NVA3F(device->chipset)) {
+		dd_emit(ctx, 1, 0);	/* ffffffff */
+		dd_emit(ctx, 1, 0);	/* 0000007f MULTISAMPLE_SAMPLES_LOG2 */
+	} else {
+		dd_emit(ctx, 1, 0);	/* 0000000f MULTISAMPLE_SAMPLES_LOG2 */
+	}
+	dd_emit(ctx, 1, 0xc);		/* 000000ff SEMANTIC_COLOR.BFC0_ID */
+	if (device->chipset != 0x50)
+		dd_emit(ctx, 1, 0);	/* 00000001 SEMANTIC_COLOR.CLMP_EN */
+	dd_emit(ctx, 1, 8);		/* 000000ff SEMANTIC_COLOR.COLR_NR */
+	dd_emit(ctx, 1, 0x14);		/* 000000ff SEMANTIC_COLOR.FFC0_ID */
+	if (device->chipset == 0x50) {
+		dd_emit(ctx, 1, 0);	/* 000000ff SEMANTIC_LAYER */
+		dd_emit(ctx, 1, 0);	/* 00000001 */
+	} else {
+		dd_emit(ctx, 1, 0);	/* 00000001 SEMANTIC_PTSZ.ENABLE */
+		dd_emit(ctx, 1, 0x29);	/* 000000ff SEMANTIC_PTSZ.PTSZ_ID */
+		dd_emit(ctx, 1, 0x27);	/* 000000ff SEMANTIC_PRIM */
+		dd_emit(ctx, 1, 0x26);	/* 000000ff SEMANTIC_LAYER */
+		dd_emit(ctx, 1, 8);	/* 0000000f SMENATIC_CLIP.CLIP_HIGH */
+		dd_emit(ctx, 1, 4);	/* 000000ff SEMANTIC_CLIP.CLIP_LO */
+		dd_emit(ctx, 1, 0x27);	/* 000000ff UNK0FD4 */
+		dd_emit(ctx, 1, 0);	/* 00000001 UNK1900 */
+	}
+	dd_emit(ctx, 1, 0);		/* 00000007 RT_CONTROL_MAP0 */
+	dd_emit(ctx, 1, 1);		/* 00000007 RT_CONTROL_MAP1 */
+	dd_emit(ctx, 1, 2);		/* 00000007 RT_CONTROL_MAP2 */
+	dd_emit(ctx, 1, 3);		/* 00000007 RT_CONTROL_MAP3 */
+	dd_emit(ctx, 1, 4);		/* 00000007 RT_CONTROL_MAP4 */
+	dd_emit(ctx, 1, 5);		/* 00000007 RT_CONTROL_MAP5 */
+	dd_emit(ctx, 1, 6);		/* 00000007 RT_CONTROL_MAP6 */
+	dd_emit(ctx, 1, 7);		/* 00000007 RT_CONTROL_MAP7 */
+	dd_emit(ctx, 1, 1);		/* 0000000f RT_CONTROL_COUNT */
+	dd_emit(ctx, 8, 0);		/* 00000001 RT_HORIZ_UNK */
+	dd_emit(ctx, 8, 0);		/* ffffffff RT_ADDRESS_LOW */
+	dd_emit(ctx, 1, 0xcf);		/* 000000ff RT_FORMAT */
+	dd_emit(ctx, 7, 0);		/* 000000ff RT_FORMAT */
+	if (device->chipset != 0x50)
+		dd_emit(ctx, 3, 0);	/* 1, 1, 1 */
+	else
+		dd_emit(ctx, 2, 0);	/* 1, 1 */
+	dd_emit(ctx, 1, 0);		/* ffffffff GP_ENABLE */
+	dd_emit(ctx, 1, 0x80);		/* 0000ffff GP_VERTEX_OUTPUT_COUNT*/
+	dd_emit(ctx, 1, 4);		/* 000000ff GP_REG_ALLOC_RESULT */
+	dd_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	if (IS_NVA3F(device->chipset)) {
+		dd_emit(ctx, 1, 3);	/* 00000003 */
+		dd_emit(ctx, 1, 0);	/* 00000001 UNK1418. Alone. */
+	}
+	if (device->chipset != 0x50)
+		dd_emit(ctx, 1, 3);	/* 00000003 UNK15AC */
+	dd_emit(ctx, 1, 1);		/* ffffffff RASTERIZE_ENABLE */
+	dd_emit(ctx, 1, 0);		/* 00000001 FP_CONTROL.EXPORTS_Z */
+	if (device->chipset != 0x50)
+		dd_emit(ctx, 1, 0);	/* 00000001 FP_CONTROL.MULTIPLE_RESULTS */
+	dd_emit(ctx, 1, 0x12);		/* 000000ff FP_INTERPOLANT_CTRL.COUNT */
+	dd_emit(ctx, 1, 0x10);		/* 000000ff FP_INTERPOLANT_CTRL.COUNT_NONFLAT */
+	dd_emit(ctx, 1, 0xc);		/* 000000ff FP_INTERPOLANT_CTRL.OFFSET */
+	dd_emit(ctx, 1, 1);		/* 00000001 FP_INTERPOLANT_CTRL.UMASK.W */
+	dd_emit(ctx, 1, 0);		/* 00000001 FP_INTERPOLANT_CTRL.UMASK.X */
+	dd_emit(ctx, 1, 0);		/* 00000001 FP_INTERPOLANT_CTRL.UMASK.Y */
+	dd_emit(ctx, 1, 0);		/* 00000001 FP_INTERPOLANT_CTRL.UMASK.Z */
+	dd_emit(ctx, 1, 4);		/* 000000ff FP_RESULT_COUNT */
+	dd_emit(ctx, 1, 2);		/* ffffffff REG_MODE */
+	dd_emit(ctx, 1, 4);		/* 000000ff FP_REG_ALLOC_TEMP */
+	if (device->chipset >= 0xa0)
+		dd_emit(ctx, 1, 0);	/* ffffffff */
+	dd_emit(ctx, 1, 0);		/* 00000001 GP_BUILTIN_RESULT_EN.LAYER_IDX */
+	dd_emit(ctx, 1, 0);		/* ffffffff STRMOUT_ENABLE */
+	dd_emit(ctx, 1, 0x3fffff);	/* 003fffff TIC_LIMIT */
+	dd_emit(ctx, 1, 0x1fff);	/* 000fffff TSC_LIMIT */
+	dd_emit(ctx, 1, 0);		/* 00000001 VERTEX_TWO_SIDE_ENABLE*/
+	if (device->chipset != 0x50)
+		dd_emit(ctx, 8, 0);	/* 00000001 */
+	if (device->chipset >= 0xa0) {
+		dd_emit(ctx, 1, 1);	/* 00000007 VTX_ATTR_DEFINE.COMP */
+		dd_emit(ctx, 1, 1);	/* 00000007 VTX_ATTR_DEFINE.SIZE */
+		dd_emit(ctx, 1, 2);	/* 00000007 VTX_ATTR_DEFINE.TYPE */
+		dd_emit(ctx, 1, 0);	/* 000000ff VTX_ATTR_DEFINE.ATTR */
+	}
+	dd_emit(ctx, 1, 4);		/* 0000007f VP_RESULT_MAP_SIZE */
+	dd_emit(ctx, 1, 0x14);		/* 0000001f ZETA_FORMAT */
+	dd_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	dd_emit(ctx, 1, 0);		/* 0000000f VP_TEXTURES_LOG2 */
+	dd_emit(ctx, 1, 0);		/* 0000000f VP_SAMPLERS_LOG2 */
+	if (IS_NVA3F(device->chipset))
+		dd_emit(ctx, 1, 0);	/* 00000001 */
+	dd_emit(ctx, 1, 2);		/* 00000003 POLYGON_MODE_BACK */
+	if (device->chipset >= 0xa0)
+		dd_emit(ctx, 1, 0);	/* 00000003 VTX_ATTR_DEFINE.SIZE - 1 */
+	dd_emit(ctx, 1, 0);		/* 0000ffff CB_ADDR_INDEX */
+	if (device->chipset >= 0xa0)
+		dd_emit(ctx, 1, 0);	/* 00000003 */
+	dd_emit(ctx, 1, 0);		/* 00000001 CULL_FACE_ENABLE */
+	dd_emit(ctx, 1, 1);		/* 00000003 CULL_FACE */
+	dd_emit(ctx, 1, 0);		/* 00000001 FRONT_FACE */
+	dd_emit(ctx, 1, 2);		/* 00000003 POLYGON_MODE_FRONT */
+	dd_emit(ctx, 1, 0x1000);	/* 00007fff UNK141C */
+	if (device->chipset != 0x50) {
+		dd_emit(ctx, 1, 0xe00);		/* 7fff */
+		dd_emit(ctx, 1, 0x1000);	/* 7fff */
+		dd_emit(ctx, 1, 0x1e00);	/* 7fff */
+	}
+	dd_emit(ctx, 1, 0);		/* 00000001 BEGIN_END_ACTIVE */
+	dd_emit(ctx, 1, 1);		/* 00000001 POLYGON_MODE_??? */
+	dd_emit(ctx, 1, 1);		/* 000000ff GP_REG_ALLOC_TEMP / 4 rounded up */
+	dd_emit(ctx, 1, 1);		/* 000000ff FP_REG_ALLOC_TEMP... without /4? */
+	dd_emit(ctx, 1, 1);		/* 000000ff VP_REG_ALLOC_TEMP / 4 rounded up */
+	dd_emit(ctx, 1, 1);		/* 00000001 */
+	dd_emit(ctx, 1, 0);		/* 00000001 */
+	dd_emit(ctx, 1, 0);		/* 00000001 VTX_ATTR_MASK_UNK0 nonempty */
+	dd_emit(ctx, 1, 0);		/* 00000001 VTX_ATTR_MASK_UNK1 nonempty */
+	dd_emit(ctx, 1, 0x200);		/* 0003ffff GP_VERTEX_OUTPUT_COUNT*GP_REG_ALLOC_RESULT */
+	if (IS_NVA3F(device->chipset))
+		dd_emit(ctx, 1, 0x200);
+	dd_emit(ctx, 1, 0);		/* 00000001 */
+	if (device->chipset < 0xa0) {
+		dd_emit(ctx, 1, 1);	/* 00000001 */
+		dd_emit(ctx, 1, 0x70);	/* 000000ff */
+		dd_emit(ctx, 1, 0x80);	/* 000000ff */
+		dd_emit(ctx, 1, 0);	/* 000000ff */
+		dd_emit(ctx, 1, 0);	/* 00000001 */
+		dd_emit(ctx, 1, 1);	/* 00000001 */
+		dd_emit(ctx, 1, 0x70);	/* 000000ff */
+		dd_emit(ctx, 1, 0x80);	/* 000000ff */
+		dd_emit(ctx, 1, 0);	/* 000000ff */
+	} else {
+		dd_emit(ctx, 1, 1);	/* 00000001 */
+		dd_emit(ctx, 1, 0xf0);	/* 000000ff */
+		dd_emit(ctx, 1, 0xff);	/* 000000ff */
+		dd_emit(ctx, 1, 0);	/* 000000ff */
+		dd_emit(ctx, 1, 0);	/* 00000001 */
+		dd_emit(ctx, 1, 1);	/* 00000001 */
+		dd_emit(ctx, 1, 0xf0);	/* 000000ff */
+		dd_emit(ctx, 1, 0xff);	/* 000000ff */
+		dd_emit(ctx, 1, 0);	/* 000000ff */
+		dd_emit(ctx, 1, 9);	/* 0000003f UNK114C.COMP,SIZE */
+	}
+
+	/* eng2d state */
+	dd_emit(ctx, 1, 0);		/* 00000001 eng2d COLOR_KEY_ENABLE */
+	dd_emit(ctx, 1, 0);		/* 00000007 eng2d COLOR_KEY_FORMAT */
+	dd_emit(ctx, 1, 1);		/* ffffffff eng2d DST_DEPTH */
+	dd_emit(ctx, 1, 0xcf);		/* 000000ff eng2d DST_FORMAT */
+	dd_emit(ctx, 1, 0);		/* ffffffff eng2d DST_LAYER */
+	dd_emit(ctx, 1, 1);		/* 00000001 eng2d DST_LINEAR */
+	dd_emit(ctx, 1, 0);		/* 00000007 eng2d PATTERN_COLOR_FORMAT */
+	dd_emit(ctx, 1, 0);		/* 00000007 eng2d OPERATION */
+	dd_emit(ctx, 1, 0);		/* 00000003 eng2d PATTERN_SELECT */
+	dd_emit(ctx, 1, 0xcf);		/* 000000ff eng2d SIFC_FORMAT */
+	dd_emit(ctx, 1, 0);		/* 00000001 eng2d SIFC_BITMAP_ENABLE */
+	dd_emit(ctx, 1, 2);		/* 00000003 eng2d SIFC_BITMAP_UNK808 */
+	dd_emit(ctx, 1, 0);		/* ffffffff eng2d BLIT_DU_DX_FRACT */
+	dd_emit(ctx, 1, 1);		/* ffffffff eng2d BLIT_DU_DX_INT */
+	dd_emit(ctx, 1, 0);		/* ffffffff eng2d BLIT_DV_DY_FRACT */
+	dd_emit(ctx, 1, 1);		/* ffffffff eng2d BLIT_DV_DY_INT */
+	dd_emit(ctx, 1, 0);		/* 00000001 eng2d BLIT_CONTROL_FILTER */
+	dd_emit(ctx, 1, 0xcf);		/* 000000ff eng2d DRAW_COLOR_FORMAT */
+	dd_emit(ctx, 1, 0xcf);		/* 000000ff eng2d SRC_FORMAT */
+	dd_emit(ctx, 1, 1);		/* 00000001 eng2d SRC_LINEAR #2 */
+
+	num = ctx->ctxvals_pos - base;
+	ctx->ctxvals_pos = base;
+	if (IS_NVA3F(device->chipset))
+		cp_ctx(ctx, 0x404800, num);
+	else
+		cp_ctx(ctx, 0x405400, num);
+}
+
+/*
+ * xfer areas. These are a pain.
+ *
+ * There are 2 xfer areas: the first one is big and contains all sorts of
+ * stuff, the second is small and contains some per-TP context.
+ *
+ * Each area is split into 8 "strands". The areas, when saved to grctx,
+ * are made of 8-word blocks. Each block contains a single word from
+ * each strand. The strands are independent of each other, their
+ * addresses are unrelated to each other, and data in them is closely
+ * packed together. The strand layout varies a bit between cards: here
+ * and there, a single word is thrown out in the middle and the whole
+ * strand is offset by a bit from corresponding one on another chipset.
+ * For this reason, addresses of stuff in strands are almost useless.
+ * Knowing sequence of stuff and size of gaps between them is much more
+ * useful, and that's how we build the strands in our generator.
+ *
+ * NVA0 takes this mess to a whole new level by cutting the old strands
+ * into a few dozen pieces [known as genes], rearranging them randomly,
+ * and putting them back together to make new strands. Hopefully these
+ * genes correspond more or less directly to the same PGRAPH subunits
+ * as in 400040 register.
+ *
+ * The most common value in default context is 0, and when the genes
+ * are separated by 0's, gene bounduaries are quite speculative...
+ * some of them can be clearly deduced, others can be guessed, and yet
+ * others won't be resolved without figuring out the real meaning of
+ * given ctxval. For the same reason, ending point of each strand
+ * is unknown. Except for strand 0, which is the longest strand and
+ * its end corresponds to end of the whole xfer.
+ *
+ * An unsolved mystery is the seek instruction: it takes an argument
+ * in bits 8-18, and that argument is clearly the place in strands to
+ * seek to... but the offsets don't seem to correspond to offsets as
+ * seen in grctx. Perhaps there's another, real, not randomly-changing
+ * addressing in strands, and the xfer insn just happens to skip over
+ * the unused bits? NV10-NV30 PIPE comes to mind...
+ *
+ * As far as I know, there's no way to access the xfer areas directly
+ * without the help of ctxprog.
+ */
+
+static void
+xf_emit(struct nouveau_grctx *ctx, int num, u32 val) {
+	int i;
+	if (val && ctx->mode == NOUVEAU_GRCTX_VALS)
+		for (i = 0; i < num; i++)
+			nv_wo32(ctx->data, 4 * (ctx->ctxvals_pos + (i << 3)), val);
+	ctx->ctxvals_pos += num << 3;
+}
+
+/* Gene declarations... */
+
+static void nv50_graph_construct_gene_dispatch(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_m2mf(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_ccache(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_unk10xx(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_unk14xx(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_zcull(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_clipid(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_unk24xx(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_vfetch(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_eng2d(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_csched(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_unk1cxx(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_strmout(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_unk34xx(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_ropm1(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_ropm2(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_gene_ropc(struct nouveau_grctx *ctx);
+static void nv50_graph_construct_xfer_tp(struct nouveau_grctx *ctx);
+
+static void
+nv50_graph_construct_xfer1(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int i;
+	int offset;
+	int size = 0;
+	u32 units = nv_rd32 (ctx->device, 0x1540);
+
+	offset = (ctx->ctxvals_pos+0x3f)&~0x3f;
+	ctx->ctxvals_base = offset;
+
+	if (device->chipset < 0xa0) {
+		/* Strand 0 */
+		ctx->ctxvals_pos = offset;
+		nv50_graph_construct_gene_dispatch(ctx);
+		nv50_graph_construct_gene_m2mf(ctx);
+		nv50_graph_construct_gene_unk24xx(ctx);
+		nv50_graph_construct_gene_clipid(ctx);
+		nv50_graph_construct_gene_zcull(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 1 */
+		ctx->ctxvals_pos = offset + 0x1;
+		nv50_graph_construct_gene_vfetch(ctx);
+		nv50_graph_construct_gene_eng2d(ctx);
+		nv50_graph_construct_gene_csched(ctx);
+		nv50_graph_construct_gene_ropm1(ctx);
+		nv50_graph_construct_gene_ropm2(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 2 */
+		ctx->ctxvals_pos = offset + 0x2;
+		nv50_graph_construct_gene_ccache(ctx);
+		nv50_graph_construct_gene_unk1cxx(ctx);
+		nv50_graph_construct_gene_strmout(ctx);
+		nv50_graph_construct_gene_unk14xx(ctx);
+		nv50_graph_construct_gene_unk10xx(ctx);
+		nv50_graph_construct_gene_unk34xx(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 3: per-ROP group state */
+		ctx->ctxvals_pos = offset + 3;
+		for (i = 0; i < 6; i++)
+			if (units & (1 << (i + 16)))
+				nv50_graph_construct_gene_ropc(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strands 4-7: per-TP state */
+		for (i = 0; i < 4; i++) {
+			ctx->ctxvals_pos = offset + 4 + i;
+			if (units & (1 << (2 * i)))
+				nv50_graph_construct_xfer_tp(ctx);
+			if (units & (1 << (2 * i + 1)))
+				nv50_graph_construct_xfer_tp(ctx);
+			if ((ctx->ctxvals_pos-offset)/8 > size)
+				size = (ctx->ctxvals_pos-offset)/8;
+		}
+	} else {
+		/* Strand 0 */
+		ctx->ctxvals_pos = offset;
+		nv50_graph_construct_gene_dispatch(ctx);
+		nv50_graph_construct_gene_m2mf(ctx);
+		nv50_graph_construct_gene_unk34xx(ctx);
+		nv50_graph_construct_gene_csched(ctx);
+		nv50_graph_construct_gene_unk1cxx(ctx);
+		nv50_graph_construct_gene_strmout(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 1 */
+		ctx->ctxvals_pos = offset + 1;
+		nv50_graph_construct_gene_unk10xx(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 2 */
+		ctx->ctxvals_pos = offset + 2;
+		if (device->chipset == 0xa0)
+			nv50_graph_construct_gene_unk14xx(ctx);
+		nv50_graph_construct_gene_unk24xx(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 3 */
+		ctx->ctxvals_pos = offset + 3;
+		nv50_graph_construct_gene_vfetch(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 4 */
+		ctx->ctxvals_pos = offset + 4;
+		nv50_graph_construct_gene_ccache(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 5 */
+		ctx->ctxvals_pos = offset + 5;
+		nv50_graph_construct_gene_ropm2(ctx);
+		nv50_graph_construct_gene_ropm1(ctx);
+		/* per-ROP context */
+		for (i = 0; i < 8; i++)
+			if (units & (1<<(i+16)))
+				nv50_graph_construct_gene_ropc(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 6 */
+		ctx->ctxvals_pos = offset + 6;
+		nv50_graph_construct_gene_zcull(ctx);
+		nv50_graph_construct_gene_clipid(ctx);
+		nv50_graph_construct_gene_eng2d(ctx);
+		if (units & (1 << 0))
+			nv50_graph_construct_xfer_tp(ctx);
+		if (units & (1 << 1))
+			nv50_graph_construct_xfer_tp(ctx);
+		if (units & (1 << 2))
+			nv50_graph_construct_xfer_tp(ctx);
+		if (units & (1 << 3))
+			nv50_graph_construct_xfer_tp(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 7 */
+		ctx->ctxvals_pos = offset + 7;
+		if (device->chipset == 0xa0) {
+			if (units & (1 << 4))
+				nv50_graph_construct_xfer_tp(ctx);
+			if (units & (1 << 5))
+				nv50_graph_construct_xfer_tp(ctx);
+			if (units & (1 << 6))
+				nv50_graph_construct_xfer_tp(ctx);
+			if (units & (1 << 7))
+				nv50_graph_construct_xfer_tp(ctx);
+			if (units & (1 << 8))
+				nv50_graph_construct_xfer_tp(ctx);
+			if (units & (1 << 9))
+				nv50_graph_construct_xfer_tp(ctx);
+		} else {
+			nv50_graph_construct_gene_unk14xx(ctx);
+		}
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+	}
+
+	ctx->ctxvals_pos = offset + size * 8;
+	ctx->ctxvals_pos = (ctx->ctxvals_pos+0x3f)&~0x3f;
+	cp_lsr (ctx, offset);
+	cp_out (ctx, CP_SET_XFER_POINTER);
+	cp_lsr (ctx, size);
+	cp_out (ctx, CP_SEEK_1);
+	cp_out (ctx, CP_XFER_1);
+	cp_wait(ctx, XFER, BUSY);
+}
+
+/*
+ * non-trivial demagiced parts of ctx init go here
+ */
+
+static void
+nv50_graph_construct_gene_dispatch(struct nouveau_grctx *ctx)
+{
+	/* start of strand 0 */
+	struct nouveau_device *device = ctx->device;
+	/* SEEK */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 5, 0);
+	else if (!IS_NVA3F(device->chipset))
+		xf_emit(ctx, 6, 0);
+	else
+		xf_emit(ctx, 4, 0);
+	/* SEEK */
+	/* the PGRAPH's internal FIFO */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 8*3, 0);
+	else
+		xf_emit(ctx, 0x100*3, 0);
+	/* and another bonus slot?!? */
+	xf_emit(ctx, 3, 0);
+	/* and YET ANOTHER bonus slot? */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 3, 0);
+	/* SEEK */
+	/* CTX_SWITCH: caches of gr objects bound to subchannels. 8 values, last used index */
+	xf_emit(ctx, 9, 0);
+	/* SEEK */
+	xf_emit(ctx, 9, 0);
+	/* SEEK */
+	xf_emit(ctx, 9, 0);
+	/* SEEK */
+	xf_emit(ctx, 9, 0);
+	/* SEEK */
+	if (device->chipset < 0x90)
+		xf_emit(ctx, 4, 0);
+	/* SEEK */
+	xf_emit(ctx, 2, 0);
+	/* SEEK */
+	xf_emit(ctx, 6*2, 0);
+	xf_emit(ctx, 2, 0);
+	/* SEEK */
+	xf_emit(ctx, 2, 0);
+	/* SEEK */
+	xf_emit(ctx, 6*2, 0);
+	xf_emit(ctx, 2, 0);
+	/* SEEK */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 0x1c, 0);
+	else if (device->chipset < 0xa0)
+		xf_emit(ctx, 0x1e, 0);
+	else
+		xf_emit(ctx, 0x22, 0);
+	/* SEEK */
+	xf_emit(ctx, 0x15, 0);
+}
+
+static void
+nv50_graph_construct_gene_m2mf(struct nouveau_grctx *ctx)
+{
+	/* Strand 0, right after dispatch */
+	struct nouveau_device *device = ctx->device;
+	int smallm2mf = 0;
+	if (device->chipset < 0x92 || device->chipset == 0x98)
+		smallm2mf = 1;
+	/* SEEK */
+	xf_emit (ctx, 1, 0);		/* DMA_NOTIFY instance >> 4 */
+	xf_emit (ctx, 1, 0);		/* DMA_BUFFER_IN instance >> 4 */
+	xf_emit (ctx, 1, 0);		/* DMA_BUFFER_OUT instance >> 4 */
+	xf_emit (ctx, 1, 0);		/* OFFSET_IN */
+	xf_emit (ctx, 1, 0);		/* OFFSET_OUT */
+	xf_emit (ctx, 1, 0);		/* PITCH_IN */
+	xf_emit (ctx, 1, 0);		/* PITCH_OUT */
+	xf_emit (ctx, 1, 0);		/* LINE_LENGTH */
+	xf_emit (ctx, 1, 0);		/* LINE_COUNT */
+	xf_emit (ctx, 1, 0x21);		/* FORMAT: bits 0-4 INPUT_INC, bits 5-9 OUTPUT_INC */
+	xf_emit (ctx, 1, 1);		/* LINEAR_IN */
+	xf_emit (ctx, 1, 0x2);		/* TILING_MODE_IN: bits 0-2 y tiling, bits 3-5 z tiling */
+	xf_emit (ctx, 1, 0x100);	/* TILING_PITCH_IN */
+	xf_emit (ctx, 1, 0x100);	/* TILING_HEIGHT_IN */
+	xf_emit (ctx, 1, 1);		/* TILING_DEPTH_IN */
+	xf_emit (ctx, 1, 0);		/* TILING_POSITION_IN_Z */
+	xf_emit (ctx, 1, 0);		/* TILING_POSITION_IN */
+	xf_emit (ctx, 1, 1);		/* LINEAR_OUT */
+	xf_emit (ctx, 1, 0x2);		/* TILING_MODE_OUT: bits 0-2 y tiling, bits 3-5 z tiling */
+	xf_emit (ctx, 1, 0x100);	/* TILING_PITCH_OUT */
+	xf_emit (ctx, 1, 0x100);	/* TILING_HEIGHT_OUT */
+	xf_emit (ctx, 1, 1);		/* TILING_DEPTH_OUT */
+	xf_emit (ctx, 1, 0);		/* TILING_POSITION_OUT_Z */
+	xf_emit (ctx, 1, 0);		/* TILING_POSITION_OUT */
+	xf_emit (ctx, 1, 0);		/* OFFSET_IN_HIGH */
+	xf_emit (ctx, 1, 0);		/* OFFSET_OUT_HIGH */
+	/* SEEK */
+	if (smallm2mf)
+		xf_emit(ctx, 0x40, 0);	/* 20 * ffffffff, 3ffff */
+	else
+		xf_emit(ctx, 0x100, 0);	/* 80 * ffffffff, 3ffff */
+	xf_emit(ctx, 4, 0);		/* 1f/7f, 0, 1f/7f, 0 [1f for smallm2mf, 7f otherwise] */
+	/* SEEK */
+	if (smallm2mf)
+		xf_emit(ctx, 0x400, 0);	/* ffffffff */
+	else
+		xf_emit(ctx, 0x800, 0);	/* ffffffff */
+	xf_emit(ctx, 4, 0);		/* ff/1ff, 0, 0, 0 [ff for smallm2mf, 1ff otherwise] */
+	/* SEEK */
+	xf_emit(ctx, 0x40, 0);		/* 20 * bits ffffffff, 3ffff */
+	xf_emit(ctx, 0x6, 0);		/* 1f, 0, 1f, 0, 1f, 0 */
+}
+
+static void
+nv50_graph_construct_gene_ccache(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	xf_emit(ctx, 2, 0);		/* RO */
+	xf_emit(ctx, 0x800, 0);		/* ffffffff */
+	switch (device->chipset) {
+	case 0x50:
+	case 0x92:
+	case 0xa0:
+		xf_emit(ctx, 0x2b, 0);
+		break;
+	case 0x84:
+		xf_emit(ctx, 0x29, 0);
+		break;
+	case 0x94:
+	case 0x96:
+	case 0xa3:
+		xf_emit(ctx, 0x27, 0);
+		break;
+	case 0x86:
+	case 0x98:
+	case 0xa5:
+	case 0xa8:
+	case 0xaa:
+	case 0xac:
+	case 0xaf:
+		xf_emit(ctx, 0x25, 0);
+		break;
+	}
+	/* CB bindings, 0x80 of them. first word is address >> 8, second is
+	 * size >> 4 | valid << 24 */
+	xf_emit(ctx, 0x100, 0);		/* ffffffff CB_DEF */
+	xf_emit(ctx, 1, 0);		/* 0000007f CB_ADDR_BUFFER */
+	xf_emit(ctx, 1, 0);		/* 0 */
+	xf_emit(ctx, 0x30, 0);		/* ff SET_PROGRAM_CB */
+	xf_emit(ctx, 1, 0);		/* 3f last SET_PROGRAM_CB */
+	xf_emit(ctx, 4, 0);		/* RO */
+	xf_emit(ctx, 0x100, 0);		/* ffffffff */
+	xf_emit(ctx, 8, 0);		/* 1f, 0, 0, ... */
+	xf_emit(ctx, 8, 0);		/* ffffffff */
+	xf_emit(ctx, 4, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 3 */
+	xf_emit(ctx, 1, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_CODE_CB */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_TIC */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_TSC */
+	xf_emit(ctx, 1, 0);		/* 00000001 LINKED_TSC */
+	xf_emit(ctx, 1, 0);		/* 000000ff TIC_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff TIC_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0x3fffff);	/* 003fffff TIC_LIMIT */
+	xf_emit(ctx, 1, 0);		/* 000000ff TSC_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff TSC_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0x1fff);	/* 000fffff TSC_LIMIT */
+	xf_emit(ctx, 1, 0);		/* 000000ff VP_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff VP_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0);		/* 00ffffff VP_START_ID */
+	xf_emit(ctx, 1, 0);		/* 000000ff CB_DEF_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff CB_DEF_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 000000ff GP_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff GP_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0);		/* 00ffffff GP_START_ID */
+	xf_emit(ctx, 1, 0);		/* 000000ff FP_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff FP_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0);		/* 00ffffff FP_START_ID */
+}
+
+static void
+nv50_graph_construct_gene_unk10xx(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int i;
+	/* end of area 2 on pre-NVA0, area 1 on NVAx */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 4);		/* 0000007f VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0x80);		/* 0000ffff GP_VERTEX_OUTPUT_COUNT */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_REG_ALLOC_RESULT */
+	xf_emit(ctx, 1, 0x80c14);	/* 01ffffff SEMANTIC_COLOR */
+	xf_emit(ctx, 1, 0);		/* 00000001 VERTEX_TWO_SIDE_ENABLE */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 1, 0x3ff);
+	else
+		xf_emit(ctx, 1, 0x7ff);	/* 000007ff */
+	xf_emit(ctx, 1, 0);		/* 111/113 */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	for (i = 0; i < 8; i++) {
+		switch (device->chipset) {
+		case 0x50:
+		case 0x86:
+		case 0x98:
+		case 0xaa:
+		case 0xac:
+			xf_emit(ctx, 0xa0, 0);	/* ffffffff */
+			break;
+		case 0x84:
+		case 0x92:
+		case 0x94:
+		case 0x96:
+			xf_emit(ctx, 0x120, 0);
+			break;
+		case 0xa5:
+		case 0xa8:
+			xf_emit(ctx, 0x100, 0);	/* ffffffff */
+			break;
+		case 0xa0:
+		case 0xa3:
+		case 0xaf:
+			xf_emit(ctx, 0x400, 0);	/* ffffffff */
+			break;
+		}
+		xf_emit(ctx, 4, 0);	/* 3f, 0, 0, 0 */
+		xf_emit(ctx, 4, 0);	/* ffffffff */
+	}
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 4);		/* 0000007f VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0x80);		/* 0000ffff GP_VERTEX_OUTPUT_COUNT */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_REG_ALLOC_TEMP */
+	xf_emit(ctx, 1, 1);		/* 00000001 RASTERIZE_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1900 */
+	xf_emit(ctx, 1, 0x27);		/* 000000ff UNK0FD4 */
+	xf_emit(ctx, 1, 0);		/* 0001ffff GP_BUILTIN_RESULT_EN */
+	xf_emit(ctx, 1, 0x26);		/* 000000ff SEMANTIC_LAYER */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+}
+
+static void
+nv50_graph_construct_gene_unk34xx(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	/* end of area 2 on pre-NVA0, area 1 on NVAx */
+	xf_emit(ctx, 1, 0);		/* 00000001 VIEWPORT_CLIP_RECTS_EN */
+	xf_emit(ctx, 1, 0);		/* 00000003 VIEWPORT_CLIP_MODE */
+	xf_emit(ctx, 0x10, 0x04000000);	/* 07ffffff VIEWPORT_CLIP_HORIZ*8, VIEWPORT_CLIP_VERT*8 */
+	xf_emit(ctx, 1, 0);		/* 00000001 POLYGON_STIPPLE_ENABLE */
+	xf_emit(ctx, 0x20, 0);		/* ffffffff POLYGON_STIPPLE */
+	xf_emit(ctx, 2, 0);		/* 00007fff WINDOW_OFFSET_XY */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 1, 0x04e3bfdf);	/* ffffffff UNK0D64 */
+	xf_emit(ctx, 1, 0x04e3bfdf);	/* ffffffff UNK0DF4 */
+	xf_emit(ctx, 1, 0);		/* 00000003 WINDOW_ORIGIN */
+	xf_emit(ctx, 1, 0);		/* 00000007 */
+	xf_emit(ctx, 1, 0x1fe21);	/* 0001ffff tesla UNK0FAC */
+	if (device->chipset >= 0xa0)
+		xf_emit(ctx, 1, 0x0fac6881);
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 1, 1);
+		xf_emit(ctx, 3, 0);
+	}
+}
+
+static void
+nv50_graph_construct_gene_unk14xx(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	/* middle of area 2 on pre-NVA0, beginning of area 2 on NVA0, area 7 on >NVA0 */
+	if (device->chipset != 0x50) {
+		xf_emit(ctx, 5, 0);		/* ffffffff */
+		xf_emit(ctx, 1, 0x80c14);	/* 01ffffff SEMANTIC_COLOR */
+		xf_emit(ctx, 1, 0);		/* 00000001 */
+		xf_emit(ctx, 1, 0);		/* 000003ff */
+		xf_emit(ctx, 1, 0x804);		/* 00000fff SEMANTIC_CLIP */
+		xf_emit(ctx, 1, 0);		/* 00000001 */
+		xf_emit(ctx, 2, 4);		/* 7f, ff */
+		xf_emit(ctx, 1, 0x8100c12);	/* 1fffffff FP_INTERPOLANT_CTRL */
+	}
+	xf_emit(ctx, 1, 0);			/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 4);			/* 0000007f VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 4);			/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);			/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0x10);			/* 7f/ff VIEW_VOLUME_CLIP_CTRL */
+	xf_emit(ctx, 1, 0);			/* 000000ff VP_CLIP_DISTANCE_ENABLE */
+	if (device->chipset != 0x50)
+		xf_emit(ctx, 1, 0);		/* 3ff */
+	xf_emit(ctx, 1, 0);			/* 000000ff tesla UNK1940 */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK0D7C */
+	xf_emit(ctx, 1, 0x804);			/* 00000fff SEMANTIC_CLIP */
+	xf_emit(ctx, 1, 1);			/* 00000001 VIEWPORT_TRANSFORM_EN */
+	xf_emit(ctx, 1, 0x1a);			/* 0000001f POLYGON_MODE */
+	if (device->chipset != 0x50)
+		xf_emit(ctx, 1, 0x7f);		/* 000000ff tesla UNK0FFC */
+	xf_emit(ctx, 1, 0);			/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 1);			/* 00000001 SHADE_MODEL */
+	xf_emit(ctx, 1, 0x80c14);		/* 01ffffff SEMANTIC_COLOR */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK1900 */
+	xf_emit(ctx, 1, 0x8100c12);		/* 1fffffff FP_INTERPOLANT_CTRL */
+	xf_emit(ctx, 1, 4);			/* 0000007f VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 4);			/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);			/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0x10);			/* 7f/ff VIEW_VOLUME_CLIP_CTRL */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK0D7C */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK0F8C */
+	xf_emit(ctx, 1, 0);			/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 1);			/* 00000001 VIEWPORT_TRANSFORM_EN */
+	xf_emit(ctx, 1, 0x8100c12);		/* 1fffffff FP_INTERPOLANT_CTRL */
+	xf_emit(ctx, 4, 0);			/* ffffffff NOPERSPECTIVE_BITMAP */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK1900 */
+	xf_emit(ctx, 1, 0);			/* 0000000f */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 1, 0x3ff);		/* 000003ff tesla UNK0D68 */
+	else
+		xf_emit(ctx, 1, 0x7ff);		/* 000007ff tesla UNK0D68 */
+	xf_emit(ctx, 1, 0x80c14);		/* 01ffffff SEMANTIC_COLOR */
+	xf_emit(ctx, 1, 0);			/* 00000001 VERTEX_TWO_SIDE_ENABLE */
+	xf_emit(ctx, 0x30, 0);			/* ffffffff VIEWPORT_SCALE: X0, Y0, Z0, X1, Y1, ... */
+	xf_emit(ctx, 3, 0);			/* f, 0, 0 */
+	xf_emit(ctx, 3, 0);			/* ffffffff last VIEWPORT_SCALE? */
+	xf_emit(ctx, 1, 0);			/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 1);			/* 00000001 VIEWPORT_TRANSFORM_EN */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK1900 */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK1924 */
+	xf_emit(ctx, 1, 0x10);			/* 000000ff VIEW_VOLUME_CLIP_CTRL */
+	xf_emit(ctx, 1, 0);			/* 00000001 */
+	xf_emit(ctx, 0x30, 0);			/* ffffffff VIEWPORT_TRANSLATE */
+	xf_emit(ctx, 3, 0);			/* f, 0, 0 */
+	xf_emit(ctx, 3, 0);			/* ffffffff */
+	xf_emit(ctx, 1, 0);			/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 2, 0x88);			/* 000001ff tesla UNK19D8 */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK1924 */
+	xf_emit(ctx, 1, 0);			/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 4);			/* 0000000f CULL_MODE */
+	xf_emit(ctx, 2, 0);			/* 07ffffff SCREEN_SCISSOR */
+	xf_emit(ctx, 2, 0);			/* 00007fff WINDOW_OFFSET_XY */
+	xf_emit(ctx, 1, 0);			/* 00000003 WINDOW_ORIGIN */
+	xf_emit(ctx, 0x10, 0);			/* 00000001 SCISSOR_ENABLE */
+	xf_emit(ctx, 1, 0);			/* 0001ffff GP_BUILTIN_RESULT_EN */
+	xf_emit(ctx, 1, 0x26);			/* 000000ff SEMANTIC_LAYER */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK1900 */
+	xf_emit(ctx, 1, 0);			/* 0000000f */
+	xf_emit(ctx, 1, 0x3f800000);		/* ffffffff LINE_WIDTH */
+	xf_emit(ctx, 1, 0);			/* 00000001 LINE_STIPPLE_ENABLE */
+	xf_emit(ctx, 1, 0);			/* 00000001 LINE_SMOOTH_ENABLE */
+	xf_emit(ctx, 1, 0);			/* 00000007 MULTISAMPLE_SAMPLES_LOG2 */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 0);		/* 00000001 */
+	xf_emit(ctx, 1, 0x1a);			/* 0000001f POLYGON_MODE */
+	xf_emit(ctx, 1, 0x10);			/* 000000ff VIEW_VOLUME_CLIP_CTRL */
+	if (device->chipset != 0x50) {
+		xf_emit(ctx, 1, 0);		/* ffffffff */
+		xf_emit(ctx, 1, 0);		/* 00000001 */
+		xf_emit(ctx, 1, 0);		/* 000003ff */
+	}
+	xf_emit(ctx, 0x20, 0);			/* 10xbits ffffffff, 3fffff. SCISSOR_* */
+	xf_emit(ctx, 1, 0);			/* f */
+	xf_emit(ctx, 1, 0);			/* 0? */
+	xf_emit(ctx, 1, 0);			/* ffffffff */
+	xf_emit(ctx, 1, 0);			/* 003fffff */
+	xf_emit(ctx, 1, 0);			/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 0x52);			/* 000001ff SEMANTIC_PTSZ */
+	xf_emit(ctx, 1, 0);			/* 0001ffff GP_BUILTIN_RESULT_EN */
+	xf_emit(ctx, 1, 0x26);			/* 000000ff SEMANTIC_LAYER */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK1900 */
+	xf_emit(ctx, 1, 4);			/* 0000007f VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 4);			/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);			/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0x1a);			/* 0000001f POLYGON_MODE */
+	xf_emit(ctx, 1, 0);			/* 00000001 LINE_SMOOTH_ENABLE */
+	xf_emit(ctx, 1, 0);			/* 00000001 LINE_STIPPLE_ENABLE */
+	xf_emit(ctx, 1, 0x00ffff00);		/* 00ffffff LINE_STIPPLE_PATTERN */
+	xf_emit(ctx, 1, 0);			/* 0000000f */
+}
+
+static void
+nv50_graph_construct_gene_zcull(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	/* end of strand 0 on pre-NVA0, beginning of strand 6 on NVAx */
+	/* SEEK */
+	xf_emit(ctx, 1, 0x3f);		/* 0000003f UNK1590 */
+	xf_emit(ctx, 1, 0);		/* 00000001 ALPHA_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000007 MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_BACK_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_FUNC_REF */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_MASK */
+	xf_emit(ctx, 3, 0);		/* 00000007 STENCIL_BACK_OP_FAIL, ZFAIL, ZPASS */
+	xf_emit(ctx, 1, 2);		/* 00000003 tesla UNK143C */
+	xf_emit(ctx, 2, 0x04000000);	/* 07ffffff tesla UNK0D6C */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 1, 0);		/* 00000001 CLIPID_ENABLE */
+	xf_emit(ctx, 2, 0);		/* ffffffff DEPTH_BOUNDS */
+	xf_emit(ctx, 1, 0);		/* 00000001 */
+	xf_emit(ctx, 1, 0);		/* 00000007 DEPTH_TEST_FUNC */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 0000000f CULL_MODE */
+	xf_emit(ctx, 1, 0);		/* 0000ffff */
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK0FB0 */
+	xf_emit(ctx, 1, 0);		/* 00000001 POLYGON_STIPPLE_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 00000007 FP_CONTROL */
+	xf_emit(ctx, 1, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 0001ffff GP_BUILTIN_RESULT_EN */
+	xf_emit(ctx, 1, 0);		/* 000000ff CLEAR_STENCIL */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_FRONT_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_FUNC_REF */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_MASK */
+	xf_emit(ctx, 3, 0);		/* 00000007 STENCIL_FRONT_OP_FAIL, ZFAIL, ZPASS */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_BACK_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffffffff CLEAR_DEPTH */
+	xf_emit(ctx, 1, 0);		/* 00000007 */
+	if (device->chipset != 0x50)
+		xf_emit(ctx, 1, 0);	/* 00000003 tesla UNK1108 */
+	xf_emit(ctx, 1, 0);		/* 00000001 SAMPLECNT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	xf_emit(ctx, 1, 0x1001);	/* 00001fff ZETA_ARRAY_MODE */
+	/* SEEK */
+	xf_emit(ctx, 4, 0xffff);	/* 0000ffff MSAA_MASK */
+	xf_emit(ctx, 0x10, 0);		/* 00000001 SCISSOR_ENABLE */
+	xf_emit(ctx, 0x10, 0);		/* ffffffff DEPTH_RANGE_NEAR */
+	xf_emit(ctx, 0x10, 0x3f800000);	/* ffffffff DEPTH_RANGE_FAR */
+	xf_emit(ctx, 1, 0x10);		/* 7f/ff/3ff VIEW_VOLUME_CLIP_CTRL */
+	xf_emit(ctx, 1, 0);		/* 00000001 VIEWPORT_CLIP_RECTS_EN */
+	xf_emit(ctx, 1, 3);		/* 00000003 FP_CTRL_UNK196C */
+	xf_emit(ctx, 1, 0);		/* 00000003 tesla UNK1968 */
+	if (device->chipset != 0x50)
+		xf_emit(ctx, 1, 0);	/* 0fffffff tesla UNK1104 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK151C */
+}
+
+static void
+nv50_graph_construct_gene_clipid(struct nouveau_grctx *ctx)
+{
+	/* middle of strand 0 on pre-NVA0 [after 24xx], middle of area 6 on NVAx */
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 00000007 UNK0FB4 */
+	/* SEEK */
+	xf_emit(ctx, 4, 0);		/* 07ffffff CLIPID_REGION_HORIZ */
+	xf_emit(ctx, 4, 0);		/* 07ffffff CLIPID_REGION_VERT */
+	xf_emit(ctx, 2, 0);		/* 07ffffff SCREEN_SCISSOR */
+	xf_emit(ctx, 2, 0x04000000);	/* 07ffffff UNK1508 */
+	xf_emit(ctx, 1, 0);		/* 00000001 CLIPID_ENABLE */
+	xf_emit(ctx, 1, 0x80);		/* 00003fff CLIPID_WIDTH */
+	xf_emit(ctx, 1, 0);		/* 000000ff CLIPID_ID */
+	xf_emit(ctx, 1, 0);		/* 000000ff CLIPID_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff CLIPID_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0x80);		/* 00003fff CLIPID_HEIGHT */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_CLIPID */
+}
+
+static void
+nv50_graph_construct_gene_unk24xx(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int i;
+	/* middle of strand 0 on pre-NVA0 [after m2mf], end of strand 2 on NVAx */
+	/* SEEK */
+	xf_emit(ctx, 0x33, 0);
+	/* SEEK */
+	xf_emit(ctx, 2, 0);
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 0000007f VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	/* SEEK */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 4, 0);	/* RO */
+		xf_emit(ctx, 0xe10, 0); /* 190 * 9: 8*ffffffff, 7ff */
+		xf_emit(ctx, 1, 0);	/* 1ff */
+		xf_emit(ctx, 8, 0);	/* 0? */
+		xf_emit(ctx, 9, 0);	/* ffffffff, 7ff */
+
+		xf_emit(ctx, 4, 0);	/* RO */
+		xf_emit(ctx, 0xe10, 0); /* 190 * 9: 8*ffffffff, 7ff */
+		xf_emit(ctx, 1, 0);	/* 1ff */
+		xf_emit(ctx, 8, 0);	/* 0? */
+		xf_emit(ctx, 9, 0);	/* ffffffff, 7ff */
+	} else {
+		xf_emit(ctx, 0xc, 0);	/* RO */
+		/* SEEK */
+		xf_emit(ctx, 0xe10, 0); /* 190 * 9: 8*ffffffff, 7ff */
+		xf_emit(ctx, 1, 0);	/* 1ff */
+		xf_emit(ctx, 8, 0);	/* 0? */
+
+		/* SEEK */
+		xf_emit(ctx, 0xc, 0);	/* RO */
+		/* SEEK */
+		xf_emit(ctx, 0xe10, 0); /* 190 * 9: 8*ffffffff, 7ff */
+		xf_emit(ctx, 1, 0);	/* 1ff */
+		xf_emit(ctx, 8, 0);	/* 0? */
+	}
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 4);		/* 0000007f VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0x8100c12);	/* 1fffffff FP_INTERPOLANT_CTRL */
+	if (device->chipset != 0x50)
+		xf_emit(ctx, 1, 3);	/* 00000003 tesla UNK1100 */
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0x8100c12);	/* 1fffffff FP_INTERPOLANT_CTRL */
+	xf_emit(ctx, 1, 0);		/* 0000000f VP_GP_BUILTIN_ATTR_EN */
+	xf_emit(ctx, 1, 0x80c14);	/* 01ffffff SEMANTIC_COLOR */
+	xf_emit(ctx, 1, 1);		/* 00000001 */
+	/* SEEK */
+	if (device->chipset >= 0xa0)
+		xf_emit(ctx, 2, 4);	/* 000000ff */
+	xf_emit(ctx, 1, 0x80c14);	/* 01ffffff SEMANTIC_COLOR */
+	xf_emit(ctx, 1, 0);		/* 00000001 VERTEX_TWO_SIDE_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 POINT_SPRITE_ENABLE */
+	xf_emit(ctx, 1, 0x8100c12);	/* 1fffffff FP_INTERPOLANT_CTRL */
+	xf_emit(ctx, 1, 0x27);		/* 000000ff SEMANTIC_PRIM_ID */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 0000000f */
+	xf_emit(ctx, 1, 1);		/* 00000001 */
+	for (i = 0; i < 10; i++) {
+		/* SEEK */
+		xf_emit(ctx, 0x40, 0);		/* ffffffff */
+		xf_emit(ctx, 0x10, 0);		/* 3, 0, 0.... */
+		xf_emit(ctx, 0x10, 0);		/* ffffffff */
+	}
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 00000001 POINT_SPRITE_CTRL */
+	xf_emit(ctx, 1, 1);		/* 00000001 */
+	xf_emit(ctx, 1, 0);		/* ffffffff */
+	xf_emit(ctx, 4, 0);		/* ffffffff NOPERSPECTIVE_BITMAP */
+	xf_emit(ctx, 0x10, 0);		/* 00ffffff POINT_COORD_REPLACE_MAP */
+	xf_emit(ctx, 1, 0);		/* 00000003 WINDOW_ORIGIN */
+	xf_emit(ctx, 1, 0x8100c12);	/* 1fffffff FP_INTERPOLANT_CTRL */
+	if (device->chipset != 0x50)
+		xf_emit(ctx, 1, 0);	/* 000003ff */
+}
+
+static void
+nv50_graph_construct_gene_vfetch(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int acnt = 0x10, rep, i;
+	/* beginning of strand 1 on pre-NVA0, strand 3 on NVAx */
+	if (IS_NVA3F(device->chipset))
+		acnt = 0x20;
+	/* SEEK */
+	if (device->chipset >= 0xa0) {
+		xf_emit(ctx, 1, 0);	/* ffffffff tesla UNK13A4 */
+		xf_emit(ctx, 1, 1);	/* 00000fff tesla UNK1318 */
+	}
+	xf_emit(ctx, 1, 0);		/* ffffffff VERTEX_BUFFER_FIRST */
+	xf_emit(ctx, 1, 0);		/* 00000001 PRIMITIVE_RESTART_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK0DE8 */
+	xf_emit(ctx, 1, 0);		/* ffffffff PRIMITIVE_RESTART_INDEX */
+	xf_emit(ctx, 1, 0xf);		/* ffffffff VP_ATTR_EN */
+	xf_emit(ctx, (acnt/8)-1, 0);	/* ffffffff VP_ATTR_EN */
+	xf_emit(ctx, acnt/8, 0);	/* ffffffff VTX_ATR_MASK_UNK0DD0 */
+	xf_emit(ctx, 1, 0);		/* 0000000f VP_GP_BUILTIN_ATTR_EN */
+	xf_emit(ctx, 1, 0x20);		/* 0000ffff tesla UNK129C */
+	xf_emit(ctx, 1, 0);		/* 000000ff turing UNK370??? */
+	xf_emit(ctx, 1, 0);		/* 0000ffff turing USER_PARAM_COUNT */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	/* SEEK */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 0xb, 0);	/* RO */
+	else if (device->chipset >= 0xa0)
+		xf_emit(ctx, 0x9, 0);	/* RO */
+	else
+		xf_emit(ctx, 0x8, 0);	/* RO */
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 00000001 EDGE_FLAG */
+	xf_emit(ctx, 1, 0);		/* 00000001 PROVOKING_VERTEX_LAST */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0x1a);		/* 0000001f POLYGON_MODE */
+	/* SEEK */
+	xf_emit(ctx, 0xc, 0);		/* RO */
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 7f/ff */
+	xf_emit(ctx, 1, 4);		/* 7f/ff VP_REG_ALLOC_RESULT */
+	xf_emit(ctx, 1, 4);		/* 7f/ff VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);		/* 0000000f VP_GP_BUILTIN_ATTR_EN */
+	xf_emit(ctx, 1, 4);		/* 000001ff UNK1A28 */
+	xf_emit(ctx, 1, 8);		/* 000001ff UNK0DF0 */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 1, 0x3ff);	/* 3ff tesla UNK0D68 */
+	else
+		xf_emit(ctx, 1, 0x7ff);	/* 7ff tesla UNK0D68 */
+	if (device->chipset == 0xa8)
+		xf_emit(ctx, 1, 0x1e00);	/* 7fff */
+	/* SEEK */
+	xf_emit(ctx, 0xc, 0);		/* RO or close */
+	/* SEEK */
+	xf_emit(ctx, 1, 0xf);		/* ffffffff VP_ATTR_EN */
+	xf_emit(ctx, (acnt/8)-1, 0);	/* ffffffff VP_ATTR_EN */
+	xf_emit(ctx, 1, 0);		/* 0000000f VP_GP_BUILTIN_ATTR_EN */
+	if (device->chipset > 0x50 && device->chipset < 0xa0)
+		xf_emit(ctx, 2, 0);	/* ffffffff */
+	else
+		xf_emit(ctx, 1, 0);	/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 00000003 tesla UNK0FD8 */
+	/* SEEK */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 0x10, 0);	/* 0? */
+		xf_emit(ctx, 2, 0);	/* weird... */
+		xf_emit(ctx, 2, 0);	/* RO */
+	} else {
+		xf_emit(ctx, 8, 0);	/* 0? */
+		xf_emit(ctx, 1, 0);	/* weird... */
+		xf_emit(ctx, 2, 0);	/* RO */
+	}
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* ffffffff VB_ELEMENT_BASE */
+	xf_emit(ctx, 1, 0);		/* ffffffff UNK1438 */
+	xf_emit(ctx, acnt, 0);		/* 1 tesla UNK1000 */
+	if (device->chipset >= 0xa0)
+		xf_emit(ctx, 1, 0);	/* ffffffff tesla UNK1118? */
+	/* SEEK */
+	xf_emit(ctx, acnt, 0);		/* ffffffff VERTEX_ARRAY_UNK90C */
+	xf_emit(ctx, 1, 0);		/* f/1f */
+	/* SEEK */
+	xf_emit(ctx, acnt, 0);		/* ffffffff VERTEX_ARRAY_UNK90C */
+	xf_emit(ctx, 1, 0);		/* f/1f */
+	/* SEEK */
+	xf_emit(ctx, acnt, 0);		/* RO */
+	xf_emit(ctx, 2, 0);		/* RO */
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK111C? */
+	xf_emit(ctx, 1, 0);		/* RO */
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 000000ff UNK15F4_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff UNK15F4_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0);		/* 000000ff UNK0F84_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff UNK0F84_ADDRESS_LOW */
+	/* SEEK */
+	xf_emit(ctx, acnt, 0);		/* 00003fff VERTEX_ARRAY_ATTRIB_OFFSET */
+	xf_emit(ctx, 3, 0);		/* f/1f */
+	/* SEEK */
+	xf_emit(ctx, acnt, 0);		/* 00000fff VERTEX_ARRAY_STRIDE */
+	xf_emit(ctx, 3, 0);		/* f/1f */
+	/* SEEK */
+	xf_emit(ctx, acnt, 0);		/* ffffffff VERTEX_ARRAY_LOW */
+	xf_emit(ctx, 3, 0);		/* f/1f */
+	/* SEEK */
+	xf_emit(ctx, acnt, 0);		/* 000000ff VERTEX_ARRAY_HIGH */
+	xf_emit(ctx, 3, 0);		/* f/1f */
+	/* SEEK */
+	xf_emit(ctx, acnt, 0);		/* ffffffff VERTEX_LIMIT_LOW */
+	xf_emit(ctx, 3, 0);		/* f/1f */
+	/* SEEK */
+	xf_emit(ctx, acnt, 0);		/* 000000ff VERTEX_LIMIT_HIGH */
+	xf_emit(ctx, 3, 0);		/* f/1f */
+	/* SEEK */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, acnt, 0);		/* f */
+		xf_emit(ctx, 3, 0);		/* f/1f */
+	}
+	/* SEEK */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 2, 0);	/* RO */
+	else
+		xf_emit(ctx, 5, 0);	/* RO */
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* ffff DMA_VTXBUF */
+	/* SEEK */
+	if (device->chipset < 0xa0) {
+		xf_emit(ctx, 0x41, 0);	/* RO */
+		/* SEEK */
+		xf_emit(ctx, 0x11, 0);	/* RO */
+	} else if (!IS_NVA3F(device->chipset))
+		xf_emit(ctx, 0x50, 0);	/* RO */
+	else
+		xf_emit(ctx, 0x58, 0);	/* RO */
+	/* SEEK */
+	xf_emit(ctx, 1, 0xf);		/* ffffffff VP_ATTR_EN */
+	xf_emit(ctx, (acnt/8)-1, 0);	/* ffffffff VP_ATTR_EN */
+	xf_emit(ctx, 1, 1);		/* 1 UNK0DEC */
+	/* SEEK */
+	xf_emit(ctx, acnt*4, 0);	/* ffffffff VTX_ATTR */
+	xf_emit(ctx, 4, 0);		/* f/1f, 0, 0, 0 */
+	/* SEEK */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 0x1d, 0);	/* RO */
+	else
+		xf_emit(ctx, 0x16, 0);	/* RO */
+	/* SEEK */
+	xf_emit(ctx, 1, 0xf);		/* ffffffff VP_ATTR_EN */
+	xf_emit(ctx, (acnt/8)-1, 0);	/* ffffffff VP_ATTR_EN */
+	/* SEEK */
+	if (device->chipset < 0xa0)
+		xf_emit(ctx, 8, 0);	/* RO */
+	else if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 0xc, 0);	/* RO */
+	else
+		xf_emit(ctx, 7, 0);	/* RO */
+	/* SEEK */
+	xf_emit(ctx, 0xa, 0);		/* RO */
+	if (device->chipset == 0xa0)
+		rep = 0xc;
+	else
+		rep = 4;
+	for (i = 0; i < rep; i++) {
+		/* SEEK */
+		if (IS_NVA3F(device->chipset))
+			xf_emit(ctx, 0x20, 0);	/* ffffffff */
+		xf_emit(ctx, 0x200, 0);	/* ffffffff */
+		xf_emit(ctx, 4, 0);	/* 7f/ff, 0, 0, 0 */
+		xf_emit(ctx, 4, 0);	/* ffffffff */
+	}
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 113/111 */
+	xf_emit(ctx, 1, 0xf);		/* ffffffff VP_ATTR_EN */
+	xf_emit(ctx, (acnt/8)-1, 0);	/* ffffffff VP_ATTR_EN */
+	xf_emit(ctx, acnt/8, 0);	/* ffffffff VTX_ATTR_MASK_UNK0DD0 */
+	xf_emit(ctx, 1, 0);		/* 0000000f VP_GP_BUILTIN_ATTR_EN */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	/* SEEK */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 7, 0);	/* weird... */
+	else
+		xf_emit(ctx, 5, 0);	/* weird... */
+}
+
+static void
+nv50_graph_construct_gene_eng2d(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	/* middle of strand 1 on pre-NVA0 [after vfetch], middle of strand 6 on NVAx */
+	/* SEEK */
+	xf_emit(ctx, 2, 0);		/* 0001ffff CLIP_X, CLIP_Y */
+	xf_emit(ctx, 2, 0);		/* 0000ffff CLIP_W, CLIP_H */
+	xf_emit(ctx, 1, 0);		/* 00000001 CLIP_ENABLE */
+	if (device->chipset < 0xa0) {
+		/* this is useless on everything but the original NV50,
+		 * guess they forgot to nuke it. Or just didn't bother. */
+		xf_emit(ctx, 2, 0);	/* 0000ffff IFC_CLIP_X, Y */
+		xf_emit(ctx, 2, 1);	/* 0000ffff IFC_CLIP_W, H */
+		xf_emit(ctx, 1, 0);	/* 00000001 IFC_CLIP_ENABLE */
+	}
+	xf_emit(ctx, 1, 1);		/* 00000001 DST_LINEAR */
+	xf_emit(ctx, 1, 0x100);		/* 0001ffff DST_WIDTH */
+	xf_emit(ctx, 1, 0x100);		/* 0001ffff DST_HEIGHT */
+	xf_emit(ctx, 1, 0x11);		/* 3f[NV50]/7f[NV84+] DST_FORMAT */
+	xf_emit(ctx, 1, 0);		/* 0001ffff DRAW_POINT_X */
+	xf_emit(ctx, 1, 8);		/* 0000000f DRAW_UNK58C */
+	xf_emit(ctx, 1, 0);		/* 000fffff SIFC_DST_X_FRACT */
+	xf_emit(ctx, 1, 0);		/* 0001ffff SIFC_DST_X_INT */
+	xf_emit(ctx, 1, 0);		/* 000fffff SIFC_DST_Y_FRACT */
+	xf_emit(ctx, 1, 0);		/* 0001ffff SIFC_DST_Y_INT */
+	xf_emit(ctx, 1, 0);		/* 000fffff SIFC_DX_DU_FRACT */
+	xf_emit(ctx, 1, 1);		/* 0001ffff SIFC_DX_DU_INT */
+	xf_emit(ctx, 1, 0);		/* 000fffff SIFC_DY_DV_FRACT */
+	xf_emit(ctx, 1, 1);		/* 0001ffff SIFC_DY_DV_INT */
+	xf_emit(ctx, 1, 1);		/* 0000ffff SIFC_WIDTH */
+	xf_emit(ctx, 1, 1);		/* 0000ffff SIFC_HEIGHT */
+	xf_emit(ctx, 1, 0xcf);		/* 000000ff SIFC_FORMAT */
+	xf_emit(ctx, 1, 2);		/* 00000003 SIFC_BITMAP_UNK808 */
+	xf_emit(ctx, 1, 0);		/* 00000003 SIFC_BITMAP_LINE_PACK_MODE */
+	xf_emit(ctx, 1, 0);		/* 00000001 SIFC_BITMAP_LSB_FIRST */
+	xf_emit(ctx, 1, 0);		/* 00000001 SIFC_BITMAP_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 0000ffff BLIT_DST_X */
+	xf_emit(ctx, 1, 0);		/* 0000ffff BLIT_DST_Y */
+	xf_emit(ctx, 1, 0);		/* 000fffff BLIT_DU_DX_FRACT */
+	xf_emit(ctx, 1, 1);		/* 0001ffff BLIT_DU_DX_INT */
+	xf_emit(ctx, 1, 0);		/* 000fffff BLIT_DV_DY_FRACT */
+	xf_emit(ctx, 1, 1);		/* 0001ffff BLIT_DV_DY_INT */
+	xf_emit(ctx, 1, 1);		/* 0000ffff BLIT_DST_W */
+	xf_emit(ctx, 1, 1);		/* 0000ffff BLIT_DST_H */
+	xf_emit(ctx, 1, 0);		/* 000fffff BLIT_SRC_X_FRACT */
+	xf_emit(ctx, 1, 0);		/* 0001ffff BLIT_SRC_X_INT */
+	xf_emit(ctx, 1, 0);		/* 000fffff BLIT_SRC_Y_FRACT */
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK888 */
+	xf_emit(ctx, 1, 4);		/* 0000003f UNK884 */
+	xf_emit(ctx, 1, 0);		/* 00000007 UNK880 */
+	xf_emit(ctx, 1, 1);		/* 0000001f tesla UNK0FB8 */
+	xf_emit(ctx, 1, 0x15);		/* 000000ff tesla UNK128C */
+	xf_emit(ctx, 2, 0);		/* 00000007, ffff0ff3 */
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK260 */
+	xf_emit(ctx, 1, 0x4444480);	/* 1fffffff UNK870 */
+	/* SEEK */
+	xf_emit(ctx, 0x10, 0);
+	/* SEEK */
+	xf_emit(ctx, 0x27, 0);
+}
+
+static void
+nv50_graph_construct_gene_csched(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	/* middle of strand 1 on pre-NVA0 [after eng2d], middle of strand 0 on NVAx */
+	/* SEEK */
+	xf_emit(ctx, 2, 0);		/* 00007fff WINDOW_OFFSET_XY... what is it doing here??? */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1924 */
+	xf_emit(ctx, 1, 0);		/* 00000003 WINDOW_ORIGIN */
+	xf_emit(ctx, 1, 0x8100c12);	/* 1fffffff FP_INTERPOLANT_CTRL */
+	xf_emit(ctx, 1, 0);		/* 000003ff */
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* ffffffff turing UNK364 */
+	xf_emit(ctx, 1, 0);		/* 0000000f turing UNK36C */
+	xf_emit(ctx, 1, 0);		/* 0000ffff USER_PARAM_COUNT */
+	xf_emit(ctx, 1, 0x100);		/* 00ffffff turing UNK384 */
+	xf_emit(ctx, 1, 0);		/* 0000000f turing UNK2A0 */
+	xf_emit(ctx, 1, 0);		/* 0000ffff GRIDID */
+	xf_emit(ctx, 1, 0x10001);	/* ffffffff GRIDDIM_XY */
+	xf_emit(ctx, 1, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0x10001);	/* ffffffff BLOCKDIM_XY */
+	xf_emit(ctx, 1, 1);		/* 0000ffff BLOCKDIM_Z */
+	xf_emit(ctx, 1, 0x10001);	/* 00ffffff BLOCK_ALLOC */
+	xf_emit(ctx, 1, 1);		/* 00000001 LANES32 */
+	xf_emit(ctx, 1, 4);		/* 000000ff FP_REG_ALLOC_TEMP */
+	xf_emit(ctx, 1, 2);		/* 00000003 REG_MODE */
+	/* SEEK */
+	xf_emit(ctx, 0x40, 0);		/* ffffffff USER_PARAM */
+	switch (device->chipset) {
+	case 0x50:
+	case 0x92:
+		xf_emit(ctx, 8, 0);	/* 7, 0, 0, 0, ... */
+		xf_emit(ctx, 0x80, 0);	/* fff */
+		xf_emit(ctx, 2, 0);	/* ff, fff */
+		xf_emit(ctx, 0x10*2, 0);	/* ffffffff, 1f */
+		break;
+	case 0x84:
+		xf_emit(ctx, 8, 0);	/* 7, 0, 0, 0, ... */
+		xf_emit(ctx, 0x60, 0);	/* fff */
+		xf_emit(ctx, 2, 0);	/* ff, fff */
+		xf_emit(ctx, 0xc*2, 0);	/* ffffffff, 1f */
+		break;
+	case 0x94:
+	case 0x96:
+		xf_emit(ctx, 8, 0);	/* 7, 0, 0, 0, ... */
+		xf_emit(ctx, 0x40, 0);	/* fff */
+		xf_emit(ctx, 2, 0);	/* ff, fff */
+		xf_emit(ctx, 8*2, 0);	/* ffffffff, 1f */
+		break;
+	case 0x86:
+	case 0x98:
+		xf_emit(ctx, 4, 0);	/* f, 0, 0, 0 */
+		xf_emit(ctx, 0x10, 0);	/* fff */
+		xf_emit(ctx, 2, 0);	/* ff, fff */
+		xf_emit(ctx, 2*2, 0);	/* ffffffff, 1f */
+		break;
+	case 0xa0:
+		xf_emit(ctx, 8, 0);	/* 7, 0, 0, 0, ... */
+		xf_emit(ctx, 0xf0, 0);	/* fff */
+		xf_emit(ctx, 2, 0);	/* ff, fff */
+		xf_emit(ctx, 0x1e*2, 0);	/* ffffffff, 1f */
+		break;
+	case 0xa3:
+		xf_emit(ctx, 8, 0);	/* 7, 0, 0, 0, ... */
+		xf_emit(ctx, 0x60, 0);	/* fff */
+		xf_emit(ctx, 2, 0);	/* ff, fff */
+		xf_emit(ctx, 0xc*2, 0);	/* ffffffff, 1f */
+		break;
+	case 0xa5:
+	case 0xaf:
+		xf_emit(ctx, 8, 0);	/* 7, 0, 0, 0, ... */
+		xf_emit(ctx, 0x30, 0);	/* fff */
+		xf_emit(ctx, 2, 0);	/* ff, fff */
+		xf_emit(ctx, 6*2, 0);	/* ffffffff, 1f */
+		break;
+	case 0xaa:
+		xf_emit(ctx, 0x12, 0);
+		break;
+	case 0xa8:
+	case 0xac:
+		xf_emit(ctx, 4, 0);	/* f, 0, 0, 0 */
+		xf_emit(ctx, 0x10, 0);	/* fff */
+		xf_emit(ctx, 2, 0);	/* ff, fff */
+		xf_emit(ctx, 2*2, 0);	/* ffffffff, 1f */
+		break;
+	}
+	xf_emit(ctx, 1, 0);		/* 0000000f */
+	xf_emit(ctx, 1, 0);		/* 00000000 */
+	xf_emit(ctx, 1, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 0000001f */
+	xf_emit(ctx, 4, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 00000003 turing UNK35C */
+	xf_emit(ctx, 1, 0);		/* ffffffff */
+	xf_emit(ctx, 4, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 00000003 turing UNK35C */
+	xf_emit(ctx, 1, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 000000ff */
+}
+
+static void
+nv50_graph_construct_gene_unk1cxx(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	xf_emit(ctx, 2, 0);		/* 00007fff WINDOW_OFFSET_XY */
+	xf_emit(ctx, 1, 0x3f800000);	/* ffffffff LINE_WIDTH */
+	xf_emit(ctx, 1, 0);		/* 00000001 LINE_SMOOTH_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1658 */
+	xf_emit(ctx, 1, 0);		/* 00000001 POLYGON_SMOOTH_ENABLE */
+	xf_emit(ctx, 3, 0);		/* 00000001 POLYGON_OFFSET_*_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 0000000f CULL_MODE */
+	xf_emit(ctx, 1, 0x1a);		/* 0000001f POLYGON_MODE */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 0);		/* 00000001 POINT_SPRITE_ENABLE */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK165C */
+	xf_emit(ctx, 0x10, 0);		/* 00000001 SCISSOR_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 1, 0);		/* 00000001 LINE_STIPPLE_ENABLE */
+	xf_emit(ctx, 1, 0x00ffff00);	/* 00ffffff LINE_STIPPLE_PATTERN */
+	xf_emit(ctx, 1, 0);		/* ffffffff POLYGON_OFFSET_UNITS */
+	xf_emit(ctx, 1, 0);		/* ffffffff POLYGON_OFFSET_FACTOR */
+	xf_emit(ctx, 1, 0);		/* 00000003 tesla UNK1668 */
+	xf_emit(ctx, 2, 0);		/* 07ffffff SCREEN_SCISSOR */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1900 */
+	xf_emit(ctx, 1, 0xf);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 7, 0);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 1, 0x0fac6881);	/* 0fffffff RT_CONTROL */
+	xf_emit(ctx, 1, 0x11);		/* 0000007f RT_FORMAT */
+	xf_emit(ctx, 7, 0);		/* 0000007f RT_FORMAT */
+	xf_emit(ctx, 8, 0);		/* 00000001 RT_HORIZ_LINEAR */
+	xf_emit(ctx, 1, 4);		/* 00000007 FP_CONTROL */
+	xf_emit(ctx, 1, 0);		/* 00000001 ALPHA_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000007 ALPHA_TEST_FUNC */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 3);	/* 00000003 UNK16B4 */
+	else if (device->chipset >= 0xa0)
+		xf_emit(ctx, 1, 1);	/* 00000001 UNK16B4 */
+	xf_emit(ctx, 1, 0);		/* 00000003 MULTISAMPLE_CTRL */
+	xf_emit(ctx, 1, 0);		/* 00000003 tesla UNK0F90 */
+	xf_emit(ctx, 1, 2);		/* 00000003 tesla UNK143C */
+	xf_emit(ctx, 2, 0x04000000);	/* 07ffffff tesla UNK0D6C */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_MASK */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 SAMPLECNT_ENABLE */
+	xf_emit(ctx, 1, 5);		/* 0000000f UNK1408 */
+	xf_emit(ctx, 1, 0x52);		/* 000001ff SEMANTIC_PTSZ */
+	xf_emit(ctx, 1, 0);		/* ffffffff POINT_SIZE */
+	xf_emit(ctx, 1, 0);		/* 00000001 */
+	xf_emit(ctx, 1, 0);		/* 00000007 tesla UNK0FB4 */
+	if (device->chipset != 0x50) {
+		xf_emit(ctx, 1, 0);	/* 3ff */
+		xf_emit(ctx, 1, 1);	/* 00000001 tesla UNK1110 */
+	}
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 0);	/* 00000003 tesla UNK1928 */
+	xf_emit(ctx, 0x10, 0);		/* ffffffff DEPTH_RANGE_NEAR */
+	xf_emit(ctx, 0x10, 0x3f800000);	/* ffffffff DEPTH_RANGE_FAR */
+	xf_emit(ctx, 1, 0x10);		/* 000000ff VIEW_VOLUME_CLIP_CTRL */
+	xf_emit(ctx, 0x20, 0);		/* 07ffffff VIEWPORT_HORIZ, then VIEWPORT_VERT. (W&0x3fff)<<13 | (X&0x1fff). */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK187C */
+	xf_emit(ctx, 1, 0);		/* 00000003 WINDOW_ORIGIN */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_BACK_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_MASK */
+	xf_emit(ctx, 1, 0x8100c12);	/* 1fffffff FP_INTERPOLANT_CTRL */
+	xf_emit(ctx, 1, 5);		/* 0000000f tesla UNK1220 */
+	xf_emit(ctx, 1, 0);		/* 00000007 MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 000000ff tesla UNK1A20 */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 VERTEX_TWO_SIDE_ENABLE */
+	xf_emit(ctx, 4, 0xffff);	/* 0000ffff MSAA_MASK */
+	if (device->chipset != 0x50)
+		xf_emit(ctx, 1, 3);	/* 00000003 tesla UNK1100 */
+	if (device->chipset < 0xa0)
+		xf_emit(ctx, 0x1c, 0);	/* RO */
+	else if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 0x9, 0);
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK1534 */
+	xf_emit(ctx, 1, 0);		/* 00000001 LINE_SMOOTH_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 LINE_STIPPLE_ENABLE */
+	xf_emit(ctx, 1, 0x00ffff00);	/* 00ffffff LINE_STIPPLE_PATTERN */
+	xf_emit(ctx, 1, 0x1a);		/* 0000001f POLYGON_MODE */
+	xf_emit(ctx, 1, 0);		/* 00000003 WINDOW_ORIGIN */
+	if (device->chipset != 0x50) {
+		xf_emit(ctx, 1, 3);	/* 00000003 tesla UNK1100 */
+		xf_emit(ctx, 1, 0);	/* 3ff */
+	}
+	/* XXX: the following block could belong either to unk1cxx, or
+	 * to STRMOUT. Rather hard to tell. */
+	if (device->chipset < 0xa0)
+		xf_emit(ctx, 0x25, 0);
+	else
+		xf_emit(ctx, 0x3b, 0);
+}
+
+static void
+nv50_graph_construct_gene_strmout(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	xf_emit(ctx, 1, 0x102);		/* 0000ffff STRMOUT_BUFFER_CTRL */
+	xf_emit(ctx, 1, 0);		/* ffffffff STRMOUT_PRIMITIVE_COUNT */
+	xf_emit(ctx, 4, 4);		/* 000000ff STRMOUT_NUM_ATTRIBS */
+	if (device->chipset >= 0xa0) {
+		xf_emit(ctx, 4, 0);	/* ffffffff UNK1A8C */
+		xf_emit(ctx, 4, 0);	/* ffffffff UNK1780 */
+	}
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 4);		/* 0000007f VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 1, 0x3ff);	/* 000003ff tesla UNK0D68 */
+	else
+		xf_emit(ctx, 1, 0x7ff);	/* 000007ff tesla UNK0D68 */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	/* SEEK */
+	xf_emit(ctx, 1, 0x102);		/* 0000ffff STRMOUT_BUFFER_CTRL */
+	xf_emit(ctx, 1, 0);		/* ffffffff STRMOUT_PRIMITIVE_COUNT */
+	xf_emit(ctx, 4, 0);		/* 000000ff STRMOUT_ADDRESS_HIGH */
+	xf_emit(ctx, 4, 0);		/* ffffffff STRMOUT_ADDRESS_LOW */
+	xf_emit(ctx, 4, 4);		/* 000000ff STRMOUT_NUM_ATTRIBS */
+	if (device->chipset >= 0xa0) {
+		xf_emit(ctx, 4, 0);	/* ffffffff UNK1A8C */
+		xf_emit(ctx, 4, 0);	/* ffffffff UNK1780 */
+	}
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_STRMOUT */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_QUERY */
+	xf_emit(ctx, 1, 0);		/* 000000ff QUERY_ADDRESS_HIGH */
+	xf_emit(ctx, 2, 0);		/* ffffffff QUERY_ADDRESS_LOW QUERY_COUNTER */
+	xf_emit(ctx, 2, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	/* SEEK */
+	xf_emit(ctx, 0x20, 0);		/* ffffffff STRMOUT_MAP */
+	xf_emit(ctx, 1, 0);		/* 0000000f */
+	xf_emit(ctx, 1, 0);		/* 00000000? */
+	xf_emit(ctx, 2, 0);		/* ffffffff */
+}
+
+static void
+nv50_graph_construct_gene_ropm1(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	xf_emit(ctx, 1, 0x4e3bfdf);	/* ffffffff UNK0D64 */
+	xf_emit(ctx, 1, 0x4e3bfdf);	/* ffffffff UNK0DF4 */
+	xf_emit(ctx, 1, 0);		/* 00000007 */
+	xf_emit(ctx, 1, 0);		/* 000003ff */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 0x11);	/* 000000ff tesla UNK1968 */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+}
+
+static void
+nv50_graph_construct_gene_ropm2(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_QUERY */
+	xf_emit(ctx, 1, 0x0fac6881);	/* 0fffffff RT_CONTROL */
+	xf_emit(ctx, 2, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 000000ff QUERY_ADDRESS_HIGH */
+	xf_emit(ctx, 2, 0);		/* ffffffff QUERY_ADDRESS_LOW, COUNTER */
+	xf_emit(ctx, 1, 0);		/* 00000001 SAMPLECNT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 7 */
+	/* SEEK */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_QUERY */
+	xf_emit(ctx, 1, 0);		/* 000000ff QUERY_ADDRESS_HIGH */
+	xf_emit(ctx, 2, 0);		/* ffffffff QUERY_ADDRESS_LOW, COUNTER */
+	xf_emit(ctx, 1, 0x4e3bfdf);	/* ffffffff UNK0D64 */
+	xf_emit(ctx, 1, 0x4e3bfdf);	/* ffffffff UNK0DF4 */
+	xf_emit(ctx, 1, 0);		/* 00000001 eng2d UNK260 */
+	xf_emit(ctx, 1, 0);		/* ff/3ff */
+	xf_emit(ctx, 1, 0);		/* 00000007 */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 0x11);	/* 000000ff tesla UNK1968 */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+}
+
+static void
+nv50_graph_construct_gene_ropc(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int magic2;
+	if (device->chipset == 0x50) {
+		magic2 = 0x00003e60;
+	} else if (!IS_NVA3F(device->chipset)) {
+		magic2 = 0x001ffe67;
+	} else {
+		magic2 = 0x00087e67;
+	}
+	xf_emit(ctx, 1, 0);		/* f/7 MUTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_BACK_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_MASK */
+	xf_emit(ctx, 3, 0);		/* 00000007 STENCIL_BACK_OP_FAIL, ZFAIL, ZPASS */
+	xf_emit(ctx, 1, 2);		/* 00000003 tesla UNK143C */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 1, magic2);	/* 001fffff tesla UNK0F78 */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_BOUNDS_EN */
+	xf_emit(ctx, 1, 0);		/* 00000007 DEPTH_TEST_FUNC */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_FRONT_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_MASK */
+	xf_emit(ctx, 3, 0);		/* 00000007 STENCIL_FRONT_OP_FAIL, ZFAIL, ZPASS */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	if (device->chipset >= 0xa0 && !IS_NVAAF(device->chipset))
+		xf_emit(ctx, 1, 0x15);	/* 000000ff */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_BACK_ENABLE */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK15B4 */
+	xf_emit(ctx, 1, 0x10);		/* 3ff/ff VIEW_VOLUME_CLIP_CTRL */
+	xf_emit(ctx, 1, 0);		/* ffffffff CLEAR_DEPTH */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+	if (device->chipset == 0x86 || device->chipset == 0x92 || device->chipset == 0x98 || device->chipset >= 0xa0) {
+		xf_emit(ctx, 3, 0);	/* ff, ffffffff, ffffffff */
+		xf_emit(ctx, 1, 4);	/* 7 */
+		xf_emit(ctx, 1, 0x400);	/* fffffff */
+		xf_emit(ctx, 1, 0x300);	/* ffff */
+		xf_emit(ctx, 1, 0x1001);	/* 1fff */
+		if (device->chipset != 0xa0) {
+			if (IS_NVA3F(device->chipset))
+				xf_emit(ctx, 1, 0);	/* 0000000f UNK15C8 */
+			else
+				xf_emit(ctx, 1, 0x15);	/* ff */
+		}
+	}
+	xf_emit(ctx, 1, 0);		/* 00000007 MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_BACK_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 1, 2);		/* 00000003 tesla UNK143C */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_BOUNDS_EN */
+	xf_emit(ctx, 1, 0);		/* 00000007 DEPTH_TEST_FUNC */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_FRONT_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_BACK_ENABLE */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK15B4 */
+	xf_emit(ctx, 1, 0x10);		/* 7f/ff VIEW_VOLUME_CLIP_CTRL */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1900 */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_BACK_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_FUNC_REF */
+	xf_emit(ctx, 2, 0);		/* ffffffff DEPTH_BOUNDS */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_BOUNDS_EN */
+	xf_emit(ctx, 1, 0);		/* 00000007 DEPTH_TEST_FUNC */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 0000000f */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK0FB0 */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_FRONT_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_FUNC_REF */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_BACK_ENABLE */
+	xf_emit(ctx, 1, 0x10);		/* 7f/ff VIEW_VOLUME_CLIP_CTRL */
+	xf_emit(ctx, 0x10, 0);		/* ffffffff DEPTH_RANGE_NEAR */
+	xf_emit(ctx, 0x10, 0x3f800000);	/* ffffffff DEPTH_RANGE_FAR */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 0);		/* 00000007 MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_BACK_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_FUNC_REF */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_MASK */
+	xf_emit(ctx, 3, 0);		/* 00000007 STENCIL_BACK_OP_FAIL, ZFAIL, ZPASS */
+	xf_emit(ctx, 2, 0);		/* ffffffff DEPTH_BOUNDS */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_BOUNDS_EN */
+	xf_emit(ctx, 1, 0);		/* 00000007 DEPTH_TEST_FUNC */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 000000ff CLEAR_STENCIL */
+	xf_emit(ctx, 1, 0);		/* 00000007 STENCIL_FRONT_FUNC_FUNC */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_FUNC_MASK */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_FUNC_REF */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_MASK */
+	xf_emit(ctx, 3, 0);		/* 00000007 STENCIL_FRONT_OP_FAIL, ZFAIL, ZPASS */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_BACK_ENABLE */
+	xf_emit(ctx, 1, 0x10);		/* 7f/ff VIEW_VOLUME_CLIP_CTRL */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 0x3f);		/* 0000003f UNK1590 */
+	xf_emit(ctx, 1, 0);		/* 00000007 MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 2, 0);		/* ffff0ff3, ffff */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK0FB0 */
+	xf_emit(ctx, 1, 0);		/* 0001ffff GP_BUILTIN_RESULT_EN */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK15B4 */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffffffff CLEAR_DEPTH */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK19CC */
+	if (device->chipset >= 0xa0) {
+		xf_emit(ctx, 2, 0);
+		xf_emit(ctx, 1, 0x1001);
+		xf_emit(ctx, 0xb, 0);
+	} else {
+		xf_emit(ctx, 1, 0);	/* 00000007 */
+		xf_emit(ctx, 1, 0);	/* 00000001 tesla UNK1534 */
+		xf_emit(ctx, 1, 0);	/* 00000007 MULTISAMPLE_SAMPLES_LOG2 */
+		xf_emit(ctx, 8, 0);	/* 00000001 BLEND_ENABLE */
+		xf_emit(ctx, 1, 0);	/* ffff0ff3 */
+	}
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 7, 0);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 1, 0xf);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 7, 0);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f */
+	xf_emit(ctx, 1, 0);		/* 00000001 LOGIC_OP_ENABLE */
+	if (device->chipset != 0x50) {
+		xf_emit(ctx, 1, 0);	/* 0000000f LOGIC_OP */
+		xf_emit(ctx, 1, 0);	/* 000000ff */
+	}
+	xf_emit(ctx, 1, 0);		/* 00000007 OPERATION */
+	xf_emit(ctx, 1, 0);		/* ff/3ff */
+	xf_emit(ctx, 1, 0);		/* 00000003 UNK0F90 */
+	xf_emit(ctx, 2, 1);		/* 00000007 BLEND_EQUATION_RGB, ALPHA */
+	xf_emit(ctx, 1, 1);		/* 00000001 UNK133C */
+	xf_emit(ctx, 1, 2);		/* 0000001f BLEND_FUNC_SRC_RGB */
+	xf_emit(ctx, 1, 1);		/* 0000001f BLEND_FUNC_DST_RGB */
+	xf_emit(ctx, 1, 2);		/* 0000001f BLEND_FUNC_SRC_ALPHA */
+	xf_emit(ctx, 1, 1);		/* 0000001f BLEND_FUNC_DST_ALPHA */
+	xf_emit(ctx, 1, 0);		/* 00000001 */
+	xf_emit(ctx, 1, magic2);	/* 001fffff tesla UNK0F78 */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+	xf_emit(ctx, 1, 0x0fac6881);	/* 0fffffff RT_CONTROL */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 1, 0);	/* 00000001 tesla UNK12E4 */
+		xf_emit(ctx, 8, 1);	/* 00000007 IBLEND_EQUATION_RGB */
+		xf_emit(ctx, 8, 1);	/* 00000007 IBLEND_EQUATION_ALPHA */
+		xf_emit(ctx, 8, 1);	/* 00000001 IBLEND_UNK00 */
+		xf_emit(ctx, 8, 2);	/* 0000001f IBLEND_FUNC_SRC_RGB */
+		xf_emit(ctx, 8, 1);	/* 0000001f IBLEND_FUNC_DST_RGB */
+		xf_emit(ctx, 8, 2);	/* 0000001f IBLEND_FUNC_SRC_ALPHA */
+		xf_emit(ctx, 8, 1);	/* 0000001f IBLEND_FUNC_DST_ALPHA */
+		xf_emit(ctx, 1, 0);	/* 00000001 tesla UNK1140 */
+		xf_emit(ctx, 2, 0);	/* 00000001 */
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+		xf_emit(ctx, 1, 0);	/* 0000000f */
+		xf_emit(ctx, 1, 0);	/* 00000003 */
+		xf_emit(ctx, 1, 0);	/* ffffffff */
+		xf_emit(ctx, 2, 0);	/* 00000001 */
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+		xf_emit(ctx, 1, 0);	/* 00000001 */
+		xf_emit(ctx, 1, 0);	/* 000003ff */
+	} else if (device->chipset >= 0xa0) {
+		xf_emit(ctx, 2, 0);	/* 00000001 */
+		xf_emit(ctx, 1, 0);	/* 00000007 */
+		xf_emit(ctx, 1, 0);	/* 00000003 */
+		xf_emit(ctx, 1, 0);	/* ffffffff */
+		xf_emit(ctx, 2, 0);	/* 00000001 */
+	} else {
+		xf_emit(ctx, 1, 0);	/* 00000007 MULTISAMPLE_SAMPLES_LOG2 */
+		xf_emit(ctx, 1, 0);	/* 00000003 tesla UNK1430 */
+		xf_emit(ctx, 1, 0);	/* ffffffff tesla UNK1A3C */
+	}
+	xf_emit(ctx, 4, 0);		/* ffffffff CLEAR_COLOR */
+	xf_emit(ctx, 4, 0);		/* ffffffff BLEND_COLOR A R G B */
+	xf_emit(ctx, 1, 0);		/* 00000fff eng2d UNK2B0 */
+	if (device->chipset >= 0xa0)
+		xf_emit(ctx, 2, 0);	/* 00000001 */
+	xf_emit(ctx, 1, 0);		/* 000003ff */
+	xf_emit(ctx, 8, 0);		/* 00000001 BLEND_ENABLE */
+	xf_emit(ctx, 1, 1);		/* 00000001 UNK133C */
+	xf_emit(ctx, 1, 2);		/* 0000001f BLEND_FUNC_SRC_RGB */
+	xf_emit(ctx, 1, 1);		/* 0000001f BLEND_FUNC_DST_RGB */
+	xf_emit(ctx, 1, 1);		/* 00000007 BLEND_EQUATION_RGB */
+	xf_emit(ctx, 1, 2);		/* 0000001f BLEND_FUNC_SRC_ALPHA */
+	xf_emit(ctx, 1, 1);		/* 0000001f BLEND_FUNC_DST_ALPHA */
+	xf_emit(ctx, 1, 1);		/* 00000007 BLEND_EQUATION_ALPHA */
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK19C0 */
+	xf_emit(ctx, 1, 0);		/* 00000001 LOGIC_OP_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 0000000f LOGIC_OP */
+	if (device->chipset >= 0xa0)
+		xf_emit(ctx, 1, 0);	/* 00000001 UNK12E4? NVA3+ only? */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 8, 1);	/* 00000001 IBLEND_UNK00 */
+		xf_emit(ctx, 8, 1);	/* 00000007 IBLEND_EQUATION_RGB */
+		xf_emit(ctx, 8, 2);	/* 0000001f IBLEND_FUNC_SRC_RGB */
+		xf_emit(ctx, 8, 1);	/* 0000001f IBLEND_FUNC_DST_RGB */
+		xf_emit(ctx, 8, 1);	/* 00000007 IBLEND_EQUATION_ALPHA */
+		xf_emit(ctx, 8, 2);	/* 0000001f IBLEND_FUNC_SRC_ALPHA */
+		xf_emit(ctx, 8, 1);	/* 0000001f IBLEND_FUNC_DST_ALPHA */
+		xf_emit(ctx, 1, 0);	/* 00000001 tesla UNK15C4 */
+		xf_emit(ctx, 1, 0);	/* 00000001 */
+		xf_emit(ctx, 1, 0);	/* 00000001 tesla UNK1140 */
+	}
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f DST_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 DST_LINEAR */
+	xf_emit(ctx, 1, 0);		/* 00000007 PATTERN_COLOR_FORMAT */
+	xf_emit(ctx, 2, 0);		/* ffffffff PATTERN_MONO_COLOR */
+	xf_emit(ctx, 1, 0);		/* 00000001 PATTERN_MONO_FORMAT */
+	xf_emit(ctx, 2, 0);		/* ffffffff PATTERN_MONO_BITMAP */
+	xf_emit(ctx, 1, 0);		/* 00000003 PATTERN_SELECT */
+	xf_emit(ctx, 1, 0);		/* 000000ff ROP */
+	xf_emit(ctx, 1, 0);		/* ffffffff BETA1 */
+	xf_emit(ctx, 1, 0);		/* ffffffff BETA4 */
+	xf_emit(ctx, 1, 0);		/* 00000007 OPERATION */
+	xf_emit(ctx, 0x50, 0);		/* 10x ffffff, ffffff, ffffff, ffffff, 3 PATTERN */
+}
+
+static void
+nv50_graph_construct_xfer_unk84xx(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int magic3;
+	switch (device->chipset) {
+	case 0x50:
+		magic3 = 0x1000;
+		break;
+	case 0x86:
+	case 0x98:
+	case 0xa8:
+	case 0xaa:
+	case 0xac:
+	case 0xaf:
+		magic3 = 0x1e00;
+		break;
+	default:
+		magic3 = 0;
+	}
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 7f/ff[NVA0+] VP_REG_ALLOC_RESULT */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 0);		/* 111/113[NVA0+] */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 0x1f, 0);	/* ffffffff */
+	else if (device->chipset >= 0xa0)
+		xf_emit(ctx, 0x0f, 0);	/* ffffffff */
+	else
+		xf_emit(ctx, 0x10, 0);	/* fffffff VP_RESULT_MAP_1 up */
+	xf_emit(ctx, 2, 0);		/* f/1f[NVA3], fffffff/ffffffff[NVA0+] */
+	xf_emit(ctx, 1, 4);		/* 7f/ff VP_REG_ALLOC_RESULT */
+	xf_emit(ctx, 1, 4);		/* 7f/ff VP_RESULT_MAP_SIZE */
+	if (device->chipset >= 0xa0)
+		xf_emit(ctx, 1, 0x03020100);	/* ffffffff */
+	else
+		xf_emit(ctx, 1, 0x00608080);	/* fffffff VP_RESULT_MAP_0 */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 2, 0);		/* 111/113, 7f/ff */
+	xf_emit(ctx, 1, 4);		/* 7f/ff VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_REG_ALLOC_RESULT */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0x80);		/* 0000ffff GP_VERTEX_OUTPUT_COUNT */
+	if (magic3)
+		xf_emit(ctx, 1, magic3);	/* 00007fff tesla UNK141C */
+	xf_emit(ctx, 1, 4);		/* 7f/ff VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 0);		/* 111/113 */
+	xf_emit(ctx, 0x1f, 0);		/* ffffffff GP_RESULT_MAP_1 up */
+	xf_emit(ctx, 1, 0);		/* 0000001f */
+	xf_emit(ctx, 1, 0);		/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_REG_ALLOC_RESULT */
+	xf_emit(ctx, 1, 0x80);		/* 0000ffff GP_VERTEX_OUTPUT_COUNT */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0x03020100);	/* ffffffff GP_RESULT_MAP_0 */
+	xf_emit(ctx, 1, 3);		/* 00000003 GP_OUTPUT_PRIMITIVE_TYPE */
+	if (magic3)
+		xf_emit(ctx, 1, magic3);	/* 7fff tesla UNK141C */
+	xf_emit(ctx, 1, 4);		/* 7f/ff VP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 0);		/* 00000001 PROVOKING_VERTEX_LAST */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 0);		/* 111/113 */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 3);		/* 00000003 GP_OUTPUT_PRIMITIVE_TYPE */
+	xf_emit(ctx, 1, 0);		/* 00000001 PROVOKING_VERTEX_LAST */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 0);		/* 00000003 tesla UNK13A0 */
+	xf_emit(ctx, 1, 4);		/* 7f/ff VP_REG_ALLOC_RESULT */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+	xf_emit(ctx, 1, 0);		/* 111/113 */
+	if (device->chipset == 0x94 || device->chipset == 0x96)
+		xf_emit(ctx, 0x1020, 0);	/* 4 x (0x400 x 0xffffffff, ff, 0, 0, 0, 4 x ffffffff) */
+	else if (device->chipset < 0xa0)
+		xf_emit(ctx, 0xa20, 0);	/* 4 x (0x280 x 0xffffffff, ff, 0, 0, 0, 4 x ffffffff) */
+	else if (!IS_NVA3F(device->chipset))
+		xf_emit(ctx, 0x210, 0);	/* ffffffff */
+	else
+		xf_emit(ctx, 0x410, 0);	/* ffffffff */
+	xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+	xf_emit(ctx, 1, 4);		/* 000000ff GP_RESULT_MAP_SIZE */
+	xf_emit(ctx, 1, 3);		/* 00000003 GP_OUTPUT_PRIMITIVE_TYPE */
+	xf_emit(ctx, 1, 0);		/* 00000001 PROVOKING_VERTEX_LAST */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+}
+
+static void
+nv50_graph_construct_xfer_tprop(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int magic1, magic2;
+	if (device->chipset == 0x50) {
+		magic1 = 0x3ff;
+		magic2 = 0x00003e60;
+	} else if (!IS_NVA3F(device->chipset)) {
+		magic1 = 0x7ff;
+		magic2 = 0x001ffe67;
+	} else {
+		magic1 = 0x7ff;
+		magic2 = 0x00087e67;
+	}
+	xf_emit(ctx, 1, 0);		/* 00000007 ALPHA_TEST_FUNC */
+	xf_emit(ctx, 1, 0);		/* ffffffff ALPHA_TEST_REF */
+	xf_emit(ctx, 1, 0);		/* 00000001 ALPHA_TEST_ENABLE */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 1);	/* 0000000f UNK16A0 */
+	xf_emit(ctx, 1, 0);		/* 7/f MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_BACK_MASK */
+	xf_emit(ctx, 3, 0);		/* 00000007 STENCIL_BACK_OP_FAIL, ZFAIL, ZPASS */
+	xf_emit(ctx, 4, 0);		/* ffffffff BLEND_COLOR */
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK19C0 */
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK0FDC */
+	xf_emit(ctx, 1, 0xf);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 7, 0);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 LOGIC_OP_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ff[NV50]/3ff[NV84+] */
+	xf_emit(ctx, 1, 4);		/* 00000007 FP_CONTROL */
+	xf_emit(ctx, 4, 0xffff);	/* 0000ffff MSAA_MASK */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_MASK */
+	xf_emit(ctx, 3, 0);		/* 00000007 STENCIL_FRONT_OP_FAIL, ZFAIL, ZPASS */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_BACK_ENABLE */
+	xf_emit(ctx, 2, 0);		/* 00007fff WINDOW_OFFSET_XY */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK19CC */
+	xf_emit(ctx, 1, 0);		/* 7 */
+	xf_emit(ctx, 1, 0);		/* 00000001 SAMPLECNT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffffffff COLOR_KEY */
+	xf_emit(ctx, 1, 0);		/* 00000001 COLOR_KEY_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000007 COLOR_KEY_FORMAT */
+	xf_emit(ctx, 2, 0);		/* ffffffff SIFC_BITMAP_COLOR */
+	xf_emit(ctx, 1, 1);		/* 00000001 SIFC_BITMAP_WRITE_BIT0_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000007 ALPHA_TEST_FUNC */
+	xf_emit(ctx, 1, 0);		/* 00000001 ALPHA_TEST_ENABLE */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 1, 3);	/* 00000003 tesla UNK16B4 */
+		xf_emit(ctx, 1, 0);	/* 00000003 */
+		xf_emit(ctx, 1, 0);	/* 00000003 tesla UNK1298 */
+	} else if (device->chipset >= 0xa0) {
+		xf_emit(ctx, 1, 1);	/* 00000001 tesla UNK16B4 */
+		xf_emit(ctx, 1, 0);	/* 00000003 */
+	} else {
+		xf_emit(ctx, 1, 0);	/* 00000003 MULTISAMPLE_CTRL */
+	}
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 8, 0);		/* 00000001 BLEND_ENABLE */
+	xf_emit(ctx, 1, 1);		/* 0000001f BLEND_FUNC_DST_ALPHA */
+	xf_emit(ctx, 1, 1);		/* 00000007 BLEND_EQUATION_ALPHA */
+	xf_emit(ctx, 1, 2);		/* 0000001f BLEND_FUNC_SRC_ALPHA */
+	xf_emit(ctx, 1, 1);		/* 0000001f BLEND_FUNC_DST_RGB */
+	xf_emit(ctx, 1, 1);		/* 00000007 BLEND_EQUATION_RGB */
+	xf_emit(ctx, 1, 2);		/* 0000001f BLEND_FUNC_SRC_RGB */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 1, 0);	/* 00000001 UNK12E4 */
+		xf_emit(ctx, 8, 1);	/* 00000007 IBLEND_EQUATION_RGB */
+		xf_emit(ctx, 8, 1);	/* 00000007 IBLEND_EQUATION_ALPHA */
+		xf_emit(ctx, 8, 1);	/* 00000001 IBLEND_UNK00 */
+		xf_emit(ctx, 8, 2);	/* 0000001f IBLEND_SRC_RGB */
+		xf_emit(ctx, 8, 1);	/* 0000001f IBLEND_DST_RGB */
+		xf_emit(ctx, 8, 2);	/* 0000001f IBLEND_SRC_ALPHA */
+		xf_emit(ctx, 8, 1);	/* 0000001f IBLEND_DST_ALPHA */
+		xf_emit(ctx, 1, 0);	/* 00000001 UNK1140 */
+	}
+	xf_emit(ctx, 1, 1);		/* 00000001 UNK133C */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 7, 0);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 1, 0x0fac6881);	/* 0fffffff RT_CONTROL */
+	xf_emit(ctx, 1, 0);		/* 00000001 LOGIC_OP_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ff/3ff */
+	xf_emit(ctx, 1, 4);		/* 00000007 FP_CONTROL */
+	xf_emit(ctx, 1, 0);		/* 00000003 UNK0F90 */
+	xf_emit(ctx, 1, 0);		/* 00000001 FRAMEBUFFER_SRGB */
+	xf_emit(ctx, 1, 0);		/* 7 */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f DST_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 DST_LINEAR */
+	xf_emit(ctx, 1, 0);		/* 00000007 OPERATION */
+	xf_emit(ctx, 1, 0xcf);		/* 000000ff SIFC_FORMAT */
+	xf_emit(ctx, 1, 0xcf);		/* 000000ff DRAW_COLOR_FORMAT */
+	xf_emit(ctx, 1, 0xcf);		/* 000000ff SRC_FORMAT */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+	xf_emit(ctx, 1, 0);		/* 7/f[NVA3] MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 8, 0);		/* 00000001 BLEND_ENABLE */
+	xf_emit(ctx, 1, 1);		/* 0000001f BLEND_FUNC_DST_ALPHA */
+	xf_emit(ctx, 1, 1);		/* 00000007 BLEND_EQUATION_ALPHA */
+	xf_emit(ctx, 1, 2);		/* 0000001f BLEND_FUNC_SRC_ALPHA */
+	xf_emit(ctx, 1, 1);		/* 0000001f BLEND_FUNC_DST_RGB */
+	xf_emit(ctx, 1, 1);		/* 00000007 BLEND_EQUATION_RGB */
+	xf_emit(ctx, 1, 2);		/* 0000001f BLEND_FUNC_SRC_RGB */
+	xf_emit(ctx, 1, 1);		/* 00000001 UNK133C */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 8, 1);		/* 00000001 UNK19E0 */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 7, 0);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 1, 0x0fac6881);	/* 0fffffff RT_CONTROL */
+	xf_emit(ctx, 1, 0xf);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 7, 0);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 1, magic2);	/* 001fffff tesla UNK0F78 */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_BOUNDS_EN */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f DST_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 DST_LINEAR */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 1, 0);	/* ff */
+	else
+		xf_emit(ctx, 3, 0);	/* 1, 7, 3ff */
+	xf_emit(ctx, 1, 4);		/* 00000007 FP_CONTROL */
+	xf_emit(ctx, 1, 0);		/* 00000003 UNK0F90 */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000007 */
+	xf_emit(ctx, 1, 0);		/* 00000001 SAMPLECNT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+	xf_emit(ctx, 1, 0);		/* 7/f MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 7, 0);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 1, 0x0fac6881);	/* 0fffffff RT_CONTROL */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_BOUNDS_EN */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f DST_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 DST_LINEAR */
+	xf_emit(ctx, 1, 0);		/* 000fffff BLIT_DU_DX_FRACT */
+	xf_emit(ctx, 1, 1);		/* 0001ffff BLIT_DU_DX_INT */
+	xf_emit(ctx, 1, 0);		/* 000fffff BLIT_DV_DY_FRACT */
+	xf_emit(ctx, 1, 1);		/* 0001ffff BLIT_DV_DY_INT */
+	xf_emit(ctx, 1, 0);		/* ff/3ff */
+	xf_emit(ctx, 1, magic1);	/* 3ff/7ff tesla UNK0D68 */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK15B4 */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000007 */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+	xf_emit(ctx, 8, 0);		/* 0000ffff DMA_COLOR */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_GLOBAL */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_LOCAL */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_STACK */
+	xf_emit(ctx, 1, 0);		/* ff/3ff */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_DST */
+	xf_emit(ctx, 1, 0);		/* 7 */
+	xf_emit(ctx, 1, 0);		/* 7/f MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 8, 0);		/* 000000ff RT_ADDRESS_HIGH */
+	xf_emit(ctx, 8, 0);		/* ffffffff RT_LAYER_STRIDE */
+	xf_emit(ctx, 8, 0);		/* ffffffff RT_ADDRESS_LOW */
+	xf_emit(ctx, 8, 8);		/* 0000007f RT_TILE_MODE */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 7, 0);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 1, 0x0fac6881);	/* 0fffffff RT_CONTROL */
+	xf_emit(ctx, 8, 0x400);		/* 0fffffff RT_HORIZ */
+	xf_emit(ctx, 8, 0x300);		/* 0000ffff RT_VERT */
+	xf_emit(ctx, 1, 1);		/* 00001fff RT_ARRAY_MODE */
+	xf_emit(ctx, 1, 0xf);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 7, 0);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 1, 0x20);		/* 00000fff DST_TILE_MODE */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f DST_FORMAT */
+	xf_emit(ctx, 1, 0x100);		/* 0001ffff DST_HEIGHT */
+	xf_emit(ctx, 1, 0);		/* 000007ff DST_LAYER */
+	xf_emit(ctx, 1, 1);		/* 00000001 DST_LINEAR */
+	xf_emit(ctx, 1, 0);		/* ffffffff DST_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0);		/* 000000ff DST_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0x40);		/* 0007ffff DST_PITCH */
+	xf_emit(ctx, 1, 0x100);		/* 0001ffff DST_WIDTH */
+	xf_emit(ctx, 1, 0);		/* 0000ffff */
+	xf_emit(ctx, 1, 3);		/* 00000003 tesla UNK15AC */
+	xf_emit(ctx, 1, 0);		/* ff/3ff */
+	xf_emit(ctx, 1, 0);		/* 0001ffff GP_BUILTIN_RESULT_EN */
+	xf_emit(ctx, 1, 0);		/* 00000003 UNK0F90 */
+	xf_emit(ctx, 1, 0);		/* 00000007 */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+	xf_emit(ctx, 1, magic2);	/* 001fffff tesla UNK0F78 */
+	xf_emit(ctx, 1, 0);		/* 7/f MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 1, 2);		/* 00000003 tesla UNK143C */
+	xf_emit(ctx, 1, 0x0fac6881);	/* 0fffffff RT_CONTROL */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_ZETA */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_BOUNDS_EN */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	xf_emit(ctx, 2, 0);		/* ffff, ff/3ff */
+	xf_emit(ctx, 1, 0);		/* 0001ffff GP_BUILTIN_RESULT_EN */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 000000ff STENCIL_FRONT_MASK */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK15B4 */
+	xf_emit(ctx, 1, 0);		/* 00000007 */
+	xf_emit(ctx, 1, 0);		/* ffffffff ZETA_LAYER_STRIDE */
+	xf_emit(ctx, 1, 0);		/* 000000ff ZETA_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);		/* ffffffff ZETA_ADDRESS_LOW */
+	xf_emit(ctx, 1, 4);		/* 00000007 ZETA_TILE_MODE */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	xf_emit(ctx, 1, 0x400);		/* 0fffffff ZETA_HORIZ */
+	xf_emit(ctx, 1, 0x300);		/* 0000ffff ZETA_VERT */
+	xf_emit(ctx, 1, 0x1001);	/* 00001fff ZETA_ARRAY_MODE */
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+	xf_emit(ctx, 1, 0);		/* 7/f MULTISAMPLE_SAMPLES_LOG2 */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 0);	/* 00000001 */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 7, 0);		/* 3f/7f RT_FORMAT */
+	xf_emit(ctx, 1, 0x0fac6881);	/* 0fffffff RT_CONTROL */
+	xf_emit(ctx, 1, 0xf);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 7, 0);		/* 0000000f COLOR_MASK */
+	xf_emit(ctx, 1, 0);		/* ff/3ff */
+	xf_emit(ctx, 8, 0);		/* 00000001 BLEND_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000003 UNK0F90 */
+	xf_emit(ctx, 1, 0);		/* 00000001 FRAMEBUFFER_SRGB */
+	xf_emit(ctx, 1, 0);		/* 7 */
+	xf_emit(ctx, 1, 0);		/* 00000001 LOGIC_OP_ENABLE */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 1, 0);	/* 00000001 UNK1140 */
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+	}
+	xf_emit(ctx, 1, 0);		/* 7/f MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK1534 */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	if (device->chipset >= 0xa0)
+		xf_emit(ctx, 1, 0x0fac6881);	/* fffffff */
+	xf_emit(ctx, 1, magic2);	/* 001fffff tesla UNK0F78 */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_BOUNDS_EN */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE_ENABLE */
+	xf_emit(ctx, 1, 0x11);		/* 3f/7f DST_FORMAT */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK0FB0 */
+	xf_emit(ctx, 1, 0);		/* ff/3ff */
+	xf_emit(ctx, 1, 4);		/* 00000007 FP_CONTROL */
+	xf_emit(ctx, 1, 0);		/* 00000001 STENCIL_FRONT_ENABLE */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK15B4 */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK19CC */
+	xf_emit(ctx, 1, 0);		/* 00000007 */
+	xf_emit(ctx, 1, 0);		/* 00000001 SAMPLECNT_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 0000000f ZETA_FORMAT */
+	xf_emit(ctx, 1, 1);		/* 00000001 ZETA_ENABLE */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+		xf_emit(ctx, 1, 0);	/* 0000000f tesla UNK15C8 */
+	}
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A3C */
+	if (device->chipset >= 0xa0) {
+		xf_emit(ctx, 3, 0);		/* 7/f, 1, ffff0ff3 */
+		xf_emit(ctx, 1, 0xfac6881);	/* fffffff */
+		xf_emit(ctx, 4, 0);		/* 1, 1, 1, 3ff */
+		xf_emit(ctx, 1, 4);		/* 7 */
+		xf_emit(ctx, 1, 0);		/* 1 */
+		xf_emit(ctx, 2, 1);		/* 1 */
+		xf_emit(ctx, 2, 0);		/* 7, f */
+		xf_emit(ctx, 1, 1);		/* 1 */
+		xf_emit(ctx, 1, 0);		/* 7/f */
+		if (IS_NVA3F(device->chipset))
+			xf_emit(ctx, 0x9, 0);	/* 1 */
+		else
+			xf_emit(ctx, 0x8, 0);	/* 1 */
+		xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+		xf_emit(ctx, 8, 1);		/* 1 */
+		xf_emit(ctx, 1, 0x11);		/* 7f */
+		xf_emit(ctx, 7, 0);		/* 7f */
+		xf_emit(ctx, 1, 0xfac6881);	/* fffffff */
+		xf_emit(ctx, 1, 0xf);		/* f */
+		xf_emit(ctx, 7, 0);		/* f */
+		xf_emit(ctx, 1, 0x11);		/* 7f */
+		xf_emit(ctx, 1, 1);		/* 1 */
+		xf_emit(ctx, 5, 0);		/* 1, 7, 3ff, 3, 7 */
+		if (IS_NVA3F(device->chipset)) {
+			xf_emit(ctx, 1, 0);	/* 00000001 UNK1140 */
+			xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+		}
+	}
+}
+
+static void
+nv50_graph_construct_xfer_tex(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	xf_emit(ctx, 2, 0);		/* 1 LINKED_TSC. yes, 2. */
+	if (device->chipset != 0x50)
+		xf_emit(ctx, 1, 0);	/* 3 */
+	xf_emit(ctx, 1, 1);		/* 1ffff BLIT_DU_DX_INT */
+	xf_emit(ctx, 1, 0);		/* fffff BLIT_DU_DX_FRACT */
+	xf_emit(ctx, 1, 1);		/* 1ffff BLIT_DV_DY_INT */
+	xf_emit(ctx, 1, 0);		/* fffff BLIT_DV_DY_FRACT */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 1, 0);	/* 3 BLIT_CONTROL */
+	else
+		xf_emit(ctx, 2, 0);	/* 3ff, 1 */
+	xf_emit(ctx, 1, 0x2a712488);	/* ffffffff SRC_TIC_0 */
+	xf_emit(ctx, 1, 0);		/* ffffffff SRC_TIC_1 */
+	xf_emit(ctx, 1, 0x4085c000);	/* ffffffff SRC_TIC_2 */
+	xf_emit(ctx, 1, 0x40);		/* ffffffff SRC_TIC_3 */
+	xf_emit(ctx, 1, 0x100);		/* ffffffff SRC_TIC_4 */
+	xf_emit(ctx, 1, 0x10100);	/* ffffffff SRC_TIC_5 */
+	xf_emit(ctx, 1, 0x02800000);	/* ffffffff SRC_TIC_6 */
+	xf_emit(ctx, 1, 0);		/* ffffffff SRC_TIC_7 */
+	if (device->chipset == 0x50) {
+		xf_emit(ctx, 1, 0);	/* 00000001 turing UNK358 */
+		xf_emit(ctx, 1, 0);	/* ffffffff tesla UNK1A34? */
+		xf_emit(ctx, 1, 0);	/* 00000003 turing UNK37C tesla UNK1690 */
+		xf_emit(ctx, 1, 0);	/* 00000003 BLIT_CONTROL */
+		xf_emit(ctx, 1, 0);	/* 00000001 turing UNK32C tesla UNK0F94 */
+	} else if (!IS_NVAAF(device->chipset)) {
+		xf_emit(ctx, 1, 0);	/* ffffffff tesla UNK1A34? */
+		xf_emit(ctx, 1, 0);	/* 00000003 */
+		xf_emit(ctx, 1, 0);	/* 000003ff */
+		xf_emit(ctx, 1, 0);	/* 00000003 */
+		xf_emit(ctx, 1, 0);	/* 000003ff */
+		xf_emit(ctx, 1, 0);	/* 00000003 tesla UNK1664 / turing UNK03E8 */
+		xf_emit(ctx, 1, 0);	/* 00000003 */
+		xf_emit(ctx, 1, 0);	/* 000003ff */
+	} else {
+		xf_emit(ctx, 0x6, 0);
+	}
+	xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A34 */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_TEXTURE */
+	xf_emit(ctx, 1, 0);		/* 0000ffff DMA_SRC */
+}
+
+static void
+nv50_graph_construct_xfer_unk8cxx(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK1534 */
+	xf_emit(ctx, 1, 0);		/* 7/f MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 2, 0);		/* 7, ffff0ff3 */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE */
+	xf_emit(ctx, 1, 0x04e3bfdf);	/* ffffffff UNK0D64 */
+	xf_emit(ctx, 1, 0x04e3bfdf);	/* ffffffff UNK0DF4 */
+	xf_emit(ctx, 1, 1);		/* 00000001 UNK15B4 */
+	xf_emit(ctx, 1, 0);		/* 00000001 LINE_STIPPLE_ENABLE */
+	xf_emit(ctx, 1, 0x00ffff00);	/* 00ffffff LINE_STIPPLE_PATTERN */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK0F98 */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 1);	/* 0000001f tesla UNK169C */
+	xf_emit(ctx, 1, 0);		/* 00000003 tesla UNK1668 */
+	xf_emit(ctx, 1, 0);		/* 00000001 LINE_STIPPLE_ENABLE */
+	xf_emit(ctx, 1, 0x00ffff00);	/* 00ffffff LINE_STIPPLE_PATTERN */
+	xf_emit(ctx, 1, 0);		/* 00000001 POLYGON_SMOOTH_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 UNK1534 */
+	xf_emit(ctx, 1, 0);		/* 7/f MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 1, 0);		/* 00000001 tesla UNK1658 */
+	xf_emit(ctx, 1, 0);		/* 00000001 LINE_SMOOTH_ENABLE */
+	xf_emit(ctx, 1, 0);		/* ffff0ff3 */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);		/* 00000001 DEPTH_WRITE */
+	xf_emit(ctx, 1, 1);		/* 00000001 UNK15B4 */
+	xf_emit(ctx, 1, 0);		/* 00000001 POINT_SPRITE_ENABLE */
+	xf_emit(ctx, 1, 1);		/* 00000001 tesla UNK165C */
+	xf_emit(ctx, 1, 0x30201000);	/* ffffffff tesla UNK1670 */
+	xf_emit(ctx, 1, 0x70605040);	/* ffffffff tesla UNK1670 */
+	xf_emit(ctx, 1, 0xb8a89888);	/* ffffffff tesla UNK1670 */
+	xf_emit(ctx, 1, 0xf8e8d8c8);	/* ffffffff tesla UNK1670 */
+	xf_emit(ctx, 1, 0);		/* 00000001 VERTEX_TWO_SIDE_ENABLE */
+	xf_emit(ctx, 1, 0x1a);		/* 0000001f POLYGON_MODE */
+}
+
+static void
+nv50_graph_construct_xfer_tp(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	if (device->chipset < 0xa0) {
+		nv50_graph_construct_xfer_unk84xx(ctx);
+		nv50_graph_construct_xfer_tprop(ctx);
+		nv50_graph_construct_xfer_tex(ctx);
+		nv50_graph_construct_xfer_unk8cxx(ctx);
+	} else {
+		nv50_graph_construct_xfer_tex(ctx);
+		nv50_graph_construct_xfer_tprop(ctx);
+		nv50_graph_construct_xfer_unk8cxx(ctx);
+		nv50_graph_construct_xfer_unk84xx(ctx);
+	}
+}
+
+static void
+nv50_graph_construct_xfer_mpc(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int i, mpcnt = 2;
+	switch (device->chipset) {
+		case 0x98:
+		case 0xaa:
+			mpcnt = 1;
+			break;
+		case 0x50:
+		case 0x84:
+		case 0x86:
+		case 0x92:
+		case 0x94:
+		case 0x96:
+		case 0xa8:
+		case 0xac:
+			mpcnt = 2;
+			break;
+		case 0xa0:
+		case 0xa3:
+		case 0xa5:
+		case 0xaf:
+			mpcnt = 3;
+			break;
+	}
+	for (i = 0; i < mpcnt; i++) {
+		xf_emit(ctx, 1, 0);		/* ff */
+		xf_emit(ctx, 1, 0x80);		/* ffffffff tesla UNK1404 */
+		xf_emit(ctx, 1, 0x80007004);	/* ffffffff tesla UNK12B0 */
+		xf_emit(ctx, 1, 0x04000400);	/* ffffffff */
+		if (device->chipset >= 0xa0)
+			xf_emit(ctx, 1, 0xc0);	/* 00007fff tesla UNK152C */
+		xf_emit(ctx, 1, 0x1000);	/* 0000ffff tesla UNK0D60 */
+		xf_emit(ctx, 1, 0);		/* ff/3ff */
+		xf_emit(ctx, 1, 0);		/* ffffffff tesla UNK1A30 */
+		if (device->chipset == 0x86 || device->chipset == 0x98 || device->chipset == 0xa8 || IS_NVAAF(device->chipset)) {
+			xf_emit(ctx, 1, 0xe00);		/* 7fff */
+			xf_emit(ctx, 1, 0x1e00);	/* 7fff */
+		}
+		xf_emit(ctx, 1, 1);		/* 000000ff VP_REG_ALLOC_TEMP */
+		xf_emit(ctx, 1, 0);		/* 00000001 LINKED_TSC */
+		xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+		if (device->chipset == 0x50)
+			xf_emit(ctx, 2, 0x1000);	/* 7fff tesla UNK141C */
+		xf_emit(ctx, 1, 1);		/* 000000ff GP_REG_ALLOC_TEMP */
+		xf_emit(ctx, 1, 0);		/* 00000001 GP_ENABLE */
+		xf_emit(ctx, 1, 4);		/* 000000ff FP_REG_ALLOC_TEMP */
+		xf_emit(ctx, 1, 2);		/* 00000003 REG_MODE */
+		if (IS_NVAAF(device->chipset))
+			xf_emit(ctx, 0xb, 0);	/* RO */
+		else if (device->chipset >= 0xa0)
+			xf_emit(ctx, 0xc, 0);	/* RO */
+		else
+			xf_emit(ctx, 0xa, 0);	/* RO */
+	}
+	xf_emit(ctx, 1, 0x08100c12);		/* 1fffffff FP_INTERPOLANT_CTRL */
+	xf_emit(ctx, 1, 0);			/* ff/3ff */
+	if (device->chipset >= 0xa0) {
+		xf_emit(ctx, 1, 0x1fe21);	/* 0003ffff tesla UNK0FAC */
+	}
+	xf_emit(ctx, 3, 0);			/* 7fff, 0, 0 */
+	xf_emit(ctx, 1, 0);			/* 00000001 tesla UNK1534 */
+	xf_emit(ctx, 1, 0);			/* 7/f MULTISAMPLE_SAMPLES_LOG2 */
+	xf_emit(ctx, 4, 0xffff);		/* 0000ffff MSAA_MASK */
+	xf_emit(ctx, 1, 1);			/* 00000001 LANES32 */
+	xf_emit(ctx, 1, 0x10001);		/* 00ffffff BLOCK_ALLOC */
+	xf_emit(ctx, 1, 0x10001);		/* ffffffff BLOCKDIM_XY */
+	xf_emit(ctx, 1, 1);			/* 0000ffff BLOCKDIM_Z */
+	xf_emit(ctx, 1, 0);			/* ffffffff SHARED_SIZE */
+	xf_emit(ctx, 1, 0x1fe21);		/* 1ffff/3ffff[NVA0+] tesla UNk0FAC */
+	xf_emit(ctx, 1, 0);			/* ffffffff tesla UNK1A34 */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 1);		/* 0000001f tesla UNK169C */
+	xf_emit(ctx, 1, 0);			/* ff/3ff */
+	xf_emit(ctx, 1, 0);			/* 1 LINKED_TSC */
+	xf_emit(ctx, 1, 0);			/* ff FP_ADDRESS_HIGH */
+	xf_emit(ctx, 1, 0);			/* ffffffff FP_ADDRESS_LOW */
+	xf_emit(ctx, 1, 0x08100c12);		/* 1fffffff FP_INTERPOLANT_CTRL */
+	xf_emit(ctx, 1, 4);			/* 00000007 FP_CONTROL */
+	xf_emit(ctx, 1, 0);			/* 000000ff FRAG_COLOR_CLAMP_EN */
+	xf_emit(ctx, 1, 2);			/* 00000003 REG_MODE */
+	xf_emit(ctx, 1, 0x11);			/* 0000007f RT_FORMAT */
+	xf_emit(ctx, 7, 0);			/* 0000007f RT_FORMAT */
+	xf_emit(ctx, 1, 0);			/* 00000007 */
+	xf_emit(ctx, 1, 0xfac6881);		/* 0fffffff RT_CONTROL */
+	xf_emit(ctx, 1, 0);			/* 00000003 MULTISAMPLE_CTRL */
+	if (IS_NVA3F(device->chipset))
+		xf_emit(ctx, 1, 3);		/* 00000003 tesla UNK16B4 */
+	xf_emit(ctx, 1, 0);			/* 00000001 ALPHA_TEST_ENABLE */
+	xf_emit(ctx, 1, 0);			/* 00000007 ALPHA_TEST_FUNC */
+	xf_emit(ctx, 1, 0);			/* 00000001 FRAMEBUFFER_SRGB */
+	xf_emit(ctx, 1, 4);			/* ffffffff tesla UNK1400 */
+	xf_emit(ctx, 8, 0);			/* 00000001 BLEND_ENABLE */
+	xf_emit(ctx, 1, 0);			/* 00000001 LOGIC_OP_ENABLE */
+	xf_emit(ctx, 1, 2);			/* 0000001f BLEND_FUNC_SRC_RGB */
+	xf_emit(ctx, 1, 1);			/* 0000001f BLEND_FUNC_DST_RGB */
+	xf_emit(ctx, 1, 1);			/* 00000007 BLEND_EQUATION_RGB */
+	xf_emit(ctx, 1, 2);			/* 0000001f BLEND_FUNC_SRC_ALPHA */
+	xf_emit(ctx, 1, 1);			/* 0000001f BLEND_FUNC_DST_ALPHA */
+	xf_emit(ctx, 1, 1);			/* 00000007 BLEND_EQUATION_ALPHA */
+	xf_emit(ctx, 1, 1);			/* 00000001 UNK133C */
+	if (IS_NVA3F(device->chipset)) {
+		xf_emit(ctx, 1, 0);		/* 00000001 UNK12E4 */
+		xf_emit(ctx, 8, 2);		/* 0000001f IBLEND_FUNC_SRC_RGB */
+		xf_emit(ctx, 8, 1);		/* 0000001f IBLEND_FUNC_DST_RGB */
+		xf_emit(ctx, 8, 1);		/* 00000007 IBLEND_EQUATION_RGB */
+		xf_emit(ctx, 8, 2);		/* 0000001f IBLEND_FUNC_SRC_ALPHA */
+		xf_emit(ctx, 8, 1);		/* 0000001f IBLEND_FUNC_DST_ALPHA */
+		xf_emit(ctx, 8, 1);		/* 00000007 IBLEND_EQUATION_ALPHA */
+		xf_emit(ctx, 8, 1);		/* 00000001 IBLEND_UNK00 */
+		xf_emit(ctx, 1, 0);		/* 00000003 tesla UNK1928 */
+		xf_emit(ctx, 1, 0);		/* 00000001 UNK1140 */
+	}
+	xf_emit(ctx, 1, 0);			/* 00000003 tesla UNK0F90 */
+	xf_emit(ctx, 1, 4);			/* 000000ff FP_RESULT_COUNT */
+	/* XXX: demagic this part some day */
+	if (device->chipset == 0x50)
+		xf_emit(ctx, 0x3a0, 0);
+	else if (device->chipset < 0x94)
+		xf_emit(ctx, 0x3a2, 0);
+	else if (device->chipset == 0x98 || device->chipset == 0xaa)
+		xf_emit(ctx, 0x39f, 0);
+	else
+		xf_emit(ctx, 0x3a3, 0);
+	xf_emit(ctx, 1, 0x11);			/* 3f/7f DST_FORMAT */
+	xf_emit(ctx, 1, 0);			/* 7 OPERATION */
+	xf_emit(ctx, 1, 1);			/* 1 DST_LINEAR */
+	xf_emit(ctx, 0x2d, 0);
+}
+
+static void
+nv50_graph_construct_xfer2(struct nouveau_grctx *ctx)
+{
+	struct nouveau_device *device = ctx->device;
+	int i;
+	u32 offset;
+	u32 units = nv_rd32 (ctx->device, 0x1540);
+	int size = 0;
+
+	offset = (ctx->ctxvals_pos+0x3f)&~0x3f;
+
+	if (device->chipset < 0xa0) {
+		for (i = 0; i < 8; i++) {
+			ctx->ctxvals_pos = offset + i;
+			/* that little bugger belongs to csched. No idea
+			 * what it's doing here. */
+			if (i == 0)
+				xf_emit(ctx, 1, 0x08100c12); /* FP_INTERPOLANT_CTRL */
+			if (units & (1 << i))
+				nv50_graph_construct_xfer_mpc(ctx);
+			if ((ctx->ctxvals_pos-offset)/8 > size)
+				size = (ctx->ctxvals_pos-offset)/8;
+		}
+	} else {
+		/* Strand 0: TPs 0, 1 */
+		ctx->ctxvals_pos = offset;
+		/* that little bugger belongs to csched. No idea
+		 * what it's doing here. */
+		xf_emit(ctx, 1, 0x08100c12); /* FP_INTERPOLANT_CTRL */
+		if (units & (1 << 0))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if (units & (1 << 1))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 1: TPs 2, 3 */
+		ctx->ctxvals_pos = offset + 1;
+		if (units & (1 << 2))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if (units & (1 << 3))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 2: TPs 4, 5, 6 */
+		ctx->ctxvals_pos = offset + 2;
+		if (units & (1 << 4))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if (units & (1 << 5))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if (units & (1 << 6))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+
+		/* Strand 3: TPs 7, 8, 9 */
+		ctx->ctxvals_pos = offset + 3;
+		if (units & (1 << 7))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if (units & (1 << 8))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if (units & (1 << 9))
+			nv50_graph_construct_xfer_mpc(ctx);
+		if ((ctx->ctxvals_pos-offset)/8 > size)
+			size = (ctx->ctxvals_pos-offset)/8;
+	}
+	ctx->ctxvals_pos = offset + size * 8;
+	ctx->ctxvals_pos = (ctx->ctxvals_pos+0x3f)&~0x3f;
+	cp_lsr (ctx, offset);
+	cp_out (ctx, CP_SET_XFER_POINTER);
+	cp_lsr (ctx, size);
+	cp_out (ctx, CP_SEEK_2);
+	cp_out (ctx, CP_XFER_2);
+	cp_wait(ctx, XFER, BUSY);
+}
diff --git a/nvkm/engine/graph/ctxnvc0.c b/nvkm/engine/graph/ctxnvc0.c
new file mode 100644
index 000000000..0b7951a85
--- /dev/null
+++ b/nvkm/engine/graph/ctxnvc0.c
@@ -0,0 +1,3039 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "nvc0.h"
+
+void
+nv_icmd(struct nvc0_graph_priv *priv, u32 icmd, u32 data)
+{
+	nv_wr32(priv, 0x400204, data);
+	nv_wr32(priv, 0x400200, icmd);
+	while (nv_rd32(priv, 0x400700) & 2) {}
+}
+
+int
+nvc0_grctx_init(struct nvc0_graph_priv *priv, struct nvc0_grctx *info)
+{
+	struct nouveau_bar *bar = nouveau_bar(priv);
+	struct nouveau_object *parent = nv_object(priv);
+	struct nouveau_gpuobj *chan;
+	u32 size = (0x80000 + priv->size + 4095) & ~4095;
+	int ret, i;
+
+	/* allocate memory to for a "channel", which we'll use to generate
+	 * the default context values
+	 */
+	ret = nouveau_gpuobj_new(parent, NULL, size, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC, &info->chan);
+	chan = info->chan;
+	if (ret) {
+		nv_error(priv, "failed to allocate channel memory, %d\n", ret);
+		return ret;
+	}
+
+	/* PGD pointer */
+	nv_wo32(chan, 0x0200, lower_32_bits(chan->addr + 0x1000));
+	nv_wo32(chan, 0x0204, upper_32_bits(chan->addr + 0x1000));
+	nv_wo32(chan, 0x0208, 0xffffffff);
+	nv_wo32(chan, 0x020c, 0x000000ff);
+
+	/* PGT[0] pointer */
+	nv_wo32(chan, 0x1000, 0x00000000);
+	nv_wo32(chan, 0x1004, 0x00000001 | (chan->addr + 0x2000) >> 8);
+
+	/* identity-map the whole "channel" into its own vm */
+	for (i = 0; i < size / 4096; i++) {
+		u64 addr = ((chan->addr + (i * 4096)) >> 8) | 1;
+		nv_wo32(chan, 0x2000 + (i * 8), lower_32_bits(addr));
+		nv_wo32(chan, 0x2004 + (i * 8), upper_32_bits(addr));
+	}
+
+	/* context pointer (virt) */
+	nv_wo32(chan, 0x0210, 0x00080004);
+	nv_wo32(chan, 0x0214, 0x00000000);
+
+	bar->flush(bar);
+
+	nv_wr32(priv, 0x100cb8, (chan->addr + 0x1000) >> 8);
+	nv_wr32(priv, 0x100cbc, 0x80000001);
+	nv_wait(priv, 0x100c80, 0x00008000, 0x00008000);
+
+	/* setup default state for mmio list construction */
+	info->data = priv->mmio_data;
+	info->mmio = priv->mmio_list;
+	info->addr = 0x2000 + (i * 8);
+	info->priv = priv;
+	info->buffer_nr = 0;
+
+	if (priv->firmware) {
+		nv_wr32(priv, 0x409840, 0x00000030);
+		nv_wr32(priv, 0x409500, 0x80000000 | chan->addr >> 12);
+		nv_wr32(priv, 0x409504, 0x00000003);
+		if (!nv_wait(priv, 0x409800, 0x00000010, 0x00000010))
+			nv_error(priv, "load_ctx timeout\n");
+
+		nv_wo32(chan, 0x8001c, 1);
+		nv_wo32(chan, 0x80020, 0);
+		nv_wo32(chan, 0x80028, 0);
+		nv_wo32(chan, 0x8002c, 0);
+		bar->flush(bar);
+		return 0;
+	}
+
+	/* HUB_FUC(SET_CHAN) */
+	nv_wr32(priv, 0x409840, 0x80000000);
+	nv_wr32(priv, 0x409500, 0x80000000 | chan->addr >> 12);
+	nv_wr32(priv, 0x409504, 0x00000001);
+	if (!nv_wait(priv, 0x409800, 0x80000000, 0x80000000)) {
+		nv_error(priv, "HUB_SET_CHAN timeout\n");
+		nvc0_graph_ctxctl_debug(priv);
+		nouveau_gpuobj_ref(NULL, &info->chan);
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+void
+nvc0_grctx_data(struct nvc0_grctx *info, u32 size, u32 align, u32 access)
+{
+	info->buffer[info->buffer_nr]  = info->addr;
+	info->buffer[info->buffer_nr] +=  (align - 1);
+	info->buffer[info->buffer_nr] &= ~(align - 1);
+	info->addr = info->buffer[info->buffer_nr++] + size;
+
+	info->data->size = size;
+	info->data->align = align;
+	info->data->access = access;
+	info->data++;
+}
+
+void
+nvc0_grctx_mmio(struct nvc0_grctx *info, u32 addr, u32 data, u32 shift, u32 buf)
+{
+	struct nvc0_graph_priv *priv = info->priv;
+
+	info->mmio->addr = addr;
+	info->mmio->data = data;
+	info->mmio->shift = shift;
+	info->mmio->buffer = buf;
+	info->mmio++;
+
+	if (shift)
+		data |= info->buffer[buf] >> shift;
+	nv_wr32(priv, addr, data);
+}
+
+int
+nvc0_grctx_fini(struct nvc0_grctx *info)
+{
+	struct nvc0_graph_priv *priv = info->priv;
+	int i;
+
+	/* trigger a context unload by unsetting the "next channel valid" bit
+	 * and faking a context switch interrupt
+	 */
+	nv_mask(priv, 0x409b04, 0x80000000, 0x00000000);
+	nv_wr32(priv, 0x409000, 0x00000100);
+	if (!nv_wait(priv, 0x409b00, 0x80000000, 0x00000000)) {
+		nv_error(priv, "grctx template channel unload timeout\n");
+		return -EBUSY;
+	}
+
+	priv->data = kmalloc(priv->size, GFP_KERNEL);
+	if (priv->data) {
+		for (i = 0; i < priv->size; i += 4)
+			priv->data[i / 4] = nv_ro32(info->chan, 0x80000 + i);
+	}
+
+	nouveau_gpuobj_ref(NULL, &info->chan);
+	return priv->data ? 0 : -ENOMEM;
+}
+
+static void
+nvc0_grctx_generate_9097(struct nvc0_graph_priv *priv)
+{
+	u32 fermi = nvc0_graph_class(priv);
+	u32 mthd;
+
+	nv_mthd(priv, 0x9097, 0x0800, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0840, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0880, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x08c0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0900, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0940, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0980, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x09c0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0804, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0844, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0884, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x08c4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0904, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0944, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0984, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x09c4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0808, 0x00000400);
+	nv_mthd(priv, 0x9097, 0x0848, 0x00000400);
+	nv_mthd(priv, 0x9097, 0x0888, 0x00000400);
+	nv_mthd(priv, 0x9097, 0x08c8, 0x00000400);
+	nv_mthd(priv, 0x9097, 0x0908, 0x00000400);
+	nv_mthd(priv, 0x9097, 0x0948, 0x00000400);
+	nv_mthd(priv, 0x9097, 0x0988, 0x00000400);
+	nv_mthd(priv, 0x9097, 0x09c8, 0x00000400);
+	nv_mthd(priv, 0x9097, 0x080c, 0x00000300);
+	nv_mthd(priv, 0x9097, 0x084c, 0x00000300);
+	nv_mthd(priv, 0x9097, 0x088c, 0x00000300);
+	nv_mthd(priv, 0x9097, 0x08cc, 0x00000300);
+	nv_mthd(priv, 0x9097, 0x090c, 0x00000300);
+	nv_mthd(priv, 0x9097, 0x094c, 0x00000300);
+	nv_mthd(priv, 0x9097, 0x098c, 0x00000300);
+	nv_mthd(priv, 0x9097, 0x09cc, 0x00000300);
+	nv_mthd(priv, 0x9097, 0x0810, 0x000000cf);
+	nv_mthd(priv, 0x9097, 0x0850, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0890, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x08d0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0910, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0950, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0990, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x09d0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0814, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x0854, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x0894, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x08d4, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x0914, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x0954, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x0994, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x09d4, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x0818, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x0858, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x0898, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x08d8, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x0918, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x0958, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x0998, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x09d8, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x081c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x085c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x089c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x08dc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x091c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x095c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x099c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x09dc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0820, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0860, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x08a0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x08e0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0920, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0960, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x09a0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x09e0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2700, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2720, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2740, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2760, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2780, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27a0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27c0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27e0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2704, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2724, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2744, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2764, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2784, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27a4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27c4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27e4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2708, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2728, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2748, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2768, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2788, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27a8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27c8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27e8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x270c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x272c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x274c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x276c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x278c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27ac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27cc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x27ec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2710, 0x00014000);
+	nv_mthd(priv, 0x9097, 0x2730, 0x00014000);
+	nv_mthd(priv, 0x9097, 0x2750, 0x00014000);
+	nv_mthd(priv, 0x9097, 0x2770, 0x00014000);
+	nv_mthd(priv, 0x9097, 0x2790, 0x00014000);
+	nv_mthd(priv, 0x9097, 0x27b0, 0x00014000);
+	nv_mthd(priv, 0x9097, 0x27d0, 0x00014000);
+	nv_mthd(priv, 0x9097, 0x27f0, 0x00014000);
+	nv_mthd(priv, 0x9097, 0x2714, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x2734, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x2754, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x2774, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x2794, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x27b4, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x27d4, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x27f4, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x1c00, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c10, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c20, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c30, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c40, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c50, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c60, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c70, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c80, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c90, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ca0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cb0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cc0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cd0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ce0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cf0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c04, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c14, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c24, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c34, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c44, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c54, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c64, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c74, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c84, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c94, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ca4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cb4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cc4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cd4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ce4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cf4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c08, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c18, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c28, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c38, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c48, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c58, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c68, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c78, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c88, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c98, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ca8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cb8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cc8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cd8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ce8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cf8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c0c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c1c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c2c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c3c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c4c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c5c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c6c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c7c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c8c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1c9c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cbc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ccc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cdc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1cfc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d00, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d10, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d20, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d30, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d40, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d50, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d60, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d70, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d80, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d90, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1da0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1db0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dc0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dd0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1de0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1df0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d04, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d14, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d24, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d34, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d44, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d54, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d64, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d74, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d84, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d94, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1da4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1db4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dc4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dd4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1de4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1df4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d08, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d18, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d28, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d38, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d48, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d58, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d68, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d78, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d88, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d98, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1da8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1db8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dc8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dd8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1de8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1df8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d0c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d1c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d2c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d3c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d4c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d5c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d6c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d7c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d8c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1d9c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dbc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dcc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ddc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1dfc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f00, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f08, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f10, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f18, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f20, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f28, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f30, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f38, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f40, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f48, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f50, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f58, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f60, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f68, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f70, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f78, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f04, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f0c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f14, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f1c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f24, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f2c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f34, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f3c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f44, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f4c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f54, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f5c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f64, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f6c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f74, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f7c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f80, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f88, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f90, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f98, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fa0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fa8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fb0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fb8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fc0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fc8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fd0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fd8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fe0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fe8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ff0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ff8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f84, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f8c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f94, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1f9c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fa4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fb4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fbc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fc4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fcc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fd4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fdc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fe4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1fec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ff4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1ffc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2200, 0x00000022);
+	nv_mthd(priv, 0x9097, 0x2210, 0x00000022);
+	nv_mthd(priv, 0x9097, 0x2220, 0x00000022);
+	nv_mthd(priv, 0x9097, 0x2230, 0x00000022);
+	nv_mthd(priv, 0x9097, 0x2240, 0x00000022);
+	nv_mthd(priv, 0x9097, 0x2000, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2040, 0x00000011);
+	nv_mthd(priv, 0x9097, 0x2080, 0x00000020);
+	nv_mthd(priv, 0x9097, 0x20c0, 0x00000030);
+	nv_mthd(priv, 0x9097, 0x2100, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x2140, 0x00000051);
+	nv_mthd(priv, 0x9097, 0x200c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x204c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x208c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x20cc, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x210c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x214c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x2010, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2050, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2090, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x20d0, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x2110, 0x00000003);
+	nv_mthd(priv, 0x9097, 0x2150, 0x00000004);
+	nv_mthd(priv, 0x9097, 0x0380, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03a0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03c0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03e0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0384, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03a4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03c4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03e4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0388, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03a8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03c8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03e8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x038c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03ac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03cc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x03ec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0700, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0710, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0720, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0730, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0704, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0714, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0724, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0734, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0708, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0718, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0728, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0738, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2800, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2804, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2808, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x280c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2810, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2814, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2818, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x281c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2820, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2824, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2828, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x282c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2830, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2834, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2838, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x283c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2840, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2844, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2848, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x284c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2850, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2854, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2858, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x285c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2860, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2864, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2868, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x286c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2870, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2874, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2878, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x287c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2880, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2884, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2888, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x288c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2890, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2894, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2898, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x289c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28a0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28a4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28a8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28ac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28b0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28b4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28b8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28bc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28c0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28c4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28c8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28cc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28d0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28d4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28d8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28dc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28e0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28e4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28e8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28ec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28f0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28f4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28f8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x28fc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2900, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2904, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2908, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x290c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2910, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2914, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2918, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x291c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2920, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2924, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2928, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x292c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2930, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2934, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2938, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x293c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2940, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2944, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2948, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x294c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2950, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2954, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2958, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x295c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2960, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2964, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2968, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x296c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2970, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2974, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2978, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x297c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2980, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2984, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2988, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x298c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2990, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2994, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2998, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x299c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29a0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29a4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29a8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29ac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29b0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29b4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29b8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29bc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29c0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29c4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29c8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29cc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29d0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29d4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29d8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29dc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29e0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29e4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29e8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29ec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29f0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29f4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29f8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x29fc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a00, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a20, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a40, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a60, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a80, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0aa0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ac0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ae0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b00, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b20, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b40, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b60, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b80, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ba0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bc0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0be0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a04, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a24, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a44, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a64, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a84, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0aa4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ac4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ae4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b04, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b24, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b44, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b64, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b84, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ba4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bc4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0be4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a08, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a28, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a48, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a68, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a88, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0aa8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ac8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ae8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b08, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b28, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b48, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b68, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b88, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ba8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bc8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0be8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a0c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a2c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a4c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a6c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a8c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0aac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0acc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0aec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b0c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b2c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b4c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b6c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b8c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bcc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a10, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a30, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a50, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a70, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a90, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ab0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ad0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0af0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b10, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b30, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b50, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b70, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b90, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bb0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bd0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bf0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a14, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a34, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a54, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a74, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0a94, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ab4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ad4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0af4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b14, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b34, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b54, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b74, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0b94, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bb4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bd4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0bf4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c00, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c10, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c20, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c30, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c40, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c50, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c60, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c70, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c80, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c90, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ca0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cb0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cc0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cd0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ce0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cf0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c04, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c14, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c24, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c34, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c44, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c54, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c64, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c74, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c84, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c94, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ca4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cb4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cc4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cd4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ce4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cf4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c08, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c18, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c28, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c38, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c48, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c58, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c68, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c78, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c88, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c98, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ca8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cb8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cc8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cd8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ce8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0cf8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0c0c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0c1c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0c2c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0c3c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0c4c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0c5c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0c6c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0c7c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0c8c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0c9c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0cac, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0cbc, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0ccc, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0cdc, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0cec, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0cfc, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0d00, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d08, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d10, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d18, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d20, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d28, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d30, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d38, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d04, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d0c, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d14, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d1c, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d24, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d2c, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d34, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d3c, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e00, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e10, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e20, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e30, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e40, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e50, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e60, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e70, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e80, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e90, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ea0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0eb0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ec0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ed0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ee0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ef0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0e04, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e14, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e24, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e34, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e44, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e54, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e64, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e74, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e84, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e94, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ea4, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0eb4, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ec4, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ed4, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ee4, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ef4, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e08, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e18, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e28, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e38, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e48, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e58, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e68, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e78, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e88, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0e98, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ea8, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0eb8, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ec8, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ed8, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ee8, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0ef8, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d40, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d48, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d50, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d58, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d44, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d4c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d54, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d5c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1e00, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e20, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e40, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e60, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e80, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ea0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ec0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ee0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e04, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e24, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e44, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e64, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e84, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ea4, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ec4, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ee4, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e08, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e28, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e48, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e68, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e88, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1ea8, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1ec8, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1ee8, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e0c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e2c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e4c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e6c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e8c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1eac, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ecc, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1eec, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e10, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e30, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e50, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e70, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e90, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1eb0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ed0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ef0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e14, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e34, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e54, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e74, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e94, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1eb4, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1ed4, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1ef4, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1e18, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e38, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e58, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e78, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1e98, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1eb8, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ed8, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1ef8, 0x00000001);
+	if (fermi == 0x9097) {
+		for (mthd = 0x3400; mthd <= 0x35fc; mthd += 4)
+			nv_mthd(priv, 0x9097, mthd, 0x00000000);
+	}
+	nv_mthd(priv, 0x9097, 0x030c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1944, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1514, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d68, 0x0000ffff);
+	nv_mthd(priv, 0x9097, 0x121c, 0x0fac6881);
+	nv_mthd(priv, 0x9097, 0x0fac, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1538, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x0fe0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0fe4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0fe8, 0x00000014);
+	nv_mthd(priv, 0x9097, 0x0fec, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x0ff0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x179c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1228, 0x00000400);
+	nv_mthd(priv, 0x9097, 0x122c, 0x00000300);
+	nv_mthd(priv, 0x9097, 0x1230, 0x00010001);
+	nv_mthd(priv, 0x9097, 0x07f8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x15b4, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x15cc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1534, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0fb0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x15d0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x153c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x16b4, 0x00000003);
+	nv_mthd(priv, 0x9097, 0x0fbc, 0x0000ffff);
+	nv_mthd(priv, 0x9097, 0x0fc0, 0x0000ffff);
+	nv_mthd(priv, 0x9097, 0x0fc4, 0x0000ffff);
+	nv_mthd(priv, 0x9097, 0x0fc8, 0x0000ffff);
+	nv_mthd(priv, 0x9097, 0x0df8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0dfc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1948, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1970, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x161c, 0x000009f0);
+	nv_mthd(priv, 0x9097, 0x0dcc, 0x00000010);
+	nv_mthd(priv, 0x9097, 0x163c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x15e4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1160, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1164, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1168, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x116c, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1170, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1174, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1178, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x117c, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1180, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1184, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1188, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x118c, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1190, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1194, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1198, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x119c, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11a0, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11a4, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11a8, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11ac, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11b0, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11b4, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11b8, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11bc, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11c0, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11c4, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11c8, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11cc, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11d0, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11d4, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11d8, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x11dc, 0x25e00040);
+	nv_mthd(priv, 0x9097, 0x1880, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1884, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1888, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x188c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1890, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1894, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1898, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x189c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18a0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18a4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18a8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18ac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18b0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18b4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18b8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18bc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18c0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18c4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18c8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18cc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18d0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18d4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18d8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18dc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18e0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18e4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18e8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18ec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18f0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18f4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18f8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x18fc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0f84, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0f88, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x17c8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x17cc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x17d0, 0x000000ff);
+	nv_mthd(priv, 0x9097, 0x17d4, 0xffffffff);
+	nv_mthd(priv, 0x9097, 0x17d8, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x17dc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x15f4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x15f8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1434, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1438, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d74, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0dec, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x13a4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1318, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1644, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0748, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0de8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1648, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x12a4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1120, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1124, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1128, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x112c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1118, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x164c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1658, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1910, 0x00000290);
+	nv_mthd(priv, 0x9097, 0x1518, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x165c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1520, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1604, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1570, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x13b0, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x13b4, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x020c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1670, 0x30201000);
+	nv_mthd(priv, 0x9097, 0x1674, 0x70605040);
+	nv_mthd(priv, 0x9097, 0x1678, 0xb8a89888);
+	nv_mthd(priv, 0x9097, 0x167c, 0xf8e8d8c8);
+	nv_mthd(priv, 0x9097, 0x166c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1680, 0x00ffff00);
+	nv_mthd(priv, 0x9097, 0x12d0, 0x00000003);
+	nv_mthd(priv, 0x9097, 0x12d4, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1684, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1688, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0dac, 0x00001b02);
+	nv_mthd(priv, 0x9097, 0x0db0, 0x00001b02);
+	nv_mthd(priv, 0x9097, 0x0db4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x168c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x15bc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x156c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x187c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1110, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x0dc0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0dc4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0dc8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1234, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1690, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x12ac, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x02c4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0790, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0794, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0798, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x079c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x07a0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x077c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1000, 0x00000010);
+	nv_mthd(priv, 0x9097, 0x10fc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1290, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0218, 0x00000010);
+	nv_mthd(priv, 0x9097, 0x12d8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x12dc, 0x00000010);
+	nv_mthd(priv, 0x9097, 0x0d94, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x155c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1560, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1564, 0x00001fff);
+	nv_mthd(priv, 0x9097, 0x1574, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1578, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x157c, 0x003fffff);
+	nv_mthd(priv, 0x9097, 0x1354, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1664, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1610, 0x00000012);
+	nv_mthd(priv, 0x9097, 0x1608, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x160c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x162c, 0x00000003);
+	nv_mthd(priv, 0x9097, 0x0210, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0320, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0324, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0328, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x032c, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0330, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0334, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0338, 0x3f800000);
+	nv_mthd(priv, 0x9097, 0x0750, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0760, 0x39291909);
+	nv_mthd(priv, 0x9097, 0x0764, 0x79695949);
+	nv_mthd(priv, 0x9097, 0x0768, 0xb9a99989);
+	nv_mthd(priv, 0x9097, 0x076c, 0xf9e9d9c9);
+	nv_mthd(priv, 0x9097, 0x0770, 0x30201000);
+	nv_mthd(priv, 0x9097, 0x0774, 0x70605040);
+	nv_mthd(priv, 0x9097, 0x0778, 0x00009080);
+	nv_mthd(priv, 0x9097, 0x0780, 0x39291909);
+	nv_mthd(priv, 0x9097, 0x0784, 0x79695949);
+	nv_mthd(priv, 0x9097, 0x0788, 0xb9a99989);
+	nv_mthd(priv, 0x9097, 0x078c, 0xf9e9d9c9);
+	nv_mthd(priv, 0x9097, 0x07d0, 0x30201000);
+	nv_mthd(priv, 0x9097, 0x07d4, 0x70605040);
+	nv_mthd(priv, 0x9097, 0x07d8, 0x00009080);
+	nv_mthd(priv, 0x9097, 0x037c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x0740, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0744, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x2600, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1918, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x191c, 0x00000900);
+	nv_mthd(priv, 0x9097, 0x1920, 0x00000405);
+	nv_mthd(priv, 0x9097, 0x1308, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1924, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x13ac, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x192c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x193c, 0x00002c1c);
+	nv_mthd(priv, 0x9097, 0x0d7c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0f8c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x02c0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1510, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1940, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ff4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0ff8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x194c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1950, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1968, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1590, 0x0000003f);
+	nv_mthd(priv, 0x9097, 0x07e8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x07ec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x07f0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x07f4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x196c, 0x00000011);
+	nv_mthd(priv, 0x9097, 0x197c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0fcc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0fd0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x02d8, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x1980, 0x00000080);
+	nv_mthd(priv, 0x9097, 0x1504, 0x00000080);
+	nv_mthd(priv, 0x9097, 0x1984, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0300, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x13a8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x12ec, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1310, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1314, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1380, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1384, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1388, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x138c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1390, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1394, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x139c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1398, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1594, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1598, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x159c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x15a0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x15a4, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x0f54, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0f58, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0f5c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x19bc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0f9c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0fa0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x12cc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x12e8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x130c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1360, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1364, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1368, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x136c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1370, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1374, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1378, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x137c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x133c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1340, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1344, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1348, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x134c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1350, 0x00000002);
+	nv_mthd(priv, 0x9097, 0x1358, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x12e4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x131c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1320, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1324, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1328, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x19c0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1140, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x19c4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x19c8, 0x00001500);
+	nv_mthd(priv, 0x9097, 0x135c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0f90, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x19e0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x19e4, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x19e8, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x19ec, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x19f0, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x19f4, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x19f8, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x19fc, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x19cc, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x15b8, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1a00, 0x00001111);
+	nv_mthd(priv, 0x9097, 0x1a04, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1a08, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1a0c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1a10, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1a14, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1a18, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1a1c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d6c, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x0d70, 0xffff0000);
+	nv_mthd(priv, 0x9097, 0x10f8, 0x00001010);
+	nv_mthd(priv, 0x9097, 0x0d80, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d84, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d88, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d8c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0d90, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0da0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1508, 0x80000000);
+	nv_mthd(priv, 0x9097, 0x150c, 0x40000000);
+	nv_mthd(priv, 0x9097, 0x1668, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0318, 0x00000008);
+	nv_mthd(priv, 0x9097, 0x031c, 0x00000008);
+	nv_mthd(priv, 0x9097, 0x0d9c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x07dc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x074c, 0x00000055);
+	nv_mthd(priv, 0x9097, 0x1420, 0x00000003);
+	nv_mthd(priv, 0x9097, 0x17bc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x17c0, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x17c4, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1008, 0x00000008);
+	nv_mthd(priv, 0x9097, 0x100c, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x1010, 0x0000012c);
+	nv_mthd(priv, 0x9097, 0x0d60, 0x00000040);
+	nv_mthd(priv, 0x9097, 0x075c, 0x00000003);
+	nv_mthd(priv, 0x9097, 0x1018, 0x00000020);
+	nv_mthd(priv, 0x9097, 0x101c, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1020, 0x00000020);
+	nv_mthd(priv, 0x9097, 0x1024, 0x00000001);
+	nv_mthd(priv, 0x9097, 0x1444, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x1448, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x144c, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0360, 0x20164010);
+	nv_mthd(priv, 0x9097, 0x0364, 0x00000020);
+	nv_mthd(priv, 0x9097, 0x0368, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0de4, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0204, 0x00000006);
+	nv_mthd(priv, 0x9097, 0x0208, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x02cc, 0x003fffff);
+	nv_mthd(priv, 0x9097, 0x02d0, 0x00000c48);
+	nv_mthd(priv, 0x9097, 0x1220, 0x00000005);
+	nv_mthd(priv, 0x9097, 0x0fdc, 0x00000000);
+	nv_mthd(priv, 0x9097, 0x0f98, 0x00300008);
+	nv_mthd(priv, 0x9097, 0x1284, 0x04000080);
+	nv_mthd(priv, 0x9097, 0x1450, 0x00300008);
+	nv_mthd(priv, 0x9097, 0x1454, 0x04000080);
+	nv_mthd(priv, 0x9097, 0x0214, 0x00000000);
+	/* in trace, right after 0x90c0, not here */
+	nv_mthd(priv, 0x9097, 0x3410, 0x80002006);
+}
+
+static void
+nvc0_grctx_generate_9197(struct nvc0_graph_priv *priv)
+{
+	u32 fermi = nvc0_graph_class(priv);
+	u32 mthd;
+
+	if (fermi == 0x9197) {
+		for (mthd = 0x3400; mthd <= 0x35fc; mthd += 4)
+			nv_mthd(priv, 0x9197, mthd, 0x00000000);
+	}
+	nv_mthd(priv, 0x9197, 0x02e4, 0x0000b001);
+}
+
+static void
+nvc0_grctx_generate_9297(struct nvc0_graph_priv *priv)
+{
+	u32 fermi = nvc0_graph_class(priv);
+	u32 mthd;
+
+	if (fermi == 0x9297) {
+		for (mthd = 0x3400; mthd <= 0x35fc; mthd += 4)
+			nv_mthd(priv, 0x9297, mthd, 0x00000000);
+	}
+	nv_mthd(priv, 0x9297, 0x036c, 0x00000000);
+	nv_mthd(priv, 0x9297, 0x0370, 0x00000000);
+	nv_mthd(priv, 0x9297, 0x07a4, 0x00000000);
+	nv_mthd(priv, 0x9297, 0x07a8, 0x00000000);
+	nv_mthd(priv, 0x9297, 0x0374, 0x00000000);
+	nv_mthd(priv, 0x9297, 0x0378, 0x00000020);
+}
+
+static void
+nvc0_grctx_generate_902d(struct nvc0_graph_priv *priv)
+{
+	nv_mthd(priv, 0x902d, 0x0200, 0x000000cf);
+	nv_mthd(priv, 0x902d, 0x0204, 0x00000001);
+	nv_mthd(priv, 0x902d, 0x0208, 0x00000020);
+	nv_mthd(priv, 0x902d, 0x020c, 0x00000001);
+	nv_mthd(priv, 0x902d, 0x0210, 0x00000000);
+	nv_mthd(priv, 0x902d, 0x0214, 0x00000080);
+	nv_mthd(priv, 0x902d, 0x0218, 0x00000100);
+	nv_mthd(priv, 0x902d, 0x021c, 0x00000100);
+	nv_mthd(priv, 0x902d, 0x0220, 0x00000000);
+	nv_mthd(priv, 0x902d, 0x0224, 0x00000000);
+	nv_mthd(priv, 0x902d, 0x0230, 0x000000cf);
+	nv_mthd(priv, 0x902d, 0x0234, 0x00000001);
+	nv_mthd(priv, 0x902d, 0x0238, 0x00000020);
+	nv_mthd(priv, 0x902d, 0x023c, 0x00000001);
+	nv_mthd(priv, 0x902d, 0x0244, 0x00000080);
+	nv_mthd(priv, 0x902d, 0x0248, 0x00000100);
+	nv_mthd(priv, 0x902d, 0x024c, 0x00000100);
+}
+
+static void
+nvc0_grctx_generate_9039(struct nvc0_graph_priv *priv)
+{
+	nv_mthd(priv, 0x9039, 0x030c, 0x00000000);
+	nv_mthd(priv, 0x9039, 0x0310, 0x00000000);
+	nv_mthd(priv, 0x9039, 0x0314, 0x00000000);
+	nv_mthd(priv, 0x9039, 0x0320, 0x00000000);
+	nv_mthd(priv, 0x9039, 0x0238, 0x00000000);
+	nv_mthd(priv, 0x9039, 0x023c, 0x00000000);
+	nv_mthd(priv, 0x9039, 0x0318, 0x00000000);
+	nv_mthd(priv, 0x9039, 0x031c, 0x00000000);
+}
+
+static void
+nvc0_grctx_generate_90c0(struct nvc0_graph_priv *priv)
+{
+	int i;
+
+	for (i = 0; nv_device(priv)->chipset == 0xd9 && i < 4; i++) {
+		nv_mthd(priv, 0x90c0, 0x2700 + (i * 0x40), 0x00000000);
+		nv_mthd(priv, 0x90c0, 0x2720 + (i * 0x40), 0x00000000);
+		nv_mthd(priv, 0x90c0, 0x2704 + (i * 0x40), 0x00000000);
+		nv_mthd(priv, 0x90c0, 0x2724 + (i * 0x40), 0x00000000);
+		nv_mthd(priv, 0x90c0, 0x2708 + (i * 0x40), 0x00000000);
+		nv_mthd(priv, 0x90c0, 0x2728 + (i * 0x40), 0x00000000);
+	}
+	nv_mthd(priv, 0x90c0, 0x270c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x272c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x274c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x276c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x278c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x27ac, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x27cc, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x27ec, 0x00000000);
+	for (i = 0; nv_device(priv)->chipset == 0xd9 && i < 4; i++) {
+		nv_mthd(priv, 0x90c0, 0x2710 + (i * 0x40), 0x00014000);
+		nv_mthd(priv, 0x90c0, 0x2730 + (i * 0x40), 0x00014000);
+		nv_mthd(priv, 0x90c0, 0x2714 + (i * 0x40), 0x00000040);
+		nv_mthd(priv, 0x90c0, 0x2734 + (i * 0x40), 0x00000040);
+	}
+	nv_mthd(priv, 0x90c0, 0x030c, 0x00000001);
+	nv_mthd(priv, 0x90c0, 0x1944, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x0758, 0x00000100);
+	nv_mthd(priv, 0x90c0, 0x02c4, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x0790, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x0794, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x0798, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x079c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x07a0, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x077c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x0204, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x0208, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x020c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x0214, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x024c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x0d94, 0x00000001);
+	nv_mthd(priv, 0x90c0, 0x1608, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x160c, 0x00000000);
+	nv_mthd(priv, 0x90c0, 0x1664, 0x00000000);
+}
+
+static void
+nvc0_grctx_generate_dispatch(struct nvc0_graph_priv *priv)
+{
+	int i;
+
+	nv_wr32(priv, 0x404004, 0x00000000);
+	nv_wr32(priv, 0x404008, 0x00000000);
+	nv_wr32(priv, 0x40400c, 0x00000000);
+	nv_wr32(priv, 0x404010, 0x00000000);
+	nv_wr32(priv, 0x404014, 0x00000000);
+	nv_wr32(priv, 0x404018, 0x00000000);
+	nv_wr32(priv, 0x40401c, 0x00000000);
+	nv_wr32(priv, 0x404020, 0x00000000);
+	nv_wr32(priv, 0x404024, 0x00000000);
+	nv_wr32(priv, 0x404028, 0x00000000);
+	nv_wr32(priv, 0x40402c, 0x00000000);
+	nv_wr32(priv, 0x404044, 0x00000000);
+	nv_wr32(priv, 0x404094, 0x00000000);
+	nv_wr32(priv, 0x404098, 0x00000000);
+	nv_wr32(priv, 0x40409c, 0x00000000);
+	nv_wr32(priv, 0x4040a0, 0x00000000);
+	nv_wr32(priv, 0x4040a4, 0x00000000);
+	nv_wr32(priv, 0x4040a8, 0x00000000);
+	nv_wr32(priv, 0x4040ac, 0x00000000);
+	nv_wr32(priv, 0x4040b0, 0x00000000);
+	nv_wr32(priv, 0x4040b4, 0x00000000);
+	nv_wr32(priv, 0x4040b8, 0x00000000);
+	nv_wr32(priv, 0x4040bc, 0x00000000);
+	nv_wr32(priv, 0x4040c0, 0x00000000);
+	nv_wr32(priv, 0x4040c4, 0x00000000);
+	nv_wr32(priv, 0x4040c8, 0xf0000087);
+	nv_wr32(priv, 0x4040d4, 0x00000000);
+	nv_wr32(priv, 0x4040d8, 0x00000000);
+	nv_wr32(priv, 0x4040dc, 0x00000000);
+	nv_wr32(priv, 0x4040e0, 0x00000000);
+	nv_wr32(priv, 0x4040e4, 0x00000000);
+	nv_wr32(priv, 0x4040e8, 0x00001000);
+	nv_wr32(priv, 0x4040f8, 0x00000000);
+	nv_wr32(priv, 0x404130, 0x00000000);
+	nv_wr32(priv, 0x404134, 0x00000000);
+	nv_wr32(priv, 0x404138, 0x20000040);
+	nv_wr32(priv, 0x404150, 0x0000002e);
+	nv_wr32(priv, 0x404154, 0x00000400);
+	nv_wr32(priv, 0x404158, 0x00000200);
+	nv_wr32(priv, 0x404164, 0x00000055);
+	nv_wr32(priv, 0x404168, 0x00000000);
+	nv_wr32(priv, 0x404174, 0x00000000);
+	nv_wr32(priv, 0x404178, 0x00000000);
+	nv_wr32(priv, 0x40417c, 0x00000000);
+	for (i = 0; i < 8; i++)
+		nv_wr32(priv, 0x404200 + (i * 4), 0x00000000); /* subc */
+}
+
+static void
+nvc0_grctx_generate_macro(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x404404, 0x00000000);
+	nv_wr32(priv, 0x404408, 0x00000000);
+	nv_wr32(priv, 0x40440c, 0x00000000);
+	nv_wr32(priv, 0x404410, 0x00000000);
+	nv_wr32(priv, 0x404414, 0x00000000);
+	nv_wr32(priv, 0x404418, 0x00000000);
+	nv_wr32(priv, 0x40441c, 0x00000000);
+	nv_wr32(priv, 0x404420, 0x00000000);
+	nv_wr32(priv, 0x404424, 0x00000000);
+	nv_wr32(priv, 0x404428, 0x00000000);
+	nv_wr32(priv, 0x40442c, 0x00000000);
+	nv_wr32(priv, 0x404430, 0x00000000);
+	nv_wr32(priv, 0x404434, 0x00000000);
+	nv_wr32(priv, 0x404438, 0x00000000);
+	nv_wr32(priv, 0x404460, 0x00000000);
+	nv_wr32(priv, 0x404464, 0x00000000);
+	nv_wr32(priv, 0x404468, 0x00ffffff);
+	nv_wr32(priv, 0x40446c, 0x00000000);
+	nv_wr32(priv, 0x404480, 0x00000001);
+	nv_wr32(priv, 0x404498, 0x00000001);
+}
+
+static void
+nvc0_grctx_generate_m2mf(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x404604, 0x00000015);
+	nv_wr32(priv, 0x404608, 0x00000000);
+	nv_wr32(priv, 0x40460c, 0x00002e00);
+	nv_wr32(priv, 0x404610, 0x00000100);
+	nv_wr32(priv, 0x404618, 0x00000000);
+	nv_wr32(priv, 0x40461c, 0x00000000);
+	nv_wr32(priv, 0x404620, 0x00000000);
+	nv_wr32(priv, 0x404624, 0x00000000);
+	nv_wr32(priv, 0x404628, 0x00000000);
+	nv_wr32(priv, 0x40462c, 0x00000000);
+	nv_wr32(priv, 0x404630, 0x00000000);
+	nv_wr32(priv, 0x404634, 0x00000000);
+	nv_wr32(priv, 0x404638, 0x00000004);
+	nv_wr32(priv, 0x40463c, 0x00000000);
+	nv_wr32(priv, 0x404640, 0x00000000);
+	nv_wr32(priv, 0x404644, 0x00000000);
+	nv_wr32(priv, 0x404648, 0x00000000);
+	nv_wr32(priv, 0x40464c, 0x00000000);
+	nv_wr32(priv, 0x404650, 0x00000000);
+	nv_wr32(priv, 0x404654, 0x00000000);
+	nv_wr32(priv, 0x404658, 0x00000000);
+	nv_wr32(priv, 0x40465c, 0x007f0100);
+	nv_wr32(priv, 0x404660, 0x00000000);
+	nv_wr32(priv, 0x404664, 0x00000000);
+	nv_wr32(priv, 0x404668, 0x00000000);
+	nv_wr32(priv, 0x40466c, 0x00000000);
+	nv_wr32(priv, 0x404670, 0x00000000);
+	nv_wr32(priv, 0x404674, 0x00000000);
+	nv_wr32(priv, 0x404678, 0x00000000);
+	nv_wr32(priv, 0x40467c, 0x00000002);
+	nv_wr32(priv, 0x404680, 0x00000000);
+	nv_wr32(priv, 0x404684, 0x00000000);
+	nv_wr32(priv, 0x404688, 0x00000000);
+	nv_wr32(priv, 0x40468c, 0x00000000);
+	nv_wr32(priv, 0x404690, 0x00000000);
+	nv_wr32(priv, 0x404694, 0x00000000);
+	nv_wr32(priv, 0x404698, 0x00000000);
+	nv_wr32(priv, 0x40469c, 0x00000000);
+	nv_wr32(priv, 0x4046a0, 0x007f0080);
+	nv_wr32(priv, 0x4046a4, 0x00000000);
+	nv_wr32(priv, 0x4046a8, 0x00000000);
+	nv_wr32(priv, 0x4046ac, 0x00000000);
+	nv_wr32(priv, 0x4046b0, 0x00000000);
+	nv_wr32(priv, 0x4046b4, 0x00000000);
+	nv_wr32(priv, 0x4046b8, 0x00000000);
+	nv_wr32(priv, 0x4046bc, 0x00000000);
+	nv_wr32(priv, 0x4046c0, 0x00000000);
+	nv_wr32(priv, 0x4046c4, 0x00000000);
+	nv_wr32(priv, 0x4046c8, 0x00000000);
+	nv_wr32(priv, 0x4046cc, 0x00000000);
+	nv_wr32(priv, 0x4046d0, 0x00000000);
+	nv_wr32(priv, 0x4046d4, 0x00000000);
+	nv_wr32(priv, 0x4046d8, 0x00000000);
+	nv_wr32(priv, 0x4046dc, 0x00000000);
+	nv_wr32(priv, 0x4046e0, 0x00000000);
+	nv_wr32(priv, 0x4046e4, 0x00000000);
+	nv_wr32(priv, 0x4046e8, 0x00000000);
+	nv_wr32(priv, 0x4046f0, 0x00000000);
+	nv_wr32(priv, 0x4046f4, 0x00000000);
+}
+
+static void
+nvc0_grctx_generate_unk47xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x404700, 0x00000000);
+	nv_wr32(priv, 0x404704, 0x00000000);
+	nv_wr32(priv, 0x404708, 0x00000000);
+	nv_wr32(priv, 0x40470c, 0x00000000);
+	nv_wr32(priv, 0x404710, 0x00000000);
+	nv_wr32(priv, 0x404714, 0x00000000);
+	nv_wr32(priv, 0x404718, 0x00000000);
+	nv_wr32(priv, 0x40471c, 0x00000000);
+	nv_wr32(priv, 0x404720, 0x00000000);
+	nv_wr32(priv, 0x404724, 0x00000000);
+	nv_wr32(priv, 0x404728, 0x00000000);
+	nv_wr32(priv, 0x40472c, 0x00000000);
+	nv_wr32(priv, 0x404730, 0x00000000);
+	nv_wr32(priv, 0x404734, 0x00000100);
+	nv_wr32(priv, 0x404738, 0x00000000);
+	nv_wr32(priv, 0x40473c, 0x00000000);
+	nv_wr32(priv, 0x404740, 0x00000000);
+	nv_wr32(priv, 0x404744, 0x00000000);
+	nv_wr32(priv, 0x404748, 0x00000000);
+	nv_wr32(priv, 0x40474c, 0x00000000);
+	nv_wr32(priv, 0x404750, 0x00000000);
+	nv_wr32(priv, 0x404754, 0x00000000);
+}
+
+static void
+nvc0_grctx_generate_shaders(struct nvc0_graph_priv *priv)
+{
+
+	if (nv_device(priv)->chipset == 0xd9) {
+		nv_wr32(priv, 0x405800, 0x0f8000bf);
+		nv_wr32(priv, 0x405830, 0x02180218);
+		nv_wr32(priv, 0x405834, 0x08000000);
+	} else
+	if (nv_device(priv)->chipset == 0xc1) {
+		nv_wr32(priv, 0x405800, 0x0f8000bf);
+		nv_wr32(priv, 0x405830, 0x02180218);
+		nv_wr32(priv, 0x405834, 0x00000000);
+	} else {
+		nv_wr32(priv, 0x405800, 0x078000bf);
+		nv_wr32(priv, 0x405830, 0x02180000);
+		nv_wr32(priv, 0x405834, 0x00000000);
+	}
+	nv_wr32(priv, 0x405838, 0x00000000);
+	nv_wr32(priv, 0x405854, 0x00000000);
+	nv_wr32(priv, 0x405870, 0x00000001);
+	nv_wr32(priv, 0x405874, 0x00000001);
+	nv_wr32(priv, 0x405878, 0x00000001);
+	nv_wr32(priv, 0x40587c, 0x00000001);
+	nv_wr32(priv, 0x405a00, 0x00000000);
+	nv_wr32(priv, 0x405a04, 0x00000000);
+	nv_wr32(priv, 0x405a18, 0x00000000);
+}
+
+static void
+nvc0_grctx_generate_unk60xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x406020, 0x000103c1);
+	nv_wr32(priv, 0x406028, 0x00000001);
+	nv_wr32(priv, 0x40602c, 0x00000001);
+	nv_wr32(priv, 0x406030, 0x00000001);
+	nv_wr32(priv, 0x406034, 0x00000001);
+}
+
+static void
+nvc0_grctx_generate_unk64xx(struct nvc0_graph_priv *priv)
+{
+
+	nv_wr32(priv, 0x4064a8, 0x00000000);
+	nv_wr32(priv, 0x4064ac, 0x00003fff);
+	nv_wr32(priv, 0x4064b4, 0x00000000);
+	nv_wr32(priv, 0x4064b8, 0x00000000);
+	if (nv_device(priv)->chipset == 0xd9)
+		nv_wr32(priv, 0x4064bc, 0x00000000);
+	if (nv_device(priv)->chipset == 0xc1 ||
+	    nv_device(priv)->chipset == 0xd9) {
+		nv_wr32(priv, 0x4064c0, 0x80140078);
+		nv_wr32(priv, 0x4064c4, 0x0086ffff);
+	}
+}
+
+static void
+nvc0_grctx_generate_tpbus(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x407804, 0x00000023);
+	nv_wr32(priv, 0x40780c, 0x0a418820);
+	nv_wr32(priv, 0x407810, 0x062080e6);
+	nv_wr32(priv, 0x407814, 0x020398a4);
+	nv_wr32(priv, 0x407818, 0x0e629062);
+	nv_wr32(priv, 0x40781c, 0x0a418820);
+	nv_wr32(priv, 0x407820, 0x000000e6);
+	nv_wr32(priv, 0x4078bc, 0x00000103);
+}
+
+static void
+nvc0_grctx_generate_ccache(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x408000, 0x00000000);
+	nv_wr32(priv, 0x408004, 0x00000000);
+	nv_wr32(priv, 0x408008, 0x00000018);
+	nv_wr32(priv, 0x40800c, 0x00000000);
+	nv_wr32(priv, 0x408010, 0x00000000);
+	nv_wr32(priv, 0x408014, 0x00000069);
+	nv_wr32(priv, 0x408018, 0xe100e100);
+	nv_wr32(priv, 0x408064, 0x00000000);
+}
+
+static void
+nvc0_grctx_generate_rop(struct nvc0_graph_priv *priv)
+{
+	int chipset = nv_device(priv)->chipset;
+
+	/* ROPC_BROADCAST */
+	nv_wr32(priv, 0x408800, 0x02802a3c);
+	nv_wr32(priv, 0x408804, 0x00000040);
+	if (chipset == 0xd9) {
+		nv_wr32(priv, 0x408808, 0x1043e005);
+		nv_wr32(priv, 0x408900, 0x3080b801);
+		nv_wr32(priv, 0x408904, 0x1043e005);
+		nv_wr32(priv, 0x408908, 0x00c8102f);
+	} else
+	if (chipset == 0xc1) {
+		nv_wr32(priv, 0x408808, 0x1003e005);
+		nv_wr32(priv, 0x408900, 0x3080b801);
+		nv_wr32(priv, 0x408904, 0x62000001);
+		nv_wr32(priv, 0x408908, 0x00c80929);
+	} else {
+		nv_wr32(priv, 0x408808, 0x0003e00d);
+		nv_wr32(priv, 0x408900, 0x3080b801);
+		nv_wr32(priv, 0x408904, 0x02000001);
+		nv_wr32(priv, 0x408908, 0x00c80929);
+	}
+	nv_wr32(priv, 0x40890c, 0x00000000);
+	nv_wr32(priv, 0x408980, 0x0000011d);
+}
+
+static void
+nvc0_grctx_generate_gpc(struct nvc0_graph_priv *priv)
+{
+	int chipset = nv_device(priv)->chipset;
+	int i;
+
+	/* GPC_BROADCAST */
+	nv_wr32(priv, 0x418380, 0x00000016);
+	nv_wr32(priv, 0x418400, 0x38004e00);
+	nv_wr32(priv, 0x418404, 0x71e0ffff);
+	nv_wr32(priv, 0x418408, 0x00000000);
+	nv_wr32(priv, 0x41840c, 0x00001008);
+	nv_wr32(priv, 0x418410, 0x0fff0fff);
+	nv_wr32(priv, 0x418414, chipset != 0xd9 ? 0x00200fff : 0x02200fff);
+	nv_wr32(priv, 0x418450, 0x00000000);
+	nv_wr32(priv, 0x418454, 0x00000000);
+	nv_wr32(priv, 0x418458, 0x00000000);
+	nv_wr32(priv, 0x41845c, 0x00000000);
+	nv_wr32(priv, 0x418460, 0x00000000);
+	nv_wr32(priv, 0x418464, 0x00000000);
+	nv_wr32(priv, 0x418468, 0x00000001);
+	nv_wr32(priv, 0x41846c, 0x00000000);
+	nv_wr32(priv, 0x418470, 0x00000000);
+	nv_wr32(priv, 0x418600, 0x0000001f);
+	nv_wr32(priv, 0x418684, 0x0000000f);
+	nv_wr32(priv, 0x418700, 0x00000002);
+	nv_wr32(priv, 0x418704, 0x00000080);
+	nv_wr32(priv, 0x418708, 0x00000000);
+	nv_wr32(priv, 0x41870c, chipset != 0xd9 ? 0x07c80000 : 0x00000000);
+	nv_wr32(priv, 0x418710, 0x00000000);
+	nv_wr32(priv, 0x418800, chipset != 0xd9 ? 0x0006860a : 0x7006860a);
+	nv_wr32(priv, 0x418808, 0x00000000);
+	nv_wr32(priv, 0x41880c, 0x00000000);
+	nv_wr32(priv, 0x418810, 0x00000000);
+	nv_wr32(priv, 0x418828, 0x00008442);
+	if (chipset == 0xc1 || chipset == 0xd9)
+		nv_wr32(priv, 0x418830, 0x10000001);
+	else
+		nv_wr32(priv, 0x418830, 0x00000001);
+	nv_wr32(priv, 0x4188d8, 0x00000008);
+	nv_wr32(priv, 0x4188e0, 0x01000000);
+	nv_wr32(priv, 0x4188e8, 0x00000000);
+	nv_wr32(priv, 0x4188ec, 0x00000000);
+	nv_wr32(priv, 0x4188f0, 0x00000000);
+	nv_wr32(priv, 0x4188f4, 0x00000000);
+	nv_wr32(priv, 0x4188f8, 0x00000000);
+	if (chipset == 0xd9)
+		nv_wr32(priv, 0x4188fc, 0x20100008);
+	else if (chipset == 0xc1)
+		nv_wr32(priv, 0x4188fc, 0x00100018);
+	else
+		nv_wr32(priv, 0x4188fc, 0x00100000);
+	nv_wr32(priv, 0x41891c, 0x00ff00ff);
+	nv_wr32(priv, 0x418924, 0x00000000);
+	nv_wr32(priv, 0x418928, 0x00ffff00);
+	nv_wr32(priv, 0x41892c, 0x0000ff00);
+	for (i = 0; i < 8; i++) {
+		nv_wr32(priv, 0x418a00 + (i * 0x20), 0x00000000);
+		nv_wr32(priv, 0x418a04 + (i * 0x20), 0x00000000);
+		nv_wr32(priv, 0x418a08 + (i * 0x20), 0x00000000);
+		nv_wr32(priv, 0x418a0c + (i * 0x20), 0x00010000);
+		nv_wr32(priv, 0x418a10 + (i * 0x20), 0x00000000);
+		nv_wr32(priv, 0x418a14 + (i * 0x20), 0x00000000);
+		nv_wr32(priv, 0x418a18 + (i * 0x20), 0x00000000);
+	}
+	nv_wr32(priv, 0x418b00, chipset != 0xd9 ? 0x00000000 : 0x00000006);
+	nv_wr32(priv, 0x418b08, 0x0a418820);
+	nv_wr32(priv, 0x418b0c, 0x062080e6);
+	nv_wr32(priv, 0x418b10, 0x020398a4);
+	nv_wr32(priv, 0x418b14, 0x0e629062);
+	nv_wr32(priv, 0x418b18, 0x0a418820);
+	nv_wr32(priv, 0x418b1c, 0x000000e6);
+	nv_wr32(priv, 0x418bb8, 0x00000103);
+	nv_wr32(priv, 0x418c08, 0x00000001);
+	nv_wr32(priv, 0x418c10, 0x00000000);
+	nv_wr32(priv, 0x418c14, 0x00000000);
+	nv_wr32(priv, 0x418c18, 0x00000000);
+	nv_wr32(priv, 0x418c1c, 0x00000000);
+	nv_wr32(priv, 0x418c20, 0x00000000);
+	nv_wr32(priv, 0x418c24, 0x00000000);
+	nv_wr32(priv, 0x418c28, 0x00000000);
+	nv_wr32(priv, 0x418c2c, 0x00000000);
+	if (chipset == 0xc1 || chipset == 0xd9)
+		nv_wr32(priv, 0x418c6c, 0x00000001);
+	nv_wr32(priv, 0x418c80, 0x20200004);
+	nv_wr32(priv, 0x418c8c, 0x00000001);
+	nv_wr32(priv, 0x419000, 0x00000780);
+	nv_wr32(priv, 0x419004, 0x00000000);
+	nv_wr32(priv, 0x419008, 0x00000000);
+	nv_wr32(priv, 0x419014, 0x00000004);
+}
+
+static void
+nvc0_grctx_generate_tp(struct nvc0_graph_priv *priv)
+{
+	int chipset = nv_device(priv)->chipset;
+
+	/* GPC_BROADCAST.TP_BROADCAST */
+	nv_wr32(priv, 0x419818, 0x00000000);
+	nv_wr32(priv, 0x41983c, 0x00038bc7);
+	nv_wr32(priv, 0x419848, 0x00000000);
+	if (chipset == 0xc1 || chipset == 0xd9)
+		nv_wr32(priv, 0x419864, 0x00000129);
+	else
+		nv_wr32(priv, 0x419864, 0x0000012a);
+	nv_wr32(priv, 0x419888, 0x00000000);
+	nv_wr32(priv, 0x419a00, 0x000001f0);
+	nv_wr32(priv, 0x419a04, 0x00000001);
+	nv_wr32(priv, 0x419a08, 0x00000023);
+	nv_wr32(priv, 0x419a0c, 0x00020000);
+	nv_wr32(priv, 0x419a10, 0x00000000);
+	nv_wr32(priv, 0x419a14, 0x00000200);
+	nv_wr32(priv, 0x419a1c, 0x00000000);
+	nv_wr32(priv, 0x419a20, 0x00000800);
+	if (chipset == 0xd9)
+		nv_wr32(priv, 0x00419ac4, 0x0017f440);
+	else if (chipset != 0xc0 && chipset != 0xc8)
+		nv_wr32(priv, 0x00419ac4, 0x0007f440);
+	nv_wr32(priv, 0x419b00, 0x0a418820);
+	nv_wr32(priv, 0x419b04, 0x062080e6);
+	nv_wr32(priv, 0x419b08, 0x020398a4);
+	nv_wr32(priv, 0x419b0c, 0x0e629062);
+	nv_wr32(priv, 0x419b10, 0x0a418820);
+	nv_wr32(priv, 0x419b14, 0x000000e6);
+	nv_wr32(priv, 0x419bd0, 0x00900103);
+	if (chipset == 0xc1 || chipset == 0xd9)
+		nv_wr32(priv, 0x419be0, 0x00400001);
+	else
+		nv_wr32(priv, 0x419be0, 0x00000001);
+	nv_wr32(priv, 0x419be4, 0x00000000);
+	nv_wr32(priv, 0x419c00, chipset != 0xd9 ? 0x00000002 : 0x0000000a);
+	nv_wr32(priv, 0x419c04, 0x00000006);
+	nv_wr32(priv, 0x419c08, 0x00000002);
+	nv_wr32(priv, 0x419c20, 0x00000000);
+	if (nv_device(priv)->chipset == 0xd9) {
+		nv_wr32(priv, 0x419c24, 0x00084210);
+		nv_wr32(priv, 0x419c28, 0x3cf3cf3c);
+		nv_wr32(priv, 0x419cb0, 0x00020048);
+	} else
+	if (chipset == 0xce || chipset == 0xcf) {
+		nv_wr32(priv, 0x419cb0, 0x00020048);
+	} else {
+		nv_wr32(priv, 0x419cb0, 0x00060048);
+	}
+	nv_wr32(priv, 0x419ce8, 0x00000000);
+	nv_wr32(priv, 0x419cf4, 0x00000183);
+	if (chipset == 0xc1 || chipset == 0xd9)
+		nv_wr32(priv, 0x419d20, 0x12180000);
+	else
+		nv_wr32(priv, 0x419d20, 0x02180000);
+	nv_wr32(priv, 0x419d24, 0x00001fff);
+	if (chipset == 0xc1 || chipset == 0xd9)
+		nv_wr32(priv, 0x419d44, 0x02180218);
+	nv_wr32(priv, 0x419e04, 0x00000000);
+	nv_wr32(priv, 0x419e08, 0x00000000);
+	nv_wr32(priv, 0x419e0c, 0x00000000);
+	nv_wr32(priv, 0x419e10, 0x00000002);
+	nv_wr32(priv, 0x419e44, 0x001beff2);
+	nv_wr32(priv, 0x419e48, 0x00000000);
+	nv_wr32(priv, 0x419e4c, 0x0000000f);
+	nv_wr32(priv, 0x419e50, 0x00000000);
+	nv_wr32(priv, 0x419e54, 0x00000000);
+	nv_wr32(priv, 0x419e58, 0x00000000);
+	nv_wr32(priv, 0x419e5c, 0x00000000);
+	nv_wr32(priv, 0x419e60, 0x00000000);
+	nv_wr32(priv, 0x419e64, 0x00000000);
+	nv_wr32(priv, 0x419e68, 0x00000000);
+	nv_wr32(priv, 0x419e6c, 0x00000000);
+	nv_wr32(priv, 0x419e70, 0x00000000);
+	nv_wr32(priv, 0x419e74, 0x00000000);
+	nv_wr32(priv, 0x419e78, 0x00000000);
+	nv_wr32(priv, 0x419e7c, 0x00000000);
+	nv_wr32(priv, 0x419e80, 0x00000000);
+	nv_wr32(priv, 0x419e84, 0x00000000);
+	nv_wr32(priv, 0x419e88, 0x00000000);
+	nv_wr32(priv, 0x419e8c, 0x00000000);
+	nv_wr32(priv, 0x419e90, 0x00000000);
+	nv_wr32(priv, 0x419e98, 0x00000000);
+	if (chipset != 0xc0 && chipset != 0xc8)
+		nv_wr32(priv, 0x419ee0, 0x00011110);
+	nv_wr32(priv, 0x419f50, 0x00000000);
+	nv_wr32(priv, 0x419f54, 0x00000000);
+	if (chipset != 0xc0 && chipset != 0xc8)
+		nv_wr32(priv, 0x419f58, 0x00000000);
+}
+
+int
+nvc0_grctx_generate(struct nvc0_graph_priv *priv)
+{
+	struct nvc0_grctx info;
+	int ret, i, gpc, tpc, id;
+	u32 fermi = nvc0_graph_class(priv);
+	u32 r000260, tmp;
+
+	ret = nvc0_grctx_init(priv, &info);
+	if (ret)
+		return ret;
+
+	r000260 = nv_rd32(priv, 0x000260);
+	nv_wr32(priv, 0x000260, r000260 & ~1);
+	nv_wr32(priv, 0x400208, 0x00000000);
+
+	nvc0_grctx_generate_dispatch(priv);
+	nvc0_grctx_generate_macro(priv);
+	nvc0_grctx_generate_m2mf(priv);
+	nvc0_grctx_generate_unk47xx(priv);
+	nvc0_grctx_generate_shaders(priv);
+	nvc0_grctx_generate_unk60xx(priv);
+	nvc0_grctx_generate_unk64xx(priv);
+	nvc0_grctx_generate_tpbus(priv);
+	nvc0_grctx_generate_ccache(priv);
+	nvc0_grctx_generate_rop(priv);
+	nvc0_grctx_generate_gpc(priv);
+	nvc0_grctx_generate_tp(priv);
+
+	nv_wr32(priv, 0x404154, 0x00000000);
+
+	/* generate per-context mmio list data */
+	mmio_data(0x002000, 0x0100, NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS);
+	mmio_data(0x008000, 0x0100, NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS);
+	mmio_data(0x060000, 0x1000, NV_MEM_ACCESS_RW);
+	mmio_list(0x408004, 0x00000000,  8, 0);
+	mmio_list(0x408008, 0x80000018,  0, 0);
+	mmio_list(0x40800c, 0x00000000,  8, 1);
+	mmio_list(0x408010, 0x80000000,  0, 0);
+	mmio_list(0x418810, 0x80000000, 12, 2);
+	mmio_list(0x419848, 0x10000000, 12, 2);
+	mmio_list(0x419004, 0x00000000,  8, 1);
+	mmio_list(0x419008, 0x00000000,  0, 0);
+	mmio_list(0x418808, 0x00000000,  8, 0);
+	mmio_list(0x41880c, 0x80000018,  0, 0);
+	if (nv_device(priv)->chipset != 0xc1) {
+		tmp = 0x02180000;
+		mmio_list(0x405830, tmp, 0, 0);
+		for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+			for (tpc = 0; tpc < priv->tpc_nr[gpc]; tpc++) {
+				u32 reg = TPC_UNIT(gpc, tpc, 0x0520);
+				mmio_list(reg, tmp, 0, 0);
+				tmp += 0x0324;
+			}
+		}
+	} else {
+		tmp = 0x02180000;
+		mmio_list(0x405830, 0x00000218 | tmp, 0, 0);
+		mmio_list(0x4064c4, 0x0086ffff, 0, 0);
+		for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+			for (tpc = 0; tpc < priv->tpc_nr[gpc]; tpc++) {
+				u32 reg = TPC_UNIT(gpc, tpc, 0x0520);
+				mmio_list(reg, 0x10000000 | tmp, 0, 0);
+				tmp += 0x0324;
+			}
+			for (tpc = 0; tpc < priv->tpc_nr[gpc]; tpc++) {
+				u32 reg = TPC_UNIT(gpc, tpc, 0x0544);
+				mmio_list(reg, tmp, 0, 0);
+				tmp += 0x0324;
+			}
+		}
+	}
+
+	for (tpc = 0, id = 0; tpc < 4; tpc++) {
+		for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+			if (tpc < priv->tpc_nr[gpc]) {
+				nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x698), id);
+				nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x4e8), id);
+				nv_wr32(priv, GPC_UNIT(gpc, 0x0c10 + tpc * 4), id);
+				nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x088), id);
+				id++;
+			}
+
+			nv_wr32(priv, GPC_UNIT(gpc, 0x0c08), priv->tpc_nr[gpc]);
+			nv_wr32(priv, GPC_UNIT(gpc, 0x0c8c), priv->tpc_nr[gpc]);
+		}
+	}
+
+	tmp = 0;
+	for (i = 0; i < priv->gpc_nr; i++)
+		tmp |= priv->tpc_nr[i] << (i * 4);
+	nv_wr32(priv, 0x406028, tmp);
+	nv_wr32(priv, 0x405870, tmp);
+
+	nv_wr32(priv, 0x40602c, 0x00000000);
+	nv_wr32(priv, 0x405874, 0x00000000);
+	nv_wr32(priv, 0x406030, 0x00000000);
+	nv_wr32(priv, 0x405878, 0x00000000);
+	nv_wr32(priv, 0x406034, 0x00000000);
+	nv_wr32(priv, 0x40587c, 0x00000000);
+
+	if (1) {
+		u8 tpcnr[GPC_MAX], data[TPC_MAX];
+
+		memcpy(tpcnr, priv->tpc_nr, sizeof(priv->tpc_nr));
+		memset(data, 0x1f, sizeof(data));
+
+		gpc = -1;
+		for (tpc = 0; tpc < priv->tpc_total; tpc++) {
+			do {
+				gpc = (gpc + 1) % priv->gpc_nr;
+			} while (!tpcnr[gpc]);
+			tpcnr[gpc]--;
+			data[tpc] = gpc;
+		}
+
+		for (i = 0; i < 4; i++)
+			nv_wr32(priv, 0x4060a8 + (i * 4), ((u32 *)data)[i]);
+	}
+
+	if (1) {
+		u32 data[6] = {}, data2[2] = {};
+		u8 tpcnr[GPC_MAX];
+		u8 shift, ntpcv;
+
+		/* calculate first set of magics */
+		memcpy(tpcnr, priv->tpc_nr, sizeof(priv->tpc_nr));
+
+		gpc = -1;
+		for (tpc = 0; tpc < priv->tpc_total; tpc++) {
+			do {
+				gpc = (gpc + 1) % priv->gpc_nr;
+			} while (!tpcnr[gpc]);
+			tpcnr[gpc]--;
+
+			data[tpc / 6] |= gpc << ((tpc % 6) * 5);
+		}
+
+		for (; tpc < 32; tpc++)
+			data[tpc / 6] |= 7 << ((tpc % 6) * 5);
+
+		/* and the second... */
+		shift = 0;
+		ntpcv = priv->tpc_total;
+		while (!(ntpcv & (1 << 4))) {
+			ntpcv <<= 1;
+			shift++;
+		}
+
+		data2[0]  = (ntpcv << 16);
+		data2[0] |= (shift << 21);
+		data2[0] |= (((1 << (0 + 5)) % ntpcv) << 24);
+		for (i = 1; i < 7; i++)
+			data2[1] |= ((1 << (i + 5)) % ntpcv) << ((i - 1) * 5);
+
+		/* GPC_BROADCAST */
+		nv_wr32(priv, 0x418bb8, (priv->tpc_total << 8) |
+					priv->magic_not_rop_nr);
+		for (i = 0; i < 6; i++)
+			nv_wr32(priv, 0x418b08 + (i * 4), data[i]);
+
+		/* GPC_BROADCAST.TP_BROADCAST */
+		nv_wr32(priv, 0x419bd0, (priv->tpc_total << 8) |
+				       priv->magic_not_rop_nr |
+				       data2[0]);
+		nv_wr32(priv, 0x419be4, data2[1]);
+		for (i = 0; i < 6; i++)
+			nv_wr32(priv, 0x419b00 + (i * 4), data[i]);
+
+		/* UNK78xx */
+		nv_wr32(priv, 0x4078bc, (priv->tpc_total << 8) |
+					priv->magic_not_rop_nr);
+		for (i = 0; i < 6; i++)
+			nv_wr32(priv, 0x40780c + (i * 4), data[i]);
+	}
+
+	if (1) {
+		u32 tpc_mask = 0, tpc_set = 0;
+		u8  tpcnr[GPC_MAX], a, b;
+
+		memcpy(tpcnr, priv->tpc_nr, sizeof(priv->tpc_nr));
+		for (gpc = 0; gpc < priv->gpc_nr; gpc++)
+			tpc_mask |= ((1 << priv->tpc_nr[gpc]) - 1) << (gpc * 8);
+
+		for (i = 0, gpc = -1, b = -1; i < 32; i++) {
+			a = (i * (priv->tpc_total - 1)) / 32;
+			if (a != b) {
+				b = a;
+				do {
+					gpc = (gpc + 1) % priv->gpc_nr;
+				} while (!tpcnr[gpc]);
+				tpc = priv->tpc_nr[gpc] - tpcnr[gpc]--;
+
+				tpc_set |= 1 << ((gpc * 8) + tpc);
+			}
+
+			nv_wr32(priv, 0x406800 + (i * 0x20), tpc_set);
+			nv_wr32(priv, 0x406c00 + (i * 0x20), tpc_set ^ tpc_mask);
+		}
+	}
+
+	nv_wr32(priv, 0x400208, 0x80000000);
+
+	nv_icmd(priv, 0x00001000, 0x00000004);
+	nv_icmd(priv, 0x000000a9, 0x0000ffff);
+	nv_icmd(priv, 0x00000038, 0x0fac6881);
+	nv_icmd(priv, 0x0000003d, 0x00000001);
+	nv_icmd(priv, 0x000000e8, 0x00000400);
+	nv_icmd(priv, 0x000000e9, 0x00000400);
+	nv_icmd(priv, 0x000000ea, 0x00000400);
+	nv_icmd(priv, 0x000000eb, 0x00000400);
+	nv_icmd(priv, 0x000000ec, 0x00000400);
+	nv_icmd(priv, 0x000000ed, 0x00000400);
+	nv_icmd(priv, 0x000000ee, 0x00000400);
+	nv_icmd(priv, 0x000000ef, 0x00000400);
+	nv_icmd(priv, 0x00000078, 0x00000300);
+	nv_icmd(priv, 0x00000079, 0x00000300);
+	nv_icmd(priv, 0x0000007a, 0x00000300);
+	nv_icmd(priv, 0x0000007b, 0x00000300);
+	nv_icmd(priv, 0x0000007c, 0x00000300);
+	nv_icmd(priv, 0x0000007d, 0x00000300);
+	nv_icmd(priv, 0x0000007e, 0x00000300);
+	nv_icmd(priv, 0x0000007f, 0x00000300);
+	nv_icmd(priv, 0x00000050, 0x00000011);
+	nv_icmd(priv, 0x00000058, 0x00000008);
+	nv_icmd(priv, 0x00000059, 0x00000008);
+	nv_icmd(priv, 0x0000005a, 0x00000008);
+	nv_icmd(priv, 0x0000005b, 0x00000008);
+	nv_icmd(priv, 0x0000005c, 0x00000008);
+	nv_icmd(priv, 0x0000005d, 0x00000008);
+	nv_icmd(priv, 0x0000005e, 0x00000008);
+	nv_icmd(priv, 0x0000005f, 0x00000008);
+	nv_icmd(priv, 0x00000208, 0x00000001);
+	nv_icmd(priv, 0x00000209, 0x00000001);
+	nv_icmd(priv, 0x0000020a, 0x00000001);
+	nv_icmd(priv, 0x0000020b, 0x00000001);
+	nv_icmd(priv, 0x0000020c, 0x00000001);
+	nv_icmd(priv, 0x0000020d, 0x00000001);
+	nv_icmd(priv, 0x0000020e, 0x00000001);
+	nv_icmd(priv, 0x0000020f, 0x00000001);
+	nv_icmd(priv, 0x00000081, 0x00000001);
+	nv_icmd(priv, 0x00000085, 0x00000004);
+	nv_icmd(priv, 0x00000088, 0x00000400);
+	nv_icmd(priv, 0x00000090, 0x00000300);
+	nv_icmd(priv, 0x00000098, 0x00001001);
+	nv_icmd(priv, 0x000000e3, 0x00000001);
+	nv_icmd(priv, 0x000000da, 0x00000001);
+	nv_icmd(priv, 0x000000f8, 0x00000003);
+	nv_icmd(priv, 0x000000fa, 0x00000001);
+	nv_icmd(priv, 0x0000009f, 0x0000ffff);
+	nv_icmd(priv, 0x000000a0, 0x0000ffff);
+	nv_icmd(priv, 0x000000a1, 0x0000ffff);
+	nv_icmd(priv, 0x000000a2, 0x0000ffff);
+	nv_icmd(priv, 0x000000b1, 0x00000001);
+	nv_icmd(priv, 0x000000b2, 0x00000000);
+	nv_icmd(priv, 0x000000b3, 0x00000000);
+	nv_icmd(priv, 0x000000b4, 0x00000000);
+	nv_icmd(priv, 0x000000b5, 0x00000000);
+	nv_icmd(priv, 0x000000b6, 0x00000000);
+	nv_icmd(priv, 0x000000b7, 0x00000000);
+	nv_icmd(priv, 0x000000b8, 0x00000000);
+	nv_icmd(priv, 0x000000b9, 0x00000000);
+	nv_icmd(priv, 0x000000ba, 0x00000000);
+	nv_icmd(priv, 0x000000bb, 0x00000000);
+	nv_icmd(priv, 0x000000bc, 0x00000000);
+	nv_icmd(priv, 0x000000bd, 0x00000000);
+	nv_icmd(priv, 0x000000be, 0x00000000);
+	nv_icmd(priv, 0x000000bf, 0x00000000);
+	nv_icmd(priv, 0x000000c0, 0x00000000);
+	nv_icmd(priv, 0x000000c1, 0x00000000);
+	nv_icmd(priv, 0x000000c2, 0x00000000);
+	nv_icmd(priv, 0x000000c3, 0x00000000);
+	nv_icmd(priv, 0x000000c4, 0x00000000);
+	nv_icmd(priv, 0x000000c5, 0x00000000);
+	nv_icmd(priv, 0x000000c6, 0x00000000);
+	nv_icmd(priv, 0x000000c7, 0x00000000);
+	nv_icmd(priv, 0x000000c8, 0x00000000);
+	nv_icmd(priv, 0x000000c9, 0x00000000);
+	nv_icmd(priv, 0x000000ca, 0x00000000);
+	nv_icmd(priv, 0x000000cb, 0x00000000);
+	nv_icmd(priv, 0x000000cc, 0x00000000);
+	nv_icmd(priv, 0x000000cd, 0x00000000);
+	nv_icmd(priv, 0x000000ce, 0x00000000);
+	nv_icmd(priv, 0x000000cf, 0x00000000);
+	nv_icmd(priv, 0x000000d0, 0x00000000);
+	nv_icmd(priv, 0x000000d1, 0x00000000);
+	nv_icmd(priv, 0x000000d2, 0x00000000);
+	nv_icmd(priv, 0x000000d3, 0x00000000);
+	nv_icmd(priv, 0x000000d4, 0x00000000);
+	nv_icmd(priv, 0x000000d5, 0x00000000);
+	nv_icmd(priv, 0x000000d6, 0x00000000);
+	nv_icmd(priv, 0x000000d7, 0x00000000);
+	nv_icmd(priv, 0x000000d8, 0x00000000);
+	nv_icmd(priv, 0x000000d9, 0x00000000);
+	nv_icmd(priv, 0x00000210, 0x00000040);
+	nv_icmd(priv, 0x00000211, 0x00000040);
+	nv_icmd(priv, 0x00000212, 0x00000040);
+	nv_icmd(priv, 0x00000213, 0x00000040);
+	nv_icmd(priv, 0x00000214, 0x00000040);
+	nv_icmd(priv, 0x00000215, 0x00000040);
+	nv_icmd(priv, 0x00000216, 0x00000040);
+	nv_icmd(priv, 0x00000217, 0x00000040);
+	if (nv_device(priv)->chipset == 0xd9) {
+		for (i = 0x0400; i <= 0x0417; i++)
+			nv_icmd(priv, i, 0x00000040);
+	}
+	nv_icmd(priv, 0x00000218, 0x0000c080);
+	nv_icmd(priv, 0x00000219, 0x0000c080);
+	nv_icmd(priv, 0x0000021a, 0x0000c080);
+	nv_icmd(priv, 0x0000021b, 0x0000c080);
+	nv_icmd(priv, 0x0000021c, 0x0000c080);
+	nv_icmd(priv, 0x0000021d, 0x0000c080);
+	nv_icmd(priv, 0x0000021e, 0x0000c080);
+	nv_icmd(priv, 0x0000021f, 0x0000c080);
+	if (nv_device(priv)->chipset == 0xd9) {
+		for (i = 0x0440; i <= 0x0457; i++)
+			nv_icmd(priv, i, 0x0000c080);
+	}
+	nv_icmd(priv, 0x000000ad, 0x0000013e);
+	nv_icmd(priv, 0x000000e1, 0x00000010);
+	nv_icmd(priv, 0x00000290, 0x00000000);
+	nv_icmd(priv, 0x00000291, 0x00000000);
+	nv_icmd(priv, 0x00000292, 0x00000000);
+	nv_icmd(priv, 0x00000293, 0x00000000);
+	nv_icmd(priv, 0x00000294, 0x00000000);
+	nv_icmd(priv, 0x00000295, 0x00000000);
+	nv_icmd(priv, 0x00000296, 0x00000000);
+	nv_icmd(priv, 0x00000297, 0x00000000);
+	nv_icmd(priv, 0x00000298, 0x00000000);
+	nv_icmd(priv, 0x00000299, 0x00000000);
+	nv_icmd(priv, 0x0000029a, 0x00000000);
+	nv_icmd(priv, 0x0000029b, 0x00000000);
+	nv_icmd(priv, 0x0000029c, 0x00000000);
+	nv_icmd(priv, 0x0000029d, 0x00000000);
+	nv_icmd(priv, 0x0000029e, 0x00000000);
+	nv_icmd(priv, 0x0000029f, 0x00000000);
+	nv_icmd(priv, 0x000003b0, 0x00000000);
+	nv_icmd(priv, 0x000003b1, 0x00000000);
+	nv_icmd(priv, 0x000003b2, 0x00000000);
+	nv_icmd(priv, 0x000003b3, 0x00000000);
+	nv_icmd(priv, 0x000003b4, 0x00000000);
+	nv_icmd(priv, 0x000003b5, 0x00000000);
+	nv_icmd(priv, 0x000003b6, 0x00000000);
+	nv_icmd(priv, 0x000003b7, 0x00000000);
+	nv_icmd(priv, 0x000003b8, 0x00000000);
+	nv_icmd(priv, 0x000003b9, 0x00000000);
+	nv_icmd(priv, 0x000003ba, 0x00000000);
+	nv_icmd(priv, 0x000003bb, 0x00000000);
+	nv_icmd(priv, 0x000003bc, 0x00000000);
+	nv_icmd(priv, 0x000003bd, 0x00000000);
+	nv_icmd(priv, 0x000003be, 0x00000000);
+	nv_icmd(priv, 0x000003bf, 0x00000000);
+	nv_icmd(priv, 0x000002a0, 0x00000000);
+	nv_icmd(priv, 0x000002a1, 0x00000000);
+	nv_icmd(priv, 0x000002a2, 0x00000000);
+	nv_icmd(priv, 0x000002a3, 0x00000000);
+	nv_icmd(priv, 0x000002a4, 0x00000000);
+	nv_icmd(priv, 0x000002a5, 0x00000000);
+	nv_icmd(priv, 0x000002a6, 0x00000000);
+	nv_icmd(priv, 0x000002a7, 0x00000000);
+	nv_icmd(priv, 0x000002a8, 0x00000000);
+	nv_icmd(priv, 0x000002a9, 0x00000000);
+	nv_icmd(priv, 0x000002aa, 0x00000000);
+	nv_icmd(priv, 0x000002ab, 0x00000000);
+	nv_icmd(priv, 0x000002ac, 0x00000000);
+	nv_icmd(priv, 0x000002ad, 0x00000000);
+	nv_icmd(priv, 0x000002ae, 0x00000000);
+	nv_icmd(priv, 0x000002af, 0x00000000);
+	nv_icmd(priv, 0x00000420, 0x00000000);
+	nv_icmd(priv, 0x00000421, 0x00000000);
+	nv_icmd(priv, 0x00000422, 0x00000000);
+	nv_icmd(priv, 0x00000423, 0x00000000);
+	nv_icmd(priv, 0x00000424, 0x00000000);
+	nv_icmd(priv, 0x00000425, 0x00000000);
+	nv_icmd(priv, 0x00000426, 0x00000000);
+	nv_icmd(priv, 0x00000427, 0x00000000);
+	nv_icmd(priv, 0x00000428, 0x00000000);
+	nv_icmd(priv, 0x00000429, 0x00000000);
+	nv_icmd(priv, 0x0000042a, 0x00000000);
+	nv_icmd(priv, 0x0000042b, 0x00000000);
+	nv_icmd(priv, 0x0000042c, 0x00000000);
+	nv_icmd(priv, 0x0000042d, 0x00000000);
+	nv_icmd(priv, 0x0000042e, 0x00000000);
+	nv_icmd(priv, 0x0000042f, 0x00000000);
+	nv_icmd(priv, 0x000002b0, 0x00000000);
+	nv_icmd(priv, 0x000002b1, 0x00000000);
+	nv_icmd(priv, 0x000002b2, 0x00000000);
+	nv_icmd(priv, 0x000002b3, 0x00000000);
+	nv_icmd(priv, 0x000002b4, 0x00000000);
+	nv_icmd(priv, 0x000002b5, 0x00000000);
+	nv_icmd(priv, 0x000002b6, 0x00000000);
+	nv_icmd(priv, 0x000002b7, 0x00000000);
+	nv_icmd(priv, 0x000002b8, 0x00000000);
+	nv_icmd(priv, 0x000002b9, 0x00000000);
+	nv_icmd(priv, 0x000002ba, 0x00000000);
+	nv_icmd(priv, 0x000002bb, 0x00000000);
+	nv_icmd(priv, 0x000002bc, 0x00000000);
+	nv_icmd(priv, 0x000002bd, 0x00000000);
+	nv_icmd(priv, 0x000002be, 0x00000000);
+	nv_icmd(priv, 0x000002bf, 0x00000000);
+	nv_icmd(priv, 0x00000430, 0x00000000);
+	nv_icmd(priv, 0x00000431, 0x00000000);
+	nv_icmd(priv, 0x00000432, 0x00000000);
+	nv_icmd(priv, 0x00000433, 0x00000000);
+	nv_icmd(priv, 0x00000434, 0x00000000);
+	nv_icmd(priv, 0x00000435, 0x00000000);
+	nv_icmd(priv, 0x00000436, 0x00000000);
+	nv_icmd(priv, 0x00000437, 0x00000000);
+	nv_icmd(priv, 0x00000438, 0x00000000);
+	nv_icmd(priv, 0x00000439, 0x00000000);
+	nv_icmd(priv, 0x0000043a, 0x00000000);
+	nv_icmd(priv, 0x0000043b, 0x00000000);
+	nv_icmd(priv, 0x0000043c, 0x00000000);
+	nv_icmd(priv, 0x0000043d, 0x00000000);
+	nv_icmd(priv, 0x0000043e, 0x00000000);
+	nv_icmd(priv, 0x0000043f, 0x00000000);
+	nv_icmd(priv, 0x000002c0, 0x00000000);
+	nv_icmd(priv, 0x000002c1, 0x00000000);
+	nv_icmd(priv, 0x000002c2, 0x00000000);
+	nv_icmd(priv, 0x000002c3, 0x00000000);
+	nv_icmd(priv, 0x000002c4, 0x00000000);
+	nv_icmd(priv, 0x000002c5, 0x00000000);
+	nv_icmd(priv, 0x000002c6, 0x00000000);
+	nv_icmd(priv, 0x000002c7, 0x00000000);
+	nv_icmd(priv, 0x000002c8, 0x00000000);
+	nv_icmd(priv, 0x000002c9, 0x00000000);
+	nv_icmd(priv, 0x000002ca, 0x00000000);
+	nv_icmd(priv, 0x000002cb, 0x00000000);
+	nv_icmd(priv, 0x000002cc, 0x00000000);
+	nv_icmd(priv, 0x000002cd, 0x00000000);
+	nv_icmd(priv, 0x000002ce, 0x00000000);
+	nv_icmd(priv, 0x000002cf, 0x00000000);
+	nv_icmd(priv, 0x000004d0, 0x00000000);
+	nv_icmd(priv, 0x000004d1, 0x00000000);
+	nv_icmd(priv, 0x000004d2, 0x00000000);
+	nv_icmd(priv, 0x000004d3, 0x00000000);
+	nv_icmd(priv, 0x000004d4, 0x00000000);
+	nv_icmd(priv, 0x000004d5, 0x00000000);
+	nv_icmd(priv, 0x000004d6, 0x00000000);
+	nv_icmd(priv, 0x000004d7, 0x00000000);
+	nv_icmd(priv, 0x000004d8, 0x00000000);
+	nv_icmd(priv, 0x000004d9, 0x00000000);
+	nv_icmd(priv, 0x000004da, 0x00000000);
+	nv_icmd(priv, 0x000004db, 0x00000000);
+	nv_icmd(priv, 0x000004dc, 0x00000000);
+	nv_icmd(priv, 0x000004dd, 0x00000000);
+	nv_icmd(priv, 0x000004de, 0x00000000);
+	nv_icmd(priv, 0x000004df, 0x00000000);
+	nv_icmd(priv, 0x00000720, 0x00000000);
+	nv_icmd(priv, 0x00000721, 0x00000000);
+	nv_icmd(priv, 0x00000722, 0x00000000);
+	nv_icmd(priv, 0x00000723, 0x00000000);
+	nv_icmd(priv, 0x00000724, 0x00000000);
+	nv_icmd(priv, 0x00000725, 0x00000000);
+	nv_icmd(priv, 0x00000726, 0x00000000);
+	nv_icmd(priv, 0x00000727, 0x00000000);
+	nv_icmd(priv, 0x00000728, 0x00000000);
+	nv_icmd(priv, 0x00000729, 0x00000000);
+	nv_icmd(priv, 0x0000072a, 0x00000000);
+	nv_icmd(priv, 0x0000072b, 0x00000000);
+	nv_icmd(priv, 0x0000072c, 0x00000000);
+	nv_icmd(priv, 0x0000072d, 0x00000000);
+	nv_icmd(priv, 0x0000072e, 0x00000000);
+	nv_icmd(priv, 0x0000072f, 0x00000000);
+	nv_icmd(priv, 0x000008c0, 0x00000000);
+	nv_icmd(priv, 0x000008c1, 0x00000000);
+	nv_icmd(priv, 0x000008c2, 0x00000000);
+	nv_icmd(priv, 0x000008c3, 0x00000000);
+	nv_icmd(priv, 0x000008c4, 0x00000000);
+	nv_icmd(priv, 0x000008c5, 0x00000000);
+	nv_icmd(priv, 0x000008c6, 0x00000000);
+	nv_icmd(priv, 0x000008c7, 0x00000000);
+	nv_icmd(priv, 0x000008c8, 0x00000000);
+	nv_icmd(priv, 0x000008c9, 0x00000000);
+	nv_icmd(priv, 0x000008ca, 0x00000000);
+	nv_icmd(priv, 0x000008cb, 0x00000000);
+	nv_icmd(priv, 0x000008cc, 0x00000000);
+	nv_icmd(priv, 0x000008cd, 0x00000000);
+	nv_icmd(priv, 0x000008ce, 0x00000000);
+	nv_icmd(priv, 0x000008cf, 0x00000000);
+	nv_icmd(priv, 0x00000890, 0x00000000);
+	nv_icmd(priv, 0x00000891, 0x00000000);
+	nv_icmd(priv, 0x00000892, 0x00000000);
+	nv_icmd(priv, 0x00000893, 0x00000000);
+	nv_icmd(priv, 0x00000894, 0x00000000);
+	nv_icmd(priv, 0x00000895, 0x00000000);
+	nv_icmd(priv, 0x00000896, 0x00000000);
+	nv_icmd(priv, 0x00000897, 0x00000000);
+	nv_icmd(priv, 0x00000898, 0x00000000);
+	nv_icmd(priv, 0x00000899, 0x00000000);
+	nv_icmd(priv, 0x0000089a, 0x00000000);
+	nv_icmd(priv, 0x0000089b, 0x00000000);
+	nv_icmd(priv, 0x0000089c, 0x00000000);
+	nv_icmd(priv, 0x0000089d, 0x00000000);
+	nv_icmd(priv, 0x0000089e, 0x00000000);
+	nv_icmd(priv, 0x0000089f, 0x00000000);
+	nv_icmd(priv, 0x000008e0, 0x00000000);
+	nv_icmd(priv, 0x000008e1, 0x00000000);
+	nv_icmd(priv, 0x000008e2, 0x00000000);
+	nv_icmd(priv, 0x000008e3, 0x00000000);
+	nv_icmd(priv, 0x000008e4, 0x00000000);
+	nv_icmd(priv, 0x000008e5, 0x00000000);
+	nv_icmd(priv, 0x000008e6, 0x00000000);
+	nv_icmd(priv, 0x000008e7, 0x00000000);
+	nv_icmd(priv, 0x000008e8, 0x00000000);
+	nv_icmd(priv, 0x000008e9, 0x00000000);
+	nv_icmd(priv, 0x000008ea, 0x00000000);
+	nv_icmd(priv, 0x000008eb, 0x00000000);
+	nv_icmd(priv, 0x000008ec, 0x00000000);
+	nv_icmd(priv, 0x000008ed, 0x00000000);
+	nv_icmd(priv, 0x000008ee, 0x00000000);
+	nv_icmd(priv, 0x000008ef, 0x00000000);
+	nv_icmd(priv, 0x000008a0, 0x00000000);
+	nv_icmd(priv, 0x000008a1, 0x00000000);
+	nv_icmd(priv, 0x000008a2, 0x00000000);
+	nv_icmd(priv, 0x000008a3, 0x00000000);
+	nv_icmd(priv, 0x000008a4, 0x00000000);
+	nv_icmd(priv, 0x000008a5, 0x00000000);
+	nv_icmd(priv, 0x000008a6, 0x00000000);
+	nv_icmd(priv, 0x000008a7, 0x00000000);
+	nv_icmd(priv, 0x000008a8, 0x00000000);
+	nv_icmd(priv, 0x000008a9, 0x00000000);
+	nv_icmd(priv, 0x000008aa, 0x00000000);
+	nv_icmd(priv, 0x000008ab, 0x00000000);
+	nv_icmd(priv, 0x000008ac, 0x00000000);
+	nv_icmd(priv, 0x000008ad, 0x00000000);
+	nv_icmd(priv, 0x000008ae, 0x00000000);
+	nv_icmd(priv, 0x000008af, 0x00000000);
+	nv_icmd(priv, 0x000008f0, 0x00000000);
+	nv_icmd(priv, 0x000008f1, 0x00000000);
+	nv_icmd(priv, 0x000008f2, 0x00000000);
+	nv_icmd(priv, 0x000008f3, 0x00000000);
+	nv_icmd(priv, 0x000008f4, 0x00000000);
+	nv_icmd(priv, 0x000008f5, 0x00000000);
+	nv_icmd(priv, 0x000008f6, 0x00000000);
+	nv_icmd(priv, 0x000008f7, 0x00000000);
+	nv_icmd(priv, 0x000008f8, 0x00000000);
+	nv_icmd(priv, 0x000008f9, 0x00000000);
+	nv_icmd(priv, 0x000008fa, 0x00000000);
+	nv_icmd(priv, 0x000008fb, 0x00000000);
+	nv_icmd(priv, 0x000008fc, 0x00000000);
+	nv_icmd(priv, 0x000008fd, 0x00000000);
+	nv_icmd(priv, 0x000008fe, 0x00000000);
+	nv_icmd(priv, 0x000008ff, 0x00000000);
+	nv_icmd(priv, 0x0000094c, 0x000000ff);
+	nv_icmd(priv, 0x0000094d, 0xffffffff);
+	nv_icmd(priv, 0x0000094e, 0x00000002);
+	nv_icmd(priv, 0x000002ec, 0x00000001);
+	nv_icmd(priv, 0x00000303, 0x00000001);
+	nv_icmd(priv, 0x000002e6, 0x00000001);
+	nv_icmd(priv, 0x00000466, 0x00000052);
+	nv_icmd(priv, 0x00000301, 0x3f800000);
+	nv_icmd(priv, 0x00000304, 0x30201000);
+	nv_icmd(priv, 0x00000305, 0x70605040);
+	nv_icmd(priv, 0x00000306, 0xb8a89888);
+	nv_icmd(priv, 0x00000307, 0xf8e8d8c8);
+	nv_icmd(priv, 0x0000030a, 0x00ffff00);
+	nv_icmd(priv, 0x0000030b, 0x0000001a);
+	nv_icmd(priv, 0x0000030c, 0x00000001);
+	nv_icmd(priv, 0x00000318, 0x00000001);
+	nv_icmd(priv, 0x00000340, 0x00000000);
+	nv_icmd(priv, 0x00000375, 0x00000001);
+	nv_icmd(priv, 0x00000351, 0x00000100);
+	nv_icmd(priv, 0x0000037d, 0x00000006);
+	nv_icmd(priv, 0x000003a0, 0x00000002);
+	nv_icmd(priv, 0x000003aa, 0x00000001);
+	nv_icmd(priv, 0x000003a9, 0x00000001);
+	nv_icmd(priv, 0x00000380, 0x00000001);
+	nv_icmd(priv, 0x00000360, 0x00000040);
+	nv_icmd(priv, 0x00000366, 0x00000000);
+	nv_icmd(priv, 0x00000367, 0x00000000);
+	nv_icmd(priv, 0x00000368, 0x00001fff);
+	nv_icmd(priv, 0x00000370, 0x00000000);
+	nv_icmd(priv, 0x00000371, 0x00000000);
+	nv_icmd(priv, 0x00000372, 0x003fffff);
+	nv_icmd(priv, 0x0000037a, 0x00000012);
+	nv_icmd(priv, 0x000005e0, 0x00000022);
+	nv_icmd(priv, 0x000005e1, 0x00000022);
+	nv_icmd(priv, 0x000005e2, 0x00000022);
+	nv_icmd(priv, 0x000005e3, 0x00000022);
+	nv_icmd(priv, 0x000005e4, 0x00000022);
+	nv_icmd(priv, 0x00000619, 0x00000003);
+	nv_icmd(priv, 0x00000811, 0x00000003);
+	nv_icmd(priv, 0x00000812, 0x00000004);
+	nv_icmd(priv, 0x00000813, 0x00000006);
+	nv_icmd(priv, 0x00000814, 0x00000008);
+	nv_icmd(priv, 0x00000815, 0x0000000b);
+	nv_icmd(priv, 0x00000800, 0x00000001);
+	nv_icmd(priv, 0x00000801, 0x00000001);
+	nv_icmd(priv, 0x00000802, 0x00000001);
+	nv_icmd(priv, 0x00000803, 0x00000001);
+	nv_icmd(priv, 0x00000804, 0x00000001);
+	nv_icmd(priv, 0x00000805, 0x00000001);
+	nv_icmd(priv, 0x00000632, 0x00000001);
+	nv_icmd(priv, 0x00000633, 0x00000002);
+	nv_icmd(priv, 0x00000634, 0x00000003);
+	nv_icmd(priv, 0x00000635, 0x00000004);
+	nv_icmd(priv, 0x00000654, 0x3f800000);
+	nv_icmd(priv, 0x00000657, 0x3f800000);
+	nv_icmd(priv, 0x00000655, 0x3f800000);
+	nv_icmd(priv, 0x00000656, 0x3f800000);
+	nv_icmd(priv, 0x000006cd, 0x3f800000);
+	nv_icmd(priv, 0x000007f5, 0x3f800000);
+	nv_icmd(priv, 0x000007dc, 0x39291909);
+	nv_icmd(priv, 0x000007dd, 0x79695949);
+	nv_icmd(priv, 0x000007de, 0xb9a99989);
+	nv_icmd(priv, 0x000007df, 0xf9e9d9c9);
+	nv_icmd(priv, 0x000007e8, 0x00003210);
+	nv_icmd(priv, 0x000007e9, 0x00007654);
+	nv_icmd(priv, 0x000007ea, 0x00000098);
+	nv_icmd(priv, 0x000007ec, 0x39291909);
+	nv_icmd(priv, 0x000007ed, 0x79695949);
+	nv_icmd(priv, 0x000007ee, 0xb9a99989);
+	nv_icmd(priv, 0x000007ef, 0xf9e9d9c9);
+	nv_icmd(priv, 0x000007f0, 0x00003210);
+	nv_icmd(priv, 0x000007f1, 0x00007654);
+	nv_icmd(priv, 0x000007f2, 0x00000098);
+	nv_icmd(priv, 0x000005a5, 0x00000001);
+	nv_icmd(priv, 0x00000980, 0x00000000);
+	nv_icmd(priv, 0x00000981, 0x00000000);
+	nv_icmd(priv, 0x00000982, 0x00000000);
+	nv_icmd(priv, 0x00000983, 0x00000000);
+	nv_icmd(priv, 0x00000984, 0x00000000);
+	nv_icmd(priv, 0x00000985, 0x00000000);
+	nv_icmd(priv, 0x00000986, 0x00000000);
+	nv_icmd(priv, 0x00000987, 0x00000000);
+	nv_icmd(priv, 0x00000988, 0x00000000);
+	nv_icmd(priv, 0x00000989, 0x00000000);
+	nv_icmd(priv, 0x0000098a, 0x00000000);
+	nv_icmd(priv, 0x0000098b, 0x00000000);
+	nv_icmd(priv, 0x0000098c, 0x00000000);
+	nv_icmd(priv, 0x0000098d, 0x00000000);
+	nv_icmd(priv, 0x0000098e, 0x00000000);
+	nv_icmd(priv, 0x0000098f, 0x00000000);
+	nv_icmd(priv, 0x00000990, 0x00000000);
+	nv_icmd(priv, 0x00000991, 0x00000000);
+	nv_icmd(priv, 0x00000992, 0x00000000);
+	nv_icmd(priv, 0x00000993, 0x00000000);
+	nv_icmd(priv, 0x00000994, 0x00000000);
+	nv_icmd(priv, 0x00000995, 0x00000000);
+	nv_icmd(priv, 0x00000996, 0x00000000);
+	nv_icmd(priv, 0x00000997, 0x00000000);
+	nv_icmd(priv, 0x00000998, 0x00000000);
+	nv_icmd(priv, 0x00000999, 0x00000000);
+	nv_icmd(priv, 0x0000099a, 0x00000000);
+	nv_icmd(priv, 0x0000099b, 0x00000000);
+	nv_icmd(priv, 0x0000099c, 0x00000000);
+	nv_icmd(priv, 0x0000099d, 0x00000000);
+	nv_icmd(priv, 0x0000099e, 0x00000000);
+	nv_icmd(priv, 0x0000099f, 0x00000000);
+	nv_icmd(priv, 0x000009a0, 0x00000000);
+	nv_icmd(priv, 0x000009a1, 0x00000000);
+	nv_icmd(priv, 0x000009a2, 0x00000000);
+	nv_icmd(priv, 0x000009a3, 0x00000000);
+	nv_icmd(priv, 0x000009a4, 0x00000000);
+	nv_icmd(priv, 0x000009a5, 0x00000000);
+	nv_icmd(priv, 0x000009a6, 0x00000000);
+	nv_icmd(priv, 0x000009a7, 0x00000000);
+	nv_icmd(priv, 0x000009a8, 0x00000000);
+	nv_icmd(priv, 0x000009a9, 0x00000000);
+	nv_icmd(priv, 0x000009aa, 0x00000000);
+	nv_icmd(priv, 0x000009ab, 0x00000000);
+	nv_icmd(priv, 0x000009ac, 0x00000000);
+	nv_icmd(priv, 0x000009ad, 0x00000000);
+	nv_icmd(priv, 0x000009ae, 0x00000000);
+	nv_icmd(priv, 0x000009af, 0x00000000);
+	nv_icmd(priv, 0x000009b0, 0x00000000);
+	nv_icmd(priv, 0x000009b1, 0x00000000);
+	nv_icmd(priv, 0x000009b2, 0x00000000);
+	nv_icmd(priv, 0x000009b3, 0x00000000);
+	nv_icmd(priv, 0x000009b4, 0x00000000);
+	nv_icmd(priv, 0x000009b5, 0x00000000);
+	nv_icmd(priv, 0x000009b6, 0x00000000);
+	nv_icmd(priv, 0x000009b7, 0x00000000);
+	nv_icmd(priv, 0x000009b8, 0x00000000);
+	nv_icmd(priv, 0x000009b9, 0x00000000);
+	nv_icmd(priv, 0x000009ba, 0x00000000);
+	nv_icmd(priv, 0x000009bb, 0x00000000);
+	nv_icmd(priv, 0x000009bc, 0x00000000);
+	nv_icmd(priv, 0x000009bd, 0x00000000);
+	nv_icmd(priv, 0x000009be, 0x00000000);
+	nv_icmd(priv, 0x000009bf, 0x00000000);
+	nv_icmd(priv, 0x000009c0, 0x00000000);
+	nv_icmd(priv, 0x000009c1, 0x00000000);
+	nv_icmd(priv, 0x000009c2, 0x00000000);
+	nv_icmd(priv, 0x000009c3, 0x00000000);
+	nv_icmd(priv, 0x000009c4, 0x00000000);
+	nv_icmd(priv, 0x000009c5, 0x00000000);
+	nv_icmd(priv, 0x000009c6, 0x00000000);
+	nv_icmd(priv, 0x000009c7, 0x00000000);
+	nv_icmd(priv, 0x000009c8, 0x00000000);
+	nv_icmd(priv, 0x000009c9, 0x00000000);
+	nv_icmd(priv, 0x000009ca, 0x00000000);
+	nv_icmd(priv, 0x000009cb, 0x00000000);
+	nv_icmd(priv, 0x000009cc, 0x00000000);
+	nv_icmd(priv, 0x000009cd, 0x00000000);
+	nv_icmd(priv, 0x000009ce, 0x00000000);
+	nv_icmd(priv, 0x000009cf, 0x00000000);
+	nv_icmd(priv, 0x000009d0, 0x00000000);
+	nv_icmd(priv, 0x000009d1, 0x00000000);
+	nv_icmd(priv, 0x000009d2, 0x00000000);
+	nv_icmd(priv, 0x000009d3, 0x00000000);
+	nv_icmd(priv, 0x000009d4, 0x00000000);
+	nv_icmd(priv, 0x000009d5, 0x00000000);
+	nv_icmd(priv, 0x000009d6, 0x00000000);
+	nv_icmd(priv, 0x000009d7, 0x00000000);
+	nv_icmd(priv, 0x000009d8, 0x00000000);
+	nv_icmd(priv, 0x000009d9, 0x00000000);
+	nv_icmd(priv, 0x000009da, 0x00000000);
+	nv_icmd(priv, 0x000009db, 0x00000000);
+	nv_icmd(priv, 0x000009dc, 0x00000000);
+	nv_icmd(priv, 0x000009dd, 0x00000000);
+	nv_icmd(priv, 0x000009de, 0x00000000);
+	nv_icmd(priv, 0x000009df, 0x00000000);
+	nv_icmd(priv, 0x000009e0, 0x00000000);
+	nv_icmd(priv, 0x000009e1, 0x00000000);
+	nv_icmd(priv, 0x000009e2, 0x00000000);
+	nv_icmd(priv, 0x000009e3, 0x00000000);
+	nv_icmd(priv, 0x000009e4, 0x00000000);
+	nv_icmd(priv, 0x000009e5, 0x00000000);
+	nv_icmd(priv, 0x000009e6, 0x00000000);
+	nv_icmd(priv, 0x000009e7, 0x00000000);
+	nv_icmd(priv, 0x000009e8, 0x00000000);
+	nv_icmd(priv, 0x000009e9, 0x00000000);
+	nv_icmd(priv, 0x000009ea, 0x00000000);
+	nv_icmd(priv, 0x000009eb, 0x00000000);
+	nv_icmd(priv, 0x000009ec, 0x00000000);
+	nv_icmd(priv, 0x000009ed, 0x00000000);
+	nv_icmd(priv, 0x000009ee, 0x00000000);
+	nv_icmd(priv, 0x000009ef, 0x00000000);
+	nv_icmd(priv, 0x000009f0, 0x00000000);
+	nv_icmd(priv, 0x000009f1, 0x00000000);
+	nv_icmd(priv, 0x000009f2, 0x00000000);
+	nv_icmd(priv, 0x000009f3, 0x00000000);
+	nv_icmd(priv, 0x000009f4, 0x00000000);
+	nv_icmd(priv, 0x000009f5, 0x00000000);
+	nv_icmd(priv, 0x000009f6, 0x00000000);
+	nv_icmd(priv, 0x000009f7, 0x00000000);
+	nv_icmd(priv, 0x000009f8, 0x00000000);
+	nv_icmd(priv, 0x000009f9, 0x00000000);
+	nv_icmd(priv, 0x000009fa, 0x00000000);
+	nv_icmd(priv, 0x000009fb, 0x00000000);
+	nv_icmd(priv, 0x000009fc, 0x00000000);
+	nv_icmd(priv, 0x000009fd, 0x00000000);
+	nv_icmd(priv, 0x000009fe, 0x00000000);
+	nv_icmd(priv, 0x000009ff, 0x00000000);
+	nv_icmd(priv, 0x00000468, 0x00000004);
+	nv_icmd(priv, 0x0000046c, 0x00000001);
+	nv_icmd(priv, 0x00000470, 0x00000000);
+	nv_icmd(priv, 0x00000471, 0x00000000);
+	nv_icmd(priv, 0x00000472, 0x00000000);
+	nv_icmd(priv, 0x00000473, 0x00000000);
+	nv_icmd(priv, 0x00000474, 0x00000000);
+	nv_icmd(priv, 0x00000475, 0x00000000);
+	nv_icmd(priv, 0x00000476, 0x00000000);
+	nv_icmd(priv, 0x00000477, 0x00000000);
+	nv_icmd(priv, 0x00000478, 0x00000000);
+	nv_icmd(priv, 0x00000479, 0x00000000);
+	nv_icmd(priv, 0x0000047a, 0x00000000);
+	nv_icmd(priv, 0x0000047b, 0x00000000);
+	nv_icmd(priv, 0x0000047c, 0x00000000);
+	nv_icmd(priv, 0x0000047d, 0x00000000);
+	nv_icmd(priv, 0x0000047e, 0x00000000);
+	nv_icmd(priv, 0x0000047f, 0x00000000);
+	nv_icmd(priv, 0x00000480, 0x00000000);
+	nv_icmd(priv, 0x00000481, 0x00000000);
+	nv_icmd(priv, 0x00000482, 0x00000000);
+	nv_icmd(priv, 0x00000483, 0x00000000);
+	nv_icmd(priv, 0x00000484, 0x00000000);
+	nv_icmd(priv, 0x00000485, 0x00000000);
+	nv_icmd(priv, 0x00000486, 0x00000000);
+	nv_icmd(priv, 0x00000487, 0x00000000);
+	nv_icmd(priv, 0x00000488, 0x00000000);
+	nv_icmd(priv, 0x00000489, 0x00000000);
+	nv_icmd(priv, 0x0000048a, 0x00000000);
+	nv_icmd(priv, 0x0000048b, 0x00000000);
+	nv_icmd(priv, 0x0000048c, 0x00000000);
+	nv_icmd(priv, 0x0000048d, 0x00000000);
+	nv_icmd(priv, 0x0000048e, 0x00000000);
+	nv_icmd(priv, 0x0000048f, 0x00000000);
+	nv_icmd(priv, 0x00000490, 0x00000000);
+	nv_icmd(priv, 0x00000491, 0x00000000);
+	nv_icmd(priv, 0x00000492, 0x00000000);
+	nv_icmd(priv, 0x00000493, 0x00000000);
+	nv_icmd(priv, 0x00000494, 0x00000000);
+	nv_icmd(priv, 0x00000495, 0x00000000);
+	nv_icmd(priv, 0x00000496, 0x00000000);
+	nv_icmd(priv, 0x00000497, 0x00000000);
+	nv_icmd(priv, 0x00000498, 0x00000000);
+	nv_icmd(priv, 0x00000499, 0x00000000);
+	nv_icmd(priv, 0x0000049a, 0x00000000);
+	nv_icmd(priv, 0x0000049b, 0x00000000);
+	nv_icmd(priv, 0x0000049c, 0x00000000);
+	nv_icmd(priv, 0x0000049d, 0x00000000);
+	nv_icmd(priv, 0x0000049e, 0x00000000);
+	nv_icmd(priv, 0x0000049f, 0x00000000);
+	nv_icmd(priv, 0x000004a0, 0x00000000);
+	nv_icmd(priv, 0x000004a1, 0x00000000);
+	nv_icmd(priv, 0x000004a2, 0x00000000);
+	nv_icmd(priv, 0x000004a3, 0x00000000);
+	nv_icmd(priv, 0x000004a4, 0x00000000);
+	nv_icmd(priv, 0x000004a5, 0x00000000);
+	nv_icmd(priv, 0x000004a6, 0x00000000);
+	nv_icmd(priv, 0x000004a7, 0x00000000);
+	nv_icmd(priv, 0x000004a8, 0x00000000);
+	nv_icmd(priv, 0x000004a9, 0x00000000);
+	nv_icmd(priv, 0x000004aa, 0x00000000);
+	nv_icmd(priv, 0x000004ab, 0x00000000);
+	nv_icmd(priv, 0x000004ac, 0x00000000);
+	nv_icmd(priv, 0x000004ad, 0x00000000);
+	nv_icmd(priv, 0x000004ae, 0x00000000);
+	nv_icmd(priv, 0x000004af, 0x00000000);
+	nv_icmd(priv, 0x000004b0, 0x00000000);
+	nv_icmd(priv, 0x000004b1, 0x00000000);
+	nv_icmd(priv, 0x000004b2, 0x00000000);
+	nv_icmd(priv, 0x000004b3, 0x00000000);
+	nv_icmd(priv, 0x000004b4, 0x00000000);
+	nv_icmd(priv, 0x000004b5, 0x00000000);
+	nv_icmd(priv, 0x000004b6, 0x00000000);
+	nv_icmd(priv, 0x000004b7, 0x00000000);
+	nv_icmd(priv, 0x000004b8, 0x00000000);
+	nv_icmd(priv, 0x000004b9, 0x00000000);
+	nv_icmd(priv, 0x000004ba, 0x00000000);
+	nv_icmd(priv, 0x000004bb, 0x00000000);
+	nv_icmd(priv, 0x000004bc, 0x00000000);
+	nv_icmd(priv, 0x000004bd, 0x00000000);
+	nv_icmd(priv, 0x000004be, 0x00000000);
+	nv_icmd(priv, 0x000004bf, 0x00000000);
+	nv_icmd(priv, 0x000004c0, 0x00000000);
+	nv_icmd(priv, 0x000004c1, 0x00000000);
+	nv_icmd(priv, 0x000004c2, 0x00000000);
+	nv_icmd(priv, 0x000004c3, 0x00000000);
+	nv_icmd(priv, 0x000004c4, 0x00000000);
+	nv_icmd(priv, 0x000004c5, 0x00000000);
+	nv_icmd(priv, 0x000004c6, 0x00000000);
+	nv_icmd(priv, 0x000004c7, 0x00000000);
+	nv_icmd(priv, 0x000004c8, 0x00000000);
+	nv_icmd(priv, 0x000004c9, 0x00000000);
+	nv_icmd(priv, 0x000004ca, 0x00000000);
+	nv_icmd(priv, 0x000004cb, 0x00000000);
+	nv_icmd(priv, 0x000004cc, 0x00000000);
+	nv_icmd(priv, 0x000004cd, 0x00000000);
+	nv_icmd(priv, 0x000004ce, 0x00000000);
+	nv_icmd(priv, 0x000004cf, 0x00000000);
+	nv_icmd(priv, 0x00000510, 0x3f800000);
+	nv_icmd(priv, 0x00000511, 0x3f800000);
+	nv_icmd(priv, 0x00000512, 0x3f800000);
+	nv_icmd(priv, 0x00000513, 0x3f800000);
+	nv_icmd(priv, 0x00000514, 0x3f800000);
+	nv_icmd(priv, 0x00000515, 0x3f800000);
+	nv_icmd(priv, 0x00000516, 0x3f800000);
+	nv_icmd(priv, 0x00000517, 0x3f800000);
+	nv_icmd(priv, 0x00000518, 0x3f800000);
+	nv_icmd(priv, 0x00000519, 0x3f800000);
+	nv_icmd(priv, 0x0000051a, 0x3f800000);
+	nv_icmd(priv, 0x0000051b, 0x3f800000);
+	nv_icmd(priv, 0x0000051c, 0x3f800000);
+	nv_icmd(priv, 0x0000051d, 0x3f800000);
+	nv_icmd(priv, 0x0000051e, 0x3f800000);
+	nv_icmd(priv, 0x0000051f, 0x3f800000);
+	nv_icmd(priv, 0x00000520, 0x000002b6);
+	nv_icmd(priv, 0x00000529, 0x00000001);
+	nv_icmd(priv, 0x00000530, 0xffff0000);
+	nv_icmd(priv, 0x00000531, 0xffff0000);
+	nv_icmd(priv, 0x00000532, 0xffff0000);
+	nv_icmd(priv, 0x00000533, 0xffff0000);
+	nv_icmd(priv, 0x00000534, 0xffff0000);
+	nv_icmd(priv, 0x00000535, 0xffff0000);
+	nv_icmd(priv, 0x00000536, 0xffff0000);
+	nv_icmd(priv, 0x00000537, 0xffff0000);
+	nv_icmd(priv, 0x00000538, 0xffff0000);
+	nv_icmd(priv, 0x00000539, 0xffff0000);
+	nv_icmd(priv, 0x0000053a, 0xffff0000);
+	nv_icmd(priv, 0x0000053b, 0xffff0000);
+	nv_icmd(priv, 0x0000053c, 0xffff0000);
+	nv_icmd(priv, 0x0000053d, 0xffff0000);
+	nv_icmd(priv, 0x0000053e, 0xffff0000);
+	nv_icmd(priv, 0x0000053f, 0xffff0000);
+	nv_icmd(priv, 0x00000585, 0x0000003f);
+	nv_icmd(priv, 0x00000576, 0x00000003);
+	if (nv_device(priv)->chipset == 0xc1 ||
+	    nv_device(priv)->chipset == 0xd9)
+		nv_icmd(priv, 0x0000057b, 0x00000059);
+	nv_icmd(priv, 0x00000586, 0x00000040);
+	nv_icmd(priv, 0x00000582, 0x00000080);
+	nv_icmd(priv, 0x00000583, 0x00000080);
+	nv_icmd(priv, 0x000005c2, 0x00000001);
+	nv_icmd(priv, 0x00000638, 0x00000001);
+	nv_icmd(priv, 0x00000639, 0x00000001);
+	nv_icmd(priv, 0x0000063a, 0x00000002);
+	nv_icmd(priv, 0x0000063b, 0x00000001);
+	nv_icmd(priv, 0x0000063c, 0x00000001);
+	nv_icmd(priv, 0x0000063d, 0x00000002);
+	nv_icmd(priv, 0x0000063e, 0x00000001);
+	nv_icmd(priv, 0x000008b8, 0x00000001);
+	nv_icmd(priv, 0x000008b9, 0x00000001);
+	nv_icmd(priv, 0x000008ba, 0x00000001);
+	nv_icmd(priv, 0x000008bb, 0x00000001);
+	nv_icmd(priv, 0x000008bc, 0x00000001);
+	nv_icmd(priv, 0x000008bd, 0x00000001);
+	nv_icmd(priv, 0x000008be, 0x00000001);
+	nv_icmd(priv, 0x000008bf, 0x00000001);
+	nv_icmd(priv, 0x00000900, 0x00000001);
+	nv_icmd(priv, 0x00000901, 0x00000001);
+	nv_icmd(priv, 0x00000902, 0x00000001);
+	nv_icmd(priv, 0x00000903, 0x00000001);
+	nv_icmd(priv, 0x00000904, 0x00000001);
+	nv_icmd(priv, 0x00000905, 0x00000001);
+	nv_icmd(priv, 0x00000906, 0x00000001);
+	nv_icmd(priv, 0x00000907, 0x00000001);
+	nv_icmd(priv, 0x00000908, 0x00000002);
+	nv_icmd(priv, 0x00000909, 0x00000002);
+	nv_icmd(priv, 0x0000090a, 0x00000002);
+	nv_icmd(priv, 0x0000090b, 0x00000002);
+	nv_icmd(priv, 0x0000090c, 0x00000002);
+	nv_icmd(priv, 0x0000090d, 0x00000002);
+	nv_icmd(priv, 0x0000090e, 0x00000002);
+	nv_icmd(priv, 0x0000090f, 0x00000002);
+	nv_icmd(priv, 0x00000910, 0x00000001);
+	nv_icmd(priv, 0x00000911, 0x00000001);
+	nv_icmd(priv, 0x00000912, 0x00000001);
+	nv_icmd(priv, 0x00000913, 0x00000001);
+	nv_icmd(priv, 0x00000914, 0x00000001);
+	nv_icmd(priv, 0x00000915, 0x00000001);
+	nv_icmd(priv, 0x00000916, 0x00000001);
+	nv_icmd(priv, 0x00000917, 0x00000001);
+	nv_icmd(priv, 0x00000918, 0x00000001);
+	nv_icmd(priv, 0x00000919, 0x00000001);
+	nv_icmd(priv, 0x0000091a, 0x00000001);
+	nv_icmd(priv, 0x0000091b, 0x00000001);
+	nv_icmd(priv, 0x0000091c, 0x00000001);
+	nv_icmd(priv, 0x0000091d, 0x00000001);
+	nv_icmd(priv, 0x0000091e, 0x00000001);
+	nv_icmd(priv, 0x0000091f, 0x00000001);
+	nv_icmd(priv, 0x00000920, 0x00000002);
+	nv_icmd(priv, 0x00000921, 0x00000002);
+	nv_icmd(priv, 0x00000922, 0x00000002);
+	nv_icmd(priv, 0x00000923, 0x00000002);
+	nv_icmd(priv, 0x00000924, 0x00000002);
+	nv_icmd(priv, 0x00000925, 0x00000002);
+	nv_icmd(priv, 0x00000926, 0x00000002);
+	nv_icmd(priv, 0x00000927, 0x00000002);
+	nv_icmd(priv, 0x00000928, 0x00000001);
+	nv_icmd(priv, 0x00000929, 0x00000001);
+	nv_icmd(priv, 0x0000092a, 0x00000001);
+	nv_icmd(priv, 0x0000092b, 0x00000001);
+	nv_icmd(priv, 0x0000092c, 0x00000001);
+	nv_icmd(priv, 0x0000092d, 0x00000001);
+	nv_icmd(priv, 0x0000092e, 0x00000001);
+	nv_icmd(priv, 0x0000092f, 0x00000001);
+	nv_icmd(priv, 0x00000648, 0x00000001);
+	nv_icmd(priv, 0x00000649, 0x00000001);
+	nv_icmd(priv, 0x0000064a, 0x00000001);
+	nv_icmd(priv, 0x0000064b, 0x00000001);
+	nv_icmd(priv, 0x0000064c, 0x00000001);
+	nv_icmd(priv, 0x0000064d, 0x00000001);
+	nv_icmd(priv, 0x0000064e, 0x00000001);
+	nv_icmd(priv, 0x0000064f, 0x00000001);
+	nv_icmd(priv, 0x00000650, 0x00000001);
+	nv_icmd(priv, 0x00000658, 0x0000000f);
+	nv_icmd(priv, 0x000007ff, 0x0000000a);
+	nv_icmd(priv, 0x0000066a, 0x40000000);
+	nv_icmd(priv, 0x0000066b, 0x10000000);
+	nv_icmd(priv, 0x0000066c, 0xffff0000);
+	nv_icmd(priv, 0x0000066d, 0xffff0000);
+	nv_icmd(priv, 0x000007af, 0x00000008);
+	nv_icmd(priv, 0x000007b0, 0x00000008);
+	nv_icmd(priv, 0x000007f6, 0x00000001);
+	nv_icmd(priv, 0x000006b2, 0x00000055);
+	nv_icmd(priv, 0x000007ad, 0x00000003);
+	nv_icmd(priv, 0x00000937, 0x00000001);
+	nv_icmd(priv, 0x00000971, 0x00000008);
+	nv_icmd(priv, 0x00000972, 0x00000040);
+	nv_icmd(priv, 0x00000973, 0x0000012c);
+	nv_icmd(priv, 0x0000097c, 0x00000040);
+	nv_icmd(priv, 0x00000979, 0x00000003);
+	nv_icmd(priv, 0x00000975, 0x00000020);
+	nv_icmd(priv, 0x00000976, 0x00000001);
+	nv_icmd(priv, 0x00000977, 0x00000020);
+	nv_icmd(priv, 0x00000978, 0x00000001);
+	nv_icmd(priv, 0x00000957, 0x00000003);
+	nv_icmd(priv, 0x0000095e, 0x20164010);
+	nv_icmd(priv, 0x0000095f, 0x00000020);
+	if (nv_device(priv)->chipset == 0xd9)
+		nv_icmd(priv, 0x0000097d, 0x00000020);
+	nv_icmd(priv, 0x00000683, 0x00000006);
+	nv_icmd(priv, 0x00000685, 0x003fffff);
+	nv_icmd(priv, 0x00000687, 0x00000c48);
+	nv_icmd(priv, 0x000006a0, 0x00000005);
+	nv_icmd(priv, 0x00000840, 0x00300008);
+	nv_icmd(priv, 0x00000841, 0x04000080);
+	nv_icmd(priv, 0x00000842, 0x00300008);
+	nv_icmd(priv, 0x00000843, 0x04000080);
+	nv_icmd(priv, 0x00000818, 0x00000000);
+	nv_icmd(priv, 0x00000819, 0x00000000);
+	nv_icmd(priv, 0x0000081a, 0x00000000);
+	nv_icmd(priv, 0x0000081b, 0x00000000);
+	nv_icmd(priv, 0x0000081c, 0x00000000);
+	nv_icmd(priv, 0x0000081d, 0x00000000);
+	nv_icmd(priv, 0x0000081e, 0x00000000);
+	nv_icmd(priv, 0x0000081f, 0x00000000);
+	nv_icmd(priv, 0x00000848, 0x00000000);
+	nv_icmd(priv, 0x00000849, 0x00000000);
+	nv_icmd(priv, 0x0000084a, 0x00000000);
+	nv_icmd(priv, 0x0000084b, 0x00000000);
+	nv_icmd(priv, 0x0000084c, 0x00000000);
+	nv_icmd(priv, 0x0000084d, 0x00000000);
+	nv_icmd(priv, 0x0000084e, 0x00000000);
+	nv_icmd(priv, 0x0000084f, 0x00000000);
+	nv_icmd(priv, 0x00000850, 0x00000000);
+	nv_icmd(priv, 0x00000851, 0x00000000);
+	nv_icmd(priv, 0x00000852, 0x00000000);
+	nv_icmd(priv, 0x00000853, 0x00000000);
+	nv_icmd(priv, 0x00000854, 0x00000000);
+	nv_icmd(priv, 0x00000855, 0x00000000);
+	nv_icmd(priv, 0x00000856, 0x00000000);
+	nv_icmd(priv, 0x00000857, 0x00000000);
+	nv_icmd(priv, 0x00000738, 0x00000000);
+	nv_icmd(priv, 0x000006aa, 0x00000001);
+	nv_icmd(priv, 0x000006ab, 0x00000002);
+	nv_icmd(priv, 0x000006ac, 0x00000080);
+	nv_icmd(priv, 0x000006ad, 0x00000100);
+	nv_icmd(priv, 0x000006ae, 0x00000100);
+	nv_icmd(priv, 0x000006b1, 0x00000011);
+	nv_icmd(priv, 0x000006bb, 0x000000cf);
+	nv_icmd(priv, 0x000006ce, 0x2a712488);
+	nv_icmd(priv, 0x00000739, 0x4085c000);
+	nv_icmd(priv, 0x0000073a, 0x00000080);
+	nv_icmd(priv, 0x00000786, 0x80000100);
+	nv_icmd(priv, 0x0000073c, 0x00010100);
+	nv_icmd(priv, 0x0000073d, 0x02800000);
+	nv_icmd(priv, 0x00000787, 0x000000cf);
+	nv_icmd(priv, 0x0000078c, 0x00000008);
+	nv_icmd(priv, 0x00000792, 0x00000001);
+	nv_icmd(priv, 0x00000794, 0x00000001);
+	nv_icmd(priv, 0x00000795, 0x00000001);
+	nv_icmd(priv, 0x00000796, 0x00000001);
+	nv_icmd(priv, 0x00000797, 0x000000cf);
+	nv_icmd(priv, 0x00000836, 0x00000001);
+	nv_icmd(priv, 0x0000079a, 0x00000002);
+	nv_icmd(priv, 0x00000833, 0x04444480);
+	nv_icmd(priv, 0x000007a1, 0x00000001);
+	nv_icmd(priv, 0x000007a3, 0x00000001);
+	nv_icmd(priv, 0x000007a4, 0x00000001);
+	nv_icmd(priv, 0x000007a5, 0x00000001);
+	nv_icmd(priv, 0x00000831, 0x00000004);
+	nv_icmd(priv, 0x0000080c, 0x00000002);
+	nv_icmd(priv, 0x0000080d, 0x00000100);
+	nv_icmd(priv, 0x0000080e, 0x00000100);
+	nv_icmd(priv, 0x0000080f, 0x00000001);
+	nv_icmd(priv, 0x00000823, 0x00000002);
+	nv_icmd(priv, 0x00000824, 0x00000100);
+	nv_icmd(priv, 0x00000825, 0x00000100);
+	nv_icmd(priv, 0x00000826, 0x00000001);
+	nv_icmd(priv, 0x0000095d, 0x00000001);
+	nv_icmd(priv, 0x0000082b, 0x00000004);
+	nv_icmd(priv, 0x00000942, 0x00010001);
+	nv_icmd(priv, 0x00000943, 0x00000001);
+	nv_icmd(priv, 0x00000944, 0x00000022);
+	nv_icmd(priv, 0x000007c5, 0x00010001);
+	nv_icmd(priv, 0x00000834, 0x00000001);
+	nv_icmd(priv, 0x000007c7, 0x00000001);
+	nv_icmd(priv, 0x0000c1b0, 0x0000000f);
+	nv_icmd(priv, 0x0000c1b1, 0x0000000f);
+	nv_icmd(priv, 0x0000c1b2, 0x0000000f);
+	nv_icmd(priv, 0x0000c1b3, 0x0000000f);
+	nv_icmd(priv, 0x0000c1b4, 0x0000000f);
+	nv_icmd(priv, 0x0000c1b5, 0x0000000f);
+	nv_icmd(priv, 0x0000c1b6, 0x0000000f);
+	nv_icmd(priv, 0x0000c1b7, 0x0000000f);
+	nv_icmd(priv, 0x0000c1b8, 0x0fac6881);
+	nv_icmd(priv, 0x0000c1b9, 0x00fac688);
+	nv_icmd(priv, 0x0001e100, 0x00000001);
+	nv_icmd(priv, 0x00001000, 0x00000002);
+	nv_icmd(priv, 0x000006aa, 0x00000001);
+	nv_icmd(priv, 0x000006ad, 0x00000100);
+	nv_icmd(priv, 0x000006ae, 0x00000100);
+	nv_icmd(priv, 0x000006b1, 0x00000011);
+	nv_icmd(priv, 0x0000078c, 0x00000008);
+	nv_icmd(priv, 0x00000792, 0x00000001);
+	nv_icmd(priv, 0x00000794, 0x00000001);
+	nv_icmd(priv, 0x00000795, 0x00000001);
+	nv_icmd(priv, 0x00000796, 0x00000001);
+	nv_icmd(priv, 0x00000797, 0x000000cf);
+	nv_icmd(priv, 0x0000079a, 0x00000002);
+	nv_icmd(priv, 0x00000833, 0x04444480);
+	nv_icmd(priv, 0x000007a1, 0x00000001);
+	nv_icmd(priv, 0x000007a3, 0x00000001);
+	nv_icmd(priv, 0x000007a4, 0x00000001);
+	nv_icmd(priv, 0x000007a5, 0x00000001);
+	nv_icmd(priv, 0x00000831, 0x00000004);
+	nv_icmd(priv, 0x0001e100, 0x00000001);
+	nv_icmd(priv, 0x00001000, 0x00000014);
+	nv_icmd(priv, 0x00000351, 0x00000100);
+	nv_icmd(priv, 0x00000957, 0x00000003);
+	nv_icmd(priv, 0x0000095d, 0x00000001);
+	nv_icmd(priv, 0x0000082b, 0x00000004);
+	nv_icmd(priv, 0x00000942, 0x00010001);
+	nv_icmd(priv, 0x00000943, 0x00000001);
+	nv_icmd(priv, 0x000007c5, 0x00010001);
+	nv_icmd(priv, 0x00000834, 0x00000001);
+	nv_icmd(priv, 0x000007c7, 0x00000001);
+	nv_icmd(priv, 0x0001e100, 0x00000001);
+	nv_icmd(priv, 0x00001000, 0x00000001);
+	nv_icmd(priv, 0x0000080c, 0x00000002);
+	nv_icmd(priv, 0x0000080d, 0x00000100);
+	nv_icmd(priv, 0x0000080e, 0x00000100);
+	nv_icmd(priv, 0x0000080f, 0x00000001);
+	nv_icmd(priv, 0x00000823, 0x00000002);
+	nv_icmd(priv, 0x00000824, 0x00000100);
+	nv_icmd(priv, 0x00000825, 0x00000100);
+	nv_icmd(priv, 0x00000826, 0x00000001);
+	nv_icmd(priv, 0x0001e100, 0x00000001);
+	nv_wr32(priv, 0x400208, 0x00000000);
+	nv_wr32(priv, 0x404154, 0x00000400);
+
+	nvc0_grctx_generate_9097(priv);
+	if (fermi >= 0x9197)
+		nvc0_grctx_generate_9197(priv);
+	if (fermi >= 0x9297)
+		nvc0_grctx_generate_9297(priv);
+	nvc0_grctx_generate_902d(priv);
+	nvc0_grctx_generate_9039(priv);
+	nvc0_grctx_generate_90c0(priv);
+
+	nv_wr32(priv, 0x000260, r000260);
+
+	return nvc0_grctx_fini(&info);
+}
diff --git a/nvkm/engine/graph/ctxnve0.c b/nvkm/engine/graph/ctxnve0.c
new file mode 100644
index 000000000..6d8c63931
--- /dev/null
+++ b/nvkm/engine/graph/ctxnve0.c
@@ -0,0 +1,2788 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "nvc0.h"
+
+static void
+nve0_grctx_generate_icmd(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x400208, 0x80000000);
+	nv_icmd(priv, 0x001000, 0x00000004);
+	nv_icmd(priv, 0x000039, 0x00000000);
+	nv_icmd(priv, 0x00003a, 0x00000000);
+	nv_icmd(priv, 0x00003b, 0x00000000);
+	nv_icmd(priv, 0x0000a9, 0x0000ffff);
+	nv_icmd(priv, 0x000038, 0x0fac6881);
+	nv_icmd(priv, 0x00003d, 0x00000001);
+	nv_icmd(priv, 0x0000e8, 0x00000400);
+	nv_icmd(priv, 0x0000e9, 0x00000400);
+	nv_icmd(priv, 0x0000ea, 0x00000400);
+	nv_icmd(priv, 0x0000eb, 0x00000400);
+	nv_icmd(priv, 0x0000ec, 0x00000400);
+	nv_icmd(priv, 0x0000ed, 0x00000400);
+	nv_icmd(priv, 0x0000ee, 0x00000400);
+	nv_icmd(priv, 0x0000ef, 0x00000400);
+	nv_icmd(priv, 0x000078, 0x00000300);
+	nv_icmd(priv, 0x000079, 0x00000300);
+	nv_icmd(priv, 0x00007a, 0x00000300);
+	nv_icmd(priv, 0x00007b, 0x00000300);
+	nv_icmd(priv, 0x00007c, 0x00000300);
+	nv_icmd(priv, 0x00007d, 0x00000300);
+	nv_icmd(priv, 0x00007e, 0x00000300);
+	nv_icmd(priv, 0x00007f, 0x00000300);
+	nv_icmd(priv, 0x000050, 0x00000011);
+	nv_icmd(priv, 0x000058, 0x00000008);
+	nv_icmd(priv, 0x000059, 0x00000008);
+	nv_icmd(priv, 0x00005a, 0x00000008);
+	nv_icmd(priv, 0x00005b, 0x00000008);
+	nv_icmd(priv, 0x00005c, 0x00000008);
+	nv_icmd(priv, 0x00005d, 0x00000008);
+	nv_icmd(priv, 0x00005e, 0x00000008);
+	nv_icmd(priv, 0x00005f, 0x00000008);
+	nv_icmd(priv, 0x000208, 0x00000001);
+	nv_icmd(priv, 0x000209, 0x00000001);
+	nv_icmd(priv, 0x00020a, 0x00000001);
+	nv_icmd(priv, 0x00020b, 0x00000001);
+	nv_icmd(priv, 0x00020c, 0x00000001);
+	nv_icmd(priv, 0x00020d, 0x00000001);
+	nv_icmd(priv, 0x00020e, 0x00000001);
+	nv_icmd(priv, 0x00020f, 0x00000001);
+	nv_icmd(priv, 0x000081, 0x00000001);
+	nv_icmd(priv, 0x000085, 0x00000004);
+	nv_icmd(priv, 0x000088, 0x00000400);
+	nv_icmd(priv, 0x000090, 0x00000300);
+	nv_icmd(priv, 0x000098, 0x00001001);
+	nv_icmd(priv, 0x0000e3, 0x00000001);
+	nv_icmd(priv, 0x0000da, 0x00000001);
+	nv_icmd(priv, 0x0000f8, 0x00000003);
+	nv_icmd(priv, 0x0000fa, 0x00000001);
+	nv_icmd(priv, 0x00009f, 0x0000ffff);
+	nv_icmd(priv, 0x0000a0, 0x0000ffff);
+	nv_icmd(priv, 0x0000a1, 0x0000ffff);
+	nv_icmd(priv, 0x0000a2, 0x0000ffff);
+	nv_icmd(priv, 0x0000b1, 0x00000001);
+	nv_icmd(priv, 0x0000ad, 0x0000013e);
+	nv_icmd(priv, 0x0000e1, 0x00000010);
+	nv_icmd(priv, 0x000290, 0x00000000);
+	nv_icmd(priv, 0x000291, 0x00000000);
+	nv_icmd(priv, 0x000292, 0x00000000);
+	nv_icmd(priv, 0x000293, 0x00000000);
+	nv_icmd(priv, 0x000294, 0x00000000);
+	nv_icmd(priv, 0x000295, 0x00000000);
+	nv_icmd(priv, 0x000296, 0x00000000);
+	nv_icmd(priv, 0x000297, 0x00000000);
+	nv_icmd(priv, 0x000298, 0x00000000);
+	nv_icmd(priv, 0x000299, 0x00000000);
+	nv_icmd(priv, 0x00029a, 0x00000000);
+	nv_icmd(priv, 0x00029b, 0x00000000);
+	nv_icmd(priv, 0x00029c, 0x00000000);
+	nv_icmd(priv, 0x00029d, 0x00000000);
+	nv_icmd(priv, 0x00029e, 0x00000000);
+	nv_icmd(priv, 0x00029f, 0x00000000);
+	nv_icmd(priv, 0x0003b0, 0x00000000);
+	nv_icmd(priv, 0x0003b1, 0x00000000);
+	nv_icmd(priv, 0x0003b2, 0x00000000);
+	nv_icmd(priv, 0x0003b3, 0x00000000);
+	nv_icmd(priv, 0x0003b4, 0x00000000);
+	nv_icmd(priv, 0x0003b5, 0x00000000);
+	nv_icmd(priv, 0x0003b6, 0x00000000);
+	nv_icmd(priv, 0x0003b7, 0x00000000);
+	nv_icmd(priv, 0x0003b8, 0x00000000);
+	nv_icmd(priv, 0x0003b9, 0x00000000);
+	nv_icmd(priv, 0x0003ba, 0x00000000);
+	nv_icmd(priv, 0x0003bb, 0x00000000);
+	nv_icmd(priv, 0x0003bc, 0x00000000);
+	nv_icmd(priv, 0x0003bd, 0x00000000);
+	nv_icmd(priv, 0x0003be, 0x00000000);
+	nv_icmd(priv, 0x0003bf, 0x00000000);
+	nv_icmd(priv, 0x0002a0, 0x00000000);
+	nv_icmd(priv, 0x0002a1, 0x00000000);
+	nv_icmd(priv, 0x0002a2, 0x00000000);
+	nv_icmd(priv, 0x0002a3, 0x00000000);
+	nv_icmd(priv, 0x0002a4, 0x00000000);
+	nv_icmd(priv, 0x0002a5, 0x00000000);
+	nv_icmd(priv, 0x0002a6, 0x00000000);
+	nv_icmd(priv, 0x0002a7, 0x00000000);
+	nv_icmd(priv, 0x0002a8, 0x00000000);
+	nv_icmd(priv, 0x0002a9, 0x00000000);
+	nv_icmd(priv, 0x0002aa, 0x00000000);
+	nv_icmd(priv, 0x0002ab, 0x00000000);
+	nv_icmd(priv, 0x0002ac, 0x00000000);
+	nv_icmd(priv, 0x0002ad, 0x00000000);
+	nv_icmd(priv, 0x0002ae, 0x00000000);
+	nv_icmd(priv, 0x0002af, 0x00000000);
+	nv_icmd(priv, 0x000420, 0x00000000);
+	nv_icmd(priv, 0x000421, 0x00000000);
+	nv_icmd(priv, 0x000422, 0x00000000);
+	nv_icmd(priv, 0x000423, 0x00000000);
+	nv_icmd(priv, 0x000424, 0x00000000);
+	nv_icmd(priv, 0x000425, 0x00000000);
+	nv_icmd(priv, 0x000426, 0x00000000);
+	nv_icmd(priv, 0x000427, 0x00000000);
+	nv_icmd(priv, 0x000428, 0x00000000);
+	nv_icmd(priv, 0x000429, 0x00000000);
+	nv_icmd(priv, 0x00042a, 0x00000000);
+	nv_icmd(priv, 0x00042b, 0x00000000);
+	nv_icmd(priv, 0x00042c, 0x00000000);
+	nv_icmd(priv, 0x00042d, 0x00000000);
+	nv_icmd(priv, 0x00042e, 0x00000000);
+	nv_icmd(priv, 0x00042f, 0x00000000);
+	nv_icmd(priv, 0x0002b0, 0x00000000);
+	nv_icmd(priv, 0x0002b1, 0x00000000);
+	nv_icmd(priv, 0x0002b2, 0x00000000);
+	nv_icmd(priv, 0x0002b3, 0x00000000);
+	nv_icmd(priv, 0x0002b4, 0x00000000);
+	nv_icmd(priv, 0x0002b5, 0x00000000);
+	nv_icmd(priv, 0x0002b6, 0x00000000);
+	nv_icmd(priv, 0x0002b7, 0x00000000);
+	nv_icmd(priv, 0x0002b8, 0x00000000);
+	nv_icmd(priv, 0x0002b9, 0x00000000);
+	nv_icmd(priv, 0x0002ba, 0x00000000);
+	nv_icmd(priv, 0x0002bb, 0x00000000);
+	nv_icmd(priv, 0x0002bc, 0x00000000);
+	nv_icmd(priv, 0x0002bd, 0x00000000);
+	nv_icmd(priv, 0x0002be, 0x00000000);
+	nv_icmd(priv, 0x0002bf, 0x00000000);
+	nv_icmd(priv, 0x000430, 0x00000000);
+	nv_icmd(priv, 0x000431, 0x00000000);
+	nv_icmd(priv, 0x000432, 0x00000000);
+	nv_icmd(priv, 0x000433, 0x00000000);
+	nv_icmd(priv, 0x000434, 0x00000000);
+	nv_icmd(priv, 0x000435, 0x00000000);
+	nv_icmd(priv, 0x000436, 0x00000000);
+	nv_icmd(priv, 0x000437, 0x00000000);
+	nv_icmd(priv, 0x000438, 0x00000000);
+	nv_icmd(priv, 0x000439, 0x00000000);
+	nv_icmd(priv, 0x00043a, 0x00000000);
+	nv_icmd(priv, 0x00043b, 0x00000000);
+	nv_icmd(priv, 0x00043c, 0x00000000);
+	nv_icmd(priv, 0x00043d, 0x00000000);
+	nv_icmd(priv, 0x00043e, 0x00000000);
+	nv_icmd(priv, 0x00043f, 0x00000000);
+	nv_icmd(priv, 0x0002c0, 0x00000000);
+	nv_icmd(priv, 0x0002c1, 0x00000000);
+	nv_icmd(priv, 0x0002c2, 0x00000000);
+	nv_icmd(priv, 0x0002c3, 0x00000000);
+	nv_icmd(priv, 0x0002c4, 0x00000000);
+	nv_icmd(priv, 0x0002c5, 0x00000000);
+	nv_icmd(priv, 0x0002c6, 0x00000000);
+	nv_icmd(priv, 0x0002c7, 0x00000000);
+	nv_icmd(priv, 0x0002c8, 0x00000000);
+	nv_icmd(priv, 0x0002c9, 0x00000000);
+	nv_icmd(priv, 0x0002ca, 0x00000000);
+	nv_icmd(priv, 0x0002cb, 0x00000000);
+	nv_icmd(priv, 0x0002cc, 0x00000000);
+	nv_icmd(priv, 0x0002cd, 0x00000000);
+	nv_icmd(priv, 0x0002ce, 0x00000000);
+	nv_icmd(priv, 0x0002cf, 0x00000000);
+	nv_icmd(priv, 0x0004d0, 0x00000000);
+	nv_icmd(priv, 0x0004d1, 0x00000000);
+	nv_icmd(priv, 0x0004d2, 0x00000000);
+	nv_icmd(priv, 0x0004d3, 0x00000000);
+	nv_icmd(priv, 0x0004d4, 0x00000000);
+	nv_icmd(priv, 0x0004d5, 0x00000000);
+	nv_icmd(priv, 0x0004d6, 0x00000000);
+	nv_icmd(priv, 0x0004d7, 0x00000000);
+	nv_icmd(priv, 0x0004d8, 0x00000000);
+	nv_icmd(priv, 0x0004d9, 0x00000000);
+	nv_icmd(priv, 0x0004da, 0x00000000);
+	nv_icmd(priv, 0x0004db, 0x00000000);
+	nv_icmd(priv, 0x0004dc, 0x00000000);
+	nv_icmd(priv, 0x0004dd, 0x00000000);
+	nv_icmd(priv, 0x0004de, 0x00000000);
+	nv_icmd(priv, 0x0004df, 0x00000000);
+	nv_icmd(priv, 0x000720, 0x00000000);
+	nv_icmd(priv, 0x000721, 0x00000000);
+	nv_icmd(priv, 0x000722, 0x00000000);
+	nv_icmd(priv, 0x000723, 0x00000000);
+	nv_icmd(priv, 0x000724, 0x00000000);
+	nv_icmd(priv, 0x000725, 0x00000000);
+	nv_icmd(priv, 0x000726, 0x00000000);
+	nv_icmd(priv, 0x000727, 0x00000000);
+	nv_icmd(priv, 0x000728, 0x00000000);
+	nv_icmd(priv, 0x000729, 0x00000000);
+	nv_icmd(priv, 0x00072a, 0x00000000);
+	nv_icmd(priv, 0x00072b, 0x00000000);
+	nv_icmd(priv, 0x00072c, 0x00000000);
+	nv_icmd(priv, 0x00072d, 0x00000000);
+	nv_icmd(priv, 0x00072e, 0x00000000);
+	nv_icmd(priv, 0x00072f, 0x00000000);
+	nv_icmd(priv, 0x0008c0, 0x00000000);
+	nv_icmd(priv, 0x0008c1, 0x00000000);
+	nv_icmd(priv, 0x0008c2, 0x00000000);
+	nv_icmd(priv, 0x0008c3, 0x00000000);
+	nv_icmd(priv, 0x0008c4, 0x00000000);
+	nv_icmd(priv, 0x0008c5, 0x00000000);
+	nv_icmd(priv, 0x0008c6, 0x00000000);
+	nv_icmd(priv, 0x0008c7, 0x00000000);
+	nv_icmd(priv, 0x0008c8, 0x00000000);
+	nv_icmd(priv, 0x0008c9, 0x00000000);
+	nv_icmd(priv, 0x0008ca, 0x00000000);
+	nv_icmd(priv, 0x0008cb, 0x00000000);
+	nv_icmd(priv, 0x0008cc, 0x00000000);
+	nv_icmd(priv, 0x0008cd, 0x00000000);
+	nv_icmd(priv, 0x0008ce, 0x00000000);
+	nv_icmd(priv, 0x0008cf, 0x00000000);
+	nv_icmd(priv, 0x000890, 0x00000000);
+	nv_icmd(priv, 0x000891, 0x00000000);
+	nv_icmd(priv, 0x000892, 0x00000000);
+	nv_icmd(priv, 0x000893, 0x00000000);
+	nv_icmd(priv, 0x000894, 0x00000000);
+	nv_icmd(priv, 0x000895, 0x00000000);
+	nv_icmd(priv, 0x000896, 0x00000000);
+	nv_icmd(priv, 0x000897, 0x00000000);
+	nv_icmd(priv, 0x000898, 0x00000000);
+	nv_icmd(priv, 0x000899, 0x00000000);
+	nv_icmd(priv, 0x00089a, 0x00000000);
+	nv_icmd(priv, 0x00089b, 0x00000000);
+	nv_icmd(priv, 0x00089c, 0x00000000);
+	nv_icmd(priv, 0x00089d, 0x00000000);
+	nv_icmd(priv, 0x00089e, 0x00000000);
+	nv_icmd(priv, 0x00089f, 0x00000000);
+	nv_icmd(priv, 0x0008e0, 0x00000000);
+	nv_icmd(priv, 0x0008e1, 0x00000000);
+	nv_icmd(priv, 0x0008e2, 0x00000000);
+	nv_icmd(priv, 0x0008e3, 0x00000000);
+	nv_icmd(priv, 0x0008e4, 0x00000000);
+	nv_icmd(priv, 0x0008e5, 0x00000000);
+	nv_icmd(priv, 0x0008e6, 0x00000000);
+	nv_icmd(priv, 0x0008e7, 0x00000000);
+	nv_icmd(priv, 0x0008e8, 0x00000000);
+	nv_icmd(priv, 0x0008e9, 0x00000000);
+	nv_icmd(priv, 0x0008ea, 0x00000000);
+	nv_icmd(priv, 0x0008eb, 0x00000000);
+	nv_icmd(priv, 0x0008ec, 0x00000000);
+	nv_icmd(priv, 0x0008ed, 0x00000000);
+	nv_icmd(priv, 0x0008ee, 0x00000000);
+	nv_icmd(priv, 0x0008ef, 0x00000000);
+	nv_icmd(priv, 0x0008a0, 0x00000000);
+	nv_icmd(priv, 0x0008a1, 0x00000000);
+	nv_icmd(priv, 0x0008a2, 0x00000000);
+	nv_icmd(priv, 0x0008a3, 0x00000000);
+	nv_icmd(priv, 0x0008a4, 0x00000000);
+	nv_icmd(priv, 0x0008a5, 0x00000000);
+	nv_icmd(priv, 0x0008a6, 0x00000000);
+	nv_icmd(priv, 0x0008a7, 0x00000000);
+	nv_icmd(priv, 0x0008a8, 0x00000000);
+	nv_icmd(priv, 0x0008a9, 0x00000000);
+	nv_icmd(priv, 0x0008aa, 0x00000000);
+	nv_icmd(priv, 0x0008ab, 0x00000000);
+	nv_icmd(priv, 0x0008ac, 0x00000000);
+	nv_icmd(priv, 0x0008ad, 0x00000000);
+	nv_icmd(priv, 0x0008ae, 0x00000000);
+	nv_icmd(priv, 0x0008af, 0x00000000);
+	nv_icmd(priv, 0x0008f0, 0x00000000);
+	nv_icmd(priv, 0x0008f1, 0x00000000);
+	nv_icmd(priv, 0x0008f2, 0x00000000);
+	nv_icmd(priv, 0x0008f3, 0x00000000);
+	nv_icmd(priv, 0x0008f4, 0x00000000);
+	nv_icmd(priv, 0x0008f5, 0x00000000);
+	nv_icmd(priv, 0x0008f6, 0x00000000);
+	nv_icmd(priv, 0x0008f7, 0x00000000);
+	nv_icmd(priv, 0x0008f8, 0x00000000);
+	nv_icmd(priv, 0x0008f9, 0x00000000);
+	nv_icmd(priv, 0x0008fa, 0x00000000);
+	nv_icmd(priv, 0x0008fb, 0x00000000);
+	nv_icmd(priv, 0x0008fc, 0x00000000);
+	nv_icmd(priv, 0x0008fd, 0x00000000);
+	nv_icmd(priv, 0x0008fe, 0x00000000);
+	nv_icmd(priv, 0x0008ff, 0x00000000);
+	nv_icmd(priv, 0x00094c, 0x000000ff);
+	nv_icmd(priv, 0x00094d, 0xffffffff);
+	nv_icmd(priv, 0x00094e, 0x00000002);
+	nv_icmd(priv, 0x0002ec, 0x00000001);
+	nv_icmd(priv, 0x000303, 0x00000001);
+	nv_icmd(priv, 0x0002e6, 0x00000001);
+	nv_icmd(priv, 0x000466, 0x00000052);
+	nv_icmd(priv, 0x000301, 0x3f800000);
+	nv_icmd(priv, 0x000304, 0x30201000);
+	nv_icmd(priv, 0x000305, 0x70605040);
+	nv_icmd(priv, 0x000306, 0xb8a89888);
+	nv_icmd(priv, 0x000307, 0xf8e8d8c8);
+	nv_icmd(priv, 0x00030a, 0x00ffff00);
+	nv_icmd(priv, 0x00030b, 0x0000001a);
+	nv_icmd(priv, 0x00030c, 0x00000001);
+	nv_icmd(priv, 0x000318, 0x00000001);
+	nv_icmd(priv, 0x000340, 0x00000000);
+	nv_icmd(priv, 0x000375, 0x00000001);
+	nv_icmd(priv, 0x00037d, 0x00000006);
+	nv_icmd(priv, 0x0003a0, 0x00000002);
+	nv_icmd(priv, 0x0003aa, 0x00000001);
+	nv_icmd(priv, 0x0003a9, 0x00000001);
+	nv_icmd(priv, 0x000380, 0x00000001);
+	nv_icmd(priv, 0x000383, 0x00000011);
+	nv_icmd(priv, 0x000360, 0x00000040);
+	nv_icmd(priv, 0x000366, 0x00000000);
+	nv_icmd(priv, 0x000367, 0x00000000);
+	nv_icmd(priv, 0x000368, 0x00000fff);
+	nv_icmd(priv, 0x000370, 0x00000000);
+	nv_icmd(priv, 0x000371, 0x00000000);
+	nv_icmd(priv, 0x000372, 0x000fffff);
+	nv_icmd(priv, 0x00037a, 0x00000012);
+	nv_icmd(priv, 0x000619, 0x00000003);
+	nv_icmd(priv, 0x000811, 0x00000003);
+	nv_icmd(priv, 0x000812, 0x00000004);
+	nv_icmd(priv, 0x000813, 0x00000006);
+	nv_icmd(priv, 0x000814, 0x00000008);
+	nv_icmd(priv, 0x000815, 0x0000000b);
+	nv_icmd(priv, 0x000800, 0x00000001);
+	nv_icmd(priv, 0x000801, 0x00000001);
+	nv_icmd(priv, 0x000802, 0x00000001);
+	nv_icmd(priv, 0x000803, 0x00000001);
+	nv_icmd(priv, 0x000804, 0x00000001);
+	nv_icmd(priv, 0x000805, 0x00000001);
+	nv_icmd(priv, 0x000632, 0x00000001);
+	nv_icmd(priv, 0x000633, 0x00000002);
+	nv_icmd(priv, 0x000634, 0x00000003);
+	nv_icmd(priv, 0x000635, 0x00000004);
+	nv_icmd(priv, 0x000654, 0x3f800000);
+	nv_icmd(priv, 0x000657, 0x3f800000);
+	nv_icmd(priv, 0x000655, 0x3f800000);
+	nv_icmd(priv, 0x000656, 0x3f800000);
+	nv_icmd(priv, 0x0006cd, 0x3f800000);
+	nv_icmd(priv, 0x0007f5, 0x3f800000);
+	nv_icmd(priv, 0x0007dc, 0x39291909);
+	nv_icmd(priv, 0x0007dd, 0x79695949);
+	nv_icmd(priv, 0x0007de, 0xb9a99989);
+	nv_icmd(priv, 0x0007df, 0xf9e9d9c9);
+	nv_icmd(priv, 0x0007e8, 0x00003210);
+	nv_icmd(priv, 0x0007e9, 0x00007654);
+	nv_icmd(priv, 0x0007ea, 0x00000098);
+	nv_icmd(priv, 0x0007ec, 0x39291909);
+	nv_icmd(priv, 0x0007ed, 0x79695949);
+	nv_icmd(priv, 0x0007ee, 0xb9a99989);
+	nv_icmd(priv, 0x0007ef, 0xf9e9d9c9);
+	nv_icmd(priv, 0x0007f0, 0x00003210);
+	nv_icmd(priv, 0x0007f1, 0x00007654);
+	nv_icmd(priv, 0x0007f2, 0x00000098);
+	nv_icmd(priv, 0x0005a5, 0x00000001);
+	nv_icmd(priv, 0x000980, 0x00000000);
+	nv_icmd(priv, 0x000981, 0x00000000);
+	nv_icmd(priv, 0x000982, 0x00000000);
+	nv_icmd(priv, 0x000983, 0x00000000);
+	nv_icmd(priv, 0x000984, 0x00000000);
+	nv_icmd(priv, 0x000985, 0x00000000);
+	nv_icmd(priv, 0x000986, 0x00000000);
+	nv_icmd(priv, 0x000987, 0x00000000);
+	nv_icmd(priv, 0x000988, 0x00000000);
+	nv_icmd(priv, 0x000989, 0x00000000);
+	nv_icmd(priv, 0x00098a, 0x00000000);
+	nv_icmd(priv, 0x00098b, 0x00000000);
+	nv_icmd(priv, 0x00098c, 0x00000000);
+	nv_icmd(priv, 0x00098d, 0x00000000);
+	nv_icmd(priv, 0x00098e, 0x00000000);
+	nv_icmd(priv, 0x00098f, 0x00000000);
+	nv_icmd(priv, 0x000990, 0x00000000);
+	nv_icmd(priv, 0x000991, 0x00000000);
+	nv_icmd(priv, 0x000992, 0x00000000);
+	nv_icmd(priv, 0x000993, 0x00000000);
+	nv_icmd(priv, 0x000994, 0x00000000);
+	nv_icmd(priv, 0x000995, 0x00000000);
+	nv_icmd(priv, 0x000996, 0x00000000);
+	nv_icmd(priv, 0x000997, 0x00000000);
+	nv_icmd(priv, 0x000998, 0x00000000);
+	nv_icmd(priv, 0x000999, 0x00000000);
+	nv_icmd(priv, 0x00099a, 0x00000000);
+	nv_icmd(priv, 0x00099b, 0x00000000);
+	nv_icmd(priv, 0x00099c, 0x00000000);
+	nv_icmd(priv, 0x00099d, 0x00000000);
+	nv_icmd(priv, 0x00099e, 0x00000000);
+	nv_icmd(priv, 0x00099f, 0x00000000);
+	nv_icmd(priv, 0x0009a0, 0x00000000);
+	nv_icmd(priv, 0x0009a1, 0x00000000);
+	nv_icmd(priv, 0x0009a2, 0x00000000);
+	nv_icmd(priv, 0x0009a3, 0x00000000);
+	nv_icmd(priv, 0x0009a4, 0x00000000);
+	nv_icmd(priv, 0x0009a5, 0x00000000);
+	nv_icmd(priv, 0x0009a6, 0x00000000);
+	nv_icmd(priv, 0x0009a7, 0x00000000);
+	nv_icmd(priv, 0x0009a8, 0x00000000);
+	nv_icmd(priv, 0x0009a9, 0x00000000);
+	nv_icmd(priv, 0x0009aa, 0x00000000);
+	nv_icmd(priv, 0x0009ab, 0x00000000);
+	nv_icmd(priv, 0x0009ac, 0x00000000);
+	nv_icmd(priv, 0x0009ad, 0x00000000);
+	nv_icmd(priv, 0x0009ae, 0x00000000);
+	nv_icmd(priv, 0x0009af, 0x00000000);
+	nv_icmd(priv, 0x0009b0, 0x00000000);
+	nv_icmd(priv, 0x0009b1, 0x00000000);
+	nv_icmd(priv, 0x0009b2, 0x00000000);
+	nv_icmd(priv, 0x0009b3, 0x00000000);
+	nv_icmd(priv, 0x0009b4, 0x00000000);
+	nv_icmd(priv, 0x0009b5, 0x00000000);
+	nv_icmd(priv, 0x0009b6, 0x00000000);
+	nv_icmd(priv, 0x0009b7, 0x00000000);
+	nv_icmd(priv, 0x0009b8, 0x00000000);
+	nv_icmd(priv, 0x0009b9, 0x00000000);
+	nv_icmd(priv, 0x0009ba, 0x00000000);
+	nv_icmd(priv, 0x0009bb, 0x00000000);
+	nv_icmd(priv, 0x0009bc, 0x00000000);
+	nv_icmd(priv, 0x0009bd, 0x00000000);
+	nv_icmd(priv, 0x0009be, 0x00000000);
+	nv_icmd(priv, 0x0009bf, 0x00000000);
+	nv_icmd(priv, 0x0009c0, 0x00000000);
+	nv_icmd(priv, 0x0009c1, 0x00000000);
+	nv_icmd(priv, 0x0009c2, 0x00000000);
+	nv_icmd(priv, 0x0009c3, 0x00000000);
+	nv_icmd(priv, 0x0009c4, 0x00000000);
+	nv_icmd(priv, 0x0009c5, 0x00000000);
+	nv_icmd(priv, 0x0009c6, 0x00000000);
+	nv_icmd(priv, 0x0009c7, 0x00000000);
+	nv_icmd(priv, 0x0009c8, 0x00000000);
+	nv_icmd(priv, 0x0009c9, 0x00000000);
+	nv_icmd(priv, 0x0009ca, 0x00000000);
+	nv_icmd(priv, 0x0009cb, 0x00000000);
+	nv_icmd(priv, 0x0009cc, 0x00000000);
+	nv_icmd(priv, 0x0009cd, 0x00000000);
+	nv_icmd(priv, 0x0009ce, 0x00000000);
+	nv_icmd(priv, 0x0009cf, 0x00000000);
+	nv_icmd(priv, 0x0009d0, 0x00000000);
+	nv_icmd(priv, 0x0009d1, 0x00000000);
+	nv_icmd(priv, 0x0009d2, 0x00000000);
+	nv_icmd(priv, 0x0009d3, 0x00000000);
+	nv_icmd(priv, 0x0009d4, 0x00000000);
+	nv_icmd(priv, 0x0009d5, 0x00000000);
+	nv_icmd(priv, 0x0009d6, 0x00000000);
+	nv_icmd(priv, 0x0009d7, 0x00000000);
+	nv_icmd(priv, 0x0009d8, 0x00000000);
+	nv_icmd(priv, 0x0009d9, 0x00000000);
+	nv_icmd(priv, 0x0009da, 0x00000000);
+	nv_icmd(priv, 0x0009db, 0x00000000);
+	nv_icmd(priv, 0x0009dc, 0x00000000);
+	nv_icmd(priv, 0x0009dd, 0x00000000);
+	nv_icmd(priv, 0x0009de, 0x00000000);
+	nv_icmd(priv, 0x0009df, 0x00000000);
+	nv_icmd(priv, 0x0009e0, 0x00000000);
+	nv_icmd(priv, 0x0009e1, 0x00000000);
+	nv_icmd(priv, 0x0009e2, 0x00000000);
+	nv_icmd(priv, 0x0009e3, 0x00000000);
+	nv_icmd(priv, 0x0009e4, 0x00000000);
+	nv_icmd(priv, 0x0009e5, 0x00000000);
+	nv_icmd(priv, 0x0009e6, 0x00000000);
+	nv_icmd(priv, 0x0009e7, 0x00000000);
+	nv_icmd(priv, 0x0009e8, 0x00000000);
+	nv_icmd(priv, 0x0009e9, 0x00000000);
+	nv_icmd(priv, 0x0009ea, 0x00000000);
+	nv_icmd(priv, 0x0009eb, 0x00000000);
+	nv_icmd(priv, 0x0009ec, 0x00000000);
+	nv_icmd(priv, 0x0009ed, 0x00000000);
+	nv_icmd(priv, 0x0009ee, 0x00000000);
+	nv_icmd(priv, 0x0009ef, 0x00000000);
+	nv_icmd(priv, 0x0009f0, 0x00000000);
+	nv_icmd(priv, 0x0009f1, 0x00000000);
+	nv_icmd(priv, 0x0009f2, 0x00000000);
+	nv_icmd(priv, 0x0009f3, 0x00000000);
+	nv_icmd(priv, 0x0009f4, 0x00000000);
+	nv_icmd(priv, 0x0009f5, 0x00000000);
+	nv_icmd(priv, 0x0009f6, 0x00000000);
+	nv_icmd(priv, 0x0009f7, 0x00000000);
+	nv_icmd(priv, 0x0009f8, 0x00000000);
+	nv_icmd(priv, 0x0009f9, 0x00000000);
+	nv_icmd(priv, 0x0009fa, 0x00000000);
+	nv_icmd(priv, 0x0009fb, 0x00000000);
+	nv_icmd(priv, 0x0009fc, 0x00000000);
+	nv_icmd(priv, 0x0009fd, 0x00000000);
+	nv_icmd(priv, 0x0009fe, 0x00000000);
+	nv_icmd(priv, 0x0009ff, 0x00000000);
+	nv_icmd(priv, 0x000468, 0x00000004);
+	nv_icmd(priv, 0x00046c, 0x00000001);
+	nv_icmd(priv, 0x000470, 0x00000000);
+	nv_icmd(priv, 0x000471, 0x00000000);
+	nv_icmd(priv, 0x000472, 0x00000000);
+	nv_icmd(priv, 0x000473, 0x00000000);
+	nv_icmd(priv, 0x000474, 0x00000000);
+	nv_icmd(priv, 0x000475, 0x00000000);
+	nv_icmd(priv, 0x000476, 0x00000000);
+	nv_icmd(priv, 0x000477, 0x00000000);
+	nv_icmd(priv, 0x000478, 0x00000000);
+	nv_icmd(priv, 0x000479, 0x00000000);
+	nv_icmd(priv, 0x00047a, 0x00000000);
+	nv_icmd(priv, 0x00047b, 0x00000000);
+	nv_icmd(priv, 0x00047c, 0x00000000);
+	nv_icmd(priv, 0x00047d, 0x00000000);
+	nv_icmd(priv, 0x00047e, 0x00000000);
+	nv_icmd(priv, 0x00047f, 0x00000000);
+	nv_icmd(priv, 0x000480, 0x00000000);
+	nv_icmd(priv, 0x000481, 0x00000000);
+	nv_icmd(priv, 0x000482, 0x00000000);
+	nv_icmd(priv, 0x000483, 0x00000000);
+	nv_icmd(priv, 0x000484, 0x00000000);
+	nv_icmd(priv, 0x000485, 0x00000000);
+	nv_icmd(priv, 0x000486, 0x00000000);
+	nv_icmd(priv, 0x000487, 0x00000000);
+	nv_icmd(priv, 0x000488, 0x00000000);
+	nv_icmd(priv, 0x000489, 0x00000000);
+	nv_icmd(priv, 0x00048a, 0x00000000);
+	nv_icmd(priv, 0x00048b, 0x00000000);
+	nv_icmd(priv, 0x00048c, 0x00000000);
+	nv_icmd(priv, 0x00048d, 0x00000000);
+	nv_icmd(priv, 0x00048e, 0x00000000);
+	nv_icmd(priv, 0x00048f, 0x00000000);
+	nv_icmd(priv, 0x000490, 0x00000000);
+	nv_icmd(priv, 0x000491, 0x00000000);
+	nv_icmd(priv, 0x000492, 0x00000000);
+	nv_icmd(priv, 0x000493, 0x00000000);
+	nv_icmd(priv, 0x000494, 0x00000000);
+	nv_icmd(priv, 0x000495, 0x00000000);
+	nv_icmd(priv, 0x000496, 0x00000000);
+	nv_icmd(priv, 0x000497, 0x00000000);
+	nv_icmd(priv, 0x000498, 0x00000000);
+	nv_icmd(priv, 0x000499, 0x00000000);
+	nv_icmd(priv, 0x00049a, 0x00000000);
+	nv_icmd(priv, 0x00049b, 0x00000000);
+	nv_icmd(priv, 0x00049c, 0x00000000);
+	nv_icmd(priv, 0x00049d, 0x00000000);
+	nv_icmd(priv, 0x00049e, 0x00000000);
+	nv_icmd(priv, 0x00049f, 0x00000000);
+	nv_icmd(priv, 0x0004a0, 0x00000000);
+	nv_icmd(priv, 0x0004a1, 0x00000000);
+	nv_icmd(priv, 0x0004a2, 0x00000000);
+	nv_icmd(priv, 0x0004a3, 0x00000000);
+	nv_icmd(priv, 0x0004a4, 0x00000000);
+	nv_icmd(priv, 0x0004a5, 0x00000000);
+	nv_icmd(priv, 0x0004a6, 0x00000000);
+	nv_icmd(priv, 0x0004a7, 0x00000000);
+	nv_icmd(priv, 0x0004a8, 0x00000000);
+	nv_icmd(priv, 0x0004a9, 0x00000000);
+	nv_icmd(priv, 0x0004aa, 0x00000000);
+	nv_icmd(priv, 0x0004ab, 0x00000000);
+	nv_icmd(priv, 0x0004ac, 0x00000000);
+	nv_icmd(priv, 0x0004ad, 0x00000000);
+	nv_icmd(priv, 0x0004ae, 0x00000000);
+	nv_icmd(priv, 0x0004af, 0x00000000);
+	nv_icmd(priv, 0x0004b0, 0x00000000);
+	nv_icmd(priv, 0x0004b1, 0x00000000);
+	nv_icmd(priv, 0x0004b2, 0x00000000);
+	nv_icmd(priv, 0x0004b3, 0x00000000);
+	nv_icmd(priv, 0x0004b4, 0x00000000);
+	nv_icmd(priv, 0x0004b5, 0x00000000);
+	nv_icmd(priv, 0x0004b6, 0x00000000);
+	nv_icmd(priv, 0x0004b7, 0x00000000);
+	nv_icmd(priv, 0x0004b8, 0x00000000);
+	nv_icmd(priv, 0x0004b9, 0x00000000);
+	nv_icmd(priv, 0x0004ba, 0x00000000);
+	nv_icmd(priv, 0x0004bb, 0x00000000);
+	nv_icmd(priv, 0x0004bc, 0x00000000);
+	nv_icmd(priv, 0x0004bd, 0x00000000);
+	nv_icmd(priv, 0x0004be, 0x00000000);
+	nv_icmd(priv, 0x0004bf, 0x00000000);
+	nv_icmd(priv, 0x0004c0, 0x00000000);
+	nv_icmd(priv, 0x0004c1, 0x00000000);
+	nv_icmd(priv, 0x0004c2, 0x00000000);
+	nv_icmd(priv, 0x0004c3, 0x00000000);
+	nv_icmd(priv, 0x0004c4, 0x00000000);
+	nv_icmd(priv, 0x0004c5, 0x00000000);
+	nv_icmd(priv, 0x0004c6, 0x00000000);
+	nv_icmd(priv, 0x0004c7, 0x00000000);
+	nv_icmd(priv, 0x0004c8, 0x00000000);
+	nv_icmd(priv, 0x0004c9, 0x00000000);
+	nv_icmd(priv, 0x0004ca, 0x00000000);
+	nv_icmd(priv, 0x0004cb, 0x00000000);
+	nv_icmd(priv, 0x0004cc, 0x00000000);
+	nv_icmd(priv, 0x0004cd, 0x00000000);
+	nv_icmd(priv, 0x0004ce, 0x00000000);
+	nv_icmd(priv, 0x0004cf, 0x00000000);
+	nv_icmd(priv, 0x000510, 0x3f800000);
+	nv_icmd(priv, 0x000511, 0x3f800000);
+	nv_icmd(priv, 0x000512, 0x3f800000);
+	nv_icmd(priv, 0x000513, 0x3f800000);
+	nv_icmd(priv, 0x000514, 0x3f800000);
+	nv_icmd(priv, 0x000515, 0x3f800000);
+	nv_icmd(priv, 0x000516, 0x3f800000);
+	nv_icmd(priv, 0x000517, 0x3f800000);
+	nv_icmd(priv, 0x000518, 0x3f800000);
+	nv_icmd(priv, 0x000519, 0x3f800000);
+	nv_icmd(priv, 0x00051a, 0x3f800000);
+	nv_icmd(priv, 0x00051b, 0x3f800000);
+	nv_icmd(priv, 0x00051c, 0x3f800000);
+	nv_icmd(priv, 0x00051d, 0x3f800000);
+	nv_icmd(priv, 0x00051e, 0x3f800000);
+	nv_icmd(priv, 0x00051f, 0x3f800000);
+	nv_icmd(priv, 0x000520, 0x000002b6);
+	nv_icmd(priv, 0x000529, 0x00000001);
+	nv_icmd(priv, 0x000530, 0xffff0000);
+	nv_icmd(priv, 0x000531, 0xffff0000);
+	nv_icmd(priv, 0x000532, 0xffff0000);
+	nv_icmd(priv, 0x000533, 0xffff0000);
+	nv_icmd(priv, 0x000534, 0xffff0000);
+	nv_icmd(priv, 0x000535, 0xffff0000);
+	nv_icmd(priv, 0x000536, 0xffff0000);
+	nv_icmd(priv, 0x000537, 0xffff0000);
+	nv_icmd(priv, 0x000538, 0xffff0000);
+	nv_icmd(priv, 0x000539, 0xffff0000);
+	nv_icmd(priv, 0x00053a, 0xffff0000);
+	nv_icmd(priv, 0x00053b, 0xffff0000);
+	nv_icmd(priv, 0x00053c, 0xffff0000);
+	nv_icmd(priv, 0x00053d, 0xffff0000);
+	nv_icmd(priv, 0x00053e, 0xffff0000);
+	nv_icmd(priv, 0x00053f, 0xffff0000);
+	nv_icmd(priv, 0x000585, 0x0000003f);
+	nv_icmd(priv, 0x000576, 0x00000003);
+	nv_icmd(priv, 0x00057b, 0x00000059);
+	nv_icmd(priv, 0x000586, 0x00000040);
+	nv_icmd(priv, 0x000582, 0x00000080);
+	nv_icmd(priv, 0x000583, 0x00000080);
+	nv_icmd(priv, 0x0005c2, 0x00000001);
+	nv_icmd(priv, 0x000638, 0x00000001);
+	nv_icmd(priv, 0x000639, 0x00000001);
+	nv_icmd(priv, 0x00063a, 0x00000002);
+	nv_icmd(priv, 0x00063b, 0x00000001);
+	nv_icmd(priv, 0x00063c, 0x00000001);
+	nv_icmd(priv, 0x00063d, 0x00000002);
+	nv_icmd(priv, 0x00063e, 0x00000001);
+	nv_icmd(priv, 0x0008b8, 0x00000001);
+	nv_icmd(priv, 0x0008b9, 0x00000001);
+	nv_icmd(priv, 0x0008ba, 0x00000001);
+	nv_icmd(priv, 0x0008bb, 0x00000001);
+	nv_icmd(priv, 0x0008bc, 0x00000001);
+	nv_icmd(priv, 0x0008bd, 0x00000001);
+	nv_icmd(priv, 0x0008be, 0x00000001);
+	nv_icmd(priv, 0x0008bf, 0x00000001);
+	nv_icmd(priv, 0x000900, 0x00000001);
+	nv_icmd(priv, 0x000901, 0x00000001);
+	nv_icmd(priv, 0x000902, 0x00000001);
+	nv_icmd(priv, 0x000903, 0x00000001);
+	nv_icmd(priv, 0x000904, 0x00000001);
+	nv_icmd(priv, 0x000905, 0x00000001);
+	nv_icmd(priv, 0x000906, 0x00000001);
+	nv_icmd(priv, 0x000907, 0x00000001);
+	nv_icmd(priv, 0x000908, 0x00000002);
+	nv_icmd(priv, 0x000909, 0x00000002);
+	nv_icmd(priv, 0x00090a, 0x00000002);
+	nv_icmd(priv, 0x00090b, 0x00000002);
+	nv_icmd(priv, 0x00090c, 0x00000002);
+	nv_icmd(priv, 0x00090d, 0x00000002);
+	nv_icmd(priv, 0x00090e, 0x00000002);
+	nv_icmd(priv, 0x00090f, 0x00000002);
+	nv_icmd(priv, 0x000910, 0x00000001);
+	nv_icmd(priv, 0x000911, 0x00000001);
+	nv_icmd(priv, 0x000912, 0x00000001);
+	nv_icmd(priv, 0x000913, 0x00000001);
+	nv_icmd(priv, 0x000914, 0x00000001);
+	nv_icmd(priv, 0x000915, 0x00000001);
+	nv_icmd(priv, 0x000916, 0x00000001);
+	nv_icmd(priv, 0x000917, 0x00000001);
+	nv_icmd(priv, 0x000918, 0x00000001);
+	nv_icmd(priv, 0x000919, 0x00000001);
+	nv_icmd(priv, 0x00091a, 0x00000001);
+	nv_icmd(priv, 0x00091b, 0x00000001);
+	nv_icmd(priv, 0x00091c, 0x00000001);
+	nv_icmd(priv, 0x00091d, 0x00000001);
+	nv_icmd(priv, 0x00091e, 0x00000001);
+	nv_icmd(priv, 0x00091f, 0x00000001);
+	nv_icmd(priv, 0x000920, 0x00000002);
+	nv_icmd(priv, 0x000921, 0x00000002);
+	nv_icmd(priv, 0x000922, 0x00000002);
+	nv_icmd(priv, 0x000923, 0x00000002);
+	nv_icmd(priv, 0x000924, 0x00000002);
+	nv_icmd(priv, 0x000925, 0x00000002);
+	nv_icmd(priv, 0x000926, 0x00000002);
+	nv_icmd(priv, 0x000927, 0x00000002);
+	nv_icmd(priv, 0x000928, 0x00000001);
+	nv_icmd(priv, 0x000929, 0x00000001);
+	nv_icmd(priv, 0x00092a, 0x00000001);
+	nv_icmd(priv, 0x00092b, 0x00000001);
+	nv_icmd(priv, 0x00092c, 0x00000001);
+	nv_icmd(priv, 0x00092d, 0x00000001);
+	nv_icmd(priv, 0x00092e, 0x00000001);
+	nv_icmd(priv, 0x00092f, 0x00000001);
+	nv_icmd(priv, 0x000648, 0x00000001);
+	nv_icmd(priv, 0x000649, 0x00000001);
+	nv_icmd(priv, 0x00064a, 0x00000001);
+	nv_icmd(priv, 0x00064b, 0x00000001);
+	nv_icmd(priv, 0x00064c, 0x00000001);
+	nv_icmd(priv, 0x00064d, 0x00000001);
+	nv_icmd(priv, 0x00064e, 0x00000001);
+	nv_icmd(priv, 0x00064f, 0x00000001);
+	nv_icmd(priv, 0x000650, 0x00000001);
+	nv_icmd(priv, 0x000658, 0x0000000f);
+	nv_icmd(priv, 0x0007ff, 0x0000000a);
+	nv_icmd(priv, 0x00066a, 0x40000000);
+	nv_icmd(priv, 0x00066b, 0x10000000);
+	nv_icmd(priv, 0x00066c, 0xffff0000);
+	nv_icmd(priv, 0x00066d, 0xffff0000);
+	nv_icmd(priv, 0x0007af, 0x00000008);
+	nv_icmd(priv, 0x0007b0, 0x00000008);
+	nv_icmd(priv, 0x0007f6, 0x00000001);
+	nv_icmd(priv, 0x0006b2, 0x00000055);
+	nv_icmd(priv, 0x0007ad, 0x00000003);
+	nv_icmd(priv, 0x000937, 0x00000001);
+	nv_icmd(priv, 0x000971, 0x00000008);
+	nv_icmd(priv, 0x000972, 0x00000040);
+	nv_icmd(priv, 0x000973, 0x0000012c);
+	nv_icmd(priv, 0x00097c, 0x00000040);
+	nv_icmd(priv, 0x000979, 0x00000003);
+	nv_icmd(priv, 0x000975, 0x00000020);
+	nv_icmd(priv, 0x000976, 0x00000001);
+	nv_icmd(priv, 0x000977, 0x00000020);
+	nv_icmd(priv, 0x000978, 0x00000001);
+	nv_icmd(priv, 0x000957, 0x00000003);
+	nv_icmd(priv, 0x00095e, 0x20164010);
+	nv_icmd(priv, 0x00095f, 0x00000020);
+	nv_icmd(priv, 0x00097d, 0x00000020);
+	nv_icmd(priv, 0x000683, 0x00000006);
+	nv_icmd(priv, 0x000685, 0x003fffff);
+	nv_icmd(priv, 0x000687, 0x003fffff);
+	nv_icmd(priv, 0x0006a0, 0x00000005);
+	nv_icmd(priv, 0x000840, 0x00400008);
+	nv_icmd(priv, 0x000841, 0x08000080);
+	nv_icmd(priv, 0x000842, 0x00400008);
+	nv_icmd(priv, 0x000843, 0x08000080);
+	nv_icmd(priv, 0x000818, 0x00000000);
+	nv_icmd(priv, 0x000819, 0x00000000);
+	nv_icmd(priv, 0x00081a, 0x00000000);
+	nv_icmd(priv, 0x00081b, 0x00000000);
+	nv_icmd(priv, 0x00081c, 0x00000000);
+	nv_icmd(priv, 0x00081d, 0x00000000);
+	nv_icmd(priv, 0x00081e, 0x00000000);
+	nv_icmd(priv, 0x00081f, 0x00000000);
+	nv_icmd(priv, 0x000848, 0x00000000);
+	nv_icmd(priv, 0x000849, 0x00000000);
+	nv_icmd(priv, 0x00084a, 0x00000000);
+	nv_icmd(priv, 0x00084b, 0x00000000);
+	nv_icmd(priv, 0x00084c, 0x00000000);
+	nv_icmd(priv, 0x00084d, 0x00000000);
+	nv_icmd(priv, 0x00084e, 0x00000000);
+	nv_icmd(priv, 0x00084f, 0x00000000);
+	nv_icmd(priv, 0x000850, 0x00000000);
+	nv_icmd(priv, 0x000851, 0x00000000);
+	nv_icmd(priv, 0x000852, 0x00000000);
+	nv_icmd(priv, 0x000853, 0x00000000);
+	nv_icmd(priv, 0x000854, 0x00000000);
+	nv_icmd(priv, 0x000855, 0x00000000);
+	nv_icmd(priv, 0x000856, 0x00000000);
+	nv_icmd(priv, 0x000857, 0x00000000);
+	nv_icmd(priv, 0x000738, 0x00000000);
+	nv_icmd(priv, 0x0006aa, 0x00000001);
+	nv_icmd(priv, 0x0006ab, 0x00000002);
+	nv_icmd(priv, 0x0006ac, 0x00000080);
+	nv_icmd(priv, 0x0006ad, 0x00000100);
+	nv_icmd(priv, 0x0006ae, 0x00000100);
+	nv_icmd(priv, 0x0006b1, 0x00000011);
+	nv_icmd(priv, 0x0006bb, 0x000000cf);
+	nv_icmd(priv, 0x0006ce, 0x2a712488);
+	nv_icmd(priv, 0x000739, 0x4085c000);
+	nv_icmd(priv, 0x00073a, 0x00000080);
+	nv_icmd(priv, 0x000786, 0x80000100);
+	nv_icmd(priv, 0x00073c, 0x00010100);
+	nv_icmd(priv, 0x00073d, 0x02800000);
+	nv_icmd(priv, 0x000787, 0x000000cf);
+	nv_icmd(priv, 0x00078c, 0x00000008);
+	nv_icmd(priv, 0x000792, 0x00000001);
+	nv_icmd(priv, 0x000794, 0x00000001);
+	nv_icmd(priv, 0x000795, 0x00000001);
+	nv_icmd(priv, 0x000796, 0x00000001);
+	nv_icmd(priv, 0x000797, 0x000000cf);
+	nv_icmd(priv, 0x000836, 0x00000001);
+	nv_icmd(priv, 0x00079a, 0x00000002);
+	nv_icmd(priv, 0x000833, 0x04444480);
+	nv_icmd(priv, 0x0007a1, 0x00000001);
+	nv_icmd(priv, 0x0007a3, 0x00000001);
+	nv_icmd(priv, 0x0007a4, 0x00000001);
+	nv_icmd(priv, 0x0007a5, 0x00000001);
+	nv_icmd(priv, 0x000831, 0x00000004);
+	nv_icmd(priv, 0x000b07, 0x00000002);
+	nv_icmd(priv, 0x000b08, 0x00000100);
+	nv_icmd(priv, 0x000b09, 0x00000100);
+	nv_icmd(priv, 0x000b0a, 0x00000001);
+	nv_icmd(priv, 0x000a04, 0x000000ff);
+	nv_icmd(priv, 0x000a0b, 0x00000040);
+	nv_icmd(priv, 0x00097f, 0x00000100);
+	nv_icmd(priv, 0x000a02, 0x00000001);
+	nv_icmd(priv, 0x000809, 0x00000007);
+	nv_icmd(priv, 0x00c221, 0x00000040);
+	nv_icmd(priv, 0x00c1b0, 0x0000000f);
+	nv_icmd(priv, 0x00c1b1, 0x0000000f);
+	nv_icmd(priv, 0x00c1b2, 0x0000000f);
+	nv_icmd(priv, 0x00c1b3, 0x0000000f);
+	nv_icmd(priv, 0x00c1b4, 0x0000000f);
+	nv_icmd(priv, 0x00c1b5, 0x0000000f);
+	nv_icmd(priv, 0x00c1b6, 0x0000000f);
+	nv_icmd(priv, 0x00c1b7, 0x0000000f);
+	nv_icmd(priv, 0x00c1b8, 0x0fac6881);
+	nv_icmd(priv, 0x00c1b9, 0x00fac688);
+	nv_icmd(priv, 0x00c401, 0x00000001);
+	nv_icmd(priv, 0x00c402, 0x00010001);
+	nv_icmd(priv, 0x00c403, 0x00000001);
+	nv_icmd(priv, 0x00c404, 0x00000001);
+	nv_icmd(priv, 0x00c40e, 0x00000020);
+	nv_icmd(priv, 0x00c500, 0x00000003);
+	nv_icmd(priv, 0x01e100, 0x00000001);
+	nv_icmd(priv, 0x001000, 0x00000002);
+	nv_icmd(priv, 0x0006aa, 0x00000001);
+	nv_icmd(priv, 0x0006ad, 0x00000100);
+	nv_icmd(priv, 0x0006ae, 0x00000100);
+	nv_icmd(priv, 0x0006b1, 0x00000011);
+	nv_icmd(priv, 0x00078c, 0x00000008);
+	nv_icmd(priv, 0x000792, 0x00000001);
+	nv_icmd(priv, 0x000794, 0x00000001);
+	nv_icmd(priv, 0x000795, 0x00000001);
+	nv_icmd(priv, 0x000796, 0x00000001);
+	nv_icmd(priv, 0x000797, 0x000000cf);
+	nv_icmd(priv, 0x00079a, 0x00000002);
+	nv_icmd(priv, 0x000833, 0x04444480);
+	nv_icmd(priv, 0x0007a1, 0x00000001);
+	nv_icmd(priv, 0x0007a3, 0x00000001);
+	nv_icmd(priv, 0x0007a4, 0x00000001);
+	nv_icmd(priv, 0x0007a5, 0x00000001);
+	nv_icmd(priv, 0x000831, 0x00000004);
+	nv_icmd(priv, 0x01e100, 0x00000001);
+	nv_icmd(priv, 0x001000, 0x00000008);
+	nv_icmd(priv, 0x000039, 0x00000000);
+	nv_icmd(priv, 0x00003a, 0x00000000);
+	nv_icmd(priv, 0x00003b, 0x00000000);
+	nv_icmd(priv, 0x000380, 0x00000001);
+	nv_icmd(priv, 0x000366, 0x00000000);
+	nv_icmd(priv, 0x000367, 0x00000000);
+	nv_icmd(priv, 0x000368, 0x00000fff);
+	nv_icmd(priv, 0x000370, 0x00000000);
+	nv_icmd(priv, 0x000371, 0x00000000);
+	nv_icmd(priv, 0x000372, 0x000fffff);
+	nv_icmd(priv, 0x000813, 0x00000006);
+	nv_icmd(priv, 0x000814, 0x00000008);
+	nv_icmd(priv, 0x000957, 0x00000003);
+	nv_icmd(priv, 0x000818, 0x00000000);
+	nv_icmd(priv, 0x000819, 0x00000000);
+	nv_icmd(priv, 0x00081a, 0x00000000);
+	nv_icmd(priv, 0x00081b, 0x00000000);
+	nv_icmd(priv, 0x00081c, 0x00000000);
+	nv_icmd(priv, 0x00081d, 0x00000000);
+	nv_icmd(priv, 0x00081e, 0x00000000);
+	nv_icmd(priv, 0x00081f, 0x00000000);
+	nv_icmd(priv, 0x000848, 0x00000000);
+	nv_icmd(priv, 0x000849, 0x00000000);
+	nv_icmd(priv, 0x00084a, 0x00000000);
+	nv_icmd(priv, 0x00084b, 0x00000000);
+	nv_icmd(priv, 0x00084c, 0x00000000);
+	nv_icmd(priv, 0x00084d, 0x00000000);
+	nv_icmd(priv, 0x00084e, 0x00000000);
+	nv_icmd(priv, 0x00084f, 0x00000000);
+	nv_icmd(priv, 0x000850, 0x00000000);
+	nv_icmd(priv, 0x000851, 0x00000000);
+	nv_icmd(priv, 0x000852, 0x00000000);
+	nv_icmd(priv, 0x000853, 0x00000000);
+	nv_icmd(priv, 0x000854, 0x00000000);
+	nv_icmd(priv, 0x000855, 0x00000000);
+	nv_icmd(priv, 0x000856, 0x00000000);
+	nv_icmd(priv, 0x000857, 0x00000000);
+	nv_icmd(priv, 0x000738, 0x00000000);
+	nv_icmd(priv, 0x000b07, 0x00000002);
+	nv_icmd(priv, 0x000b08, 0x00000100);
+	nv_icmd(priv, 0x000b09, 0x00000100);
+	nv_icmd(priv, 0x000b0a, 0x00000001);
+	nv_icmd(priv, 0x000a04, 0x000000ff);
+	nv_icmd(priv, 0x00097f, 0x00000100);
+	nv_icmd(priv, 0x000a02, 0x00000001);
+	nv_icmd(priv, 0x000809, 0x00000007);
+	nv_icmd(priv, 0x00c221, 0x00000040);
+	nv_icmd(priv, 0x00c401, 0x00000001);
+	nv_icmd(priv, 0x00c402, 0x00010001);
+	nv_icmd(priv, 0x00c403, 0x00000001);
+	nv_icmd(priv, 0x00c404, 0x00000001);
+	nv_icmd(priv, 0x00c40e, 0x00000020);
+	nv_icmd(priv, 0x00c500, 0x00000003);
+	nv_icmd(priv, 0x01e100, 0x00000001);
+	nv_icmd(priv, 0x001000, 0x00000001);
+	nv_icmd(priv, 0x000b07, 0x00000002);
+	nv_icmd(priv, 0x000b08, 0x00000100);
+	nv_icmd(priv, 0x000b09, 0x00000100);
+	nv_icmd(priv, 0x000b0a, 0x00000001);
+	nv_icmd(priv, 0x01e100, 0x00000001);
+	nv_wr32(priv, 0x400208, 0x00000000);
+}
+
+static void
+nve0_grctx_generate_a097(struct nvc0_graph_priv *priv)
+{
+	nv_mthd(priv, 0xa097, 0x0800, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0840, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0880, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x08c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0900, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0940, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0980, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x09c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0804, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0844, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0884, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x08c4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0904, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0944, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0984, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x09c4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0808, 0x00000400);
+	nv_mthd(priv, 0xa097, 0x0848, 0x00000400);
+	nv_mthd(priv, 0xa097, 0x0888, 0x00000400);
+	nv_mthd(priv, 0xa097, 0x08c8, 0x00000400);
+	nv_mthd(priv, 0xa097, 0x0908, 0x00000400);
+	nv_mthd(priv, 0xa097, 0x0948, 0x00000400);
+	nv_mthd(priv, 0xa097, 0x0988, 0x00000400);
+	nv_mthd(priv, 0xa097, 0x09c8, 0x00000400);
+	nv_mthd(priv, 0xa097, 0x080c, 0x00000300);
+	nv_mthd(priv, 0xa097, 0x084c, 0x00000300);
+	nv_mthd(priv, 0xa097, 0x088c, 0x00000300);
+	nv_mthd(priv, 0xa097, 0x08cc, 0x00000300);
+	nv_mthd(priv, 0xa097, 0x090c, 0x00000300);
+	nv_mthd(priv, 0xa097, 0x094c, 0x00000300);
+	nv_mthd(priv, 0xa097, 0x098c, 0x00000300);
+	nv_mthd(priv, 0xa097, 0x09cc, 0x00000300);
+	nv_mthd(priv, 0xa097, 0x0810, 0x000000cf);
+	nv_mthd(priv, 0xa097, 0x0850, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0890, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x08d0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0910, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0950, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0990, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x09d0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0814, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x0854, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x0894, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x08d4, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x0914, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x0954, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x0994, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x09d4, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x0818, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0858, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0898, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x08d8, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0918, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0958, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0998, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x09d8, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x081c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x085c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x089c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x08dc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x091c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x095c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x099c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x09dc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0820, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0860, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x08a0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x08e0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0920, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0960, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x09a0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x09e0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c00, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c10, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c20, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c30, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c40, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c50, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c60, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c70, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c80, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c90, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ca0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cb0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cc0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cd0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ce0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cf0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c04, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c14, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c24, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c34, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c44, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c54, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c64, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c74, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c84, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c94, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ca4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cb4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cc4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cd4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ce4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cf4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c08, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c18, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c28, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c38, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c48, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c58, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c68, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c78, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c88, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c98, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ca8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cb8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cc8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cd8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ce8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cf8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c0c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c1c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c2c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c3c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c4c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c5c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c6c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c7c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c8c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1c9c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cbc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ccc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cdc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1cfc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d00, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d10, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d20, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d30, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d40, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d50, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d60, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d70, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d80, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d90, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1da0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1db0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dc0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dd0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1de0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1df0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d04, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d14, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d24, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d34, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d44, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d54, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d64, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d74, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d84, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d94, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1da4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1db4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dc4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dd4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1de4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1df4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d08, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d18, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d28, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d38, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d48, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d58, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d68, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d78, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d88, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d98, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1da8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1db8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dc8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dd8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1de8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1df8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d0c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d1c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d2c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d3c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d4c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d5c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d6c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d7c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d8c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1d9c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dbc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dcc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ddc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1dfc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f00, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f08, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f10, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f18, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f20, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f28, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f30, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f38, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f40, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f48, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f50, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f58, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f60, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f68, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f70, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f78, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f04, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f0c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f14, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f1c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f24, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f2c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f34, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f3c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f44, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f4c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f54, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f5c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f64, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f6c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f74, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f7c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f80, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f88, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f90, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f98, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fa0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fa8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fb0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fb8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fc0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fc8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fd0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fd8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fe0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fe8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ff0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ff8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f84, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f8c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f94, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1f9c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fa4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fb4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fbc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fc4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fcc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fd4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fdc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fe4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1fec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ff4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1ffc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2000, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2040, 0x00000011);
+	nv_mthd(priv, 0xa097, 0x2080, 0x00000020);
+	nv_mthd(priv, 0xa097, 0x20c0, 0x00000030);
+	nv_mthd(priv, 0xa097, 0x2100, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x2140, 0x00000051);
+	nv_mthd(priv, 0xa097, 0x200c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x204c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x208c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x20cc, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x210c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x214c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x2010, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2050, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2090, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x20d0, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x2110, 0x00000003);
+	nv_mthd(priv, 0xa097, 0x2150, 0x00000004);
+	nv_mthd(priv, 0xa097, 0x0380, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03a0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03e0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0384, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03a4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03c4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03e4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0388, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03a8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03c8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03e8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x038c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03ac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03cc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x03ec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0700, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0710, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0720, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0730, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0704, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0714, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0724, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0734, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0708, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0718, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0728, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0738, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2800, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2804, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2808, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x280c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2810, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2814, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2818, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x281c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2820, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2824, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2828, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x282c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2830, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2834, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2838, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x283c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2840, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2844, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2848, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x284c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2850, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2854, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2858, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x285c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2860, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2864, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2868, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x286c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2870, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2874, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2878, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x287c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2880, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2884, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2888, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x288c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2890, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2894, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2898, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x289c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28a0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28a4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28a8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28ac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28b0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28b4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28b8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28bc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28c4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28c8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28cc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28d0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28d4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28d8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28dc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28e0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28e4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28e8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28ec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28f0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28f4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28f8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x28fc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2900, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2904, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2908, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x290c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2910, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2914, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2918, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x291c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2920, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2924, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2928, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x292c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2930, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2934, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2938, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x293c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2940, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2944, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2948, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x294c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2950, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2954, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2958, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x295c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2960, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2964, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2968, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x296c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2970, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2974, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2978, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x297c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2980, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2984, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2988, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x298c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2990, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2994, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2998, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x299c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29a0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29a4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29a8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29ac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29b0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29b4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29b8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29bc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29c4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29c8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29cc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29d0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29d4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29d8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29dc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29e0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29e4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29e8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29ec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29f0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29f4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29f8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x29fc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a00, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a20, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a40, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a60, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a80, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0aa0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ac0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ae0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b00, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b20, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b40, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b60, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b80, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ba0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bc0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0be0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a04, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a24, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a44, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a64, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a84, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0aa4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ac4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ae4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b04, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b24, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b44, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b64, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b84, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ba4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bc4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0be4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a08, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a28, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a48, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a68, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a88, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0aa8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ac8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ae8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b08, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b28, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b48, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b68, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b88, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ba8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bc8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0be8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a0c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a2c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a4c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a6c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a8c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0aac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0acc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0aec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b0c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b2c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b4c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b6c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b8c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bcc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a10, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a30, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a50, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a70, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a90, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ab0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ad0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0af0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b10, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b30, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b50, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b70, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b90, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bb0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bd0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bf0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a14, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a34, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a54, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a74, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0a94, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ab4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ad4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0af4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b14, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b34, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b54, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b74, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0b94, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bb4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bd4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0bf4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c00, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c10, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c20, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c30, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c40, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c50, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c60, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c70, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c80, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c90, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ca0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cb0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cc0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cd0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ce0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cf0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c04, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c14, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c24, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c34, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c44, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c54, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c64, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c74, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c84, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c94, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ca4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cb4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cc4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cd4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ce4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cf4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c08, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c18, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c28, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c38, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c48, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c58, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c68, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c78, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c88, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c98, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ca8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cb8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cc8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cd8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ce8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0cf8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0c0c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0c1c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0c2c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0c3c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0c4c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0c5c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0c6c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0c7c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0c8c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0c9c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0cac, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0cbc, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0ccc, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0cdc, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0cec, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0cfc, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0d00, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d08, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d10, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d18, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d20, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d28, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d30, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d38, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d04, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d0c, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d14, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d1c, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d24, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d2c, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d34, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d3c, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e00, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e10, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e20, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e30, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e40, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e50, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e60, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e70, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e80, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e90, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ea0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0eb0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ec0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ed0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ee0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ef0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0e04, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e14, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e24, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e34, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e44, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e54, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e64, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e74, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e84, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e94, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ea4, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0eb4, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ec4, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ed4, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ee4, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ef4, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e08, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e18, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e28, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e38, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e48, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e58, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e68, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e78, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e88, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0e98, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ea8, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0eb8, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ec8, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ed8, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ee8, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0ef8, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d40, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d48, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d50, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d58, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d44, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d4c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d54, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d5c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1e00, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e20, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e40, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e60, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e80, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ea0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ec0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ee0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e04, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e24, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e44, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e64, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e84, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ea4, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ec4, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ee4, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e08, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e28, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e48, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e68, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e88, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1ea8, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1ec8, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1ee8, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e0c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e2c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e4c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e6c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e8c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1eac, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ecc, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1eec, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e10, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e30, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e50, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e70, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e90, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1eb0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ed0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ef0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e14, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e34, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e54, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e74, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e94, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1eb4, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1ed4, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1ef4, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1e18, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e38, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e58, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e78, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1e98, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1eb8, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ed8, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1ef8, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x3400, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3404, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3408, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x340c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3410, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3414, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3418, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x341c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3420, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3424, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3428, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x342c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3430, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3434, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3438, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x343c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3440, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3444, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3448, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x344c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3450, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3454, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3458, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x345c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3460, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3464, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3468, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x346c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3470, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3474, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3478, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x347c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3480, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3484, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3488, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x348c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3490, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3494, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3498, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x349c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34a0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34a4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34a8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34ac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34b0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34b4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34b8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34bc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34c4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34c8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34cc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34d0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34d4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34d8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34dc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34e0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34e4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34e8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34ec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34f0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34f4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34f8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x34fc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3500, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3504, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3508, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x350c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3510, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3514, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3518, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x351c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3520, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3524, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3528, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x352c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3530, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3534, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3538, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x353c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3540, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3544, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3548, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x354c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3550, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3554, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3558, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x355c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3560, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3564, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3568, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x356c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3570, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3574, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3578, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x357c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3580, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3584, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3588, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x358c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3590, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3594, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x3598, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x359c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35a0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35a4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35a8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35ac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35b0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35b4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35b8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35bc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35c4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35c8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35cc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35d0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35d4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35d8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35dc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35e0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35e4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35e8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35ec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35f0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35f4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35f8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x35fc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x030c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1944, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1514, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d68, 0x0000ffff);
+	nv_mthd(priv, 0xa097, 0x121c, 0x0fac6881);
+	nv_mthd(priv, 0xa097, 0x0fac, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1538, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0fe0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0fe4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0fe8, 0x00000014);
+	nv_mthd(priv, 0xa097, 0x0fec, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x0ff0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x179c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1228, 0x00000400);
+	nv_mthd(priv, 0xa097, 0x122c, 0x00000300);
+	nv_mthd(priv, 0xa097, 0x1230, 0x00010001);
+	nv_mthd(priv, 0xa097, 0x07f8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x15b4, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x15cc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1534, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0fb0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x15d0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x153c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x16b4, 0x00000003);
+	nv_mthd(priv, 0xa097, 0x0fbc, 0x0000ffff);
+	nv_mthd(priv, 0xa097, 0x0fc0, 0x0000ffff);
+	nv_mthd(priv, 0xa097, 0x0fc4, 0x0000ffff);
+	nv_mthd(priv, 0xa097, 0x0fc8, 0x0000ffff);
+	nv_mthd(priv, 0xa097, 0x0df8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0dfc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1948, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1970, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x161c, 0x000009f0);
+	nv_mthd(priv, 0xa097, 0x0dcc, 0x00000010);
+	nv_mthd(priv, 0xa097, 0x163c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x15e4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1160, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1164, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1168, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x116c, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1170, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1174, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1178, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x117c, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1180, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1184, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1188, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x118c, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1190, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1194, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1198, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x119c, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11a0, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11a4, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11a8, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11ac, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11b0, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11b4, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11b8, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11bc, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11c0, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11c4, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11c8, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11cc, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11d0, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11d4, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11d8, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x11dc, 0x25e00040);
+	nv_mthd(priv, 0xa097, 0x1880, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1884, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1888, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x188c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1890, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1894, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1898, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x189c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18a0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18a4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18a8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18ac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18b0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18b4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18b8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18bc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18c4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18c8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18cc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18d0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18d4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18d8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18dc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18e0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18e4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18e8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18ec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18f0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18f4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18f8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x18fc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0f84, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0f88, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x17c8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x17cc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x17d0, 0x000000ff);
+	nv_mthd(priv, 0xa097, 0x17d4, 0xffffffff);
+	nv_mthd(priv, 0xa097, 0x17d8, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x17dc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x15f4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x15f8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1434, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1438, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d74, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0dec, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x13a4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1318, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1644, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0748, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0de8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1648, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x12a4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1120, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1124, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1128, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x112c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1118, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x164c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1658, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1910, 0x00000290);
+	nv_mthd(priv, 0xa097, 0x1518, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x165c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1520, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1604, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1570, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x13b0, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x13b4, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x020c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1670, 0x30201000);
+	nv_mthd(priv, 0xa097, 0x1674, 0x70605040);
+	nv_mthd(priv, 0xa097, 0x1678, 0xb8a89888);
+	nv_mthd(priv, 0xa097, 0x167c, 0xf8e8d8c8);
+	nv_mthd(priv, 0xa097, 0x166c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1680, 0x00ffff00);
+	nv_mthd(priv, 0xa097, 0x12d0, 0x00000003);
+	nv_mthd(priv, 0xa097, 0x12d4, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1684, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1688, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0dac, 0x00001b02);
+	nv_mthd(priv, 0xa097, 0x0db0, 0x00001b02);
+	nv_mthd(priv, 0xa097, 0x0db4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x168c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x15bc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x156c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x187c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1110, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0dc0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0dc4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0dc8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1234, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1690, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x12ac, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0790, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0794, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0798, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x079c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x07a0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x077c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1000, 0x00000010);
+	nv_mthd(priv, 0xa097, 0x10fc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1290, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0218, 0x00000010);
+	nv_mthd(priv, 0xa097, 0x12d8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x12dc, 0x00000010);
+	nv_mthd(priv, 0xa097, 0x0d94, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x155c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1560, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1564, 0x00000fff);
+	nv_mthd(priv, 0xa097, 0x1574, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1578, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x157c, 0x000fffff);
+	nv_mthd(priv, 0xa097, 0x1354, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1610, 0x00000012);
+	nv_mthd(priv, 0xa097, 0x1608, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x160c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x260c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x07ac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x162c, 0x00000003);
+	nv_mthd(priv, 0xa097, 0x0210, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0320, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0324, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0328, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x032c, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0330, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0334, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0338, 0x3f800000);
+	nv_mthd(priv, 0xa097, 0x0750, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0760, 0x39291909);
+	nv_mthd(priv, 0xa097, 0x0764, 0x79695949);
+	nv_mthd(priv, 0xa097, 0x0768, 0xb9a99989);
+	nv_mthd(priv, 0xa097, 0x076c, 0xf9e9d9c9);
+	nv_mthd(priv, 0xa097, 0x0770, 0x30201000);
+	nv_mthd(priv, 0xa097, 0x0774, 0x70605040);
+	nv_mthd(priv, 0xa097, 0x0778, 0x00009080);
+	nv_mthd(priv, 0xa097, 0x0780, 0x39291909);
+	nv_mthd(priv, 0xa097, 0x0784, 0x79695949);
+	nv_mthd(priv, 0xa097, 0x0788, 0xb9a99989);
+	nv_mthd(priv, 0xa097, 0x078c, 0xf9e9d9c9);
+	nv_mthd(priv, 0xa097, 0x07d0, 0x30201000);
+	nv_mthd(priv, 0xa097, 0x07d4, 0x70605040);
+	nv_mthd(priv, 0xa097, 0x07d8, 0x00009080);
+	nv_mthd(priv, 0xa097, 0x037c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0740, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0744, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x2600, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1918, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x191c, 0x00000900);
+	nv_mthd(priv, 0xa097, 0x1920, 0x00000405);
+	nv_mthd(priv, 0xa097, 0x1308, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1924, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x13ac, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x192c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x193c, 0x00002c1c);
+	nv_mthd(priv, 0xa097, 0x0d7c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0f8c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x02c0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1510, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1940, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ff4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0ff8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x194c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1950, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1968, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1590, 0x0000003f);
+	nv_mthd(priv, 0xa097, 0x07e8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x07ec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x07f0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x07f4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x196c, 0x00000011);
+	nv_mthd(priv, 0xa097, 0x02e4, 0x0000b001);
+	nv_mthd(priv, 0xa097, 0x036c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0370, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x197c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0fcc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0fd0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x02d8, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x1980, 0x00000080);
+	nv_mthd(priv, 0xa097, 0x1504, 0x00000080);
+	nv_mthd(priv, 0xa097, 0x1984, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0300, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x13a8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x12ec, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1310, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1314, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1380, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1384, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1388, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x138c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1390, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1394, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x139c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1398, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1594, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1598, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x159c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x15a0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x15a4, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0f54, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0f58, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0f5c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x19bc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0f9c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0fa0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x12cc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x12e8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x130c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1360, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1364, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1368, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x136c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1370, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1374, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1378, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x137c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x133c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1340, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1344, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1348, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x134c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1350, 0x00000002);
+	nv_mthd(priv, 0xa097, 0x1358, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x12e4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x131c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1320, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1324, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1328, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x19c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1140, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x19c4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x19c8, 0x00001500);
+	nv_mthd(priv, 0xa097, 0x135c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0f90, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x19e0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x19e4, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x19e8, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x19ec, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x19f0, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x19f4, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x19f8, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x19fc, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x19cc, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x15b8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1a00, 0x00001111);
+	nv_mthd(priv, 0xa097, 0x1a04, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1a08, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1a0c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1a10, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1a14, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1a18, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1a1c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d6c, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x0d70, 0xffff0000);
+	nv_mthd(priv, 0xa097, 0x10f8, 0x00001010);
+	nv_mthd(priv, 0xa097, 0x0d80, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d84, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d88, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d8c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0d90, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0da0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x07a4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x07a8, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1508, 0x80000000);
+	nv_mthd(priv, 0xa097, 0x150c, 0x40000000);
+	nv_mthd(priv, 0xa097, 0x1668, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0318, 0x00000008);
+	nv_mthd(priv, 0xa097, 0x031c, 0x00000008);
+	nv_mthd(priv, 0xa097, 0x0d9c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x0374, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0378, 0x00000020);
+	nv_mthd(priv, 0xa097, 0x07dc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x074c, 0x00000055);
+	nv_mthd(priv, 0xa097, 0x1420, 0x00000003);
+	nv_mthd(priv, 0xa097, 0x17bc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x17c0, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x17c4, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1008, 0x00000008);
+	nv_mthd(priv, 0xa097, 0x100c, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x1010, 0x0000012c);
+	nv_mthd(priv, 0xa097, 0x0d60, 0x00000040);
+	nv_mthd(priv, 0xa097, 0x075c, 0x00000003);
+	nv_mthd(priv, 0xa097, 0x1018, 0x00000020);
+	nv_mthd(priv, 0xa097, 0x101c, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1020, 0x00000020);
+	nv_mthd(priv, 0xa097, 0x1024, 0x00000001);
+	nv_mthd(priv, 0xa097, 0x1444, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x1448, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x144c, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0360, 0x20164010);
+	nv_mthd(priv, 0xa097, 0x0364, 0x00000020);
+	nv_mthd(priv, 0xa097, 0x0368, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0de4, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0204, 0x00000006);
+	nv_mthd(priv, 0xa097, 0x0208, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x02cc, 0x003fffff);
+	nv_mthd(priv, 0xa097, 0x02d0, 0x003fffff);
+	nv_mthd(priv, 0xa097, 0x1220, 0x00000005);
+	nv_mthd(priv, 0xa097, 0x0fdc, 0x00000000);
+	nv_mthd(priv, 0xa097, 0x0f98, 0x00400008);
+	nv_mthd(priv, 0xa097, 0x1284, 0x08000080);
+	nv_mthd(priv, 0xa097, 0x1450, 0x00400008);
+	nv_mthd(priv, 0xa097, 0x1454, 0x08000080);
+	nv_mthd(priv, 0xa097, 0x0214, 0x00000000);
+}
+
+static void
+nve0_grctx_generate_902d(struct nvc0_graph_priv *priv)
+{
+	nv_mthd(priv, 0x902d, 0x0200, 0x000000cf);
+	nv_mthd(priv, 0x902d, 0x0204, 0x00000001);
+	nv_mthd(priv, 0x902d, 0x0208, 0x00000020);
+	nv_mthd(priv, 0x902d, 0x020c, 0x00000001);
+	nv_mthd(priv, 0x902d, 0x0210, 0x00000000);
+	nv_mthd(priv, 0x902d, 0x0214, 0x00000080);
+	nv_mthd(priv, 0x902d, 0x0218, 0x00000100);
+	nv_mthd(priv, 0x902d, 0x021c, 0x00000100);
+	nv_mthd(priv, 0x902d, 0x0220, 0x00000000);
+	nv_mthd(priv, 0x902d, 0x0224, 0x00000000);
+	nv_mthd(priv, 0x902d, 0x0230, 0x000000cf);
+	nv_mthd(priv, 0x902d, 0x0234, 0x00000001);
+	nv_mthd(priv, 0x902d, 0x0238, 0x00000020);
+	nv_mthd(priv, 0x902d, 0x023c, 0x00000001);
+	nv_mthd(priv, 0x902d, 0x0244, 0x00000080);
+	nv_mthd(priv, 0x902d, 0x0248, 0x00000100);
+	nv_mthd(priv, 0x902d, 0x024c, 0x00000100);
+	nv_mthd(priv, 0x902d, 0x3410, 0x00000000);
+}
+
+static void
+nve0_graph_generate_unk40xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x404010, 0x0);
+	nv_wr32(priv, 0x404014, 0x0);
+	nv_wr32(priv, 0x404018, 0x0);
+	nv_wr32(priv, 0x40401c, 0x0);
+	nv_wr32(priv, 0x404020, 0x0);
+	nv_wr32(priv, 0x404024, 0xe000);
+	nv_wr32(priv, 0x404028, 0x0);
+	nv_wr32(priv, 0x4040a8, 0x0);
+	nv_wr32(priv, 0x4040ac, 0x0);
+	nv_wr32(priv, 0x4040b0, 0x0);
+	nv_wr32(priv, 0x4040b4, 0x0);
+	nv_wr32(priv, 0x4040b8, 0x0);
+	nv_wr32(priv, 0x4040bc, 0x0);
+	nv_wr32(priv, 0x4040c0, 0x0);
+	nv_wr32(priv, 0x4040c4, 0x0);
+	nv_wr32(priv, 0x4040c8, 0xf800008f);
+	nv_wr32(priv, 0x4040d0, 0x0);
+	nv_wr32(priv, 0x4040d4, 0x0);
+	nv_wr32(priv, 0x4040d8, 0x0);
+	nv_wr32(priv, 0x4040dc, 0x0);
+	nv_wr32(priv, 0x4040e0, 0x0);
+	nv_wr32(priv, 0x4040e4, 0x0);
+	nv_wr32(priv, 0x4040e8, 0x1000);
+	nv_wr32(priv, 0x4040f8, 0x0);
+	nv_wr32(priv, 0x404130, 0x0);
+	nv_wr32(priv, 0x404134, 0x0);
+	nv_wr32(priv, 0x404138, 0x20000040);
+	nv_wr32(priv, 0x404150, 0x2e);
+	nv_wr32(priv, 0x404154, 0x400);
+	nv_wr32(priv, 0x404158, 0x200);
+	nv_wr32(priv, 0x404164, 0x55);
+	nv_wr32(priv, 0x4041a0, 0x0);
+	nv_wr32(priv, 0x4041a4, 0x0);
+	nv_wr32(priv, 0x4041a8, 0x0);
+	nv_wr32(priv, 0x4041ac, 0x0);
+	nv_wr32(priv, 0x404200, 0x0);
+	nv_wr32(priv, 0x404204, 0x0);
+	nv_wr32(priv, 0x404208, 0x0);
+	nv_wr32(priv, 0x40420c, 0x0);
+}
+
+static void
+nve0_graph_generate_unk44xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x404404, 0x0);
+	nv_wr32(priv, 0x404408, 0x0);
+	nv_wr32(priv, 0x40440c, 0x0);
+	nv_wr32(priv, 0x404410, 0x0);
+	nv_wr32(priv, 0x404414, 0x0);
+	nv_wr32(priv, 0x404418, 0x0);
+	nv_wr32(priv, 0x40441c, 0x0);
+	nv_wr32(priv, 0x404420, 0x0);
+	nv_wr32(priv, 0x404424, 0x0);
+	nv_wr32(priv, 0x404428, 0x0);
+	nv_wr32(priv, 0x40442c, 0x0);
+	nv_wr32(priv, 0x404430, 0x0);
+	nv_wr32(priv, 0x404434, 0x0);
+	nv_wr32(priv, 0x404438, 0x0);
+	nv_wr32(priv, 0x404460, 0x0);
+	nv_wr32(priv, 0x404464, 0x0);
+	nv_wr32(priv, 0x404468, 0xffffff);
+	nv_wr32(priv, 0x40446c, 0x0);
+	nv_wr32(priv, 0x404480, 0x1);
+	nv_wr32(priv, 0x404498, 0x1);
+}
+
+static void
+nve0_graph_generate_unk46xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x404604, 0x14);
+	nv_wr32(priv, 0x404608, 0x0);
+	nv_wr32(priv, 0x40460c, 0x3fff);
+	nv_wr32(priv, 0x404610, 0x100);
+	nv_wr32(priv, 0x404618, 0x0);
+	nv_wr32(priv, 0x40461c, 0x0);
+	nv_wr32(priv, 0x404620, 0x0);
+	nv_wr32(priv, 0x404624, 0x0);
+	nv_wr32(priv, 0x40462c, 0x0);
+	nv_wr32(priv, 0x404630, 0x0);
+	nv_wr32(priv, 0x404640, 0x0);
+	nv_wr32(priv, 0x404654, 0x0);
+	nv_wr32(priv, 0x404660, 0x0);
+	nv_wr32(priv, 0x404678, 0x0);
+	nv_wr32(priv, 0x40467c, 0x2);
+	nv_wr32(priv, 0x404680, 0x0);
+	nv_wr32(priv, 0x404684, 0x0);
+	nv_wr32(priv, 0x404688, 0x0);
+	nv_wr32(priv, 0x40468c, 0x0);
+	nv_wr32(priv, 0x404690, 0x0);
+	nv_wr32(priv, 0x404694, 0x0);
+	nv_wr32(priv, 0x404698, 0x0);
+	nv_wr32(priv, 0x40469c, 0x0);
+	nv_wr32(priv, 0x4046a0, 0x7f0080);
+	nv_wr32(priv, 0x4046a4, 0x0);
+	nv_wr32(priv, 0x4046a8, 0x0);
+	nv_wr32(priv, 0x4046ac, 0x0);
+	nv_wr32(priv, 0x4046b0, 0x0);
+	nv_wr32(priv, 0x4046b4, 0x0);
+	nv_wr32(priv, 0x4046b8, 0x0);
+	nv_wr32(priv, 0x4046bc, 0x0);
+	nv_wr32(priv, 0x4046c0, 0x0);
+	nv_wr32(priv, 0x4046c8, 0x0);
+	nv_wr32(priv, 0x4046cc, 0x0);
+	nv_wr32(priv, 0x4046d0, 0x0);
+}
+
+static void
+nve0_graph_generate_unk47xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x404700, 0x0);
+	nv_wr32(priv, 0x404704, 0x0);
+	nv_wr32(priv, 0x404708, 0x0);
+	nv_wr32(priv, 0x404718, 0x0);
+	nv_wr32(priv, 0x40471c, 0x0);
+	nv_wr32(priv, 0x404720, 0x0);
+	nv_wr32(priv, 0x404724, 0x0);
+	nv_wr32(priv, 0x404728, 0x0);
+	nv_wr32(priv, 0x40472c, 0x0);
+	nv_wr32(priv, 0x404730, 0x0);
+	nv_wr32(priv, 0x404734, 0x100);
+	nv_wr32(priv, 0x404738, 0x0);
+	nv_wr32(priv, 0x40473c, 0x0);
+	nv_wr32(priv, 0x404744, 0x0);
+	nv_wr32(priv, 0x404748, 0x0);
+	nv_wr32(priv, 0x404754, 0x0);
+}
+
+static void
+nve0_graph_generate_unk58xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x405800, 0xf8000bf);
+	nv_wr32(priv, 0x405830, 0x2180648);
+	nv_wr32(priv, 0x405834, 0x8000000);
+	nv_wr32(priv, 0x405838, 0x0);
+	nv_wr32(priv, 0x405854, 0x0);
+	nv_wr32(priv, 0x405870, 0x1);
+	nv_wr32(priv, 0x405874, 0x1);
+	nv_wr32(priv, 0x405878, 0x1);
+	nv_wr32(priv, 0x40587c, 0x1);
+	nv_wr32(priv, 0x405a00, 0x0);
+	nv_wr32(priv, 0x405a04, 0x0);
+	nv_wr32(priv, 0x405a18, 0x0);
+	nv_wr32(priv, 0x405b00, 0x0);
+	nv_wr32(priv, 0x405b10, 0x1000);
+}
+
+static void
+nve0_graph_generate_unk60xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x406020, 0x4103c1);
+	nv_wr32(priv, 0x406028, 0x1);
+	nv_wr32(priv, 0x40602c, 0x1);
+	nv_wr32(priv, 0x406030, 0x1);
+	nv_wr32(priv, 0x406034, 0x1);
+}
+
+static void
+nve0_graph_generate_unk64xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x4064a8, 0x0);
+	nv_wr32(priv, 0x4064ac, 0x3fff);
+	nv_wr32(priv, 0x4064b4, 0x0);
+	nv_wr32(priv, 0x4064b8, 0x0);
+	nv_wr32(priv, 0x4064c0, 0x801a00f0);
+	nv_wr32(priv, 0x4064c4, 0x192ffff);
+	nv_wr32(priv, 0x4064c8, 0x1800600);
+	nv_wr32(priv, 0x4064cc, 0x0);
+	nv_wr32(priv, 0x4064d0, 0x0);
+	nv_wr32(priv, 0x4064d4, 0x0);
+	nv_wr32(priv, 0x4064d8, 0x0);
+	nv_wr32(priv, 0x4064dc, 0x0);
+	nv_wr32(priv, 0x4064e0, 0x0);
+	nv_wr32(priv, 0x4064e4, 0x0);
+	nv_wr32(priv, 0x4064e8, 0x0);
+	nv_wr32(priv, 0x4064ec, 0x0);
+	nv_wr32(priv, 0x4064fc, 0x22a);
+}
+
+static void
+nve0_graph_generate_unk70xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x407040, 0x0);
+}
+
+static void
+nve0_graph_generate_unk78xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x407804, 0x23);
+	nv_wr32(priv, 0x40780c, 0xa418820);
+	nv_wr32(priv, 0x407810, 0x62080e6);
+	nv_wr32(priv, 0x407814, 0x20398a4);
+	nv_wr32(priv, 0x407818, 0xe629062);
+	nv_wr32(priv, 0x40781c, 0xa418820);
+	nv_wr32(priv, 0x407820, 0xe6);
+	nv_wr32(priv, 0x4078bc, 0x103);
+}
+
+static void
+nve0_graph_generate_unk80xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x408000, 0x0);
+	nv_wr32(priv, 0x408004, 0x0);
+	nv_wr32(priv, 0x408008, 0x30);
+	nv_wr32(priv, 0x40800c, 0x0);
+	nv_wr32(priv, 0x408010, 0x0);
+	nv_wr32(priv, 0x408014, 0x69);
+	nv_wr32(priv, 0x408018, 0xe100e100);
+	nv_wr32(priv, 0x408064, 0x0);
+}
+
+static void
+nve0_graph_generate_unk88xx(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x408800, 0x2802a3c);
+	nv_wr32(priv, 0x408804, 0x40);
+	nv_wr32(priv, 0x408808, 0x1043e005);
+	nv_wr32(priv, 0x408840, 0xb);
+	nv_wr32(priv, 0x408900, 0x3080b801);
+	nv_wr32(priv, 0x408904, 0x62000001);
+	nv_wr32(priv, 0x408908, 0xc8102f);
+	nv_wr32(priv, 0x408980, 0x11d);
+}
+
+static void
+nve0_graph_generate_gpc(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x418380, 0x16);
+	nv_wr32(priv, 0x418400, 0x38004e00);
+	nv_wr32(priv, 0x418404, 0x71e0ffff);
+	nv_wr32(priv, 0x41840c, 0x1008);
+	nv_wr32(priv, 0x418410, 0xfff0fff);
+	nv_wr32(priv, 0x418414, 0x2200fff);
+	nv_wr32(priv, 0x418450, 0x0);
+	nv_wr32(priv, 0x418454, 0x0);
+	nv_wr32(priv, 0x418458, 0x0);
+	nv_wr32(priv, 0x41845c, 0x0);
+	nv_wr32(priv, 0x418460, 0x0);
+	nv_wr32(priv, 0x418464, 0x0);
+	nv_wr32(priv, 0x418468, 0x1);
+	nv_wr32(priv, 0x41846c, 0x0);
+	nv_wr32(priv, 0x418470, 0x0);
+	nv_wr32(priv, 0x418600, 0x1f);
+	nv_wr32(priv, 0x418684, 0xf);
+	nv_wr32(priv, 0x418700, 0x2);
+	nv_wr32(priv, 0x418704, 0x80);
+	nv_wr32(priv, 0x418708, 0x0);
+	nv_wr32(priv, 0x41870c, 0x0);
+	nv_wr32(priv, 0x418710, 0x0);
+	nv_wr32(priv, 0x418800, 0x7006860a);
+	nv_wr32(priv, 0x418808, 0x0);
+	nv_wr32(priv, 0x41880c, 0x0);
+	nv_wr32(priv, 0x418810, 0x0);
+	nv_wr32(priv, 0x418828, 0x44);
+	nv_wr32(priv, 0x418830, 0x10000001);
+	nv_wr32(priv, 0x4188d8, 0x8);
+	nv_wr32(priv, 0x4188e0, 0x1000000);
+	nv_wr32(priv, 0x4188e8, 0x0);
+	nv_wr32(priv, 0x4188ec, 0x0);
+	nv_wr32(priv, 0x4188f0, 0x0);
+	nv_wr32(priv, 0x4188f4, 0x0);
+	nv_wr32(priv, 0x4188f8, 0x0);
+	nv_wr32(priv, 0x4188fc, 0x20100018);
+	nv_wr32(priv, 0x41891c, 0xff00ff);
+	nv_wr32(priv, 0x418924, 0x0);
+	nv_wr32(priv, 0x418928, 0xffff00);
+	nv_wr32(priv, 0x41892c, 0xff00);
+	nv_wr32(priv, 0x418a00, 0x0);
+	nv_wr32(priv, 0x418a04, 0x0);
+	nv_wr32(priv, 0x418a08, 0x0);
+	nv_wr32(priv, 0x418a0c, 0x10000);
+	nv_wr32(priv, 0x418a10, 0x0);
+	nv_wr32(priv, 0x418a14, 0x0);
+	nv_wr32(priv, 0x418a18, 0x0);
+	nv_wr32(priv, 0x418a20, 0x0);
+	nv_wr32(priv, 0x418a24, 0x0);
+	nv_wr32(priv, 0x418a28, 0x0);
+	nv_wr32(priv, 0x418a2c, 0x10000);
+	nv_wr32(priv, 0x418a30, 0x0);
+	nv_wr32(priv, 0x418a34, 0x0);
+	nv_wr32(priv, 0x418a38, 0x0);
+	nv_wr32(priv, 0x418a40, 0x0);
+	nv_wr32(priv, 0x418a44, 0x0);
+	nv_wr32(priv, 0x418a48, 0x0);
+	nv_wr32(priv, 0x418a4c, 0x10000);
+	nv_wr32(priv, 0x418a50, 0x0);
+	nv_wr32(priv, 0x418a54, 0x0);
+	nv_wr32(priv, 0x418a58, 0x0);
+	nv_wr32(priv, 0x418a60, 0x0);
+	nv_wr32(priv, 0x418a64, 0x0);
+	nv_wr32(priv, 0x418a68, 0x0);
+	nv_wr32(priv, 0x418a6c, 0x10000);
+	nv_wr32(priv, 0x418a70, 0x0);
+	nv_wr32(priv, 0x418a74, 0x0);
+	nv_wr32(priv, 0x418a78, 0x0);
+	nv_wr32(priv, 0x418a80, 0x0);
+	nv_wr32(priv, 0x418a84, 0x0);
+	nv_wr32(priv, 0x418a88, 0x0);
+	nv_wr32(priv, 0x418a8c, 0x10000);
+	nv_wr32(priv, 0x418a90, 0x0);
+	nv_wr32(priv, 0x418a94, 0x0);
+	nv_wr32(priv, 0x418a98, 0x0);
+	nv_wr32(priv, 0x418aa0, 0x0);
+	nv_wr32(priv, 0x418aa4, 0x0);
+	nv_wr32(priv, 0x418aa8, 0x0);
+	nv_wr32(priv, 0x418aac, 0x10000);
+	nv_wr32(priv, 0x418ab0, 0x0);
+	nv_wr32(priv, 0x418ab4, 0x0);
+	nv_wr32(priv, 0x418ab8, 0x0);
+	nv_wr32(priv, 0x418ac0, 0x0);
+	nv_wr32(priv, 0x418ac4, 0x0);
+	nv_wr32(priv, 0x418ac8, 0x0);
+	nv_wr32(priv, 0x418acc, 0x10000);
+	nv_wr32(priv, 0x418ad0, 0x0);
+	nv_wr32(priv, 0x418ad4, 0x0);
+	nv_wr32(priv, 0x418ad8, 0x0);
+	nv_wr32(priv, 0x418ae0, 0x0);
+	nv_wr32(priv, 0x418ae4, 0x0);
+	nv_wr32(priv, 0x418ae8, 0x0);
+	nv_wr32(priv, 0x418aec, 0x10000);
+	nv_wr32(priv, 0x418af0, 0x0);
+	nv_wr32(priv, 0x418af4, 0x0);
+	nv_wr32(priv, 0x418af8, 0x0);
+	nv_wr32(priv, 0x418b00, 0x6);
+	nv_wr32(priv, 0x418b08, 0xa418820);
+	nv_wr32(priv, 0x418b0c, 0x62080e6);
+	nv_wr32(priv, 0x418b10, 0x20398a4);
+	nv_wr32(priv, 0x418b14, 0xe629062);
+	nv_wr32(priv, 0x418b18, 0xa418820);
+	nv_wr32(priv, 0x418b1c, 0xe6);
+	nv_wr32(priv, 0x418bb8, 0x103);
+	nv_wr32(priv, 0x418c08, 0x1);
+	nv_wr32(priv, 0x418c10, 0x0);
+	nv_wr32(priv, 0x418c14, 0x0);
+	nv_wr32(priv, 0x418c18, 0x0);
+	nv_wr32(priv, 0x418c1c, 0x0);
+	nv_wr32(priv, 0x418c20, 0x0);
+	nv_wr32(priv, 0x418c24, 0x0);
+	nv_wr32(priv, 0x418c28, 0x0);
+	nv_wr32(priv, 0x418c2c, 0x0);
+	nv_wr32(priv, 0x418c40, 0xffffffff);
+	nv_wr32(priv, 0x418c6c, 0x1);
+	nv_wr32(priv, 0x418c80, 0x20200004);
+	nv_wr32(priv, 0x418c8c, 0x1);
+	nv_wr32(priv, 0x419000, 0x780);
+	nv_wr32(priv, 0x419004, 0x0);
+	nv_wr32(priv, 0x419008, 0x0);
+	nv_wr32(priv, 0x419014, 0x4);
+}
+
+static void
+nve0_graph_generate_tpc(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x419848, 0x0);
+	nv_wr32(priv, 0x419864, 0x129);
+	nv_wr32(priv, 0x419888, 0x0);
+	nv_wr32(priv, 0x419a00, 0xf0);
+	nv_wr32(priv, 0x419a04, 0x1);
+	nv_wr32(priv, 0x419a08, 0x21);
+	nv_wr32(priv, 0x419a0c, 0x20000);
+	nv_wr32(priv, 0x419a10, 0x0);
+	nv_wr32(priv, 0x419a14, 0x200);
+	nv_wr32(priv, 0x419a1c, 0xc000);
+	nv_wr32(priv, 0x419a20, 0x800);
+	nv_wr32(priv, 0x419a30, 0x1);
+	nv_wr32(priv, 0x419ac4, 0x37f440);
+	nv_wr32(priv, 0x419c00, 0xa);
+	nv_wr32(priv, 0x419c04, 0x80000006);
+	nv_wr32(priv, 0x419c08, 0x2);
+	nv_wr32(priv, 0x419c20, 0x0);
+	nv_wr32(priv, 0x419c24, 0x84210);
+	nv_wr32(priv, 0x419c28, 0x3efbefbe);
+	nv_wr32(priv, 0x419ce8, 0x0);
+	nv_wr32(priv, 0x419cf4, 0x3203);
+	nv_wr32(priv, 0x419e04, 0x0);
+	nv_wr32(priv, 0x419e08, 0x0);
+	nv_wr32(priv, 0x419e0c, 0x0);
+	nv_wr32(priv, 0x419e10, 0x402);
+	nv_wr32(priv, 0x419e44, 0x13eff2);
+	nv_wr32(priv, 0x419e48, 0x0);
+	nv_wr32(priv, 0x419e4c, 0x7f);
+	nv_wr32(priv, 0x419e50, 0x0);
+	nv_wr32(priv, 0x419e54, 0x0);
+	nv_wr32(priv, 0x419e58, 0x0);
+	nv_wr32(priv, 0x419e5c, 0x0);
+	nv_wr32(priv, 0x419e60, 0x0);
+	nv_wr32(priv, 0x419e64, 0x0);
+	nv_wr32(priv, 0x419e68, 0x0);
+	nv_wr32(priv, 0x419e6c, 0x0);
+	nv_wr32(priv, 0x419e70, 0x0);
+	nv_wr32(priv, 0x419e74, 0x0);
+	nv_wr32(priv, 0x419e78, 0x0);
+	nv_wr32(priv, 0x419e7c, 0x0);
+	nv_wr32(priv, 0x419e80, 0x0);
+	nv_wr32(priv, 0x419e84, 0x0);
+	nv_wr32(priv, 0x419e88, 0x0);
+	nv_wr32(priv, 0x419e8c, 0x0);
+	nv_wr32(priv, 0x419e90, 0x0);
+	nv_wr32(priv, 0x419e94, 0x0);
+	nv_wr32(priv, 0x419e98, 0x0);
+	nv_wr32(priv, 0x419eac, 0x1fcf);
+	nv_wr32(priv, 0x419eb0, 0xd3f);
+	nv_wr32(priv, 0x419ec8, 0x1304f);
+	nv_wr32(priv, 0x419f30, 0x0);
+	nv_wr32(priv, 0x419f34, 0x0);
+	nv_wr32(priv, 0x419f38, 0x0);
+	nv_wr32(priv, 0x419f3c, 0x0);
+	nv_wr32(priv, 0x419f40, 0x0);
+	nv_wr32(priv, 0x419f44, 0x0);
+	nv_wr32(priv, 0x419f48, 0x0);
+	nv_wr32(priv, 0x419f4c, 0x0);
+	nv_wr32(priv, 0x419f58, 0x0);
+	nv_wr32(priv, 0x419f78, 0xb);
+}
+
+static void
+nve0_graph_generate_tpcunk(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x41be24, 0x6);
+	nv_wr32(priv, 0x41bec0, 0x12180000);
+	nv_wr32(priv, 0x41bec4, 0x37f7f);
+	nv_wr32(priv, 0x41bee4, 0x6480430);
+	nv_wr32(priv, 0x41bf00, 0xa418820);
+	nv_wr32(priv, 0x41bf04, 0x62080e6);
+	nv_wr32(priv, 0x41bf08, 0x20398a4);
+	nv_wr32(priv, 0x41bf0c, 0xe629062);
+	nv_wr32(priv, 0x41bf10, 0xa418820);
+	nv_wr32(priv, 0x41bf14, 0xe6);
+	nv_wr32(priv, 0x41bfd0, 0x900103);
+	nv_wr32(priv, 0x41bfe0, 0x400001);
+	nv_wr32(priv, 0x41bfe4, 0x0);
+}
+
+int
+nve0_grctx_generate(struct nvc0_graph_priv *priv)
+{
+	struct nvc0_grctx info;
+	int ret, i, gpc, tpc, id;
+	u32 data[6] = {}, data2[2] = {}, tmp;
+	u32 tpc_set = 0, tpc_mask = 0;
+	u32 magic[GPC_MAX][2], offset;
+	u8 tpcnr[GPC_MAX], a, b;
+	u8 shift, ntpcv;
+
+	ret = nvc0_grctx_init(priv, &info);
+	if (ret)
+		return ret;
+
+	nv_mask(priv, 0x000260, 0x00000001, 0x00000000);
+	nv_wr32(priv, 0x400204, 0x00000000);
+	nv_wr32(priv, 0x400208, 0x00000000);
+
+	nve0_graph_generate_unk40xx(priv);
+	nve0_graph_generate_unk44xx(priv);
+	nve0_graph_generate_unk46xx(priv);
+	nve0_graph_generate_unk47xx(priv);
+	nve0_graph_generate_unk58xx(priv);
+	nve0_graph_generate_unk60xx(priv);
+	nve0_graph_generate_unk64xx(priv);
+	nve0_graph_generate_unk70xx(priv);
+	nve0_graph_generate_unk78xx(priv);
+	nve0_graph_generate_unk80xx(priv);
+	nve0_graph_generate_unk88xx(priv);
+	nve0_graph_generate_gpc(priv);
+	nve0_graph_generate_tpc(priv);
+	nve0_graph_generate_tpcunk(priv);
+
+	nv_wr32(priv, 0x404154, 0x0);
+
+	mmio_data(0x003000, 0x0100, NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS);
+	mmio_data(0x008000, 0x0100, NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS);
+	mmio_data(0x060000, 0x1000, NV_MEM_ACCESS_RW);
+	mmio_list(0x40800c, 0x00000000,  8, 1);
+	mmio_list(0x408010, 0x80000000,  0, 0);
+	mmio_list(0x419004, 0x00000000,  8, 1);
+	mmio_list(0x419008, 0x00000000,  0, 0);
+	mmio_list(0x4064cc, 0x80000000,  0, 0);
+	mmio_list(0x408004, 0x00000000,  8, 0);
+	mmio_list(0x408008, 0x80000030,  0, 0);
+	mmio_list(0x418808, 0x00000000,  8, 0);
+	mmio_list(0x41880c, 0x80000030,  0, 0);
+	mmio_list(0x4064c8, 0x01800600,  0, 0);
+	mmio_list(0x418810, 0x80000000, 12, 2);
+	mmio_list(0x419848, 0x10000000, 12, 2);
+	mmio_list(0x405830, 0x02180648,  0, 0);
+	mmio_list(0x4064c4, 0x0192ffff,  0, 0);
+	for (gpc = 0, offset = 0; gpc < priv->gpc_nr; gpc++) {
+		u16 magic0 = 0x0218 * priv->tpc_nr[gpc];
+		u16 magic1 = 0x0648 * priv->tpc_nr[gpc];
+		magic[gpc][0]  = 0x10000000 | (magic0 << 16) | offset;
+		magic[gpc][1]  = 0x00000000 | (magic1 << 16);
+		offset += 0x0324 * priv->tpc_nr[gpc];
+	}
+	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+		mmio_list(GPC_UNIT(gpc, 0x30c0), magic[gpc][0], 0, 0);
+		mmio_list(GPC_UNIT(gpc, 0x30e4), magic[gpc][1] | offset, 0, 0);
+		offset += 0x07ff * priv->tpc_nr[gpc];
+	}
+	mmio_list(0x17e91c, 0x06060609, 0, 0);
+	mmio_list(0x17e920, 0x00090a05, 0, 0);
+
+	nv_wr32(priv, 0x418c6c, 0x1);
+	nv_wr32(priv, 0x41980c, 0x10);
+	nv_wr32(priv, 0x41be08, 0x4);
+	nv_wr32(priv, 0x4064c0, 0x801a00f0);
+	nv_wr32(priv, 0x405800, 0xf8000bf);
+	nv_wr32(priv, 0x419c00, 0xa);
+
+	for (tpc = 0, id = 0; tpc < 4; tpc++) {
+		for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+			if (tpc < priv->tpc_nr[gpc]) {
+				nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0698), id);
+				nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x04e8), id);
+				nv_wr32(priv, GPC_UNIT(gpc, 0x0c10 + tpc * 4), id);
+				nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0088), id++);
+			}
+
+			nv_wr32(priv, GPC_UNIT(gpc, 0x0c08), priv->tpc_nr[gpc]);
+			nv_wr32(priv, GPC_UNIT(gpc, 0x0c8c), priv->tpc_nr[gpc]);
+		}
+	}
+
+	tmp = 0;
+	for (i = 0; i < priv->gpc_nr; i++)
+		tmp |= priv->tpc_nr[i] << (i * 4);
+	nv_wr32(priv, 0x406028, tmp);
+	nv_wr32(priv, 0x405870, tmp);
+
+	nv_wr32(priv, 0x40602c, 0x0);
+	nv_wr32(priv, 0x405874, 0x0);
+	nv_wr32(priv, 0x406030, 0x0);
+	nv_wr32(priv, 0x405878, 0x0);
+	nv_wr32(priv, 0x406034, 0x0);
+	nv_wr32(priv, 0x40587c, 0x0);
+
+	/* calculate first set of magics */
+	memcpy(tpcnr, priv->tpc_nr, sizeof(priv->tpc_nr));
+
+	gpc = -1;
+	for (tpc = 0; tpc < priv->tpc_total; tpc++) {
+		do {
+			gpc = (gpc + 1) % priv->gpc_nr;
+		} while (!tpcnr[gpc]);
+		tpcnr[gpc]--;
+
+		data[tpc / 6] |= gpc << ((tpc % 6) * 5);
+	}
+
+	for (; tpc < 32; tpc++)
+		data[tpc / 6] |= 7 << ((tpc % 6) * 5);
+
+	/* and the second... */
+	shift = 0;
+	ntpcv = priv->tpc_total;
+	while (!(ntpcv & (1 << 4))) {
+		ntpcv <<= 1;
+		shift++;
+	}
+
+	data2[0]  = ntpcv << 16;
+	data2[0] |= shift << 21;
+	data2[0] |= (((1 << (0 + 5)) % ntpcv) << 24);
+	data2[0] |= priv->tpc_total << 8;
+	data2[0] |= priv->magic_not_rop_nr;
+	for (i = 1; i < 7; i++)
+		data2[1] |= ((1 << (i + 5)) % ntpcv) << ((i - 1) * 5);
+
+	/* and write it all the various parts of PGRAPH */
+	nv_wr32(priv, 0x418bb8, (priv->tpc_total << 8) | priv->magic_not_rop_nr);
+	for (i = 0; i < 6; i++)
+		nv_wr32(priv, 0x418b08 + (i * 4), data[i]);
+
+	nv_wr32(priv, 0x41bfd0, data2[0]);
+	nv_wr32(priv, 0x41bfe4, data2[1]);
+	for (i = 0; i < 6; i++)
+		nv_wr32(priv, 0x41bf00 + (i * 4), data[i]);
+
+	nv_wr32(priv, 0x4078bc, (priv->tpc_total << 8) | priv->magic_not_rop_nr);
+	for (i = 0; i < 6; i++)
+		nv_wr32(priv, 0x40780c + (i * 4), data[i]);
+
+
+	memcpy(tpcnr, priv->tpc_nr, sizeof(priv->tpc_nr));
+	for (gpc = 0; gpc < priv->gpc_nr; gpc++)
+		tpc_mask |= ((1 << priv->tpc_nr[gpc]) - 1) << (gpc * 8);
+
+	for (i = 0, gpc = -1, b = -1; i < 32; i++) {
+		a = (i * (priv->tpc_total - 1)) / 32;
+		if (a != b) {
+			b = a;
+			do {
+				gpc = (gpc + 1) % priv->gpc_nr;
+			} while (!tpcnr[gpc]);
+			tpc = priv->tpc_nr[gpc] - tpcnr[gpc]--;
+
+			tpc_set |= 1 << ((gpc * 8) + tpc);
+		}
+
+		nv_wr32(priv, 0x406800 + (i * 0x20), tpc_set);
+		nv_wr32(priv, 0x406c00 + (i * 0x20), tpc_set ^ tpc_mask);
+	}
+
+	for (i = 0; i < 8; i++)
+		nv_wr32(priv, 0x4064d0 + (i * 0x04), 0x00000000);
+
+	nv_wr32(priv, 0x405b00, 0x201);
+	nv_wr32(priv, 0x408850, 0x2);
+	nv_wr32(priv, 0x408958, 0x2);
+	nv_wr32(priv, 0x419f78, 0xa);
+
+	nve0_grctx_generate_icmd(priv);
+	nve0_grctx_generate_a097(priv);
+	nve0_grctx_generate_902d(priv);
+
+	nv_mask(priv, 0x000260, 0x00000001, 0x00000001);
+	nv_wr32(priv, 0x418800, 0x7026860a); //XXX
+	nv_wr32(priv, 0x41be10, 0x00bb8bc7); //XXX
+	return nvc0_grctx_fini(&info);
+}
diff --git a/nvkm/engine/graph/fuc/gpcnvc0.fuc b/nvkm/engine/graph/fuc/gpcnvc0.fuc
new file mode 100644
index 000000000..b86cc60dc
--- /dev/null
+++ b/nvkm/engine/graph/fuc/gpcnvc0.fuc
@@ -0,0 +1,539 @@
+/* fuc microcode for nvc0 PGRAPH/GPC
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+/* To build:
+ *    m4 gpcnvc0.fuc | envyas -a -w -m fuc -V fuc3 -o gpcnvc0.fuc.h
+ */
+
+/* TODO
+ * - bracket certain functions with scratch writes, useful for debugging
+ * - watchdog timer around ctx operations
+ */
+
+.section #nvc0_grgpc_data
+include(`nvc0.fuc')
+gpc_id:			.b32 0
+gpc_mmio_list_head:	.b32 0
+gpc_mmio_list_tail:	.b32 0
+
+tpc_count:		.b32 0
+tpc_mask:		.b32 0
+tpc_mmio_list_head:	.b32 0
+tpc_mmio_list_tail:	.b32 0
+
+cmd_queue:		queue_init
+
+// chipset descriptions
+chipsets:
+.b8  0xc0 0 0 0
+.b16 #nvc0_gpc_mmio_head
+.b16 #nvc0_gpc_mmio_tail
+.b16 #nvc0_tpc_mmio_head
+.b16 #nvc0_tpc_mmio_tail
+.b8  0xc1 0 0 0
+.b16 #nvc0_gpc_mmio_head
+.b16 #nvc1_gpc_mmio_tail
+.b16 #nvc0_tpc_mmio_head
+.b16 #nvc1_tpc_mmio_tail
+.b8  0xc3 0 0 0
+.b16 #nvc0_gpc_mmio_head
+.b16 #nvc0_gpc_mmio_tail
+.b16 #nvc0_tpc_mmio_head
+.b16 #nvc3_tpc_mmio_tail
+.b8  0xc4 0 0 0
+.b16 #nvc0_gpc_mmio_head
+.b16 #nvc0_gpc_mmio_tail
+.b16 #nvc0_tpc_mmio_head
+.b16 #nvc3_tpc_mmio_tail
+.b8  0xc8 0 0 0
+.b16 #nvc0_gpc_mmio_head
+.b16 #nvc0_gpc_mmio_tail
+.b16 #nvc0_tpc_mmio_head
+.b16 #nvc0_tpc_mmio_tail
+.b8  0xce 0 0 0
+.b16 #nvc0_gpc_mmio_head
+.b16 #nvc0_gpc_mmio_tail
+.b16 #nvc0_tpc_mmio_head
+.b16 #nvc3_tpc_mmio_tail
+.b8  0xcf 0 0 0
+.b16 #nvc0_gpc_mmio_head
+.b16 #nvc0_gpc_mmio_tail
+.b16 #nvc0_tpc_mmio_head
+.b16 #nvcf_tpc_mmio_tail
+.b8  0xd9 0 0 0
+.b16 #nvd9_gpc_mmio_head
+.b16 #nvd9_gpc_mmio_tail
+.b16 #nvd9_tpc_mmio_head
+.b16 #nvd9_tpc_mmio_tail
+.b8  0 0 0 0
+
+// GPC mmio lists
+nvc0_gpc_mmio_head:
+mmctx_data(0x000380, 1)
+mmctx_data(0x000400, 6)
+mmctx_data(0x000450, 9)
+mmctx_data(0x000600, 1)
+mmctx_data(0x000684, 1)
+mmctx_data(0x000700, 5)
+mmctx_data(0x000800, 1)
+mmctx_data(0x000808, 3)
+mmctx_data(0x000828, 1)
+mmctx_data(0x000830, 1)
+mmctx_data(0x0008d8, 1)
+mmctx_data(0x0008e0, 1)
+mmctx_data(0x0008e8, 6)
+mmctx_data(0x00091c, 1)
+mmctx_data(0x000924, 3)
+mmctx_data(0x000b00, 1)
+mmctx_data(0x000b08, 6)
+mmctx_data(0x000bb8, 1)
+mmctx_data(0x000c08, 1)
+mmctx_data(0x000c10, 8)
+mmctx_data(0x000c80, 1)
+mmctx_data(0x000c8c, 1)
+mmctx_data(0x001000, 3)
+mmctx_data(0x001014, 1)
+nvc0_gpc_mmio_tail:
+mmctx_data(0x000c6c, 1);
+nvc1_gpc_mmio_tail:
+
+nvd9_gpc_mmio_head:
+mmctx_data(0x000380, 1)
+mmctx_data(0x000400, 2)
+mmctx_data(0x00040c, 3)
+mmctx_data(0x000450, 9)
+mmctx_data(0x000600, 1)
+mmctx_data(0x000684, 1)
+mmctx_data(0x000700, 5)
+mmctx_data(0x000800, 1)
+mmctx_data(0x000808, 3)
+mmctx_data(0x000828, 1)
+mmctx_data(0x000830, 1)
+mmctx_data(0x0008d8, 1)
+mmctx_data(0x0008e0, 1)
+mmctx_data(0x0008e8, 6)
+mmctx_data(0x00091c, 1)
+mmctx_data(0x000924, 3)
+mmctx_data(0x000b00, 1)
+mmctx_data(0x000b08, 6)
+mmctx_data(0x000bb8, 1)
+mmctx_data(0x000c08, 1)
+mmctx_data(0x000c10, 8)
+mmctx_data(0x000c6c, 1)
+mmctx_data(0x000c80, 1)
+mmctx_data(0x000c8c, 1)
+mmctx_data(0x001000, 3)
+mmctx_data(0x001014, 1)
+nvd9_gpc_mmio_tail:
+
+// TPC mmio lists
+nvc0_tpc_mmio_head:
+mmctx_data(0x000018, 1)
+mmctx_data(0x00003c, 1)
+mmctx_data(0x000048, 1)
+mmctx_data(0x000064, 1)
+mmctx_data(0x000088, 1)
+mmctx_data(0x000200, 6)
+mmctx_data(0x00021c, 2)
+mmctx_data(0x000300, 6)
+mmctx_data(0x0003d0, 1)
+mmctx_data(0x0003e0, 2)
+mmctx_data(0x000400, 3)
+mmctx_data(0x000420, 1)
+mmctx_data(0x0004b0, 1)
+mmctx_data(0x0004e8, 1)
+mmctx_data(0x0004f4, 1)
+mmctx_data(0x000520, 2)
+mmctx_data(0x000604, 4)
+mmctx_data(0x000644, 20)
+mmctx_data(0x000698, 1)
+mmctx_data(0x000750, 2)
+nvc0_tpc_mmio_tail:
+mmctx_data(0x000758, 1)
+mmctx_data(0x0002c4, 1)
+mmctx_data(0x0006e0, 1)
+nvcf_tpc_mmio_tail:
+mmctx_data(0x0004bc, 1)
+nvc3_tpc_mmio_tail:
+mmctx_data(0x000544, 1)
+nvc1_tpc_mmio_tail:
+
+nvd9_tpc_mmio_head:
+mmctx_data(0x000018, 1)
+mmctx_data(0x00003c, 1)
+mmctx_data(0x000048, 1)
+mmctx_data(0x000064, 1)
+mmctx_data(0x000088, 1)
+mmctx_data(0x000200, 6)
+mmctx_data(0x00021c, 2)
+mmctx_data(0x0002c4, 1)
+mmctx_data(0x000300, 6)
+mmctx_data(0x0003d0, 1)
+mmctx_data(0x0003e0, 2)
+mmctx_data(0x000400, 3)
+mmctx_data(0x000420, 3)
+mmctx_data(0x0004b0, 1)
+mmctx_data(0x0004e8, 1)
+mmctx_data(0x0004f4, 1)
+mmctx_data(0x000520, 2)
+mmctx_data(0x000544, 1)
+mmctx_data(0x000604, 4)
+mmctx_data(0x000644, 20)
+mmctx_data(0x000698, 1)
+mmctx_data(0x0006e0, 1)
+mmctx_data(0x000750, 3)
+nvd9_tpc_mmio_tail:
+
+.section #nvc0_grgpc_code
+bra #init
+define(`include_code')
+include(`nvc0.fuc')
+
+// reports an exception to the host
+//
+// In: $r15 error code (see nvc0.fuc)
+//
+error:
+	push $r14
+	mov $r14 -0x67ec 	// 0x9814
+	sethi $r14 0x400000
+	call #nv_wr32		// HUB_CTXCTL_CC_SCRATCH[5] = error code
+	add b32 $r14 0x41c
+	mov $r15 1
+	call #nv_wr32		// HUB_CTXCTL_INTR_UP_SET
+	pop $r14
+	ret
+
+// GPC fuc initialisation, executed by triggering ucode start, will
+// fall through to main loop after completion.
+//
+// Input:
+//   CC_SCRATCH[0]: chipset (PMC_BOOT_0 read returns 0x0bad0bad... sigh)
+//   CC_SCRATCH[1]: context base
+//
+// Output:
+//   CC_SCRATCH[0]:
+//	     31:31: set to signal completion
+//   CC_SCRATCH[1]:
+//	      31:0: GPC context size
+//
+init:
+	clear b32 $r0
+	mov $sp $r0
+
+	// enable fifo access
+	mov $r1 0x1200
+	mov $r2 2
+	iowr I[$r1 + 0x000] $r2		// FIFO_ENABLE
+
+	// setup i0 handler, and route all interrupts to it
+	mov $r1 #ih
+	mov $iv0 $r1
+	mov $r1 0x400
+	iowr I[$r1 + 0x300] $r0		// INTR_DISPATCH
+
+	// enable fifo interrupt
+	mov $r2 4
+	iowr I[$r1 + 0x000] $r2		// INTR_EN_SET
+
+	// enable interrupts
+	bset $flags ie0
+
+	// figure out which GPC we are, and how many TPCs we have
+	mov $r1 0x608
+	shl b32 $r1 6
+	iord $r2 I[$r1 + 0x000]		// UNITS
+	mov $r3 1
+	and $r2 0x1f
+	shl b32 $r3 $r2
+	sub b32 $r3 1
+	st b32 D[$r0 + #tpc_count] $r2
+	st b32 D[$r0 + #tpc_mask] $r3
+	add b32 $r1 0x400
+	iord $r2 I[$r1 + 0x000]		// MYINDEX
+	st b32 D[$r0 + #gpc_id] $r2
+
+	// find context data for this chipset
+	mov $r2 0x800
+	shl b32 $r2 6
+	iord $r2 I[$r2 + 0x000]		// CC_SCRATCH[0]
+	mov $r1 #chipsets - 12
+	init_find_chipset:
+		add b32 $r1 12
+		ld b32 $r3 D[$r1 + 0x00]
+		cmpu b32 $r3 $r2
+		bra e #init_context
+		cmpu b32 $r3 0
+		bra ne #init_find_chipset
+		// unknown chipset
+		ret
+
+	// initialise context base, and size tracking
+	init_context:
+	mov $r2 0x800
+	shl b32 $r2 6
+	iord $r2 I[$r2 + 0x100]	// CC_SCRATCH[1], initial base
+	clear b32 $r3		// track GPC context size here
+
+	// set mmctx base addresses now so we don't have to do it later,
+	// they don't currently ever change
+	mov $r4 0x700
+	shl b32 $r4 6
+	shr b32 $r5 $r2 8
+	iowr I[$r4 + 0x000] $r5		// MMCTX_SAVE_SWBASE
+	iowr I[$r4 + 0x100] $r5		// MMCTX_LOAD_SWBASE
+
+	// calculate GPC mmio context size, store the chipset-specific
+	// mmio list pointers somewhere we can get at them later without
+	// re-parsing the chipset list
+	clear b32 $r14
+	clear b32 $r15
+	ld b16 $r14 D[$r1 + 4]
+	ld b16 $r15 D[$r1 + 6]
+	st b16 D[$r0 + #gpc_mmio_list_head] $r14
+	st b16 D[$r0 + #gpc_mmio_list_tail] $r15
+	call #mmctx_size
+	add b32 $r2 $r15
+	add b32 $r3 $r15
+
+	// calculate per-TPC mmio context size, store the list pointers
+	ld b16 $r14 D[$r1 + 8]
+	ld b16 $r15 D[$r1 + 10]
+	st b16 D[$r0 + #tpc_mmio_list_head] $r14
+	st b16 D[$r0 + #tpc_mmio_list_tail] $r15
+	call #mmctx_size
+	ld b32 $r14 D[$r0 + #tpc_count]
+	mulu $r14 $r15
+	add b32 $r2 $r14
+	add b32 $r3 $r14
+
+	// round up base/size to 256 byte boundary (for strand SWBASE)
+	add b32 $r4 0x1300
+	shr b32 $r3 2
+	iowr I[$r4 + 0x000] $r3		// MMCTX_LOAD_COUNT, wtf for?!?
+	shr b32 $r2 8
+	shr b32 $r3 6
+	add b32 $r2 1
+	add b32 $r3 1
+	shl b32 $r2 8
+	shl b32 $r3 8
+
+	// calculate size of strand context data
+	mov b32 $r15 $r2
+	call #strand_ctx_init
+	add b32 $r3 $r15
+
+	// save context size, and tell HUB we're done
+	mov $r1 0x800
+	shl b32 $r1 6
+	iowr I[$r1 + 0x100] $r3		// CC_SCRATCH[1]  = context size
+	add b32 $r1 0x800
+	clear b32 $r2
+	bset $r2 31
+	iowr I[$r1 + 0x000] $r2		// CC_SCRATCH[0] |= 0x80000000
+
+// Main program loop, very simple, sleeps until woken up by the interrupt
+// handler, pulls a command from the queue and executes its handler
+//
+main:
+	bset $flags $p0
+	sleep $p0
+	mov $r13 #cmd_queue
+	call #queue_get
+	bra $p1 #main
+
+	// 0x0000-0x0003 are all context transfers
+	cmpu b32 $r14 0x04
+	bra nc #main_not_ctx_xfer
+		// fetch $flags and mask off $p1/$p2
+		mov $r1 $flags
+		mov $r2 0x0006
+		not b32 $r2
+		and $r1 $r2
+		// set $p1/$p2 according to transfer type
+		shl b32 $r14 1
+		or $r1 $r14
+		mov $flags $r1
+		// transfer context data
+		call #ctx_xfer
+		bra #main
+
+	main_not_ctx_xfer:
+	shl b32 $r15 $r14 16
+	or $r15 E_BAD_COMMAND
+	call #error
+	bra #main
+
+// interrupt handler
+ih:
+	push $r8
+	mov $r8 $flags
+	push $r8
+	push $r9
+	push $r10
+	push $r11
+	push $r13
+	push $r14
+	push $r15
+
+	// incoming fifo command?
+	iord $r10 I[$r0 + 0x200]	// INTR
+	and $r11 $r10 0x00000004
+	bra e #ih_no_fifo
+		// queue incoming fifo command for later processing
+		mov $r11 0x1900
+		mov $r13 #cmd_queue
+		iord $r14 I[$r11 + 0x100]	// FIFO_CMD
+		iord $r15 I[$r11 + 0x000]	// FIFO_DATA
+		call #queue_put
+		add b32 $r11 0x400
+		mov $r14 1
+		iowr I[$r11 + 0x000] $r14	// FIFO_ACK
+
+	// ack, and wake up main()
+	ih_no_fifo:
+	iowr I[$r0 + 0x100] $r10	// INTR_ACK
+
+	pop $r15
+	pop $r14
+	pop $r13
+	pop $r11
+	pop $r10
+	pop $r9
+	pop $r8
+	mov $flags $r8
+	pop $r8
+	bclr $flags $p0
+	iret
+
+// Set this GPC's bit in HUB_BAR, used to signal completion of various
+// activities to the HUB fuc
+//
+hub_barrier_done:
+	mov $r15 1
+	ld b32 $r14 D[$r0 + #gpc_id]
+	shl b32 $r15 $r14
+	mov $r14 -0x6be8 	// 0x409418 - HUB_BAR_SET
+	sethi $r14 0x400000
+	call #nv_wr32
+	ret
+
+// Disables various things, waits a bit, and re-enables them..
+//
+// Not sure how exactly this helps, perhaps "ENABLE" is not such a
+// good description for the bits we turn off?  Anyways, without this,
+// funny things happen.
+//
+ctx_redswitch:
+	mov $r14 0x614
+	shl b32 $r14 6
+	mov $r15 0x020
+	iowr I[$r14] $r15	// GPC_RED_SWITCH = POWER
+	mov $r15 8
+	ctx_redswitch_delay:
+		sub b32 $r15 1
+		bra ne #ctx_redswitch_delay
+	mov $r15 0xa20
+	iowr I[$r14] $r15	// GPC_RED_SWITCH = UNK11, ENABLE, POWER
+	ret
+
+// Transfer GPC context data between GPU and storage area
+//
+// In: $r15 context base address
+//     $p1 clear on save, set on load
+//     $p2 set if opposite direction done/will be done, so:
+//		on save it means: "a load will follow this save"
+//		on load it means: "a save preceeded this load"
+//
+ctx_xfer:
+	// set context base address
+	mov $r1 0xa04
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r15// MEM_BASE
+	bra not $p1 #ctx_xfer_not_load
+		call #ctx_redswitch
+	ctx_xfer_not_load:
+
+	// strands
+	mov $r1 0x4afc
+	sethi $r1 0x20000
+	mov $r2 0xc
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x0c
+	call #strand_wait
+	mov $r2 0x47fc
+	sethi $r2 0x20000
+	iowr I[$r2] $r0		// STRAND_FIRST_GENE(0x3f) = 0x00
+	xbit $r2 $flags $p1
+	add b32 $r2 3
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x03/0x04 (SAVE/LOAD)
+
+	// mmio context
+	xbit $r10 $flags $p1	// direction
+	or $r10 2		// first
+	mov $r11 0x0000
+	sethi $r11 0x500000
+	ld b32 $r12 D[$r0 + #gpc_id]
+	shl b32 $r12 15
+	add b32 $r11 $r12	// base = NV_PGRAPH_GPCn
+	ld b32 $r12 D[$r0 + #gpc_mmio_list_head]
+	ld b32 $r13 D[$r0 + #gpc_mmio_list_tail]
+	mov $r14 0		// not multi
+	call #mmctx_xfer
+
+	// per-TPC mmio context
+	xbit $r10 $flags $p1	// direction
+	or $r10 4		// last
+	mov $r11 0x4000
+	sethi $r11 0x500000	// base = NV_PGRAPH_GPC0_TPC0
+	ld b32 $r12 D[$r0 + #gpc_id]
+	shl b32 $r12 15
+	add b32 $r11 $r12	// base = NV_PGRAPH_GPCn_TPC0
+	ld b32 $r12 D[$r0 + #tpc_mmio_list_head]
+	ld b32 $r13 D[$r0 + #tpc_mmio_list_tail]
+	ld b32 $r15 D[$r0 + #tpc_mask]
+	mov $r14 0x800		// stride = 0x800
+	call #mmctx_xfer
+
+	// wait for strands to finish
+	call #strand_wait
+
+	// if load, or a save without a load following, do some
+	// unknown stuff that's done after finishing a block of
+	// strand commands
+	bra $p1 #ctx_xfer_post
+	bra not $p2 #ctx_xfer_done
+	ctx_xfer_post:
+		mov $r1 0x4afc
+		sethi $r1 0x20000
+		mov $r2 0xd
+		iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x0d
+		call #strand_wait
+
+	// mark completion in HUB's barrier
+	ctx_xfer_done:
+	call #hub_barrier_done
+	ret
+
+.align 256
diff --git a/nvkm/engine/graph/fuc/gpcnve0.fuc b/nvkm/engine/graph/fuc/gpcnve0.fuc
new file mode 100644
index 000000000..62ab231cd
--- /dev/null
+++ b/nvkm/engine/graph/fuc/gpcnve0.fuc
@@ -0,0 +1,456 @@
+/* fuc microcode for nve0 PGRAPH/GPC
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+/* To build:
+ *    m4 nve0_grgpc.fuc | envyas -a -w -m fuc -V nva3 -o nve0_grgpc.fuc.h
+ */
+
+/* TODO
+ * - bracket certain functions with scratch writes, useful for debugging
+ * - watchdog timer around ctx operations
+ */
+
+.section #nve0_grgpc_data
+include(`nve0.fuc')
+gpc_id:			.b32 0
+gpc_mmio_list_head:	.b32 0
+gpc_mmio_list_tail:	.b32 0
+
+tpc_count:		.b32 0
+tpc_mask:		.b32 0
+tpc_mmio_list_head:	.b32 0
+tpc_mmio_list_tail:	.b32 0
+
+cmd_queue:		queue_init
+
+// chipset descriptions
+chipsets:
+.b8  0xe4 0 0 0
+.b16 #nve4_gpc_mmio_head
+.b16 #nve4_gpc_mmio_tail
+.b16 #nve4_tpc_mmio_head
+.b16 #nve4_tpc_mmio_tail
+.b8  0xe7 0 0 0
+.b16 #nve4_gpc_mmio_head
+.b16 #nve4_gpc_mmio_tail
+.b16 #nve4_tpc_mmio_head
+.b16 #nve4_tpc_mmio_tail
+.b8  0xe6 0 0 0
+.b16 #nve4_gpc_mmio_head
+.b16 #nve4_gpc_mmio_tail
+.b16 #nve4_tpc_mmio_head
+.b16 #nve4_tpc_mmio_tail
+.b8  0 0 0 0
+
+// GPC mmio lists
+nve4_gpc_mmio_head:
+mmctx_data(0x000380, 1)
+mmctx_data(0x000400, 2)
+mmctx_data(0x00040c, 3)
+mmctx_data(0x000450, 9)
+mmctx_data(0x000600, 1)
+mmctx_data(0x000684, 1)
+mmctx_data(0x000700, 5)
+mmctx_data(0x000800, 1)
+mmctx_data(0x000808, 3)
+mmctx_data(0x000828, 1)
+mmctx_data(0x000830, 1)
+mmctx_data(0x0008d8, 1)
+mmctx_data(0x0008e0, 1)
+mmctx_data(0x0008e8, 6)
+mmctx_data(0x00091c, 1)
+mmctx_data(0x000924, 3)
+mmctx_data(0x000b00, 1)
+mmctx_data(0x000b08, 6)
+mmctx_data(0x000bb8, 1)
+mmctx_data(0x000c08, 1)
+mmctx_data(0x000c10, 8)
+mmctx_data(0x000c40, 1)
+mmctx_data(0x000c6c, 1)
+mmctx_data(0x000c80, 1)
+mmctx_data(0x000c8c, 1)
+mmctx_data(0x001000, 3)
+mmctx_data(0x001014, 1)
+mmctx_data(0x003024, 1)
+mmctx_data(0x0030c0, 2)
+mmctx_data(0x0030e4, 1)
+mmctx_data(0x003100, 6)
+mmctx_data(0x0031d0, 1)
+mmctx_data(0x0031e0, 2)
+nve4_gpc_mmio_tail:
+
+// TPC mmio lists
+nve4_tpc_mmio_head:
+mmctx_data(0x000048, 1)
+mmctx_data(0x000064, 1)
+mmctx_data(0x000088, 1)
+mmctx_data(0x000200, 6)
+mmctx_data(0x00021c, 2)
+mmctx_data(0x000230, 1)
+mmctx_data(0x0002c4, 1)
+mmctx_data(0x000400, 3)
+mmctx_data(0x000420, 3)
+mmctx_data(0x0004e8, 1)
+mmctx_data(0x0004f4, 1)
+mmctx_data(0x000604, 4)
+mmctx_data(0x000644, 22)
+mmctx_data(0x0006ac, 2)
+mmctx_data(0x0006c8, 1)
+mmctx_data(0x000730, 8)
+mmctx_data(0x000758, 1)
+mmctx_data(0x000778, 1)
+nve4_tpc_mmio_tail:
+
+.section #nve0_grgpc_code
+bra #init
+define(`include_code')
+include(`nve0.fuc')
+
+// reports an exception to the host
+//
+// In: $r15 error code (see nve0.fuc)
+//
+error:
+	push $r14
+	mov $r14 -0x67ec 	// 0x9814
+	sethi $r14 0x400000
+	call #nv_wr32		// HUB_CTXCTL_CC_SCRATCH[5] = error code
+	add b32 $r14 0x41c
+	mov $r15 1
+	call #nv_wr32		// HUB_CTXCTL_INTR_UP_SET
+	pop $r14
+	ret
+
+// GPC fuc initialisation, executed by triggering ucode start, will
+// fall through to main loop after completion.
+//
+// Input:
+//   CC_SCRATCH[0]: chipset (PMC_BOOT_0 read returns 0x0bad0bad... sigh)
+//   CC_SCRATCH[1]: context base
+//
+// Output:
+//   CC_SCRATCH[0]:
+//	     31:31: set to signal completion
+//   CC_SCRATCH[1]:
+//	      31:0: GPC context size
+//
+init:
+	clear b32 $r0
+	mov $sp $r0
+
+	// enable fifo access
+	mov $r1 0x1200
+	mov $r2 2
+	iowr I[$r1 + 0x000] $r2		// FIFO_ENABLE
+
+	// setup i0 handler, and route all interrupts to it
+	mov $r1 #ih
+	mov $iv0 $r1
+	mov $r1 0x400
+	iowr I[$r1 + 0x300] $r0		// INTR_DISPATCH
+
+	// enable fifo interrupt
+	mov $r2 4
+	iowr I[$r1 + 0x000] $r2		// INTR_EN_SET
+
+	// enable interrupts
+	bset $flags ie0
+
+	// figure out which GPC we are, and how many TPCs we have
+	mov $r1 0x608
+	shl b32 $r1 6
+	iord $r2 I[$r1 + 0x000]		// UNITS
+	mov $r3 1
+	and $r2 0x1f
+	shl b32 $r3 $r2
+	sub b32 $r3 1
+	st b32 D[$r0 + #tpc_count] $r2
+	st b32 D[$r0 + #tpc_mask] $r3
+	add b32 $r1 0x400
+	iord $r2 I[$r1 + 0x000]		// MYINDEX
+	st b32 D[$r0 + #gpc_id] $r2
+
+	// find context data for this chipset
+	mov $r2 0x800
+	shl b32 $r2 6
+	iord $r2 I[$r2 + 0x000]		// CC_SCRATCH[0]
+	mov $r1 #chipsets - 12
+	init_find_chipset:
+		add b32 $r1 12
+		ld b32 $r3 D[$r1 + 0x00]
+		cmpu b32 $r3 $r2
+		bra e #init_context
+		cmpu b32 $r3 0
+		bra ne #init_find_chipset
+		// unknown chipset
+		ret
+
+	// initialise context base, and size tracking
+	init_context:
+	mov $r2 0x800
+	shl b32 $r2 6
+	iord $r2 I[$r2 + 0x100]	// CC_SCRATCH[1], initial base
+	clear b32 $r3		// track GPC context size here
+
+	// set mmctx base addresses now so we don't have to do it later,
+	// they don't currently ever change
+	mov $r4 0x700
+	shl b32 $r4 6
+	shr b32 $r5 $r2 8
+	iowr I[$r4 + 0x000] $r5		// MMCTX_SAVE_SWBASE
+	iowr I[$r4 + 0x100] $r5		// MMCTX_LOAD_SWBASE
+
+	// calculate GPC mmio context size, store the chipset-specific
+	// mmio list pointers somewhere we can get at them later without
+	// re-parsing the chipset list
+	clear b32 $r14
+	clear b32 $r15
+	ld b16 $r14 D[$r1 + 4]
+	ld b16 $r15 D[$r1 + 6]
+	st b16 D[$r0 + #gpc_mmio_list_head] $r14
+	st b16 D[$r0 + #gpc_mmio_list_tail] $r15
+	call #mmctx_size
+	add b32 $r2 $r15
+	add b32 $r3 $r15
+
+	// calculate per-TPC mmio context size, store the list pointers
+	ld b16 $r14 D[$r1 + 8]
+	ld b16 $r15 D[$r1 + 10]
+	st b16 D[$r0 + #tpc_mmio_list_head] $r14
+	st b16 D[$r0 + #tpc_mmio_list_tail] $r15
+	call #mmctx_size
+	ld b32 $r14 D[$r0 + #tpc_count]
+	mulu $r14 $r15
+	add b32 $r2 $r14
+	add b32 $r3 $r14
+
+	// round up base/size to 256 byte boundary (for strand SWBASE)
+	add b32 $r4 0x1300
+	shr b32 $r3 2
+	iowr I[$r4 + 0x000] $r3		// MMCTX_LOAD_COUNT, wtf for?!?
+	shr b32 $r2 8
+	shr b32 $r3 6
+	add b32 $r2 1
+	add b32 $r3 1
+	shl b32 $r2 8
+	shl b32 $r3 8
+
+	// calculate size of strand context data
+	mov b32 $r15 $r2
+	call #strand_ctx_init
+	add b32 $r3 $r15
+
+	// save context size, and tell HUB we're done
+	mov $r1 0x800
+	shl b32 $r1 6
+	iowr I[$r1 + 0x100] $r3		// CC_SCRATCH[1]  = context size
+	add b32 $r1 0x800
+	clear b32 $r2
+	bset $r2 31
+	iowr I[$r1 + 0x000] $r2		// CC_SCRATCH[0] |= 0x80000000
+
+// Main program loop, very simple, sleeps until woken up by the interrupt
+// handler, pulls a command from the queue and executes its handler
+//
+main:
+	bset $flags $p0
+	sleep $p0
+	mov $r13 #cmd_queue
+	call #queue_get
+	bra $p1 #main
+
+	// 0x0000-0x0003 are all context transfers
+	cmpu b32 $r14 0x04
+	bra nc #main_not_ctx_xfer
+		// fetch $flags and mask off $p1/$p2
+		mov $r1 $flags
+		mov $r2 0x0006
+		not b32 $r2
+		and $r1 $r2
+		// set $p1/$p2 according to transfer type
+		shl b32 $r14 1
+		or $r1 $r14
+		mov $flags $r1
+		// transfer context data
+		call #ctx_xfer
+		bra #main
+
+	main_not_ctx_xfer:
+	shl b32 $r15 $r14 16
+	or $r15 E_BAD_COMMAND
+	call #error
+	bra #main
+
+// interrupt handler
+ih:
+	push $r8
+	mov $r8 $flags
+	push $r8
+	push $r9
+	push $r10
+	push $r11
+	push $r13
+	push $r14
+	push $r15
+
+	// incoming fifo command?
+	iord $r10 I[$r0 + 0x200]	// INTR
+	and $r11 $r10 0x00000004
+	bra e #ih_no_fifo
+		// queue incoming fifo command for later processing
+		mov $r11 0x1900
+		mov $r13 #cmd_queue
+		iord $r14 I[$r11 + 0x100]	// FIFO_CMD
+		iord $r15 I[$r11 + 0x000]	// FIFO_DATA
+		call #queue_put
+		add b32 $r11 0x400
+		mov $r14 1
+		iowr I[$r11 + 0x000] $r14	// FIFO_ACK
+
+	// ack, and wake up main()
+	ih_no_fifo:
+	iowr I[$r0 + 0x100] $r10	// INTR_ACK
+
+	pop $r15
+	pop $r14
+	pop $r13
+	pop $r11
+	pop $r10
+	pop $r9
+	pop $r8
+	mov $flags $r8
+	pop $r8
+	bclr $flags $p0
+	iret
+
+// Set this GPC's bit in HUB_BAR, used to signal completion of various
+// activities to the HUB fuc
+//
+hub_barrier_done:
+	mov $r15 1
+	ld b32 $r14 D[$r0 + #gpc_id]
+	shl b32 $r15 $r14
+	mov $r14 -0x6be8 	// 0x409418 - HUB_BAR_SET
+	sethi $r14 0x400000
+	call #nv_wr32
+	ret
+
+// Disables various things, waits a bit, and re-enables them..
+//
+// Not sure how exactly this helps, perhaps "ENABLE" is not such a
+// good description for the bits we turn off?  Anyways, without this,
+// funny things happen.
+//
+ctx_redswitch:
+	mov $r14 0x614
+	shl b32 $r14 6
+	mov $r15 0x020
+	iowr I[$r14] $r15	// GPC_RED_SWITCH = POWER
+	mov $r15 8
+	ctx_redswitch_delay:
+		sub b32 $r15 1
+		bra ne #ctx_redswitch_delay
+	mov $r15 0xa20
+	iowr I[$r14] $r15	// GPC_RED_SWITCH = UNK11, ENABLE, POWER
+	ret
+
+// Transfer GPC context data between GPU and storage area
+//
+// In: $r15 context base address
+//     $p1 clear on save, set on load
+//     $p2 set if opposite direction done/will be done, so:
+//		on save it means: "a load will follow this save"
+//		on load it means: "a save preceeded this load"
+//
+ctx_xfer:
+	// set context base address
+	mov $r1 0xa04
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r15// MEM_BASE
+	bra not $p1 #ctx_xfer_not_load
+		call #ctx_redswitch
+	ctx_xfer_not_load:
+
+	// strands
+	mov $r1 0x4afc
+	sethi $r1 0x20000
+	mov $r2 0xc
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x0c
+	call #strand_wait
+	mov $r2 0x47fc
+	sethi $r2 0x20000
+	iowr I[$r2] $r0		// STRAND_FIRST_GENE(0x3f) = 0x00
+	xbit $r2 $flags $p1
+	add b32 $r2 3
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x03/0x04 (SAVE/LOAD)
+
+	// mmio context
+	xbit $r10 $flags $p1	// direction
+	or $r10 2		// first
+	mov $r11 0x0000
+	sethi $r11 0x500000
+	ld b32 $r12 D[$r0 + #gpc_id]
+	shl b32 $r12 15
+	add b32 $r11 $r12	// base = NV_PGRAPH_GPCn
+	ld b32 $r12 D[$r0 + #gpc_mmio_list_head]
+	ld b32 $r13 D[$r0 + #gpc_mmio_list_tail]
+	mov $r14 0		// not multi
+	call #mmctx_xfer
+
+	// per-TPC mmio context
+	xbit $r10 $flags $p1	// direction
+	or $r10 4		// last
+	mov $r11 0x4000
+	sethi $r11 0x500000	// base = NV_PGRAPH_GPC0_TPC0
+	ld b32 $r12 D[$r0 + #gpc_id]
+	shl b32 $r12 15
+	add b32 $r11 $r12	// base = NV_PGRAPH_GPCn_TPC0
+	ld b32 $r12 D[$r0 + #tpc_mmio_list_head]
+	ld b32 $r13 D[$r0 + #tpc_mmio_list_tail]
+	ld b32 $r15 D[$r0 + #tpc_mask]
+	mov $r14 0x800		// stride = 0x800
+	call #mmctx_xfer
+
+	// wait for strands to finish
+	call #strand_wait
+
+	// if load, or a save without a load following, do some
+	// unknown stuff that's done after finishing a block of
+	// strand commands
+	bra $p1 #ctx_xfer_post
+	bra not $p2 #ctx_xfer_done
+	ctx_xfer_post:
+		mov $r1 0x4afc
+		sethi $r1 0x20000
+		mov $r2 0xd
+		iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x0d
+		call #strand_wait
+
+	// mark completion in HUB's barrier
+	ctx_xfer_done:
+	call #hub_barrier_done
+	ret
+
+.align 256
diff --git a/nvkm/engine/graph/fuc/hubnvc0.fuc b/nvkm/engine/graph/fuc/hubnvc0.fuc
new file mode 100644
index 000000000..0bcfa4d44
--- /dev/null
+++ b/nvkm/engine/graph/fuc/hubnvc0.fuc
@@ -0,0 +1,866 @@
+/* fuc microcode for nvc0 PGRAPH/HUB
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+/* To build:
+ *    m4 hubnvc0.fuc | envyas -a -w -m fuc -V fuc3 -o hubnvc0.fuc.h
+ */
+
+.section #nvc0_grhub_data
+include(`nvc0.fuc')
+gpc_count:		.b32 0
+rop_count:		.b32 0
+cmd_queue:		queue_init
+hub_mmio_list_head:	.b32 0
+hub_mmio_list_tail:	.b32 0
+
+ctx_current:		.b32 0
+
+chipsets:
+.b8  0xc0 0 0 0
+.b16 #nvc0_hub_mmio_head
+.b16 #nvc0_hub_mmio_tail
+.b8  0xc1 0 0 0
+.b16 #nvc0_hub_mmio_head
+.b16 #nvc1_hub_mmio_tail
+.b8  0xc3 0 0 0
+.b16 #nvc0_hub_mmio_head
+.b16 #nvc0_hub_mmio_tail
+.b8  0xc4 0 0 0
+.b16 #nvc0_hub_mmio_head
+.b16 #nvc0_hub_mmio_tail
+.b8  0xc8 0 0 0
+.b16 #nvc0_hub_mmio_head
+.b16 #nvc0_hub_mmio_tail
+.b8  0xce 0 0 0
+.b16 #nvc0_hub_mmio_head
+.b16 #nvc0_hub_mmio_tail
+.b8  0xcf 0 0 0
+.b16 #nvc0_hub_mmio_head
+.b16 #nvc0_hub_mmio_tail
+.b8  0xd9 0 0 0
+.b16 #nvd9_hub_mmio_head
+.b16 #nvd9_hub_mmio_tail
+.b8  0 0 0 0
+
+nvc0_hub_mmio_head:
+mmctx_data(0x17e91c, 2)
+mmctx_data(0x400204, 2)
+mmctx_data(0x404004, 11)
+mmctx_data(0x404044, 1)
+mmctx_data(0x404094, 14)
+mmctx_data(0x4040d0, 7)
+mmctx_data(0x4040f8, 1)
+mmctx_data(0x404130, 3)
+mmctx_data(0x404150, 3)
+mmctx_data(0x404164, 2)
+mmctx_data(0x404174, 3)
+mmctx_data(0x404200, 8)
+mmctx_data(0x404404, 14)
+mmctx_data(0x404460, 4)
+mmctx_data(0x404480, 1)
+mmctx_data(0x404498, 1)
+mmctx_data(0x404604, 4)
+mmctx_data(0x404618, 32)
+mmctx_data(0x404698, 21)
+mmctx_data(0x4046f0, 2)
+mmctx_data(0x404700, 22)
+mmctx_data(0x405800, 1)
+mmctx_data(0x405830, 3)
+mmctx_data(0x405854, 1)
+mmctx_data(0x405870, 4)
+mmctx_data(0x405a00, 2)
+mmctx_data(0x405a18, 1)
+mmctx_data(0x406020, 1)
+mmctx_data(0x406028, 4)
+mmctx_data(0x4064a8, 2)
+mmctx_data(0x4064b4, 2)
+mmctx_data(0x407804, 1)
+mmctx_data(0x40780c, 6)
+mmctx_data(0x4078bc, 1)
+mmctx_data(0x408000, 7)
+mmctx_data(0x408064, 1)
+mmctx_data(0x408800, 3)
+mmctx_data(0x408900, 4)
+mmctx_data(0x408980, 1)
+nvc0_hub_mmio_tail:
+mmctx_data(0x4064c0, 2)
+nvc1_hub_mmio_tail:
+
+nvd9_hub_mmio_head:
+mmctx_data(0x17e91c, 2)
+mmctx_data(0x400204, 2)
+mmctx_data(0x404004, 10)
+mmctx_data(0x404044, 1)
+mmctx_data(0x404094, 14)
+mmctx_data(0x4040d0, 7)
+mmctx_data(0x4040f8, 1)
+mmctx_data(0x404130, 3)
+mmctx_data(0x404150, 3)
+mmctx_data(0x404164, 2)
+mmctx_data(0x404178, 2)
+mmctx_data(0x404200, 8)
+mmctx_data(0x404404, 14)
+mmctx_data(0x404460, 4)
+mmctx_data(0x404480, 1)
+mmctx_data(0x404498, 1)
+mmctx_data(0x404604, 4)
+mmctx_data(0x404618, 32)
+mmctx_data(0x404698, 21)
+mmctx_data(0x4046f0, 2)
+mmctx_data(0x404700, 22)
+mmctx_data(0x405800, 1)
+mmctx_data(0x405830, 3)
+mmctx_data(0x405854, 1)
+mmctx_data(0x405870, 4)
+mmctx_data(0x405a00, 2)
+mmctx_data(0x405a18, 1)
+mmctx_data(0x406020, 1)
+mmctx_data(0x406028, 4)
+mmctx_data(0x4064a8, 2)
+mmctx_data(0x4064b4, 5)
+mmctx_data(0x407804, 1)
+mmctx_data(0x40780c, 6)
+mmctx_data(0x4078bc, 1)
+mmctx_data(0x408000, 7)
+mmctx_data(0x408064, 1)
+mmctx_data(0x408800, 3)
+mmctx_data(0x408900, 4)
+mmctx_data(0x408980, 1)
+nvd9_hub_mmio_tail:
+
+.align 256
+chan_data:
+chan_mmio_count:	.b32 0
+chan_mmio_address:	.b32 0
+
+.align 256
+xfer_data: 		.b32 0
+
+.section #nvc0_grhub_code
+bra #init
+define(`include_code')
+include(`nvc0.fuc')
+
+// reports an exception to the host
+//
+// In: $r15 error code (see nvc0.fuc)
+//
+error:
+	push $r14
+	mov $r14 0x814
+	shl b32 $r14 6
+	iowr I[$r14 + 0x000] $r15	// CC_SCRATCH[5] = error code
+	mov $r14 0xc1c
+	shl b32 $r14 6
+	mov $r15 1
+	iowr I[$r14 + 0x000] $r15	// INTR_UP_SET
+	pop $r14
+	ret
+
+// HUB fuc initialisation, executed by triggering ucode start, will
+// fall through to main loop after completion.
+//
+// Input:
+//   CC_SCRATCH[0]: chipset (PMC_BOOT_0 read returns 0x0bad0bad... sigh)
+//
+// Output:
+//   CC_SCRATCH[0]:
+//	     31:31: set to signal completion
+//   CC_SCRATCH[1]:
+//	      31:0: total PGRAPH context size
+//
+init:
+	clear b32 $r0
+	mov $sp $r0
+	mov $xdbase $r0
+
+	// enable fifo access
+	mov $r1 0x1200
+	mov $r2 2
+	iowr I[$r1 + 0x000] $r2	// FIFO_ENABLE
+
+	// setup i0 handler, and route all interrupts to it
+	mov $r1 #ih
+	mov $iv0 $r1
+	mov $r1 0x400
+	iowr I[$r1 + 0x300] $r0	// INTR_DISPATCH
+
+	// route HUB_CHANNEL_SWITCH to fuc interrupt 8
+	mov $r3 0x404
+	shl b32 $r3 6
+	mov $r2 0x2003		// { HUB_CHANNEL_SWITCH, ZERO } -> intr 8
+	iowr I[$r3 + 0x000] $r2
+
+	// not sure what these are, route them because NVIDIA does, and
+	// the IRQ handler will signal the host if we ever get one.. we
+	// may find out if/why we need to handle these if so..
+	//
+	mov $r2 0x2004
+	iowr I[$r3 + 0x004] $r2 // { 0x04, ZERO } -> intr 9
+	mov $r2 0x200b
+	iowr I[$r3 + 0x008] $r2 // { 0x0b, ZERO } -> intr 10
+	mov $r2 0x200c
+	iowr I[$r3 + 0x01c] $r2 // { 0x0c, ZERO } -> intr 15
+
+	// enable all INTR_UP interrupts
+	mov $r2 0xc24
+	shl b32 $r2 6
+	not b32 $r3 $r0
+	iowr I[$r2] $r3
+
+	// enable fifo, ctxsw, 9, 10, 15 interrupts
+	mov $r2 -0x78fc		// 0x8704
+	sethi $r2 0
+	iowr I[$r1 + 0x000] $r2	// INTR_EN_SET
+
+	// fifo level triggered, rest edge
+	sub b32 $r1 0x100
+	mov $r2 4
+	iowr I[$r1] $r2
+
+	// enable interrupts
+	bset $flags ie0
+
+	// fetch enabled GPC/ROP counts
+	mov $r14 -0x69fc	// 0x409604
+	sethi $r14 0x400000
+	call #nv_rd32
+	extr $r1 $r15 16:20
+	st b32 D[$r0 + #rop_count] $r1
+	and $r15 0x1f
+	st b32 D[$r0 + #gpc_count] $r15
+
+	// set BAR_REQMASK to GPC mask
+	mov $r1 1
+	shl b32 $r1 $r15
+	sub b32 $r1 1
+	mov $r2 0x40c
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r1
+	iowr I[$r2 + 0x100] $r1
+
+	// find context data for this chipset
+	mov $r2 0x800
+	shl b32 $r2 6
+	iord $r2 I[$r2 + 0x000]		// CC_SCRATCH[0]
+	mov $r15 #chipsets - 8
+	init_find_chipset:
+		add b32 $r15 8
+		ld b32 $r3 D[$r15 + 0x00]
+		cmpu b32 $r3 $r2
+		bra e #init_context
+		cmpu b32 $r3 0
+		bra ne #init_find_chipset
+		// unknown chipset
+		ret
+
+	// context size calculation, reserve first 256 bytes for use by fuc
+	init_context:
+	mov $r1 256
+
+	// calculate size of mmio context data
+	ld b16 $r14 D[$r15 + 4]
+	ld b16 $r15 D[$r15 + 6]
+	sethi $r14 0
+	st b32 D[$r0 + #hub_mmio_list_head] $r14
+	st b32 D[$r0 + #hub_mmio_list_tail] $r15
+	call #mmctx_size
+
+	// set mmctx base addresses now so we don't have to do it later,
+	// they don't (currently) ever change
+	mov $r3 0x700
+	shl b32 $r3 6
+	shr b32 $r4 $r1 8
+	iowr I[$r3 + 0x000] $r4		// MMCTX_SAVE_SWBASE
+	iowr I[$r3 + 0x100] $r4		// MMCTX_LOAD_SWBASE
+	add b32 $r3 0x1300
+	add b32 $r1 $r15
+	shr b32 $r15 2
+	iowr I[$r3 + 0x000] $r15	// MMCTX_LOAD_COUNT, wtf for?!?
+
+	// strands, base offset needs to be aligned to 256 bytes
+	shr b32 $r1 8
+	add b32 $r1 1
+	shl b32 $r1 8
+	mov b32 $r15 $r1
+	call #strand_ctx_init
+	add b32 $r1 $r15
+
+	// initialise each GPC in sequence by passing in the offset of its
+	// context data in GPCn_CC_SCRATCH[1], and starting its FUC (which
+	// has previously been uploaded by the host) running.
+	//
+	// the GPC fuc init sequence will set GPCn_CC_SCRATCH[0] bit 31
+	// when it has completed, and return the size of its context data
+	// in GPCn_CC_SCRATCH[1]
+	//
+	ld b32 $r3 D[$r0 + #gpc_count]
+	mov $r4 0x2000
+	sethi $r4 0x500000
+	init_gpc:
+		// setup, and start GPC ucode running
+		add b32 $r14 $r4 0x804
+		mov b32 $r15 $r1
+		call #nv_wr32			// CC_SCRATCH[1] = ctx offset
+		add b32 $r14 $r4 0x800
+		mov b32 $r15 $r2
+		call #nv_wr32			// CC_SCRATCH[0] = chipset
+		add b32 $r14 $r4 0x10c
+		clear b32 $r15
+		call #nv_wr32
+		add b32 $r14 $r4 0x104
+		call #nv_wr32			// ENTRY
+		add b32 $r14 $r4 0x100
+		mov $r15 2			// CTRL_START_TRIGGER
+		call #nv_wr32			// CTRL
+
+		// wait for it to complete, and adjust context size
+		add b32 $r14 $r4 0x800
+		init_gpc_wait:
+			call #nv_rd32
+			xbit $r15 $r15 31
+			bra e #init_gpc_wait
+		add b32 $r14 $r4 0x804
+		call #nv_rd32
+		add b32 $r1 $r15
+
+		// next!
+		add b32 $r4 0x8000
+		sub b32 $r3 1
+		bra ne #init_gpc
+
+	// save context size, and tell host we're ready
+	mov $r2 0x800
+	shl b32 $r2 6
+	iowr I[$r2 + 0x100] $r1		// CC_SCRATCH[1]  = context size
+	add b32 $r2 0x800
+	clear b32 $r1
+	bset $r1 31
+	iowr I[$r2 + 0x000] $r1		// CC_SCRATCH[0] |= 0x80000000
+
+// Main program loop, very simple, sleeps until woken up by the interrupt
+// handler, pulls a command from the queue and executes its handler
+//
+main:
+	// sleep until we have something to do
+	bset $flags $p0
+	sleep $p0
+	mov $r13 #cmd_queue
+	call #queue_get
+	bra $p1 #main
+
+	// context switch, requested by GPU?
+	cmpu b32 $r14 0x4001
+	bra ne #main_not_ctx_switch
+		trace_set(T_AUTO)
+		mov $r1 0xb00
+		shl b32 $r1 6
+		iord $r2 I[$r1 + 0x100]		// CHAN_NEXT
+		iord $r1 I[$r1 + 0x000]		// CHAN_CUR
+
+		xbit $r3 $r1 31
+		bra e #chsw_no_prev
+			xbit $r3 $r2 31
+			bra e #chsw_prev_no_next
+				push $r2
+				mov b32 $r2 $r1
+				trace_set(T_SAVE)
+				bclr $flags $p1
+				bset $flags $p2
+				call #ctx_xfer
+				trace_clr(T_SAVE);
+				pop $r2
+				trace_set(T_LOAD);
+				bset $flags $p1
+				call #ctx_xfer
+				trace_clr(T_LOAD);
+				bra #chsw_done
+			chsw_prev_no_next:
+				push $r2
+				mov b32 $r2 $r1
+				bclr $flags $p1
+				bclr $flags $p2
+				call #ctx_xfer
+				pop $r2
+				mov $r1 0xb00
+				shl b32 $r1 6
+				iowr I[$r1] $r2
+				bra #chsw_done
+		chsw_no_prev:
+			xbit $r3 $r2 31
+			bra e #chsw_done
+				bset $flags $p1
+				bclr $flags $p2
+				call #ctx_xfer
+
+		// ack the context switch request
+		chsw_done:
+		mov $r1 0xb0c
+		shl b32 $r1 6
+		mov $r2 1
+		iowr I[$r1 + 0x000] $r2		// 0x409b0c
+		trace_clr(T_AUTO)
+		bra #main
+
+	// request to set current channel? (*not* a context switch)
+	main_not_ctx_switch:
+	cmpu b32 $r14 0x0001
+	bra ne #main_not_ctx_chan
+		mov b32 $r2 $r15
+		call #ctx_chan
+		bra #main_done
+
+	// request to store current channel context?
+	main_not_ctx_chan:
+	cmpu b32 $r14 0x0002
+	bra ne #main_not_ctx_save
+		trace_set(T_SAVE)
+		bclr $flags $p1
+		bclr $flags $p2
+		call #ctx_xfer
+		trace_clr(T_SAVE)
+		bra #main_done
+
+	main_not_ctx_save:
+		shl b32 $r15 $r14 16
+		or $r15 E_BAD_COMMAND
+		call #error
+		bra #main
+
+	main_done:
+	mov $r1 0x820
+	shl b32 $r1 6
+	clear b32 $r2
+	bset $r2 31
+	iowr I[$r1 + 0x000] $r2		// CC_SCRATCH[0] |= 0x80000000
+	bra #main
+
+// interrupt handler
+ih:
+	push $r8
+	mov $r8 $flags
+	push $r8
+	push $r9
+	push $r10
+	push $r11
+	push $r13
+	push $r14
+	push $r15
+
+	// incoming fifo command?
+	iord $r10 I[$r0 + 0x200]	// INTR
+	and $r11 $r10 0x00000004
+	bra e #ih_no_fifo
+		// queue incoming fifo command for later processing
+		mov $r11 0x1900
+		mov $r13 #cmd_queue
+		iord $r14 I[$r11 + 0x100]	// FIFO_CMD
+		iord $r15 I[$r11 + 0x000]	// FIFO_DATA
+		call #queue_put
+		add b32 $r11 0x400
+		mov $r14 1
+		iowr I[$r11 + 0x000] $r14	// FIFO_ACK
+
+	// context switch request?
+	ih_no_fifo:
+	and $r11 $r10 0x00000100
+	bra e #ih_no_ctxsw
+		// enqueue a context switch for later processing
+		mov $r13 #cmd_queue
+		mov $r14 0x4001
+		call #queue_put
+
+	// anything we didn't handle, bring it to the host's attention
+	ih_no_ctxsw:
+	mov $r11 0x104
+	not b32 $r11
+	and $r11 $r10 $r11
+	bra e #ih_no_other
+		mov $r10 0xc1c
+		shl b32 $r10 6
+		iowr I[$r10] $r11	// INTR_UP_SET
+
+	// ack, and wake up main()
+	ih_no_other:
+	iowr I[$r0 + 0x100] $r10	// INTR_ACK
+
+	pop $r15
+	pop $r14
+	pop $r13
+	pop $r11
+	pop $r10
+	pop $r9
+	pop $r8
+	mov $flags $r8
+	pop $r8
+	bclr $flags $p0
+	iret
+
+// Not real sure, but, MEM_CMD 7 will hang forever if this isn't done
+ctx_4160s:
+	mov $r14 0x4160
+	sethi $r14 0x400000
+	mov $r15 1
+	call #nv_wr32
+	ctx_4160s_wait:
+		call #nv_rd32
+		xbit $r15 $r15 4
+		bra e #ctx_4160s_wait
+	ret
+
+// Without clearing again at end of xfer, some things cause PGRAPH
+// to hang with STATUS=0x00000007 until it's cleared.. fbcon can
+// still function with it set however...
+ctx_4160c:
+	mov $r14 0x4160
+	sethi $r14 0x400000
+	clear b32 $r15
+	call #nv_wr32
+	ret
+
+// Again, not real sure
+//
+// In: $r15 value to set 0x404170 to
+//
+ctx_4170s:
+	mov $r14 0x4170
+	sethi $r14 0x400000
+	or $r15 0x10
+	call #nv_wr32
+	ret
+
+// Waits for a ctx_4170s() call to complete
+//
+ctx_4170w:
+	mov $r14 0x4170
+	sethi $r14 0x400000
+	call #nv_rd32
+	and $r15 0x10
+	bra ne #ctx_4170w
+	ret
+
+// Disables various things, waits a bit, and re-enables them..
+//
+// Not sure how exactly this helps, perhaps "ENABLE" is not such a
+// good description for the bits we turn off?  Anyways, without this,
+// funny things happen.
+//
+ctx_redswitch:
+	mov $r14 0x614
+	shl b32 $r14 6
+	mov $r15 0x270
+	iowr I[$r14] $r15	// HUB_RED_SWITCH = ENABLE_GPC, POWER_ALL
+	mov $r15 8
+	ctx_redswitch_delay:
+		sub b32 $r15 1
+		bra ne #ctx_redswitch_delay
+	mov $r15 0x770
+	iowr I[$r14] $r15	// HUB_RED_SWITCH = ENABLE_ALL, POWER_ALL
+	ret
+
+// Not a clue what this is for, except that unless the value is 0x10, the
+// strand context is saved (and presumably restored) incorrectly..
+//
+// In: $r15 value to set to (0x00/0x10 are used)
+//
+ctx_86c:
+	mov $r14 0x86c
+	shl b32 $r14 6
+	iowr I[$r14] $r15	// HUB(0x86c) = val
+	mov $r14 -0x75ec
+	sethi $r14 0x400000
+	call #nv_wr32		// ROP(0xa14) = val
+	mov $r14 -0x5794
+	sethi $r14 0x410000
+	call #nv_wr32		// GPC(0x86c) = val
+	ret
+
+// ctx_load - load's a channel's ctxctl data, and selects its vm
+//
+// In: $r2 channel address
+//
+ctx_load:
+	trace_set(T_CHAN)
+
+	// switch to channel, somewhat magic in parts..
+	mov $r10 12		// DONE_UNK12
+	call #wait_donez
+	mov $r1 0xa24
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r0	// 0x409a24
+	mov $r3 0xb00
+	shl b32 $r3 6
+	iowr I[$r3 + 0x100] $r2	// CHAN_NEXT
+	mov $r1 0xa0c
+	shl b32 $r1 6
+	mov $r4 7
+	iowr I[$r1 + 0x000] $r2 // MEM_CHAN
+	iowr I[$r1 + 0x100] $r4	// MEM_CMD
+	ctx_chan_wait_0:
+		iord $r4 I[$r1 + 0x100]
+		and $r4 0x1f
+		bra ne #ctx_chan_wait_0
+	iowr I[$r3 + 0x000] $r2	// CHAN_CUR
+
+	// load channel header, fetch PGRAPH context pointer
+	mov $xtargets $r0
+	bclr $r2 31
+	shl b32 $r2 4
+	add b32 $r2 2
+
+	trace_set(T_LCHAN)
+	mov $r1 0xa04
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r2		// MEM_BASE
+	mov $r1 0xa20
+	shl b32 $r1 6
+	mov $r2 0x0002
+	sethi $r2 0x80000000
+	iowr I[$r1 + 0x000] $r2		// MEM_TARGET = vram
+	mov $r1 0x10			// chan + 0x0210
+	mov $r2 #xfer_data
+	sethi $r2 0x00020000		// 16 bytes
+	xdld $r1 $r2
+	xdwait
+	trace_clr(T_LCHAN)
+
+	// update current context
+	ld b32 $r1 D[$r0 + #xfer_data + 4]
+	shl b32 $r1 24
+	ld b32 $r2 D[$r0 + #xfer_data + 0]
+	shr b32 $r2 8
+	or $r1 $r2
+	st b32 D[$r0 + #ctx_current] $r1
+
+	// set transfer base to start of context, and fetch context header
+	trace_set(T_LCTXH)
+	mov $r2 0xa04
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r1		// MEM_BASE
+	mov $r2 1
+	mov $r1 0xa20
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r2		// MEM_TARGET = vm
+	mov $r1 #chan_data
+	sethi $r1 0x00060000		// 256 bytes
+	xdld $r0 $r1
+	xdwait
+	trace_clr(T_LCTXH)
+
+	trace_clr(T_CHAN)
+	ret
+
+// ctx_chan - handler for HUB_SET_CHAN command, will set a channel as
+//            the active channel for ctxctl, but not actually transfer
+//            any context data.  intended for use only during initial
+//            context construction.
+//
+// In: $r2 channel address
+//
+ctx_chan:
+	call #ctx_4160s
+	call #ctx_load
+	mov $r10 12			// DONE_UNK12
+	call #wait_donez
+	mov $r1 0xa10
+	shl b32 $r1 6
+	mov $r2 5
+	iowr I[$r1 + 0x000] $r2		// MEM_CMD = 5 (???)
+	ctx_chan_wait:
+		iord $r2 I[$r1 + 0x000]
+		or $r2 $r2
+		bra ne #ctx_chan_wait
+	call #ctx_4160c
+	ret
+
+// Execute per-context state overrides list
+//
+// Only executed on the first load of a channel.  Might want to look into
+// removing this and having the host directly modify the channel's context
+// to change this state...  The nouveau DRM already builds this list as
+// it's definitely needed for NVIDIA's, so we may as well use it for now
+//
+// Input: $r1 mmio list length
+//
+ctx_mmio_exec:
+	// set transfer base to be the mmio list
+	ld b32 $r3 D[$r0 + #chan_mmio_address]
+	mov $r2 0xa04
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r3		// MEM_BASE
+
+	clear b32 $r3
+	ctx_mmio_loop:
+		// fetch next 256 bytes of mmio list if necessary
+		and $r4 $r3 0xff
+		bra ne #ctx_mmio_pull
+			mov $r5 #xfer_data
+			sethi $r5 0x00060000	// 256 bytes
+			xdld $r3 $r5
+			xdwait
+
+		// execute a single list entry
+		ctx_mmio_pull:
+		ld b32 $r14 D[$r4 + #xfer_data + 0x00]
+		ld b32 $r15 D[$r4 + #xfer_data + 0x04]
+		call #nv_wr32
+
+		// next!
+		add b32 $r3 8
+		sub b32 $r1 1
+		bra ne #ctx_mmio_loop
+
+	// set transfer base back to the current context
+	ctx_mmio_done:
+	ld b32 $r3 D[$r0 + #ctx_current]
+	iowr I[$r2 + 0x000] $r3		// MEM_BASE
+
+	// disable the mmio list now, we don't need/want to execute it again
+	st b32 D[$r0 + #chan_mmio_count] $r0
+	mov $r1 #chan_data
+	sethi $r1 0x00060000		// 256 bytes
+	xdst $r0 $r1
+	xdwait
+	ret
+
+// Transfer HUB context data between GPU and storage area
+//
+// In: $r2 channel address
+//     $p1 clear on save, set on load
+//     $p2 set if opposite direction done/will be done, so:
+//		on save it means: "a load will follow this save"
+//		on load it means: "a save preceeded this load"
+//
+ctx_xfer:
+	// according to mwk, some kind of wait for idle
+	mov $r15 0xc00
+	shl b32 $r15 6
+	mov $r14 4
+	iowr I[$r15 + 0x200] $r14
+	ctx_xfer_idle:
+		iord $r14 I[$r15 + 0x000]
+		and $r14 0x2000
+		bra ne #ctx_xfer_idle
+
+	bra not $p1 #ctx_xfer_pre
+	bra $p2 #ctx_xfer_pre_load
+	ctx_xfer_pre:
+		mov $r15 0x10
+		call #ctx_86c
+		call #ctx_4160s
+		bra not $p1 #ctx_xfer_exec
+
+	ctx_xfer_pre_load:
+		mov $r15 2
+		call #ctx_4170s
+		call #ctx_4170w
+		call #ctx_redswitch
+		clear b32 $r15
+		call #ctx_4170s
+		call #ctx_load
+
+	// fetch context pointer, and initiate xfer on all GPCs
+	ctx_xfer_exec:
+	ld b32 $r1 D[$r0 + #ctx_current]
+	mov $r2 0x414
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r0	// BAR_STATUS = reset
+	mov $r14 -0x5b00
+	sethi $r14 0x410000
+	mov b32 $r15 $r1
+	call #nv_wr32		// GPC_BCAST_WRCMD_DATA = ctx pointer
+	add b32 $r14 4
+	xbit $r15 $flags $p1
+	xbit $r2 $flags $p2
+	shl b32 $r2 1
+	or $r15 $r2
+	call #nv_wr32		// GPC_BCAST_WRCMD_CMD = GPC_XFER(type)
+
+	// strands
+	mov $r1 0x4afc
+	sethi $r1 0x20000
+	mov $r2 0xc
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x0c
+	call #strand_wait
+	mov $r2 0x47fc
+	sethi $r2 0x20000
+	iowr I[$r2] $r0		// STRAND_FIRST_GENE(0x3f) = 0x00
+	xbit $r2 $flags $p1
+	add b32 $r2 3
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x03/0x04 (SAVE/LOAD)
+
+	// mmio context
+	xbit $r10 $flags $p1	// direction
+	or $r10 6		// first, last
+	mov $r11 0		// base = 0
+	ld b32 $r12 D[$r0 + #hub_mmio_list_head]
+	ld b32 $r13 D[$r0 + #hub_mmio_list_tail]
+	mov $r14 0		// not multi
+	call #mmctx_xfer
+
+	// wait for GPCs to all complete
+	mov $r10 8		// DONE_BAR
+	call #wait_doneo
+
+	// wait for strand xfer to complete
+	call #strand_wait
+
+	// post-op
+	bra $p1 #ctx_xfer_post
+		mov $r10 12		// DONE_UNK12
+		call #wait_donez
+		mov $r1 0xa10
+		shl b32 $r1 6
+		mov $r2 5
+		iowr I[$r1] $r2		// MEM_CMD
+		ctx_xfer_post_save_wait:
+			iord $r2 I[$r1]
+			or $r2 $r2
+			bra ne #ctx_xfer_post_save_wait
+
+	bra $p2 #ctx_xfer_done
+	ctx_xfer_post:
+		mov $r15 2
+		call #ctx_4170s
+		clear b32 $r15
+		call #ctx_86c
+		call #strand_post
+		call #ctx_4170w
+		clear b32 $r15
+		call #ctx_4170s
+
+		bra not $p1 #ctx_xfer_no_post_mmio
+		ld b32 $r1 D[$r0 + #chan_mmio_count]
+		or $r1 $r1
+		bra e #ctx_xfer_no_post_mmio
+			call #ctx_mmio_exec
+
+		ctx_xfer_no_post_mmio:
+		call #ctx_4160c
+
+	ctx_xfer_done:
+	ret
+
+.align 256
diff --git a/nvkm/engine/graph/fuc/hubnve0.fuc b/nvkm/engine/graph/fuc/hubnve0.fuc
new file mode 100644
index 000000000..7fe9d7cf4
--- /dev/null
+++ b/nvkm/engine/graph/fuc/hubnve0.fuc
@@ -0,0 +1,793 @@
+/* fuc microcode for nve0 PGRAPH/HUB
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+/* To build:
+ *    m4 nve0_grhub.fuc | envyas -a -w -m fuc -V nva3 -o nve0_grhub.fuc.h
+ */
+
+.section #nve0_grhub_data
+include(`nve0.fuc')
+gpc_count:		.b32 0
+rop_count:		.b32 0
+cmd_queue:		queue_init
+hub_mmio_list_head:	.b32 0
+hub_mmio_list_tail:	.b32 0
+
+ctx_current:		.b32 0
+
+chipsets:
+.b8  0xe4 0 0 0
+.b16 #nve4_hub_mmio_head
+.b16 #nve4_hub_mmio_tail
+.b8  0xe7 0 0 0
+.b16 #nve4_hub_mmio_head
+.b16 #nve4_hub_mmio_tail
+.b8  0xe6 0 0 0
+.b16 #nve4_hub_mmio_head
+.b16 #nve4_hub_mmio_tail
+.b8  0 0 0 0
+
+nve4_hub_mmio_head:
+mmctx_data(0x17e91c, 2)
+mmctx_data(0x400204, 2)
+mmctx_data(0x404010, 7)
+mmctx_data(0x4040a8, 9)
+mmctx_data(0x4040d0, 7)
+mmctx_data(0x4040f8, 1)
+mmctx_data(0x404130, 3)
+mmctx_data(0x404150, 3)
+mmctx_data(0x404164, 1)
+mmctx_data(0x4041a0, 4)
+mmctx_data(0x404200, 4)
+mmctx_data(0x404404, 14)
+mmctx_data(0x404460, 4)
+mmctx_data(0x404480, 1)
+mmctx_data(0x404498, 1)
+mmctx_data(0x404604, 4)
+mmctx_data(0x404618, 4)
+mmctx_data(0x40462c, 2)
+mmctx_data(0x404640, 1)
+mmctx_data(0x404654, 1)
+mmctx_data(0x404660, 1)
+mmctx_data(0x404678, 19)
+mmctx_data(0x4046c8, 3)
+mmctx_data(0x404700, 3)
+mmctx_data(0x404718, 10)
+mmctx_data(0x404744, 2)
+mmctx_data(0x404754, 1)
+mmctx_data(0x405800, 1)
+mmctx_data(0x405830, 3)
+mmctx_data(0x405854, 1)
+mmctx_data(0x405870, 4)
+mmctx_data(0x405a00, 2)
+mmctx_data(0x405a18, 1)
+mmctx_data(0x405b00, 1)
+mmctx_data(0x405b10, 1)
+mmctx_data(0x406020, 1)
+mmctx_data(0x406028, 4)
+mmctx_data(0x4064a8, 2)
+mmctx_data(0x4064b4, 2)
+mmctx_data(0x4064c0, 12)
+mmctx_data(0x4064fc, 1)
+mmctx_data(0x407040, 1)
+mmctx_data(0x407804, 1)
+mmctx_data(0x40780c, 6)
+mmctx_data(0x4078bc, 1)
+mmctx_data(0x408000, 7)
+mmctx_data(0x408064, 1)
+mmctx_data(0x408800, 3)
+mmctx_data(0x408840, 1)
+mmctx_data(0x408900, 3)
+mmctx_data(0x408980, 1)
+nve4_hub_mmio_tail:
+
+.align 256
+chan_data:
+chan_mmio_count:	.b32 0
+chan_mmio_address:	.b32 0
+
+.align 256
+xfer_data: 		.b32 0
+
+.section #nve0_grhub_code
+bra #init
+define(`include_code')
+include(`nve0.fuc')
+
+// reports an exception to the host
+//
+// In: $r15 error code (see nve0.fuc)
+//
+error:
+	push $r14
+	mov $r14 0x814
+	shl b32 $r14 6
+	iowr I[$r14 + 0x000] $r15	// CC_SCRATCH[5] = error code
+	mov $r14 0xc1c
+	shl b32 $r14 6
+	mov $r15 1
+	iowr I[$r14 + 0x000] $r15	// INTR_UP_SET
+	pop $r14
+	ret
+
+// HUB fuc initialisation, executed by triggering ucode start, will
+// fall through to main loop after completion.
+//
+// Input:
+//   CC_SCRATCH[0]: chipset (PMC_BOOT_0 read returns 0x0bad0bad... sigh)
+//
+// Output:
+//   CC_SCRATCH[0]:
+//	     31:31: set to signal completion
+//   CC_SCRATCH[1]:
+//	      31:0: total PGRAPH context size
+//
+init:
+	clear b32 $r0
+	mov $sp $r0
+	mov $xdbase $r0
+
+	// enable fifo access
+	mov $r1 0x1200
+	mov $r2 2
+	iowr I[$r1 + 0x000] $r2	// FIFO_ENABLE
+
+	// setup i0 handler, and route all interrupts to it
+	mov $r1 #ih
+	mov $iv0 $r1
+	mov $r1 0x400
+	iowr I[$r1 + 0x300] $r0	// INTR_DISPATCH
+
+	// route HUB_CHANNEL_SWITCH to fuc interrupt 8
+	mov $r3 0x404
+	shl b32 $r3 6
+	mov $r2 0x2003		// { HUB_CHANNEL_SWITCH, ZERO } -> intr 8
+	iowr I[$r3 + 0x000] $r2
+
+	// not sure what these are, route them because NVIDIA does, and
+	// the IRQ handler will signal the host if we ever get one.. we
+	// may find out if/why we need to handle these if so..
+	//
+	mov $r2 0x2004
+	iowr I[$r3 + 0x004] $r2 // { 0x04, ZERO } -> intr 9
+	mov $r2 0x200b
+	iowr I[$r3 + 0x008] $r2 // { 0x0b, ZERO } -> intr 10
+	mov $r2 0x200c
+	iowr I[$r3 + 0x01c] $r2 // { 0x0c, ZERO } -> intr 15
+
+	// enable all INTR_UP interrupts
+	mov $r2 0xc24
+	shl b32 $r2 6
+	not b32 $r3 $r0
+	iowr I[$r2] $r3
+
+	// enable fifo, ctxsw, 9, 10, 15 interrupts
+	mov $r2 -0x78fc		// 0x8704
+	sethi $r2 0
+	iowr I[$r1 + 0x000] $r2	// INTR_EN_SET
+
+	// fifo level triggered, rest edge
+	sub b32 $r1 0x100
+	mov $r2 4
+	iowr I[$r1] $r2
+
+	// enable interrupts
+	bset $flags ie0
+
+	// fetch enabled GPC/ROP counts
+	mov $r14 -0x69fc	// 0x409604
+	sethi $r14 0x400000
+	call #nv_rd32
+	extr $r1 $r15 16:20
+	st b32 D[$r0 + #rop_count] $r1
+	and $r15 0x1f
+	st b32 D[$r0 + #gpc_count] $r15
+
+	// set BAR_REQMASK to GPC mask
+	mov $r1 1
+	shl b32 $r1 $r15
+	sub b32 $r1 1
+	mov $r2 0x40c
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r1
+	iowr I[$r2 + 0x100] $r1
+
+	// find context data for this chipset
+	mov $r2 0x800
+	shl b32 $r2 6
+	iord $r2 I[$r2 + 0x000]		// CC_SCRATCH[0]
+	mov $r15 #chipsets - 8
+	init_find_chipset:
+		add b32 $r15 8
+		ld b32 $r3 D[$r15 + 0x00]
+		cmpu b32 $r3 $r2
+		bra e #init_context
+		cmpu b32 $r3 0
+		bra ne #init_find_chipset
+		// unknown chipset
+		ret
+
+	// context size calculation, reserve first 256 bytes for use by fuc
+	init_context:
+	mov $r1 256
+
+	// calculate size of mmio context data
+	ld b16 $r14 D[$r15 + 4]
+	ld b16 $r15 D[$r15 + 6]
+	sethi $r14 0
+	st b32 D[$r0 + #hub_mmio_list_head] $r14
+	st b32 D[$r0 + #hub_mmio_list_tail] $r15
+	call #mmctx_size
+
+	// set mmctx base addresses now so we don't have to do it later,
+	// they don't (currently) ever change
+	mov $r3 0x700
+	shl b32 $r3 6
+	shr b32 $r4 $r1 8
+	iowr I[$r3 + 0x000] $r4		// MMCTX_SAVE_SWBASE
+	iowr I[$r3 + 0x100] $r4		// MMCTX_LOAD_SWBASE
+	add b32 $r3 0x1300
+	add b32 $r1 $r15
+	shr b32 $r15 2
+	iowr I[$r3 + 0x000] $r15	// MMCTX_LOAD_COUNT, wtf for?!?
+
+	// strands, base offset needs to be aligned to 256 bytes
+	shr b32 $r1 8
+	add b32 $r1 1
+	shl b32 $r1 8
+	mov b32 $r15 $r1
+	call #strand_ctx_init
+	add b32 $r1 $r15
+
+	// initialise each GPC in sequence by passing in the offset of its
+	// context data in GPCn_CC_SCRATCH[1], and starting its FUC (which
+	// has previously been uploaded by the host) running.
+	//
+	// the GPC fuc init sequence will set GPCn_CC_SCRATCH[0] bit 31
+	// when it has completed, and return the size of its context data
+	// in GPCn_CC_SCRATCH[1]
+	//
+	ld b32 $r3 D[$r0 + #gpc_count]
+	mov $r4 0x2000
+	sethi $r4 0x500000
+	init_gpc:
+		// setup, and start GPC ucode running
+		add b32 $r14 $r4 0x804
+		mov b32 $r15 $r1
+		call #nv_wr32			// CC_SCRATCH[1] = ctx offset
+		add b32 $r14 $r4 0x800
+		mov b32 $r15 $r2
+		call #nv_wr32			// CC_SCRATCH[0] = chipset
+		add b32 $r14 $r4 0x10c
+		clear b32 $r15
+		call #nv_wr32
+		add b32 $r14 $r4 0x104
+		call #nv_wr32			// ENTRY
+		add b32 $r14 $r4 0x100
+		mov $r15 2			// CTRL_START_TRIGGER
+		call #nv_wr32			// CTRL
+
+		// wait for it to complete, and adjust context size
+		add b32 $r14 $r4 0x800
+		init_gpc_wait:
+			call #nv_rd32
+			xbit $r15 $r15 31
+			bra e #init_gpc_wait
+		add b32 $r14 $r4 0x804
+		call #nv_rd32
+		add b32 $r1 $r15
+
+		// next!
+		add b32 $r4 0x8000
+		sub b32 $r3 1
+		bra ne #init_gpc
+
+	// save context size, and tell host we're ready
+	mov $r2 0x800
+	shl b32 $r2 6
+	iowr I[$r2 + 0x100] $r1		// CC_SCRATCH[1]  = context size
+	add b32 $r2 0x800
+	clear b32 $r1
+	bset $r1 31
+	iowr I[$r2 + 0x000] $r1		// CC_SCRATCH[0] |= 0x80000000
+
+// Main program loop, very simple, sleeps until woken up by the interrupt
+// handler, pulls a command from the queue and executes its handler
+//
+main:
+	// sleep until we have something to do
+	bset $flags $p0
+	sleep $p0
+	mov $r13 #cmd_queue
+	call #queue_get
+	bra $p1 #main
+
+	// context switch, requested by GPU?
+	cmpu b32 $r14 0x4001
+	bra ne #main_not_ctx_switch
+		trace_set(T_AUTO)
+		mov $r1 0xb00
+		shl b32 $r1 6
+		iord $r2 I[$r1 + 0x100]		// CHAN_NEXT
+		iord $r1 I[$r1 + 0x000]		// CHAN_CUR
+
+		xbit $r3 $r1 31
+		bra e #chsw_no_prev
+			xbit $r3 $r2 31
+			bra e #chsw_prev_no_next
+				push $r2
+				mov b32 $r2 $r1
+				trace_set(T_SAVE)
+				bclr $flags $p1
+				bset $flags $p2
+				call #ctx_xfer
+				trace_clr(T_SAVE);
+				pop $r2
+				trace_set(T_LOAD);
+				bset $flags $p1
+				call #ctx_xfer
+				trace_clr(T_LOAD);
+				bra #chsw_done
+			chsw_prev_no_next:
+				push $r2
+				mov b32 $r2 $r1
+				bclr $flags $p1
+				bclr $flags $p2
+				call #ctx_xfer
+				pop $r2
+				mov $r1 0xb00
+				shl b32 $r1 6
+				iowr I[$r1] $r2
+				bra #chsw_done
+		chsw_no_prev:
+			xbit $r3 $r2 31
+			bra e #chsw_done
+				bset $flags $p1
+				bclr $flags $p2
+				call #ctx_xfer
+
+		// ack the context switch request
+		chsw_done:
+		mov $r1 0xb0c
+		shl b32 $r1 6
+		mov $r2 1
+		iowr I[$r1 + 0x000] $r2		// 0x409b0c
+		trace_clr(T_AUTO)
+		bra #main
+
+	// request to set current channel? (*not* a context switch)
+	main_not_ctx_switch:
+	cmpu b32 $r14 0x0001
+	bra ne #main_not_ctx_chan
+		mov b32 $r2 $r15
+		call #ctx_chan
+		bra #main_done
+
+	// request to store current channel context?
+	main_not_ctx_chan:
+	cmpu b32 $r14 0x0002
+	bra ne #main_not_ctx_save
+		trace_set(T_SAVE)
+		bclr $flags $p1
+		bclr $flags $p2
+		call #ctx_xfer
+		trace_clr(T_SAVE)
+		bra #main_done
+
+	main_not_ctx_save:
+		shl b32 $r15 $r14 16
+		or $r15 E_BAD_COMMAND
+		call #error
+		bra #main
+
+	main_done:
+	mov $r1 0x820
+	shl b32 $r1 6
+	clear b32 $r2
+	bset $r2 31
+	iowr I[$r1 + 0x000] $r2		// CC_SCRATCH[0] |= 0x80000000
+	bra #main
+
+// interrupt handler
+ih:
+	push $r8
+	mov $r8 $flags
+	push $r8
+	push $r9
+	push $r10
+	push $r11
+	push $r13
+	push $r14
+	push $r15
+
+	// incoming fifo command?
+	iord $r10 I[$r0 + 0x200]	// INTR
+	and $r11 $r10 0x00000004
+	bra e #ih_no_fifo
+		// queue incoming fifo command for later processing
+		mov $r11 0x1900
+		mov $r13 #cmd_queue
+		iord $r14 I[$r11 + 0x100]	// FIFO_CMD
+		iord $r15 I[$r11 + 0x000]	// FIFO_DATA
+		call #queue_put
+		add b32 $r11 0x400
+		mov $r14 1
+		iowr I[$r11 + 0x000] $r14	// FIFO_ACK
+
+	// context switch request?
+	ih_no_fifo:
+	and $r11 $r10 0x00000100
+	bra e #ih_no_ctxsw
+		// enqueue a context switch for later processing
+		mov $r13 #cmd_queue
+		mov $r14 0x4001
+		call #queue_put
+
+	// anything we didn't handle, bring it to the host's attention
+	ih_no_ctxsw:
+	mov $r11 0x104
+	not b32 $r11
+	and $r11 $r10 $r11
+	bra e #ih_no_other
+		mov $r10 0xc1c
+		shl b32 $r10 6
+		iowr I[$r10] $r11	// INTR_UP_SET
+
+	// ack, and wake up main()
+	ih_no_other:
+	iowr I[$r0 + 0x100] $r10	// INTR_ACK
+
+	pop $r15
+	pop $r14
+	pop $r13
+	pop $r11
+	pop $r10
+	pop $r9
+	pop $r8
+	mov $flags $r8
+	pop $r8
+	bclr $flags $p0
+	iret
+
+// Again, not real sure
+//
+// In: $r15 value to set 0x404170 to
+//
+ctx_4170s:
+	mov $r14 0x4170
+	sethi $r14 0x400000
+	or $r15 0x10
+	call #nv_wr32
+	ret
+
+// Waits for a ctx_4170s() call to complete
+//
+ctx_4170w:
+	mov $r14 0x4170
+	sethi $r14 0x400000
+	call #nv_rd32
+	and $r15 0x10
+	bra ne #ctx_4170w
+	ret
+
+// Disables various things, waits a bit, and re-enables them..
+//
+// Not sure how exactly this helps, perhaps "ENABLE" is not such a
+// good description for the bits we turn off?  Anyways, without this,
+// funny things happen.
+//
+ctx_redswitch:
+	mov $r14 0x614
+	shl b32 $r14 6
+	mov $r15 0x270
+	iowr I[$r14] $r15	// HUB_RED_SWITCH = ENABLE_GPC, POWER_ALL
+	mov $r15 8
+	ctx_redswitch_delay:
+		sub b32 $r15 1
+		bra ne #ctx_redswitch_delay
+	mov $r15 0x770
+	iowr I[$r14] $r15	// HUB_RED_SWITCH = ENABLE_ALL, POWER_ALL
+	ret
+
+// Not a clue what this is for, except that unless the value is 0x10, the
+// strand context is saved (and presumably restored) incorrectly..
+//
+// In: $r15 value to set to (0x00/0x10 are used)
+//
+ctx_86c:
+	mov $r14 0x86c
+	shl b32 $r14 6
+	iowr I[$r14] $r15	// HUB(0x86c) = val
+	mov $r14 -0x75ec
+	sethi $r14 0x400000
+	call #nv_wr32		// ROP(0xa14) = val
+	mov $r14 -0x5794
+	sethi $r14 0x410000
+	call #nv_wr32		// GPC(0x86c) = val
+	ret
+
+// ctx_load - load's a channel's ctxctl data, and selects its vm
+//
+// In: $r2 channel address
+//
+ctx_load:
+	trace_set(T_CHAN)
+
+	// switch to channel, somewhat magic in parts..
+	mov $r10 12		// DONE_UNK12
+	call #wait_donez
+	mov $r1 0xa24
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r0	// 0x409a24
+	mov $r3 0xb00
+	shl b32 $r3 6
+	iowr I[$r3 + 0x100] $r2	// CHAN_NEXT
+	mov $r1 0xa0c
+	shl b32 $r1 6
+	mov $r4 7
+	iowr I[$r1 + 0x000] $r2 // MEM_CHAN
+	iowr I[$r1 + 0x100] $r4	// MEM_CMD
+	ctx_chan_wait_0:
+		iord $r4 I[$r1 + 0x100]
+		and $r4 0x1f
+		bra ne #ctx_chan_wait_0
+	iowr I[$r3 + 0x000] $r2	// CHAN_CUR
+
+	// load channel header, fetch PGRAPH context pointer
+	mov $xtargets $r0
+	bclr $r2 31
+	shl b32 $r2 4
+	add b32 $r2 2
+
+	trace_set(T_LCHAN)
+	mov $r1 0xa04
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r2		// MEM_BASE
+	mov $r1 0xa20
+	shl b32 $r1 6
+	mov $r2 0x0002
+	sethi $r2 0x80000000
+	iowr I[$r1 + 0x000] $r2		// MEM_TARGET = vram
+	mov $r1 0x10			// chan + 0x0210
+	mov $r2 #xfer_data
+	sethi $r2 0x00020000		// 16 bytes
+	xdld $r1 $r2
+	xdwait
+	trace_clr(T_LCHAN)
+
+	// update current context
+	ld b32 $r1 D[$r0 + #xfer_data + 4]
+	shl b32 $r1 24
+	ld b32 $r2 D[$r0 + #xfer_data + 0]
+	shr b32 $r2 8
+	or $r1 $r2
+	st b32 D[$r0 + #ctx_current] $r1
+
+	// set transfer base to start of context, and fetch context header
+	trace_set(T_LCTXH)
+	mov $r2 0xa04
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r1		// MEM_BASE
+	mov $r2 1
+	mov $r1 0xa20
+	shl b32 $r1 6
+	iowr I[$r1 + 0x000] $r2		// MEM_TARGET = vm
+	mov $r1 #chan_data
+	sethi $r1 0x00060000		// 256 bytes
+	xdld $r0 $r1
+	xdwait
+	trace_clr(T_LCTXH)
+
+	trace_clr(T_CHAN)
+	ret
+
+// ctx_chan - handler for HUB_SET_CHAN command, will set a channel as
+//            the active channel for ctxctl, but not actually transfer
+//            any context data.  intended for use only during initial
+//            context construction.
+//
+// In: $r2 channel address
+//
+ctx_chan:
+	call #ctx_load
+	mov $r10 12			// DONE_UNK12
+	call #wait_donez
+	mov $r1 0xa10
+	shl b32 $r1 6
+	mov $r2 5
+	iowr I[$r1 + 0x000] $r2		// MEM_CMD = 5 (???)
+	ctx_chan_wait:
+		iord $r2 I[$r1 + 0x000]
+		or $r2 $r2
+		bra ne #ctx_chan_wait
+	ret
+
+// Execute per-context state overrides list
+//
+// Only executed on the first load of a channel.  Might want to look into
+// removing this and having the host directly modify the channel's context
+// to change this state...  The nouveau DRM already builds this list as
+// it's definitely needed for NVIDIA's, so we may as well use it for now
+//
+// Input: $r1 mmio list length
+//
+ctx_mmio_exec:
+	// set transfer base to be the mmio list
+	ld b32 $r3 D[$r0 + #chan_mmio_address]
+	mov $r2 0xa04
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r3		// MEM_BASE
+
+	clear b32 $r3
+	ctx_mmio_loop:
+		// fetch next 256 bytes of mmio list if necessary
+		and $r4 $r3 0xff
+		bra ne #ctx_mmio_pull
+			mov $r5 #xfer_data
+			sethi $r5 0x00060000	// 256 bytes
+			xdld $r3 $r5
+			xdwait
+
+		// execute a single list entry
+		ctx_mmio_pull:
+		ld b32 $r14 D[$r4 + #xfer_data + 0x00]
+		ld b32 $r15 D[$r4 + #xfer_data + 0x04]
+		call #nv_wr32
+
+		// next!
+		add b32 $r3 8
+		sub b32 $r1 1
+		bra ne #ctx_mmio_loop
+
+	// set transfer base back to the current context
+	ctx_mmio_done:
+	ld b32 $r3 D[$r0 + #ctx_current]
+	iowr I[$r2 + 0x000] $r3		// MEM_BASE
+
+	// disable the mmio list now, we don't need/want to execute it again
+	st b32 D[$r0 + #chan_mmio_count] $r0
+	mov $r1 #chan_data
+	sethi $r1 0x00060000		// 256 bytes
+	xdst $r0 $r1
+	xdwait
+	ret
+
+// Transfer HUB context data between GPU and storage area
+//
+// In: $r2 channel address
+//     $p1 clear on save, set on load
+//     $p2 set if opposite direction done/will be done, so:
+//		on save it means: "a load will follow this save"
+//		on load it means: "a save preceeded this load"
+//
+ctx_xfer:
+	// according to mwk, some kind of wait for idle
+	mov $r15 0xc00
+	shl b32 $r15 6
+	mov $r14 4
+	iowr I[$r15 + 0x200] $r14
+	ctx_xfer_idle:
+		iord $r14 I[$r15 + 0x000]
+		and $r14 0x2000
+		bra ne #ctx_xfer_idle
+
+	bra not $p1 #ctx_xfer_pre
+	bra $p2 #ctx_xfer_pre_load
+	ctx_xfer_pre:
+		mov $r15 0x10
+		call #ctx_86c
+		bra not $p1 #ctx_xfer_exec
+
+	ctx_xfer_pre_load:
+		mov $r15 2
+		call #ctx_4170s
+		call #ctx_4170w
+		call #ctx_redswitch
+		clear b32 $r15
+		call #ctx_4170s
+		call #ctx_load
+
+	// fetch context pointer, and initiate xfer on all GPCs
+	ctx_xfer_exec:
+	ld b32 $r1 D[$r0 + #ctx_current]
+	mov $r2 0x414
+	shl b32 $r2 6
+	iowr I[$r2 + 0x000] $r0	// BAR_STATUS = reset
+	mov $r14 -0x5b00
+	sethi $r14 0x410000
+	mov b32 $r15 $r1
+	call #nv_wr32		// GPC_BCAST_WRCMD_DATA = ctx pointer
+	add b32 $r14 4
+	xbit $r15 $flags $p1
+	xbit $r2 $flags $p2
+	shl b32 $r2 1
+	or $r15 $r2
+	call #nv_wr32		// GPC_BCAST_WRCMD_CMD = GPC_XFER(type)
+
+	// strands
+	mov $r1 0x4afc
+	sethi $r1 0x20000
+	mov $r2 0xc
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x0c
+	call #strand_wait
+	mov $r2 0x47fc
+	sethi $r2 0x20000
+	iowr I[$r2] $r0		// STRAND_FIRST_GENE(0x3f) = 0x00
+	xbit $r2 $flags $p1
+	add b32 $r2 3
+	iowr I[$r1] $r2		// STRAND_CMD(0x3f) = 0x03/0x04 (SAVE/LOAD)
+
+	// mmio context
+	xbit $r10 $flags $p1	// direction
+	or $r10 6		// first, last
+	mov $r11 0		// base = 0
+	ld b32 $r12 D[$r0 + #hub_mmio_list_head]
+	ld b32 $r13 D[$r0 + #hub_mmio_list_tail]
+	mov $r14 0		// not multi
+	call #mmctx_xfer
+
+	// wait for GPCs to all complete
+	mov $r10 8		// DONE_BAR
+	call #wait_doneo
+
+	// wait for strand xfer to complete
+	call #strand_wait
+
+	// post-op
+	bra $p1 #ctx_xfer_post
+		mov $r10 12		// DONE_UNK12
+		call #wait_donez
+		mov $r1 0xa10
+		shl b32 $r1 6
+		mov $r2 5
+		iowr I[$r1] $r2		// MEM_CMD
+		ctx_xfer_post_save_wait:
+			iord $r2 I[$r1]
+			or $r2 $r2
+			bra ne #ctx_xfer_post_save_wait
+
+	bra $p2 #ctx_xfer_done
+	ctx_xfer_post:
+		mov $r15 2
+		call #ctx_4170s
+		clear b32 $r15
+		call #ctx_86c
+		call #strand_post
+		call #ctx_4170w
+		clear b32 $r15
+		call #ctx_4170s
+
+		bra not $p1 #ctx_xfer_no_post_mmio
+		ld b32 $r1 D[$r0 + #chan_mmio_count]
+		or $r1 $r1
+		bra e #ctx_xfer_no_post_mmio
+			call #ctx_mmio_exec
+
+		ctx_xfer_no_post_mmio:
+
+	ctx_xfer_done:
+	ret
+
+.align 256
diff --git a/nvkm/engine/graph/fuc/nvc0.fuc b/nvkm/engine/graph/fuc/nvc0.fuc
new file mode 100644
index 000000000..e6b228844
--- /dev/null
+++ b/nvkm/engine/graph/fuc/nvc0.fuc
@@ -0,0 +1,400 @@
+/* fuc microcode util functions for nvc0 PGRAPH
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+define(`mmctx_data', `.b32 eval((($2 - 1) << 26) | $1)')
+define(`queue_init', `.skip eval((2 * 4) + ((8 * 4) * 2))')
+
+ifdef(`include_code', `
+// Error codes
+define(`E_BAD_COMMAND', 0x01)
+define(`E_CMD_OVERFLOW', 0x02)
+
+// Util macros to help with debugging ucode hangs etc
+define(`T_WAIT', 0)
+define(`T_MMCTX', 1)
+define(`T_STRWAIT', 2)
+define(`T_STRINIT', 3)
+define(`T_AUTO', 4)
+define(`T_CHAN', 5)
+define(`T_LOAD', 6)
+define(`T_SAVE', 7)
+define(`T_LCHAN', 8)
+define(`T_LCTXH', 9)
+
+define(`trace_set', `
+	mov $r8 0x83c
+	shl b32 $r8 6
+	clear b32 $r9
+	bset $r9 $1
+	iowr I[$r8 + 0x000] $r9		// CC_SCRATCH[7]
+')
+
+define(`trace_clr', `
+	mov $r8 0x85c
+	shl b32 $r8 6
+	clear b32 $r9
+	bset $r9 $1
+	iowr I[$r8 + 0x000] $r9		// CC_SCRATCH[7]
+')
+
+// queue_put - add request to queue
+//
+// In : $r13 queue pointer
+//	$r14 command
+//	$r15 data
+//
+queue_put:
+	// make sure we have space..
+	ld b32 $r8 D[$r13 + 0x0]	// GET
+	ld b32 $r9 D[$r13 + 0x4]	// PUT
+	xor $r8 8
+	cmpu b32 $r8 $r9
+	bra ne #queue_put_next
+		mov $r15 E_CMD_OVERFLOW
+		call #error
+		ret
+
+	// store cmd/data on queue
+	queue_put_next:
+	and $r8 $r9 7
+	shl b32 $r8 3
+	add b32 $r8 $r13
+	add b32 $r8 8
+	st b32 D[$r8 + 0x0] $r14
+	st b32 D[$r8 + 0x4] $r15
+
+	// update PUT
+	add b32 $r9 1
+	and $r9 0xf
+	st b32 D[$r13 + 0x4] $r9
+	ret
+
+// queue_get - fetch request from queue
+//
+// In : $r13 queue pointer
+//
+// Out:	$p1  clear on success (data available)
+//	$r14 command
+// 	$r15 data
+//
+queue_get:
+	bset $flags $p1
+	ld b32 $r8 D[$r13 + 0x0]	// GET
+	ld b32 $r9 D[$r13 + 0x4]	// PUT
+	cmpu b32 $r8 $r9
+	bra e #queue_get_done
+		// fetch first cmd/data pair
+		and $r9 $r8 7
+		shl b32 $r9 3
+		add b32 $r9 $r13
+		add b32 $r9 8
+		ld b32 $r14 D[$r9 + 0x0]
+		ld b32 $r15 D[$r9 + 0x4]
+
+		// update GET
+		add b32 $r8 1
+		and $r8 0xf
+		st b32 D[$r13 + 0x0] $r8
+		bclr $flags $p1
+queue_get_done:
+	ret
+
+// nv_rd32 - read 32-bit value from nv register
+//
+// In : $r14 register
+// Out: $r15 value
+//
+nv_rd32:
+	mov $r11 0x728
+	shl b32 $r11 6
+	mov b32 $r12 $r14
+	bset $r12 31			// MMIO_CTRL_PENDING
+	iowr I[$r11 + 0x000] $r12	// MMIO_CTRL
+	nv_rd32_wait:
+		iord $r12 I[$r11 + 0x000]
+		xbit $r12 $r12 31
+		bra ne #nv_rd32_wait
+	mov $r10 6			// DONE_MMIO_RD
+	call #wait_doneo
+	iord $r15 I[$r11 + 0x100]	// MMIO_RDVAL
+	ret
+
+// nv_wr32 - write 32-bit value to nv register
+//
+// In : $r14 register
+//      $r15 value
+//
+nv_wr32:
+	mov $r11 0x728
+	shl b32 $r11 6
+	iowr I[$r11 + 0x200] $r15	// MMIO_WRVAL
+	mov b32 $r12 $r14
+	bset $r12 31			// MMIO_CTRL_PENDING
+	bset $r12 30			// MMIO_CTRL_WRITE
+	iowr I[$r11 + 0x000] $r12	// MMIO_CTRL
+	nv_wr32_wait:
+		iord $r12 I[$r11 + 0x000]
+		xbit $r12 $r12 31
+		bra ne #nv_wr32_wait
+	ret
+
+// (re)set watchdog timer
+//
+// In : $r15 timeout
+//
+watchdog_reset:
+	mov $r8 0x430
+	shl b32 $r8 6
+	bset $r15 31
+	iowr I[$r8 + 0x000] $r15
+	ret
+
+// clear watchdog timer
+watchdog_clear:
+	mov $r8 0x430
+	shl b32 $r8 6
+	iowr I[$r8 + 0x000] $r0
+	ret
+
+// wait_done{z,o} - wait on FUC_DONE bit to become clear/set
+//
+// In : $r10 bit to wait on
+//
+define(`wait_done', `
+$1:
+	trace_set(T_WAIT);
+	mov $r8 0x818
+	shl b32 $r8 6
+	iowr I[$r8 + 0x000] $r10	// CC_SCRATCH[6] = wait bit
+	wait_done_$1:
+		mov $r8 0x400
+		shl b32 $r8 6
+		iord $r8 I[$r8 + 0x000]	// DONE
+		xbit $r8 $r8 $r10
+		bra $2 #wait_done_$1
+	trace_clr(T_WAIT)
+	ret
+')
+wait_done(wait_donez, ne)
+wait_done(wait_doneo, e)
+
+// mmctx_size - determine size of a mmio list transfer
+//
+// In : $r14 mmio list head
+//      $r15 mmio list tail
+// Out: $r15 transfer size (in bytes)
+//
+mmctx_size:
+	clear b32 $r9
+	nv_mmctx_size_loop:
+		ld b32 $r8 D[$r14]
+		shr b32 $r8 26
+		add b32 $r8 1
+		shl b32 $r8 2
+		add b32 $r9 $r8
+		add b32 $r14 4
+		cmpu b32 $r14 $r15
+		bra ne #nv_mmctx_size_loop
+	mov b32 $r15 $r9
+	ret
+
+// mmctx_xfer - execute a list of mmio transfers
+//
+// In : $r10 flags
+//		bit 0: direction (0 = save, 1 = load)
+//		bit 1: set if first transfer
+//		bit 2: set if last transfer
+//	$r11 base
+//	$r12 mmio list head
+//	$r13 mmio list tail
+//	$r14 multi_stride
+//	$r15 multi_mask
+//
+mmctx_xfer:
+	trace_set(T_MMCTX)
+	mov $r8 0x710
+	shl b32 $r8 6
+	clear b32 $r9
+	or $r11 $r11
+	bra e #mmctx_base_disabled
+		iowr I[$r8 + 0x000] $r11	// MMCTX_BASE
+		bset $r9 0			// BASE_EN
+	mmctx_base_disabled:
+	or $r14 $r14
+	bra e #mmctx_multi_disabled
+		iowr I[$r8 + 0x200] $r14 	// MMCTX_MULTI_STRIDE
+		iowr I[$r8 + 0x300] $r15 	// MMCTX_MULTI_MASK
+		bset $r9 1			// MULTI_EN
+	mmctx_multi_disabled:
+	add b32 $r8 0x100
+
+	xbit $r11 $r10 0
+	shl b32 $r11 16			// DIR
+	bset $r11 12			// QLIMIT = 0x10
+	xbit $r14 $r10 1
+	shl b32 $r14 17
+	or $r11 $r14			// START_TRIGGER
+	iowr I[$r8 + 0x000] $r11	// MMCTX_CTRL
+
+	// loop over the mmio list, and send requests to the hw
+	mmctx_exec_loop:
+		// wait for space in mmctx queue
+		mmctx_wait_free:
+			iord $r14 I[$r8 + 0x000] // MMCTX_CTRL
+			and $r14 0x1f
+			bra e #mmctx_wait_free
+
+		// queue up an entry
+		ld b32 $r14 D[$r12]
+		or $r14 $r9
+		iowr I[$r8 + 0x300] $r14
+		add b32 $r12 4
+		cmpu b32 $r12 $r13
+		bra ne #mmctx_exec_loop
+
+	xbit $r11 $r10 2
+	bra ne #mmctx_stop
+		// wait for queue to empty
+		mmctx_fini_wait:
+			iord $r11 I[$r8 + 0x000]	// MMCTX_CTRL
+			and $r11 0x1f
+			cmpu b32 $r11 0x10
+			bra ne #mmctx_fini_wait
+		mov $r10 2				// DONE_MMCTX
+		call #wait_donez
+		bra #mmctx_done
+	mmctx_stop:
+		xbit $r11 $r10 0
+		shl b32 $r11 16			// DIR
+		bset $r11 12			// QLIMIT = 0x10
+		bset $r11 18			// STOP_TRIGGER
+		iowr I[$r8 + 0x000] $r11	// MMCTX_CTRL
+		mmctx_stop_wait:
+			// wait for STOP_TRIGGER to clear
+			iord $r11 I[$r8 + 0x000] // MMCTX_CTRL
+			xbit $r11 $r11 18
+			bra ne #mmctx_stop_wait
+	mmctx_done:
+	trace_clr(T_MMCTX)
+	ret
+
+// Wait for DONE_STRAND
+//
+strand_wait:
+	push $r10
+	mov $r10 2
+	call #wait_donez
+	pop $r10
+	ret
+
+// unknown - call before issuing strand commands
+//
+strand_pre:
+	mov $r8 0x4afc
+	sethi $r8 0x20000
+	mov $r9 0xc
+	iowr I[$r8] $r9
+	call #strand_wait
+	ret
+
+// unknown - call after issuing strand commands
+//
+strand_post:
+	mov $r8 0x4afc
+	sethi $r8 0x20000
+	mov $r9 0xd
+	iowr I[$r8] $r9
+	call #strand_wait
+	ret
+
+// Selects strand set?!
+//
+// In: $r14 id
+//
+strand_set:
+	mov $r10 0x4ffc
+	sethi $r10 0x20000
+	sub b32 $r11 $r10 0x500
+	mov $r12 0xf
+	iowr I[$r10 + 0x000] $r12		// 0x93c = 0xf
+	mov $r12 0xb
+	iowr I[$r11 + 0x000] $r12		// 0x928 = 0xb
+	call #strand_wait
+	iowr I[$r10 + 0x000] $r14		// 0x93c = <id>
+	mov $r12 0xa
+	iowr I[$r11 + 0x000] $r12		// 0x928 = 0xa
+	call #strand_wait
+	ret
+
+// Initialise strand context data
+//
+// In : $r15 context base
+// Out: $r15 context size (in bytes)
+//
+// Strandset(?) 3 hardcoded currently
+//
+strand_ctx_init:
+	trace_set(T_STRINIT)
+	call #strand_pre
+	mov $r14 3
+	call #strand_set
+	mov $r10 0x46fc
+	sethi $r10 0x20000
+	add b32 $r11 $r10 0x400
+	iowr I[$r10 + 0x100] $r0	// STRAND_FIRST_GENE = 0
+	mov $r12 1
+	iowr I[$r11 + 0x000] $r12	// STRAND_CMD = LATCH_FIRST_GENE
+	call #strand_wait
+	sub b32 $r12 $r0 1
+	iowr I[$r10 + 0x000] $r12	// STRAND_GENE_CNT = 0xffffffff
+	mov $r12 2
+	iowr I[$r11 + 0x000] $r12	// STRAND_CMD = LATCH_GENE_CNT
+	call #strand_wait
+	call #strand_post
+
+	// read the size of each strand, poke the context offset of
+	// each into STRAND_{SAVE,LOAD}_SWBASE now, no need to worry
+	// about it later then.
+	mov $r8 0x880
+	shl b32 $r8 6
+	iord $r9 I[$r8 + 0x000]		// STRANDS
+	add b32 $r8 0x2200
+	shr b32 $r14 $r15 8
+	ctx_init_strand_loop:
+		iowr I[$r8 + 0x000] $r14	// STRAND_SAVE_SWBASE
+		iowr I[$r8 + 0x100] $r14	// STRAND_LOAD_SWBASE
+		iord $r10 I[$r8 + 0x200]	// STRAND_SIZE
+		shr b32 $r10 6
+		add b32 $r10 1
+		add b32 $r14 $r10
+		add b32 $r8 4
+		sub b32 $r9 1
+		bra ne #ctx_init_strand_loop
+
+	shl b32 $r14 8
+	sub b32 $r15 $r14 $r15
+	trace_clr(T_STRINIT)
+	ret
+')
diff --git a/nvkm/engine/graph/fuc/nve0.fuc b/nvkm/engine/graph/fuc/nve0.fuc
new file mode 100644
index 000000000..f16a5d533
--- /dev/null
+++ b/nvkm/engine/graph/fuc/nve0.fuc
@@ -0,0 +1,400 @@
+/* fuc microcode util functions for nve0 PGRAPH
+ *
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+define(`mmctx_data', `.b32 eval((($2 - 1) << 26) | $1)')
+define(`queue_init', `.skip eval((2 * 4) + ((8 * 4) * 2))')
+
+ifdef(`include_code', `
+// Error codes
+define(`E_BAD_COMMAND', 0x01)
+define(`E_CMD_OVERFLOW', 0x02)
+
+// Util macros to help with debugging ucode hangs etc
+define(`T_WAIT', 0)
+define(`T_MMCTX', 1)
+define(`T_STRWAIT', 2)
+define(`T_STRINIT', 3)
+define(`T_AUTO', 4)
+define(`T_CHAN', 5)
+define(`T_LOAD', 6)
+define(`T_SAVE', 7)
+define(`T_LCHAN', 8)
+define(`T_LCTXH', 9)
+
+define(`trace_set', `
+	mov $r8 0x83c
+	shl b32 $r8 6
+	clear b32 $r9
+	bset $r9 $1
+	iowr I[$r8 + 0x000] $r9		// CC_SCRATCH[7]
+')
+
+define(`trace_clr', `
+	mov $r8 0x85c
+	shl b32 $r8 6
+	clear b32 $r9
+	bset $r9 $1
+	iowr I[$r8 + 0x000] $r9		// CC_SCRATCH[7]
+')
+
+// queue_put - add request to queue
+//
+// In : $r13 queue pointer
+//	$r14 command
+//	$r15 data
+//
+queue_put:
+	// make sure we have space..
+	ld b32 $r8 D[$r13 + 0x0]	// GET
+	ld b32 $r9 D[$r13 + 0x4]	// PUT
+	xor $r8 8
+	cmpu b32 $r8 $r9
+	bra ne #queue_put_next
+		mov $r15 E_CMD_OVERFLOW
+		call #error
+		ret
+
+	// store cmd/data on queue
+	queue_put_next:
+	and $r8 $r9 7
+	shl b32 $r8 3
+	add b32 $r8 $r13
+	add b32 $r8 8
+	st b32 D[$r8 + 0x0] $r14
+	st b32 D[$r8 + 0x4] $r15
+
+	// update PUT
+	add b32 $r9 1
+	and $r9 0xf
+	st b32 D[$r13 + 0x4] $r9
+	ret
+
+// queue_get - fetch request from queue
+//
+// In : $r13 queue pointer
+//
+// Out:	$p1  clear on success (data available)
+//	$r14 command
+// 	$r15 data
+//
+queue_get:
+	bset $flags $p1
+	ld b32 $r8 D[$r13 + 0x0]	// GET
+	ld b32 $r9 D[$r13 + 0x4]	// PUT
+	cmpu b32 $r8 $r9
+	bra e #queue_get_done
+		// fetch first cmd/data pair
+		and $r9 $r8 7
+		shl b32 $r9 3
+		add b32 $r9 $r13
+		add b32 $r9 8
+		ld b32 $r14 D[$r9 + 0x0]
+		ld b32 $r15 D[$r9 + 0x4]
+
+		// update GET
+		add b32 $r8 1
+		and $r8 0xf
+		st b32 D[$r13 + 0x0] $r8
+		bclr $flags $p1
+queue_get_done:
+	ret
+
+// nv_rd32 - read 32-bit value from nv register
+//
+// In : $r14 register
+// Out: $r15 value
+//
+nv_rd32:
+	mov $r11 0x728
+	shl b32 $r11 6
+	mov b32 $r12 $r14
+	bset $r12 31			// MMIO_CTRL_PENDING
+	iowr I[$r11 + 0x000] $r12	// MMIO_CTRL
+	nv_rd32_wait:
+		iord $r12 I[$r11 + 0x000]
+		xbit $r12 $r12 31
+		bra ne #nv_rd32_wait
+	mov $r10 6			// DONE_MMIO_RD
+	call #wait_doneo
+	iord $r15 I[$r11 + 0x100]	// MMIO_RDVAL
+	ret
+
+// nv_wr32 - write 32-bit value to nv register
+//
+// In : $r14 register
+//      $r15 value
+//
+nv_wr32:
+	mov $r11 0x728
+	shl b32 $r11 6
+	iowr I[$r11 + 0x200] $r15	// MMIO_WRVAL
+	mov b32 $r12 $r14
+	bset $r12 31			// MMIO_CTRL_PENDING
+	bset $r12 30			// MMIO_CTRL_WRITE
+	iowr I[$r11 + 0x000] $r12	// MMIO_CTRL
+	nv_wr32_wait:
+		iord $r12 I[$r11 + 0x000]
+		xbit $r12 $r12 31
+		bra ne #nv_wr32_wait
+	ret
+
+// (re)set watchdog timer
+//
+// In : $r15 timeout
+//
+watchdog_reset:
+	mov $r8 0x430
+	shl b32 $r8 6
+	bset $r15 31
+	iowr I[$r8 + 0x000] $r15
+	ret
+
+// clear watchdog timer
+watchdog_clear:
+	mov $r8 0x430
+	shl b32 $r8 6
+	iowr I[$r8 + 0x000] $r0
+	ret
+
+// wait_done{z,o} - wait on FUC_DONE bit to become clear/set
+//
+// In : $r10 bit to wait on
+//
+define(`wait_done', `
+$1:
+	trace_set(T_WAIT);
+	mov $r8 0x818
+	shl b32 $r8 6
+	iowr I[$r8 + 0x000] $r10	// CC_SCRATCH[6] = wait bit
+	wait_done_$1:
+		mov $r8 0x400
+		shl b32 $r8 6
+		iord $r8 I[$r8 + 0x000]	// DONE
+		xbit $r8 $r8 $r10
+		bra $2 #wait_done_$1
+	trace_clr(T_WAIT)
+	ret
+')
+wait_done(wait_donez, ne)
+wait_done(wait_doneo, e)
+
+// mmctx_size - determine size of a mmio list transfer
+//
+// In : $r14 mmio list head
+//      $r15 mmio list tail
+// Out: $r15 transfer size (in bytes)
+//
+mmctx_size:
+	clear b32 $r9
+	nv_mmctx_size_loop:
+		ld b32 $r8 D[$r14]
+		shr b32 $r8 26
+		add b32 $r8 1
+		shl b32 $r8 2
+		add b32 $r9 $r8
+		add b32 $r14 4
+		cmpu b32 $r14 $r15
+		bra ne #nv_mmctx_size_loop
+	mov b32 $r15 $r9
+	ret
+
+// mmctx_xfer - execute a list of mmio transfers
+//
+// In : $r10 flags
+//		bit 0: direction (0 = save, 1 = load)
+//		bit 1: set if first transfer
+//		bit 2: set if last transfer
+//	$r11 base
+//	$r12 mmio list head
+//	$r13 mmio list tail
+//	$r14 multi_stride
+//	$r15 multi_mask
+//
+mmctx_xfer:
+	trace_set(T_MMCTX)
+	mov $r8 0x710
+	shl b32 $r8 6
+	clear b32 $r9
+	or $r11 $r11
+	bra e #mmctx_base_disabled
+		iowr I[$r8 + 0x000] $r11	// MMCTX_BASE
+		bset $r9 0			// BASE_EN
+	mmctx_base_disabled:
+	or $r14 $r14
+	bra e #mmctx_multi_disabled
+		iowr I[$r8 + 0x200] $r14 	// MMCTX_MULTI_STRIDE
+		iowr I[$r8 + 0x300] $r15 	// MMCTX_MULTI_MASK
+		bset $r9 1			// MULTI_EN
+	mmctx_multi_disabled:
+	add b32 $r8 0x100
+
+	xbit $r11 $r10 0
+	shl b32 $r11 16			// DIR
+	bset $r11 12			// QLIMIT = 0x10
+	xbit $r14 $r10 1
+	shl b32 $r14 17
+	or $r11 $r14			// START_TRIGGER
+	iowr I[$r8 + 0x000] $r11	// MMCTX_CTRL
+
+	// loop over the mmio list, and send requests to the hw
+	mmctx_exec_loop:
+		// wait for space in mmctx queue
+		mmctx_wait_free:
+			iord $r14 I[$r8 + 0x000] // MMCTX_CTRL
+			and $r14 0x1f
+			bra e #mmctx_wait_free
+
+		// queue up an entry
+		ld b32 $r14 D[$r12]
+		or $r14 $r9
+		iowr I[$r8 + 0x300] $r14
+		add b32 $r12 4
+		cmpu b32 $r12 $r13
+		bra ne #mmctx_exec_loop
+
+	xbit $r11 $r10 2
+	bra ne #mmctx_stop
+		// wait for queue to empty
+		mmctx_fini_wait:
+			iord $r11 I[$r8 + 0x000]	// MMCTX_CTRL
+			and $r11 0x1f
+			cmpu b32 $r11 0x10
+			bra ne #mmctx_fini_wait
+		mov $r10 2				// DONE_MMCTX
+		call #wait_donez
+		bra #mmctx_done
+	mmctx_stop:
+		xbit $r11 $r10 0
+		shl b32 $r11 16			// DIR
+		bset $r11 12			// QLIMIT = 0x10
+		bset $r11 18			// STOP_TRIGGER
+		iowr I[$r8 + 0x000] $r11	// MMCTX_CTRL
+		mmctx_stop_wait:
+			// wait for STOP_TRIGGER to clear
+			iord $r11 I[$r8 + 0x000] // MMCTX_CTRL
+			xbit $r11 $r11 18
+			bra ne #mmctx_stop_wait
+	mmctx_done:
+	trace_clr(T_MMCTX)
+	ret
+
+// Wait for DONE_STRAND
+//
+strand_wait:
+	push $r10
+	mov $r10 2
+	call #wait_donez
+	pop $r10
+	ret
+
+// unknown - call before issuing strand commands
+//
+strand_pre:
+	mov $r8 0x4afc
+	sethi $r8 0x20000
+	mov $r9 0xc
+	iowr I[$r8] $r9
+	call #strand_wait
+	ret
+
+// unknown - call after issuing strand commands
+//
+strand_post:
+	mov $r8 0x4afc
+	sethi $r8 0x20000
+	mov $r9 0xd
+	iowr I[$r8] $r9
+	call #strand_wait
+	ret
+
+// Selects strand set?!
+//
+// In: $r14 id
+//
+strand_set:
+	mov $r10 0x4ffc
+	sethi $r10 0x20000
+	sub b32 $r11 $r10 0x500
+	mov $r12 0xf
+	iowr I[$r10 + 0x000] $r12		// 0x93c = 0xf
+	mov $r12 0xb
+	iowr I[$r11 + 0x000] $r12		// 0x928 = 0xb
+	call #strand_wait
+	iowr I[$r10 + 0x000] $r14		// 0x93c = <id>
+	mov $r12 0xa
+	iowr I[$r11 + 0x000] $r12		// 0x928 = 0xa
+	call #strand_wait
+	ret
+
+// Initialise strand context data
+//
+// In : $r15 context base
+// Out: $r15 context size (in bytes)
+//
+// Strandset(?) 3 hardcoded currently
+//
+strand_ctx_init:
+	trace_set(T_STRINIT)
+	call #strand_pre
+	mov $r14 3
+	call #strand_set
+	mov $r10 0x46fc
+	sethi $r10 0x20000
+	add b32 $r11 $r10 0x400
+	iowr I[$r10 + 0x100] $r0	// STRAND_FIRST_GENE = 0
+	mov $r12 1
+	iowr I[$r11 + 0x000] $r12	// STRAND_CMD = LATCH_FIRST_GENE
+	call #strand_wait
+	sub b32 $r12 $r0 1
+	iowr I[$r10 + 0x000] $r12	// STRAND_GENE_CNT = 0xffffffff
+	mov $r12 2
+	iowr I[$r11 + 0x000] $r12	// STRAND_CMD = LATCH_GENE_CNT
+	call #strand_wait
+	call #strand_post
+
+	// read the size of each strand, poke the context offset of
+	// each into STRAND_{SAVE,LOAD}_SWBASE now, no need to worry
+	// about it later then.
+	mov $r8 0x880
+	shl b32 $r8 6
+	iord $r9 I[$r8 + 0x000]		// STRANDS
+	add b32 $r8 0x2200
+	shr b32 $r14 $r15 8
+	ctx_init_strand_loop:
+		iowr I[$r8 + 0x000] $r14	// STRAND_SAVE_SWBASE
+		iowr I[$r8 + 0x100] $r14	// STRAND_LOAD_SWBASE
+		iord $r10 I[$r8 + 0x200]	// STRAND_SIZE
+		shr b32 $r10 6
+		add b32 $r10 1
+		add b32 $r14 $r10
+		add b32 $r8 4
+		sub b32 $r9 1
+		bra ne #ctx_init_strand_loop
+
+	shl b32 $r14 8
+	sub b32 $r15 $r14 $r15
+	trace_clr(T_STRINIT)
+	ret
+')
diff --git a/nvkm/engine/graph/nv04.c b/nvkm/engine/graph/nv04.c
new file mode 100644
index 000000000..e30a9c5ff
--- /dev/null
+++ b/nvkm/engine/graph/nv04.c
@@ -0,0 +1,1387 @@
+/*
+ * Copyright 2007 Stephane Marchesin
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/handle.h>
+#include <core/namedb.h>
+
+#include <subdev/fb.h>
+#include <subdev/instmem.h>
+#include <subdev/timer.h>
+
+#include <engine/fifo.h>
+#include <engine/graph.h>
+
+#include "regs.h"
+
+static u32
+nv04_graph_ctx_regs[] = {
+	0x0040053c,
+	0x00400544,
+	0x00400540,
+	0x00400548,
+	NV04_PGRAPH_CTX_SWITCH1,
+	NV04_PGRAPH_CTX_SWITCH2,
+	NV04_PGRAPH_CTX_SWITCH3,
+	NV04_PGRAPH_CTX_SWITCH4,
+	NV04_PGRAPH_CTX_CACHE1,
+	NV04_PGRAPH_CTX_CACHE2,
+	NV04_PGRAPH_CTX_CACHE3,
+	NV04_PGRAPH_CTX_CACHE4,
+	0x00400184,
+	0x004001a4,
+	0x004001c4,
+	0x004001e4,
+	0x00400188,
+	0x004001a8,
+	0x004001c8,
+	0x004001e8,
+	0x0040018c,
+	0x004001ac,
+	0x004001cc,
+	0x004001ec,
+	0x00400190,
+	0x004001b0,
+	0x004001d0,
+	0x004001f0,
+	0x00400194,
+	0x004001b4,
+	0x004001d4,
+	0x004001f4,
+	0x00400198,
+	0x004001b8,
+	0x004001d8,
+	0x004001f8,
+	0x0040019c,
+	0x004001bc,
+	0x004001dc,
+	0x004001fc,
+	0x00400174,
+	NV04_PGRAPH_DMA_START_0,
+	NV04_PGRAPH_DMA_START_1,
+	NV04_PGRAPH_DMA_LENGTH,
+	NV04_PGRAPH_DMA_MISC,
+	NV04_PGRAPH_DMA_PITCH,
+	NV04_PGRAPH_BOFFSET0,
+	NV04_PGRAPH_BBASE0,
+	NV04_PGRAPH_BLIMIT0,
+	NV04_PGRAPH_BOFFSET1,
+	NV04_PGRAPH_BBASE1,
+	NV04_PGRAPH_BLIMIT1,
+	NV04_PGRAPH_BOFFSET2,
+	NV04_PGRAPH_BBASE2,
+	NV04_PGRAPH_BLIMIT2,
+	NV04_PGRAPH_BOFFSET3,
+	NV04_PGRAPH_BBASE3,
+	NV04_PGRAPH_BLIMIT3,
+	NV04_PGRAPH_BOFFSET4,
+	NV04_PGRAPH_BBASE4,
+	NV04_PGRAPH_BLIMIT4,
+	NV04_PGRAPH_BOFFSET5,
+	NV04_PGRAPH_BBASE5,
+	NV04_PGRAPH_BLIMIT5,
+	NV04_PGRAPH_BPITCH0,
+	NV04_PGRAPH_BPITCH1,
+	NV04_PGRAPH_BPITCH2,
+	NV04_PGRAPH_BPITCH3,
+	NV04_PGRAPH_BPITCH4,
+	NV04_PGRAPH_SURFACE,
+	NV04_PGRAPH_STATE,
+	NV04_PGRAPH_BSWIZZLE2,
+	NV04_PGRAPH_BSWIZZLE5,
+	NV04_PGRAPH_BPIXEL,
+	NV04_PGRAPH_NOTIFY,
+	NV04_PGRAPH_PATT_COLOR0,
+	NV04_PGRAPH_PATT_COLOR1,
+	NV04_PGRAPH_PATT_COLORRAM+0x00,
+	NV04_PGRAPH_PATT_COLORRAM+0x04,
+	NV04_PGRAPH_PATT_COLORRAM+0x08,
+	NV04_PGRAPH_PATT_COLORRAM+0x0c,
+	NV04_PGRAPH_PATT_COLORRAM+0x10,
+	NV04_PGRAPH_PATT_COLORRAM+0x14,
+	NV04_PGRAPH_PATT_COLORRAM+0x18,
+	NV04_PGRAPH_PATT_COLORRAM+0x1c,
+	NV04_PGRAPH_PATT_COLORRAM+0x20,
+	NV04_PGRAPH_PATT_COLORRAM+0x24,
+	NV04_PGRAPH_PATT_COLORRAM+0x28,
+	NV04_PGRAPH_PATT_COLORRAM+0x2c,
+	NV04_PGRAPH_PATT_COLORRAM+0x30,
+	NV04_PGRAPH_PATT_COLORRAM+0x34,
+	NV04_PGRAPH_PATT_COLORRAM+0x38,
+	NV04_PGRAPH_PATT_COLORRAM+0x3c,
+	NV04_PGRAPH_PATT_COLORRAM+0x40,
+	NV04_PGRAPH_PATT_COLORRAM+0x44,
+	NV04_PGRAPH_PATT_COLORRAM+0x48,
+	NV04_PGRAPH_PATT_COLORRAM+0x4c,
+	NV04_PGRAPH_PATT_COLORRAM+0x50,
+	NV04_PGRAPH_PATT_COLORRAM+0x54,
+	NV04_PGRAPH_PATT_COLORRAM+0x58,
+	NV04_PGRAPH_PATT_COLORRAM+0x5c,
+	NV04_PGRAPH_PATT_COLORRAM+0x60,
+	NV04_PGRAPH_PATT_COLORRAM+0x64,
+	NV04_PGRAPH_PATT_COLORRAM+0x68,
+	NV04_PGRAPH_PATT_COLORRAM+0x6c,
+	NV04_PGRAPH_PATT_COLORRAM+0x70,
+	NV04_PGRAPH_PATT_COLORRAM+0x74,
+	NV04_PGRAPH_PATT_COLORRAM+0x78,
+	NV04_PGRAPH_PATT_COLORRAM+0x7c,
+	NV04_PGRAPH_PATT_COLORRAM+0x80,
+	NV04_PGRAPH_PATT_COLORRAM+0x84,
+	NV04_PGRAPH_PATT_COLORRAM+0x88,
+	NV04_PGRAPH_PATT_COLORRAM+0x8c,
+	NV04_PGRAPH_PATT_COLORRAM+0x90,
+	NV04_PGRAPH_PATT_COLORRAM+0x94,
+	NV04_PGRAPH_PATT_COLORRAM+0x98,
+	NV04_PGRAPH_PATT_COLORRAM+0x9c,
+	NV04_PGRAPH_PATT_COLORRAM+0xa0,
+	NV04_PGRAPH_PATT_COLORRAM+0xa4,
+	NV04_PGRAPH_PATT_COLORRAM+0xa8,
+	NV04_PGRAPH_PATT_COLORRAM+0xac,
+	NV04_PGRAPH_PATT_COLORRAM+0xb0,
+	NV04_PGRAPH_PATT_COLORRAM+0xb4,
+	NV04_PGRAPH_PATT_COLORRAM+0xb8,
+	NV04_PGRAPH_PATT_COLORRAM+0xbc,
+	NV04_PGRAPH_PATT_COLORRAM+0xc0,
+	NV04_PGRAPH_PATT_COLORRAM+0xc4,
+	NV04_PGRAPH_PATT_COLORRAM+0xc8,
+	NV04_PGRAPH_PATT_COLORRAM+0xcc,
+	NV04_PGRAPH_PATT_COLORRAM+0xd0,
+	NV04_PGRAPH_PATT_COLORRAM+0xd4,
+	NV04_PGRAPH_PATT_COLORRAM+0xd8,
+	NV04_PGRAPH_PATT_COLORRAM+0xdc,
+	NV04_PGRAPH_PATT_COLORRAM+0xe0,
+	NV04_PGRAPH_PATT_COLORRAM+0xe4,
+	NV04_PGRAPH_PATT_COLORRAM+0xe8,
+	NV04_PGRAPH_PATT_COLORRAM+0xec,
+	NV04_PGRAPH_PATT_COLORRAM+0xf0,
+	NV04_PGRAPH_PATT_COLORRAM+0xf4,
+	NV04_PGRAPH_PATT_COLORRAM+0xf8,
+	NV04_PGRAPH_PATT_COLORRAM+0xfc,
+	NV04_PGRAPH_PATTERN,
+	0x0040080c,
+	NV04_PGRAPH_PATTERN_SHAPE,
+	0x00400600,
+	NV04_PGRAPH_ROP3,
+	NV04_PGRAPH_CHROMA,
+	NV04_PGRAPH_BETA_AND,
+	NV04_PGRAPH_BETA_PREMULT,
+	NV04_PGRAPH_CONTROL0,
+	NV04_PGRAPH_CONTROL1,
+	NV04_PGRAPH_CONTROL2,
+	NV04_PGRAPH_BLEND,
+	NV04_PGRAPH_STORED_FMT,
+	NV04_PGRAPH_SOURCE_COLOR,
+	0x00400560,
+	0x00400568,
+	0x00400564,
+	0x0040056c,
+	0x00400400,
+	0x00400480,
+	0x00400404,
+	0x00400484,
+	0x00400408,
+	0x00400488,
+	0x0040040c,
+	0x0040048c,
+	0x00400410,
+	0x00400490,
+	0x00400414,
+	0x00400494,
+	0x00400418,
+	0x00400498,
+	0x0040041c,
+	0x0040049c,
+	0x00400420,
+	0x004004a0,
+	0x00400424,
+	0x004004a4,
+	0x00400428,
+	0x004004a8,
+	0x0040042c,
+	0x004004ac,
+	0x00400430,
+	0x004004b0,
+	0x00400434,
+	0x004004b4,
+	0x00400438,
+	0x004004b8,
+	0x0040043c,
+	0x004004bc,
+	0x00400440,
+	0x004004c0,
+	0x00400444,
+	0x004004c4,
+	0x00400448,
+	0x004004c8,
+	0x0040044c,
+	0x004004cc,
+	0x00400450,
+	0x004004d0,
+	0x00400454,
+	0x004004d4,
+	0x00400458,
+	0x004004d8,
+	0x0040045c,
+	0x004004dc,
+	0x00400460,
+	0x004004e0,
+	0x00400464,
+	0x004004e4,
+	0x00400468,
+	0x004004e8,
+	0x0040046c,
+	0x004004ec,
+	0x00400470,
+	0x004004f0,
+	0x00400474,
+	0x004004f4,
+	0x00400478,
+	0x004004f8,
+	0x0040047c,
+	0x004004fc,
+	0x00400534,
+	0x00400538,
+	0x00400514,
+	0x00400518,
+	0x0040051c,
+	0x00400520,
+	0x00400524,
+	0x00400528,
+	0x0040052c,
+	0x00400530,
+	0x00400d00,
+	0x00400d40,
+	0x00400d80,
+	0x00400d04,
+	0x00400d44,
+	0x00400d84,
+	0x00400d08,
+	0x00400d48,
+	0x00400d88,
+	0x00400d0c,
+	0x00400d4c,
+	0x00400d8c,
+	0x00400d10,
+	0x00400d50,
+	0x00400d90,
+	0x00400d14,
+	0x00400d54,
+	0x00400d94,
+	0x00400d18,
+	0x00400d58,
+	0x00400d98,
+	0x00400d1c,
+	0x00400d5c,
+	0x00400d9c,
+	0x00400d20,
+	0x00400d60,
+	0x00400da0,
+	0x00400d24,
+	0x00400d64,
+	0x00400da4,
+	0x00400d28,
+	0x00400d68,
+	0x00400da8,
+	0x00400d2c,
+	0x00400d6c,
+	0x00400dac,
+	0x00400d30,
+	0x00400d70,
+	0x00400db0,
+	0x00400d34,
+	0x00400d74,
+	0x00400db4,
+	0x00400d38,
+	0x00400d78,
+	0x00400db8,
+	0x00400d3c,
+	0x00400d7c,
+	0x00400dbc,
+	0x00400590,
+	0x00400594,
+	0x00400598,
+	0x0040059c,
+	0x004005a8,
+	0x004005ac,
+	0x004005b0,
+	0x004005b4,
+	0x004005c0,
+	0x004005c4,
+	0x004005c8,
+	0x004005cc,
+	0x004005d0,
+	0x004005d4,
+	0x004005d8,
+	0x004005dc,
+	0x004005e0,
+	NV04_PGRAPH_PASSTHRU_0,
+	NV04_PGRAPH_PASSTHRU_1,
+	NV04_PGRAPH_PASSTHRU_2,
+	NV04_PGRAPH_DVD_COLORFMT,
+	NV04_PGRAPH_SCALED_FORMAT,
+	NV04_PGRAPH_MISC24_0,
+	NV04_PGRAPH_MISC24_1,
+	NV04_PGRAPH_MISC24_2,
+	0x00400500,
+	0x00400504,
+	NV04_PGRAPH_VALID1,
+	NV04_PGRAPH_VALID2,
+	NV04_PGRAPH_DEBUG_3
+};
+
+struct nv04_graph_priv {
+	struct nouveau_graph base;
+	struct nv04_graph_chan *chan[16];
+	spinlock_t lock;
+};
+
+struct nv04_graph_chan {
+	struct nouveau_object base;
+	int chid;
+	u32 nv04[ARRAY_SIZE(nv04_graph_ctx_regs)];
+};
+
+
+static inline struct nv04_graph_priv *
+nv04_graph_priv(struct nv04_graph_chan *chan)
+{
+	return (void *)nv_object(chan)->engine;
+}
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+/*
+ * Software methods, why they are needed, and how they all work:
+ *
+ * NV04 and NV05 keep most of the state in PGRAPH context itself, but some
+ * 2d engine settings are kept inside the grobjs themselves. The grobjs are
+ * 3 words long on both. grobj format on NV04 is:
+ *
+ * word 0:
+ *  - bits 0-7: class
+ *  - bit 12: color key active
+ *  - bit 13: clip rect active
+ *  - bit 14: if set, destination surface is swizzled and taken from buffer 5
+ *            [set by NV04_SWIZZLED_SURFACE], otherwise it's linear and taken
+ *            from buffer 0 [set by NV04_CONTEXT_SURFACES_2D or
+ *            NV03_CONTEXT_SURFACE_DST].
+ *  - bits 15-17: 2d operation [aka patch config]
+ *  - bit 24: patch valid [enables rendering using this object]
+ *  - bit 25: surf3d valid [for tex_tri and multitex_tri only]
+ * word 1:
+ *  - bits 0-1: mono format
+ *  - bits 8-13: color format
+ *  - bits 16-31: DMA_NOTIFY instance
+ * word 2:
+ *  - bits 0-15: DMA_A instance
+ *  - bits 16-31: DMA_B instance
+ *
+ * On NV05 it's:
+ *
+ * word 0:
+ *  - bits 0-7: class
+ *  - bit 12: color key active
+ *  - bit 13: clip rect active
+ *  - bit 14: if set, destination surface is swizzled and taken from buffer 5
+ *            [set by NV04_SWIZZLED_SURFACE], otherwise it's linear and taken
+ *            from buffer 0 [set by NV04_CONTEXT_SURFACES_2D or
+ *            NV03_CONTEXT_SURFACE_DST].
+ *  - bits 15-17: 2d operation [aka patch config]
+ *  - bits 20-22: dither mode
+ *  - bit 24: patch valid [enables rendering using this object]
+ *  - bit 25: surface_dst/surface_color/surf2d/surf3d valid
+ *  - bit 26: surface_src/surface_zeta valid
+ *  - bit 27: pattern valid
+ *  - bit 28: rop valid
+ *  - bit 29: beta1 valid
+ *  - bit 30: beta4 valid
+ * word 1:
+ *  - bits 0-1: mono format
+ *  - bits 8-13: color format
+ *  - bits 16-31: DMA_NOTIFY instance
+ * word 2:
+ *  - bits 0-15: DMA_A instance
+ *  - bits 16-31: DMA_B instance
+ *
+ * NV05 will set/unset the relevant valid bits when you poke the relevant
+ * object-binding methods with object of the proper type, or with the NULL
+ * type. It'll only allow rendering using the grobj if all needed objects
+ * are bound. The needed set of objects depends on selected operation: for
+ * example rop object is needed by ROP_AND, but not by SRCCOPY_AND.
+ *
+ * NV04 doesn't have these methods implemented at all, and doesn't have the
+ * relevant bits in grobj. Instead, it'll allow rendering whenever bit 24
+ * is set. So we have to emulate them in software, internally keeping the
+ * same bits as NV05 does. Since grobjs are aligned to 16 bytes on nv04,
+ * but the last word isn't actually used for anything, we abuse it for this
+ * purpose.
+ *
+ * Actually, NV05 can optionally check bit 24 too, but we disable this since
+ * there's no use for it.
+ *
+ * For unknown reasons, NV04 implements surf3d binding in hardware as an
+ * exception. Also for unknown reasons, NV04 doesn't implement the clipping
+ * methods on the surf3d object, so we have to emulate them too.
+ */
+
+static void
+nv04_graph_set_ctx1(struct nouveau_object *object, u32 mask, u32 value)
+{
+	struct nv04_graph_priv *priv = (void *)object->engine;
+	int subc = (nv_rd32(priv, NV04_PGRAPH_TRAPPED_ADDR) >> 13) & 0x7;
+	u32 tmp;
+
+	tmp  = nv_ro32(object, 0x00);
+	tmp &= ~mask;
+	tmp |= value;
+	nv_wo32(object, 0x00, tmp);
+
+	nv_wr32(priv, NV04_PGRAPH_CTX_SWITCH1, tmp);
+	nv_wr32(priv, NV04_PGRAPH_CTX_CACHE1 + (subc<<2), tmp);
+}
+
+static void
+nv04_graph_set_ctx_val(struct nouveau_object *object, u32 mask, u32 value)
+{
+	int class, op, valid = 1;
+	u32 tmp, ctx1;
+
+	ctx1 = nv_ro32(object, 0x00);
+	class = ctx1 & 0xff;
+	op = (ctx1 >> 15) & 7;
+
+	tmp = nv_ro32(object, 0x0c);
+	tmp &= ~mask;
+	tmp |= value;
+	nv_wo32(object, 0x0c, tmp);
+
+	/* check for valid surf2d/surf_dst/surf_color */
+	if (!(tmp & 0x02000000))
+		valid = 0;
+	/* check for valid surf_src/surf_zeta */
+	if ((class == 0x1f || class == 0x48) && !(tmp & 0x04000000))
+		valid = 0;
+
+	switch (op) {
+	/* SRCCOPY_AND, SRCCOPY: no extra objects required */
+	case 0:
+	case 3:
+		break;
+	/* ROP_AND: requires pattern and rop */
+	case 1:
+		if (!(tmp & 0x18000000))
+			valid = 0;
+		break;
+	/* BLEND_AND: requires beta1 */
+	case 2:
+		if (!(tmp & 0x20000000))
+			valid = 0;
+		break;
+	/* SRCCOPY_PREMULT, BLEND_PREMULT: beta4 required */
+	case 4:
+	case 5:
+		if (!(tmp & 0x40000000))
+			valid = 0;
+		break;
+	}
+
+	nv04_graph_set_ctx1(object, 0x01000000, valid << 24);
+}
+
+static int
+nv04_graph_mthd_set_operation(struct nouveau_object *object, u32 mthd,
+			      void *args, u32 size)
+{
+	u32 class = nv_ro32(object, 0) & 0xff;
+	u32 data = *(u32 *)args;
+	if (data > 5)
+		return 1;
+	/* Old versions of the objects only accept first three operations. */
+	if (data > 2 && class < 0x40)
+		return 1;
+	nv04_graph_set_ctx1(object, 0x00038000, data << 15);
+	/* changing operation changes set of objects needed for validation */
+	nv04_graph_set_ctx_val(object, 0, 0);
+	return 0;
+}
+
+static int
+nv04_graph_mthd_surf3d_clip_h(struct nouveau_object *object, u32 mthd,
+			      void *args, u32 size)
+{
+	struct nv04_graph_priv *priv = (void *)object->engine;
+	u32 data = *(u32 *)args;
+	u32 min = data & 0xffff, max;
+	u32 w = data >> 16;
+	if (min & 0x8000)
+		/* too large */
+		return 1;
+	if (w & 0x8000)
+		/* yes, it accepts negative for some reason. */
+		w |= 0xffff0000;
+	max = min + w;
+	max &= 0x3ffff;
+	nv_wr32(priv, 0x40053c, min);
+	nv_wr32(priv, 0x400544, max);
+	return 0;
+}
+
+static int
+nv04_graph_mthd_surf3d_clip_v(struct nouveau_object *object, u32 mthd,
+			      void *args, u32 size)
+{
+	struct nv04_graph_priv *priv = (void *)object->engine;
+	u32 data = *(u32 *)args;
+	u32 min = data & 0xffff, max;
+	u32 w = data >> 16;
+	if (min & 0x8000)
+		/* too large */
+		return 1;
+	if (w & 0x8000)
+		/* yes, it accepts negative for some reason. */
+		w |= 0xffff0000;
+	max = min + w;
+	max &= 0x3ffff;
+	nv_wr32(priv, 0x400540, min);
+	nv_wr32(priv, 0x400548, max);
+	return 0;
+}
+
+static u16
+nv04_graph_mthd_bind_class(struct nouveau_object *object, u32 *args, u32 size)
+{
+	struct nouveau_instmem *imem = nouveau_instmem(object);
+	u32 inst = *(u32 *)args << 4;
+	return nv_ro32(imem, inst);
+}
+
+static int
+nv04_graph_mthd_bind_surf2d(struct nouveau_object *object, u32 mthd,
+			    void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx1(object, 0x00004000, 0);
+		nv04_graph_set_ctx_val(object, 0x02000000, 0);
+		return 0;
+	case 0x42:
+		nv04_graph_set_ctx1(object, 0x00004000, 0);
+		nv04_graph_set_ctx_val(object, 0x02000000, 0x02000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv04_graph_mthd_bind_surf2d_swzsurf(struct nouveau_object *object, u32 mthd,
+				    void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx1(object, 0x00004000, 0);
+		nv04_graph_set_ctx_val(object, 0x02000000, 0);
+		return 0;
+	case 0x42:
+		nv04_graph_set_ctx1(object, 0x00004000, 0);
+		nv04_graph_set_ctx_val(object, 0x02000000, 0x02000000);
+		return 0;
+	case 0x52:
+		nv04_graph_set_ctx1(object, 0x00004000, 0x00004000);
+		nv04_graph_set_ctx_val(object, 0x02000000, 0x02000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv01_graph_mthd_bind_patt(struct nouveau_object *object, u32 mthd,
+			  void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx_val(object, 0x08000000, 0);
+		return 0;
+	case 0x18:
+		nv04_graph_set_ctx_val(object, 0x08000000, 0x08000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv04_graph_mthd_bind_patt(struct nouveau_object *object, u32 mthd,
+			  void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx_val(object, 0x08000000, 0);
+		return 0;
+	case 0x44:
+		nv04_graph_set_ctx_val(object, 0x08000000, 0x08000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv04_graph_mthd_bind_rop(struct nouveau_object *object, u32 mthd,
+			 void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx_val(object, 0x10000000, 0);
+		return 0;
+	case 0x43:
+		nv04_graph_set_ctx_val(object, 0x10000000, 0x10000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv04_graph_mthd_bind_beta1(struct nouveau_object *object, u32 mthd,
+			   void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx_val(object, 0x20000000, 0);
+		return 0;
+	case 0x12:
+		nv04_graph_set_ctx_val(object, 0x20000000, 0x20000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv04_graph_mthd_bind_beta4(struct nouveau_object *object, u32 mthd,
+			   void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx_val(object, 0x40000000, 0);
+		return 0;
+	case 0x72:
+		nv04_graph_set_ctx_val(object, 0x40000000, 0x40000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv04_graph_mthd_bind_surf_dst(struct nouveau_object *object, u32 mthd,
+			      void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx_val(object, 0x02000000, 0);
+		return 0;
+	case 0x58:
+		nv04_graph_set_ctx_val(object, 0x02000000, 0x02000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv04_graph_mthd_bind_surf_src(struct nouveau_object *object, u32 mthd,
+			      void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx_val(object, 0x04000000, 0);
+		return 0;
+	case 0x59:
+		nv04_graph_set_ctx_val(object, 0x04000000, 0x04000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv04_graph_mthd_bind_surf_color(struct nouveau_object *object, u32 mthd,
+				void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx_val(object, 0x02000000, 0);
+		return 0;
+	case 0x5a:
+		nv04_graph_set_ctx_val(object, 0x02000000, 0x02000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv04_graph_mthd_bind_surf_zeta(struct nouveau_object *object, u32 mthd,
+			       void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx_val(object, 0x04000000, 0);
+		return 0;
+	case 0x5b:
+		nv04_graph_set_ctx_val(object, 0x04000000, 0x04000000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv01_graph_mthd_bind_clip(struct nouveau_object *object, u32 mthd,
+			  void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx1(object, 0x2000, 0);
+		return 0;
+	case 0x19:
+		nv04_graph_set_ctx1(object, 0x2000, 0x2000);
+		return 0;
+	}
+	return 1;
+}
+
+static int
+nv01_graph_mthd_bind_chroma(struct nouveau_object *object, u32 mthd,
+			    void *args, u32 size)
+{
+	switch (nv04_graph_mthd_bind_class(object, args, size)) {
+	case 0x30:
+		nv04_graph_set_ctx1(object, 0x1000, 0);
+		return 0;
+	/* Yes, for some reason even the old versions of objects
+	 * accept 0x57 and not 0x17. Consistency be damned.
+	 */
+	case 0x57:
+		nv04_graph_set_ctx1(object, 0x1000, 0x1000);
+		return 0;
+	}
+	return 1;
+}
+
+static struct nouveau_omthds
+nv03_graph_gdi_omthds[] = {
+	{ 0x0184, 0x0184, nv01_graph_mthd_bind_patt },
+	{ 0x0188, 0x0188, nv04_graph_mthd_bind_rop },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_beta1 },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_surf_dst },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv04_graph_gdi_omthds[] = {
+	{ 0x0188, 0x0188, nv04_graph_mthd_bind_patt },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_rop },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_beta1 },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_beta4 },
+	{ 0x0198, 0x0198, nv04_graph_mthd_bind_surf2d },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv01_graph_blit_omthds[] = {
+	{ 0x0184, 0x0184, nv01_graph_mthd_bind_chroma },
+	{ 0x0188, 0x0188, nv01_graph_mthd_bind_clip },
+	{ 0x018c, 0x018c, nv01_graph_mthd_bind_patt },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_rop },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_beta1 },
+	{ 0x0198, 0x0198, nv04_graph_mthd_bind_surf_dst },
+	{ 0x019c, 0x019c, nv04_graph_mthd_bind_surf_src },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv04_graph_blit_omthds[] = {
+	{ 0x0184, 0x0184, nv01_graph_mthd_bind_chroma },
+	{ 0x0188, 0x0188, nv01_graph_mthd_bind_clip },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_patt },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_rop },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_beta1 },
+	{ 0x0198, 0x0198, nv04_graph_mthd_bind_beta4 },
+	{ 0x019c, 0x019c, nv04_graph_mthd_bind_surf2d },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv04_graph_iifc_omthds[] = {
+	{ 0x0188, 0x0188, nv01_graph_mthd_bind_chroma },
+	{ 0x018c, 0x018c, nv01_graph_mthd_bind_clip },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_patt },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_rop },
+	{ 0x0198, 0x0198, nv04_graph_mthd_bind_beta1 },
+	{ 0x019c, 0x019c, nv04_graph_mthd_bind_beta4 },
+	{ 0x01a0, 0x01a0, nv04_graph_mthd_bind_surf2d_swzsurf },
+	{ 0x03e4, 0x03e4, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv01_graph_ifc_omthds[] = {
+	{ 0x0184, 0x0184, nv01_graph_mthd_bind_chroma },
+	{ 0x0188, 0x0188, nv01_graph_mthd_bind_clip },
+	{ 0x018c, 0x018c, nv01_graph_mthd_bind_patt },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_rop },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_beta1 },
+	{ 0x0198, 0x0198, nv04_graph_mthd_bind_surf_dst },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv04_graph_ifc_omthds[] = {
+	{ 0x0184, 0x0184, nv01_graph_mthd_bind_chroma },
+	{ 0x0188, 0x0188, nv01_graph_mthd_bind_clip },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_patt },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_rop },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_beta1 },
+	{ 0x0198, 0x0198, nv04_graph_mthd_bind_beta4 },
+	{ 0x019c, 0x019c, nv04_graph_mthd_bind_surf2d },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv03_graph_sifc_omthds[] = {
+	{ 0x0184, 0x0184, nv01_graph_mthd_bind_chroma },
+	{ 0x0188, 0x0188, nv01_graph_mthd_bind_patt },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_rop },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_beta1 },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_surf_dst },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv04_graph_sifc_omthds[] = {
+	{ 0x0184, 0x0184, nv01_graph_mthd_bind_chroma },
+	{ 0x0188, 0x0188, nv04_graph_mthd_bind_patt },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_rop },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_beta1 },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_beta4 },
+	{ 0x0198, 0x0198, nv04_graph_mthd_bind_surf2d },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv03_graph_sifm_omthds[] = {
+	{ 0x0188, 0x0188, nv01_graph_mthd_bind_patt },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_rop },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_beta1 },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_surf_dst },
+	{ 0x0304, 0x0304, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv04_graph_sifm_omthds[] = {
+	{ 0x0188, 0x0188, nv04_graph_mthd_bind_patt },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_rop },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_beta1 },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_beta4 },
+	{ 0x0198, 0x0198, nv04_graph_mthd_bind_surf2d },
+	{ 0x0304, 0x0304, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv04_graph_surf3d_omthds[] = {
+	{ 0x02f8, 0x02f8, nv04_graph_mthd_surf3d_clip_h },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_surf3d_clip_v },
+	{}
+};
+
+static struct nouveau_omthds
+nv03_graph_ttri_omthds[] = {
+	{ 0x0188, 0x0188, nv01_graph_mthd_bind_clip },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_surf_color },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_surf_zeta },
+	{}
+};
+
+static struct nouveau_omthds
+nv01_graph_prim_omthds[] = {
+	{ 0x0184, 0x0184, nv01_graph_mthd_bind_clip },
+	{ 0x0188, 0x0188, nv01_graph_mthd_bind_patt },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_rop },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_beta1 },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_surf_dst },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static struct nouveau_omthds
+nv04_graph_prim_omthds[] = {
+	{ 0x0184, 0x0184, nv01_graph_mthd_bind_clip },
+	{ 0x0188, 0x0188, nv04_graph_mthd_bind_patt },
+	{ 0x018c, 0x018c, nv04_graph_mthd_bind_rop },
+	{ 0x0190, 0x0190, nv04_graph_mthd_bind_beta1 },
+	{ 0x0194, 0x0194, nv04_graph_mthd_bind_beta4 },
+	{ 0x0198, 0x0198, nv04_graph_mthd_bind_surf2d },
+	{ 0x02fc, 0x02fc, nv04_graph_mthd_set_operation },
+	{}
+};
+
+static int
+nv04_graph_object_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nouveau_gpuobj *obj;
+	int ret;
+
+	ret = nouveau_gpuobj_create(parent, engine, oclass, 0, parent,
+				    16, 16, 0, &obj);
+	*pobject = nv_object(obj);
+	if (ret)
+		return ret;
+
+	nv_wo32(obj, 0x00, nv_mclass(obj));
+#ifdef __BIG_ENDIAN
+	nv_mo32(obj, 0x00, 0x00080000, 0x00080000);
+#endif
+	nv_wo32(obj, 0x04, 0x00000000);
+	nv_wo32(obj, 0x08, 0x00000000);
+	nv_wo32(obj, 0x0c, 0x00000000);
+	return 0;
+}
+
+struct nouveau_ofuncs
+nv04_graph_ofuncs = {
+	.ctor = nv04_graph_object_ctor,
+	.dtor = _nouveau_gpuobj_dtor,
+	.init = _nouveau_gpuobj_init,
+	.fini = _nouveau_gpuobj_fini,
+	.rd32 = _nouveau_gpuobj_rd32,
+	.wr32 = _nouveau_gpuobj_wr32,
+};
+
+static struct nouveau_oclass
+nv04_graph_sclass[] = {
+	{ 0x0012, &nv04_graph_ofuncs }, /* beta1 */
+	{ 0x0017, &nv04_graph_ofuncs }, /* chroma */
+	{ 0x0018, &nv04_graph_ofuncs }, /* pattern (nv01) */
+	{ 0x0019, &nv04_graph_ofuncs }, /* clip */
+	{ 0x001c, &nv04_graph_ofuncs, nv01_graph_prim_omthds }, /* line */
+	{ 0x001d, &nv04_graph_ofuncs, nv01_graph_prim_omthds }, /* tri */
+	{ 0x001e, &nv04_graph_ofuncs, nv01_graph_prim_omthds }, /* rect */
+	{ 0x001f, &nv04_graph_ofuncs, nv01_graph_blit_omthds },
+	{ 0x0021, &nv04_graph_ofuncs, nv01_graph_ifc_omthds },
+	{ 0x0030, &nv04_graph_ofuncs }, /* null */
+	{ 0x0036, &nv04_graph_ofuncs, nv03_graph_sifc_omthds },
+	{ 0x0037, &nv04_graph_ofuncs, nv03_graph_sifm_omthds },
+	{ 0x0038, &nv04_graph_ofuncs }, /* dvd subpicture */
+	{ 0x0039, &nv04_graph_ofuncs }, /* m2mf */
+	{ 0x0042, &nv04_graph_ofuncs }, /* surf2d */
+	{ 0x0043, &nv04_graph_ofuncs }, /* rop */
+	{ 0x0044, &nv04_graph_ofuncs }, /* pattern */
+	{ 0x0048, &nv04_graph_ofuncs, nv03_graph_ttri_omthds },
+	{ 0x004a, &nv04_graph_ofuncs, nv04_graph_gdi_omthds },
+	{ 0x004b, &nv04_graph_ofuncs, nv03_graph_gdi_omthds },
+	{ 0x0052, &nv04_graph_ofuncs }, /* swzsurf */
+	{ 0x0053, &nv04_graph_ofuncs, nv04_graph_surf3d_omthds },
+	{ 0x0054, &nv04_graph_ofuncs }, /* ttri */
+	{ 0x0055, &nv04_graph_ofuncs }, /* mtri */
+	{ 0x0057, &nv04_graph_ofuncs }, /* chroma */
+	{ 0x0058, &nv04_graph_ofuncs }, /* surf_dst */
+	{ 0x0059, &nv04_graph_ofuncs }, /* surf_src */
+	{ 0x005a, &nv04_graph_ofuncs }, /* surf_color */
+	{ 0x005b, &nv04_graph_ofuncs }, /* surf_zeta */
+	{ 0x005c, &nv04_graph_ofuncs, nv04_graph_prim_omthds }, /* line */
+	{ 0x005d, &nv04_graph_ofuncs, nv04_graph_prim_omthds }, /* tri */
+	{ 0x005e, &nv04_graph_ofuncs, nv04_graph_prim_omthds }, /* rect */
+	{ 0x005f, &nv04_graph_ofuncs, nv04_graph_blit_omthds },
+	{ 0x0060, &nv04_graph_ofuncs, nv04_graph_iifc_omthds },
+	{ 0x0061, &nv04_graph_ofuncs, nv04_graph_ifc_omthds },
+	{ 0x0064, &nv04_graph_ofuncs }, /* iifc (nv05) */
+	{ 0x0065, &nv04_graph_ofuncs }, /* ifc (nv05) */
+	{ 0x0066, &nv04_graph_ofuncs }, /* sifc (nv05) */
+	{ 0x0072, &nv04_graph_ofuncs }, /* beta4 */
+	{ 0x0076, &nv04_graph_ofuncs, nv04_graph_sifc_omthds },
+	{ 0x0077, &nv04_graph_ofuncs, nv04_graph_sifm_omthds },
+	{},
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static struct nv04_graph_chan *
+nv04_graph_channel(struct nv04_graph_priv *priv)
+{
+	struct nv04_graph_chan *chan = NULL;
+	if (nv_rd32(priv, NV04_PGRAPH_CTX_CONTROL) & 0x00010000) {
+		int chid = nv_rd32(priv, NV04_PGRAPH_CTX_USER) >> 24;
+		if (chid < ARRAY_SIZE(priv->chan))
+			chan = priv->chan[chid];
+	}
+	return chan;
+}
+
+static int
+nv04_graph_load_context(struct nv04_graph_chan *chan, int chid)
+{
+	struct nv04_graph_priv *priv = nv04_graph_priv(chan);
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(nv04_graph_ctx_regs); i++)
+		nv_wr32(priv, nv04_graph_ctx_regs[i], chan->nv04[i]);
+
+	nv_wr32(priv, NV04_PGRAPH_CTX_CONTROL, 0x10010100);
+	nv_mask(priv, NV04_PGRAPH_CTX_USER, 0xff000000, chid << 24);
+	nv_mask(priv, NV04_PGRAPH_FFINTFC_ST2, 0xfff00000, 0x00000000);
+	return 0;
+}
+
+static int
+nv04_graph_unload_context(struct nv04_graph_chan *chan)
+{
+	struct nv04_graph_priv *priv = nv04_graph_priv(chan);
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(nv04_graph_ctx_regs); i++)
+		chan->nv04[i] = nv_rd32(priv, nv04_graph_ctx_regs[i]);
+
+	nv_wr32(priv, NV04_PGRAPH_CTX_CONTROL, 0x10000000);
+	nv_mask(priv, NV04_PGRAPH_CTX_USER, 0xff000000, 0x0f000000);
+	return 0;
+}
+
+static void
+nv04_graph_context_switch(struct nv04_graph_priv *priv)
+{
+	struct nv04_graph_chan *prev = NULL;
+	struct nv04_graph_chan *next = NULL;
+	unsigned long flags;
+	int chid;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	nv04_graph_idle(priv);
+
+	/* If previous context is valid, we need to save it */
+	prev = nv04_graph_channel(priv);
+	if (prev)
+		nv04_graph_unload_context(prev);
+
+	/* load context for next channel */
+	chid = (nv_rd32(priv, NV04_PGRAPH_TRAPPED_ADDR) >> 24) & 0x0f;
+	next = priv->chan[chid];
+	if (next)
+		nv04_graph_load_context(next, chid);
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+static u32 *ctx_reg(struct nv04_graph_chan *chan, u32 reg)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(nv04_graph_ctx_regs); i++) {
+		if (nv04_graph_ctx_regs[i] == reg)
+			return &chan->nv04[i];
+	}
+
+	return NULL;
+}
+
+static int
+nv04_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nouveau_fifo_chan *fifo = (void *)parent;
+	struct nv04_graph_priv *priv = (void *)engine;
+	struct nv04_graph_chan *chan;
+	unsigned long flags;
+	int ret;
+
+	ret = nouveau_object_create(parent, engine, oclass, 0, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	if (priv->chan[fifo->chid]) {
+		*pobject = nv_object(priv->chan[fifo->chid]);
+		atomic_inc(&(*pobject)->refcount);
+		spin_unlock_irqrestore(&priv->lock, flags);
+		nouveau_object_destroy(&chan->base);
+		return 1;
+	}
+
+	*ctx_reg(chan, NV04_PGRAPH_DEBUG_3) = 0xfad4ff31;
+
+	priv->chan[fifo->chid] = chan;
+	chan->chid = fifo->chid;
+	spin_unlock_irqrestore(&priv->lock, flags);
+	return 0;
+}
+
+static void
+nv04_graph_context_dtor(struct nouveau_object *object)
+{
+	struct nv04_graph_priv *priv = (void *)object->engine;
+	struct nv04_graph_chan *chan = (void *)object;
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	priv->chan[chan->chid] = NULL;
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	nouveau_object_destroy(&chan->base);
+}
+
+static int
+nv04_graph_context_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv04_graph_priv *priv = (void *)object->engine;
+	struct nv04_graph_chan *chan = (void *)object;
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	nv_mask(priv, NV04_PGRAPH_FIFO, 0x00000001, 0x00000000);
+	if (nv04_graph_channel(priv) == chan)
+		nv04_graph_unload_context(chan);
+	nv_mask(priv, NV04_PGRAPH_FIFO, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	return nouveau_object_fini(&chan->base, suspend);
+}
+
+static struct nouveau_oclass
+nv04_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_graph_context_ctor,
+		.dtor = nv04_graph_context_dtor,
+		.init = nouveau_object_init,
+		.fini = nv04_graph_context_fini,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+bool
+nv04_graph_idle(void *obj)
+{
+	struct nouveau_graph *graph = nouveau_graph(obj);
+	u32 mask = 0xffffffff;
+
+	if (nv_device(obj)->card_type == NV_40)
+		mask &= ~NV40_PGRAPH_STATUS_SYNC_STALL;
+
+	if (!nv_wait(graph, NV04_PGRAPH_STATUS, mask, 0)) {
+		nv_error(graph, "idle timed out with status 0x%08x\n",
+			 nv_rd32(graph, NV04_PGRAPH_STATUS));
+		return false;
+	}
+
+	return true;
+}
+
+static const struct nouveau_bitfield
+nv04_graph_intr_name[] = {
+	{ NV_PGRAPH_INTR_NOTIFY, "NOTIFY" },
+	{}
+};
+
+static const struct nouveau_bitfield
+nv04_graph_nstatus[] = {
+	{ NV04_PGRAPH_NSTATUS_STATE_IN_USE,       "STATE_IN_USE" },
+	{ NV04_PGRAPH_NSTATUS_INVALID_STATE,      "INVALID_STATE" },
+	{ NV04_PGRAPH_NSTATUS_BAD_ARGUMENT,       "BAD_ARGUMENT" },
+	{ NV04_PGRAPH_NSTATUS_PROTECTION_FAULT,   "PROTECTION_FAULT" },
+	{}
+};
+
+const struct nouveau_bitfield
+nv04_graph_nsource[] = {
+	{ NV03_PGRAPH_NSOURCE_NOTIFICATION,       "NOTIFICATION" },
+	{ NV03_PGRAPH_NSOURCE_DATA_ERROR,         "DATA_ERROR" },
+	{ NV03_PGRAPH_NSOURCE_PROTECTION_ERROR,   "PROTECTION_ERROR" },
+	{ NV03_PGRAPH_NSOURCE_RANGE_EXCEPTION,    "RANGE_EXCEPTION" },
+	{ NV03_PGRAPH_NSOURCE_LIMIT_COLOR,        "LIMIT_COLOR" },
+	{ NV03_PGRAPH_NSOURCE_LIMIT_ZETA,         "LIMIT_ZETA" },
+	{ NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD,       "ILLEGAL_MTHD" },
+	{ NV03_PGRAPH_NSOURCE_DMA_R_PROTECTION,   "DMA_R_PROTECTION" },
+	{ NV03_PGRAPH_NSOURCE_DMA_W_PROTECTION,   "DMA_W_PROTECTION" },
+	{ NV03_PGRAPH_NSOURCE_FORMAT_EXCEPTION,   "FORMAT_EXCEPTION" },
+	{ NV03_PGRAPH_NSOURCE_PATCH_EXCEPTION,    "PATCH_EXCEPTION" },
+	{ NV03_PGRAPH_NSOURCE_STATE_INVALID,      "STATE_INVALID" },
+	{ NV03_PGRAPH_NSOURCE_DOUBLE_NOTIFY,      "DOUBLE_NOTIFY" },
+	{ NV03_PGRAPH_NSOURCE_NOTIFY_IN_USE,      "NOTIFY_IN_USE" },
+	{ NV03_PGRAPH_NSOURCE_METHOD_CNT,         "METHOD_CNT" },
+	{ NV03_PGRAPH_NSOURCE_BFR_NOTIFICATION,   "BFR_NOTIFICATION" },
+	{ NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION, "DMA_VTX_PROTECTION" },
+	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_A,        "DMA_WIDTH_A" },
+	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_B,        "DMA_WIDTH_B" },
+	{}
+};
+
+static void
+nv04_graph_intr(struct nouveau_subdev *subdev)
+{
+	struct nv04_graph_priv *priv = (void *)subdev;
+	struct nv04_graph_chan *chan = NULL;
+	struct nouveau_namedb *namedb = NULL;
+	struct nouveau_handle *handle = NULL;
+	u32 stat = nv_rd32(priv, NV03_PGRAPH_INTR);
+	u32 nsource = nv_rd32(priv, NV03_PGRAPH_NSOURCE);
+	u32 nstatus = nv_rd32(priv, NV03_PGRAPH_NSTATUS);
+	u32 addr = nv_rd32(priv, NV04_PGRAPH_TRAPPED_ADDR);
+	u32 chid = (addr & 0x0f000000) >> 24;
+	u32 subc = (addr & 0x0000e000) >> 13;
+	u32 mthd = (addr & 0x00001ffc);
+	u32 data = nv_rd32(priv, NV04_PGRAPH_TRAPPED_DATA);
+	u32 class = nv_rd32(priv, 0x400180 + subc * 4) & 0xff;
+	u32 inst = (nv_rd32(priv, 0x40016c) & 0xffff) << 4;
+	u32 show = stat;
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	chan = priv->chan[chid];
+	if (chan)
+		namedb = (void *)nv_pclass(nv_object(chan), NV_NAMEDB_CLASS);
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	if (stat & NV_PGRAPH_INTR_NOTIFY) {
+		if (chan && (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD)) {
+			handle = nouveau_namedb_get_vinst(namedb, inst);
+			if (handle && !nv_call(handle->object, mthd, data))
+				show &= ~NV_PGRAPH_INTR_NOTIFY;
+		}
+	}
+
+	if (stat & NV_PGRAPH_INTR_CONTEXT_SWITCH) {
+		nv_wr32(priv, NV03_PGRAPH_INTR, NV_PGRAPH_INTR_CONTEXT_SWITCH);
+		stat &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+		show &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+		nv04_graph_context_switch(priv);
+	}
+
+	nv_wr32(priv, NV03_PGRAPH_INTR, stat);
+	nv_wr32(priv, NV04_PGRAPH_FIFO, 0x00000001);
+
+	if (show) {
+		nv_error(priv, "");
+		nouveau_bitfield_print(nv04_graph_intr_name, show);
+		printk(" nsource:");
+		nouveau_bitfield_print(nv04_graph_nsource, nsource);
+		printk(" nstatus:");
+		nouveau_bitfield_print(nv04_graph_nstatus, nstatus);
+		printk("\n");
+		nv_error(priv, "ch %d/%d class 0x%04x "
+			       "mthd 0x%04x data 0x%08x\n",
+			 chid, subc, class, mthd, data);
+	}
+
+	nouveau_namedb_put(handle);
+}
+
+static int
+nv04_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv04_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00001000;
+	nv_subdev(priv)->intr = nv04_graph_intr;
+	nv_engine(priv)->cclass = &nv04_graph_cclass;
+	nv_engine(priv)->sclass = nv04_graph_sclass;
+	spin_lock_init(&priv->lock);
+	return 0;
+}
+
+static int
+nv04_graph_init(struct nouveau_object *object)
+{
+	struct nouveau_engine *engine = nv_engine(object);
+	struct nv04_graph_priv *priv = (void *)engine;
+	int ret;
+
+	ret = nouveau_graph_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* Enable PGRAPH interrupts */
+	nv_wr32(priv, NV03_PGRAPH_INTR, 0xFFFFFFFF);
+	nv_wr32(priv, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
+
+	nv_wr32(priv, NV04_PGRAPH_VALID1, 0);
+	nv_wr32(priv, NV04_PGRAPH_VALID2, 0);
+	/*nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0x000001FF);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0x001FFFFF);*/
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0x1231c000);
+	/*1231C000 blob, 001 haiku*/
+	/*V_WRITE(NV04_PGRAPH_DEBUG_1, 0xf2d91100);*/
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_1, 0x72111100);
+	/*0x72111100 blob , 01 haiku*/
+	/*nv_wr32(priv, NV04_PGRAPH_DEBUG_2, 0x11d5f870);*/
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_2, 0x11d5f071);
+	/*haiku same*/
+
+	/*nv_wr32(priv, NV04_PGRAPH_DEBUG_3, 0xfad4ff31);*/
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_3, 0xf0d4ff31);
+	/*haiku and blob 10d4*/
+
+	nv_wr32(priv, NV04_PGRAPH_STATE        , 0xFFFFFFFF);
+	nv_wr32(priv, NV04_PGRAPH_CTX_CONTROL  , 0x10000100);
+	nv_mask(priv, NV04_PGRAPH_CTX_USER, 0xff000000, 0x0f000000);
+
+	/* These don't belong here, they're part of a per-channel context */
+	nv_wr32(priv, NV04_PGRAPH_PATTERN_SHAPE, 0x00000000);
+	nv_wr32(priv, NV04_PGRAPH_BETA_AND     , 0xFFFFFFFF);
+	return 0;
+}
+
+struct nouveau_oclass
+nv04_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_graph_ctor,
+		.dtor = _nouveau_graph_dtor,
+		.init = nv04_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv10.c b/nvkm/engine/graph/nv10.c
new file mode 100644
index 000000000..5c0f843ea
--- /dev/null
+++ b/nvkm/engine/graph/nv10.c
@@ -0,0 +1,1314 @@
+/*
+ * Copyright 2007 Matthieu CASTET <castet.matthieu@free.fr>
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * PRECISION INSIGHT AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+ * DEALINGS IN THE SOFTWARE.
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/handle.h>
+
+#include <subdev/fb.h>
+
+#include <engine/fifo.h>
+#include <engine/graph.h>
+
+#include "regs.h"
+
+struct pipe_state {
+	u32 pipe_0x0000[0x040/4];
+	u32 pipe_0x0040[0x010/4];
+	u32 pipe_0x0200[0x0c0/4];
+	u32 pipe_0x4400[0x080/4];
+	u32 pipe_0x6400[0x3b0/4];
+	u32 pipe_0x6800[0x2f0/4];
+	u32 pipe_0x6c00[0x030/4];
+	u32 pipe_0x7000[0x130/4];
+	u32 pipe_0x7400[0x0c0/4];
+	u32 pipe_0x7800[0x0c0/4];
+};
+
+static int nv10_graph_ctx_regs[] = {
+	NV10_PGRAPH_CTX_SWITCH(0),
+	NV10_PGRAPH_CTX_SWITCH(1),
+	NV10_PGRAPH_CTX_SWITCH(2),
+	NV10_PGRAPH_CTX_SWITCH(3),
+	NV10_PGRAPH_CTX_SWITCH(4),
+	NV10_PGRAPH_CTX_CACHE(0, 0),
+	NV10_PGRAPH_CTX_CACHE(0, 1),
+	NV10_PGRAPH_CTX_CACHE(0, 2),
+	NV10_PGRAPH_CTX_CACHE(0, 3),
+	NV10_PGRAPH_CTX_CACHE(0, 4),
+	NV10_PGRAPH_CTX_CACHE(1, 0),
+	NV10_PGRAPH_CTX_CACHE(1, 1),
+	NV10_PGRAPH_CTX_CACHE(1, 2),
+	NV10_PGRAPH_CTX_CACHE(1, 3),
+	NV10_PGRAPH_CTX_CACHE(1, 4),
+	NV10_PGRAPH_CTX_CACHE(2, 0),
+	NV10_PGRAPH_CTX_CACHE(2, 1),
+	NV10_PGRAPH_CTX_CACHE(2, 2),
+	NV10_PGRAPH_CTX_CACHE(2, 3),
+	NV10_PGRAPH_CTX_CACHE(2, 4),
+	NV10_PGRAPH_CTX_CACHE(3, 0),
+	NV10_PGRAPH_CTX_CACHE(3, 1),
+	NV10_PGRAPH_CTX_CACHE(3, 2),
+	NV10_PGRAPH_CTX_CACHE(3, 3),
+	NV10_PGRAPH_CTX_CACHE(3, 4),
+	NV10_PGRAPH_CTX_CACHE(4, 0),
+	NV10_PGRAPH_CTX_CACHE(4, 1),
+	NV10_PGRAPH_CTX_CACHE(4, 2),
+	NV10_PGRAPH_CTX_CACHE(4, 3),
+	NV10_PGRAPH_CTX_CACHE(4, 4),
+	NV10_PGRAPH_CTX_CACHE(5, 0),
+	NV10_PGRAPH_CTX_CACHE(5, 1),
+	NV10_PGRAPH_CTX_CACHE(5, 2),
+	NV10_PGRAPH_CTX_CACHE(5, 3),
+	NV10_PGRAPH_CTX_CACHE(5, 4),
+	NV10_PGRAPH_CTX_CACHE(6, 0),
+	NV10_PGRAPH_CTX_CACHE(6, 1),
+	NV10_PGRAPH_CTX_CACHE(6, 2),
+	NV10_PGRAPH_CTX_CACHE(6, 3),
+	NV10_PGRAPH_CTX_CACHE(6, 4),
+	NV10_PGRAPH_CTX_CACHE(7, 0),
+	NV10_PGRAPH_CTX_CACHE(7, 1),
+	NV10_PGRAPH_CTX_CACHE(7, 2),
+	NV10_PGRAPH_CTX_CACHE(7, 3),
+	NV10_PGRAPH_CTX_CACHE(7, 4),
+	NV10_PGRAPH_CTX_USER,
+	NV04_PGRAPH_DMA_START_0,
+	NV04_PGRAPH_DMA_START_1,
+	NV04_PGRAPH_DMA_LENGTH,
+	NV04_PGRAPH_DMA_MISC,
+	NV10_PGRAPH_DMA_PITCH,
+	NV04_PGRAPH_BOFFSET0,
+	NV04_PGRAPH_BBASE0,
+	NV04_PGRAPH_BLIMIT0,
+	NV04_PGRAPH_BOFFSET1,
+	NV04_PGRAPH_BBASE1,
+	NV04_PGRAPH_BLIMIT1,
+	NV04_PGRAPH_BOFFSET2,
+	NV04_PGRAPH_BBASE2,
+	NV04_PGRAPH_BLIMIT2,
+	NV04_PGRAPH_BOFFSET3,
+	NV04_PGRAPH_BBASE3,
+	NV04_PGRAPH_BLIMIT3,
+	NV04_PGRAPH_BOFFSET4,
+	NV04_PGRAPH_BBASE4,
+	NV04_PGRAPH_BLIMIT4,
+	NV04_PGRAPH_BOFFSET5,
+	NV04_PGRAPH_BBASE5,
+	NV04_PGRAPH_BLIMIT5,
+	NV04_PGRAPH_BPITCH0,
+	NV04_PGRAPH_BPITCH1,
+	NV04_PGRAPH_BPITCH2,
+	NV04_PGRAPH_BPITCH3,
+	NV04_PGRAPH_BPITCH4,
+	NV10_PGRAPH_SURFACE,
+	NV10_PGRAPH_STATE,
+	NV04_PGRAPH_BSWIZZLE2,
+	NV04_PGRAPH_BSWIZZLE5,
+	NV04_PGRAPH_BPIXEL,
+	NV10_PGRAPH_NOTIFY,
+	NV04_PGRAPH_PATT_COLOR0,
+	NV04_PGRAPH_PATT_COLOR1,
+	NV04_PGRAPH_PATT_COLORRAM, /* 64 values from 0x400900 to 0x4009fc */
+	0x00400904,
+	0x00400908,
+	0x0040090c,
+	0x00400910,
+	0x00400914,
+	0x00400918,
+	0x0040091c,
+	0x00400920,
+	0x00400924,
+	0x00400928,
+	0x0040092c,
+	0x00400930,
+	0x00400934,
+	0x00400938,
+	0x0040093c,
+	0x00400940,
+	0x00400944,
+	0x00400948,
+	0x0040094c,
+	0x00400950,
+	0x00400954,
+	0x00400958,
+	0x0040095c,
+	0x00400960,
+	0x00400964,
+	0x00400968,
+	0x0040096c,
+	0x00400970,
+	0x00400974,
+	0x00400978,
+	0x0040097c,
+	0x00400980,
+	0x00400984,
+	0x00400988,
+	0x0040098c,
+	0x00400990,
+	0x00400994,
+	0x00400998,
+	0x0040099c,
+	0x004009a0,
+	0x004009a4,
+	0x004009a8,
+	0x004009ac,
+	0x004009b0,
+	0x004009b4,
+	0x004009b8,
+	0x004009bc,
+	0x004009c0,
+	0x004009c4,
+	0x004009c8,
+	0x004009cc,
+	0x004009d0,
+	0x004009d4,
+	0x004009d8,
+	0x004009dc,
+	0x004009e0,
+	0x004009e4,
+	0x004009e8,
+	0x004009ec,
+	0x004009f0,
+	0x004009f4,
+	0x004009f8,
+	0x004009fc,
+	NV04_PGRAPH_PATTERN,	/* 2 values from 0x400808 to 0x40080c */
+	0x0040080c,
+	NV04_PGRAPH_PATTERN_SHAPE,
+	NV03_PGRAPH_MONO_COLOR0,
+	NV04_PGRAPH_ROP3,
+	NV04_PGRAPH_CHROMA,
+	NV04_PGRAPH_BETA_AND,
+	NV04_PGRAPH_BETA_PREMULT,
+	0x00400e70,
+	0x00400e74,
+	0x00400e78,
+	0x00400e7c,
+	0x00400e80,
+	0x00400e84,
+	0x00400e88,
+	0x00400e8c,
+	0x00400ea0,
+	0x00400ea4,
+	0x00400ea8,
+	0x00400e90,
+	0x00400e94,
+	0x00400e98,
+	0x00400e9c,
+	NV10_PGRAPH_WINDOWCLIP_HORIZONTAL, /* 8 values from 0x400f00-0x400f1c */
+	NV10_PGRAPH_WINDOWCLIP_VERTICAL,   /* 8 values from 0x400f20-0x400f3c */
+	0x00400f04,
+	0x00400f24,
+	0x00400f08,
+	0x00400f28,
+	0x00400f0c,
+	0x00400f2c,
+	0x00400f10,
+	0x00400f30,
+	0x00400f14,
+	0x00400f34,
+	0x00400f18,
+	0x00400f38,
+	0x00400f1c,
+	0x00400f3c,
+	NV10_PGRAPH_XFMODE0,
+	NV10_PGRAPH_XFMODE1,
+	NV10_PGRAPH_GLOBALSTATE0,
+	NV10_PGRAPH_GLOBALSTATE1,
+	NV04_PGRAPH_STORED_FMT,
+	NV04_PGRAPH_SOURCE_COLOR,
+	NV03_PGRAPH_ABS_X_RAM,	/* 32 values from 0x400400 to 0x40047c */
+	NV03_PGRAPH_ABS_Y_RAM,	/* 32 values from 0x400480 to 0x4004fc */
+	0x00400404,
+	0x00400484,
+	0x00400408,
+	0x00400488,
+	0x0040040c,
+	0x0040048c,
+	0x00400410,
+	0x00400490,
+	0x00400414,
+	0x00400494,
+	0x00400418,
+	0x00400498,
+	0x0040041c,
+	0x0040049c,
+	0x00400420,
+	0x004004a0,
+	0x00400424,
+	0x004004a4,
+	0x00400428,
+	0x004004a8,
+	0x0040042c,
+	0x004004ac,
+	0x00400430,
+	0x004004b0,
+	0x00400434,
+	0x004004b4,
+	0x00400438,
+	0x004004b8,
+	0x0040043c,
+	0x004004bc,
+	0x00400440,
+	0x004004c0,
+	0x00400444,
+	0x004004c4,
+	0x00400448,
+	0x004004c8,
+	0x0040044c,
+	0x004004cc,
+	0x00400450,
+	0x004004d0,
+	0x00400454,
+	0x004004d4,
+	0x00400458,
+	0x004004d8,
+	0x0040045c,
+	0x004004dc,
+	0x00400460,
+	0x004004e0,
+	0x00400464,
+	0x004004e4,
+	0x00400468,
+	0x004004e8,
+	0x0040046c,
+	0x004004ec,
+	0x00400470,
+	0x004004f0,
+	0x00400474,
+	0x004004f4,
+	0x00400478,
+	0x004004f8,
+	0x0040047c,
+	0x004004fc,
+	NV03_PGRAPH_ABS_UCLIP_XMIN,
+	NV03_PGRAPH_ABS_UCLIP_XMAX,
+	NV03_PGRAPH_ABS_UCLIP_YMIN,
+	NV03_PGRAPH_ABS_UCLIP_YMAX,
+	0x00400550,
+	0x00400558,
+	0x00400554,
+	0x0040055c,
+	NV03_PGRAPH_ABS_UCLIPA_XMIN,
+	NV03_PGRAPH_ABS_UCLIPA_XMAX,
+	NV03_PGRAPH_ABS_UCLIPA_YMIN,
+	NV03_PGRAPH_ABS_UCLIPA_YMAX,
+	NV03_PGRAPH_ABS_ICLIP_XMAX,
+	NV03_PGRAPH_ABS_ICLIP_YMAX,
+	NV03_PGRAPH_XY_LOGIC_MISC0,
+	NV03_PGRAPH_XY_LOGIC_MISC1,
+	NV03_PGRAPH_XY_LOGIC_MISC2,
+	NV03_PGRAPH_XY_LOGIC_MISC3,
+	NV03_PGRAPH_CLIPX_0,
+	NV03_PGRAPH_CLIPX_1,
+	NV03_PGRAPH_CLIPY_0,
+	NV03_PGRAPH_CLIPY_1,
+	NV10_PGRAPH_COMBINER0_IN_ALPHA,
+	NV10_PGRAPH_COMBINER1_IN_ALPHA,
+	NV10_PGRAPH_COMBINER0_IN_RGB,
+	NV10_PGRAPH_COMBINER1_IN_RGB,
+	NV10_PGRAPH_COMBINER_COLOR0,
+	NV10_PGRAPH_COMBINER_COLOR1,
+	NV10_PGRAPH_COMBINER0_OUT_ALPHA,
+	NV10_PGRAPH_COMBINER1_OUT_ALPHA,
+	NV10_PGRAPH_COMBINER0_OUT_RGB,
+	NV10_PGRAPH_COMBINER1_OUT_RGB,
+	NV10_PGRAPH_COMBINER_FINAL0,
+	NV10_PGRAPH_COMBINER_FINAL1,
+	0x00400e00,
+	0x00400e04,
+	0x00400e08,
+	0x00400e0c,
+	0x00400e10,
+	0x00400e14,
+	0x00400e18,
+	0x00400e1c,
+	0x00400e20,
+	0x00400e24,
+	0x00400e28,
+	0x00400e2c,
+	0x00400e30,
+	0x00400e34,
+	0x00400e38,
+	0x00400e3c,
+	NV04_PGRAPH_PASSTHRU_0,
+	NV04_PGRAPH_PASSTHRU_1,
+	NV04_PGRAPH_PASSTHRU_2,
+	NV10_PGRAPH_DIMX_TEXTURE,
+	NV10_PGRAPH_WDIMX_TEXTURE,
+	NV10_PGRAPH_DVD_COLORFMT,
+	NV10_PGRAPH_SCALED_FORMAT,
+	NV04_PGRAPH_MISC24_0,
+	NV04_PGRAPH_MISC24_1,
+	NV04_PGRAPH_MISC24_2,
+	NV03_PGRAPH_X_MISC,
+	NV03_PGRAPH_Y_MISC,
+	NV04_PGRAPH_VALID1,
+	NV04_PGRAPH_VALID2,
+};
+
+static int nv17_graph_ctx_regs[] = {
+	NV10_PGRAPH_DEBUG_4,
+	0x004006b0,
+	0x00400eac,
+	0x00400eb0,
+	0x00400eb4,
+	0x00400eb8,
+	0x00400ebc,
+	0x00400ec0,
+	0x00400ec4,
+	0x00400ec8,
+	0x00400ecc,
+	0x00400ed0,
+	0x00400ed4,
+	0x00400ed8,
+	0x00400edc,
+	0x00400ee0,
+	0x00400a00,
+	0x00400a04,
+};
+
+struct nv10_graph_priv {
+	struct nouveau_graph base;
+	struct nv10_graph_chan *chan[32];
+	spinlock_t lock;
+};
+
+struct nv10_graph_chan {
+	struct nouveau_object base;
+	int chid;
+	int nv10[ARRAY_SIZE(nv10_graph_ctx_regs)];
+	int nv17[ARRAY_SIZE(nv17_graph_ctx_regs)];
+	struct pipe_state pipe_state;
+	u32 lma_window[4];
+};
+
+
+static inline struct nv10_graph_priv *
+nv10_graph_priv(struct nv10_graph_chan *chan)
+{
+	return (void *)nv_object(chan)->engine;
+}
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+#define PIPE_SAVE(priv, state, addr)					\
+	do {								\
+		int __i;						\
+		nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, addr);		\
+		for (__i = 0; __i < ARRAY_SIZE(state); __i++)		\
+			state[__i] = nv_rd32(priv, NV10_PGRAPH_PIPE_DATA); \
+	} while (0)
+
+#define PIPE_RESTORE(priv, state, addr)					\
+	do {								\
+		int __i;						\
+		nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, addr);		\
+		for (__i = 0; __i < ARRAY_SIZE(state); __i++)		\
+			nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, state[__i]); \
+	} while (0)
+
+static struct nouveau_oclass
+nv10_graph_sclass[] = {
+	{ 0x0012, &nv04_graph_ofuncs }, /* beta1 */
+	{ 0x0019, &nv04_graph_ofuncs }, /* clip */
+	{ 0x0030, &nv04_graph_ofuncs }, /* null */
+	{ 0x0039, &nv04_graph_ofuncs }, /* m2mf */
+	{ 0x0043, &nv04_graph_ofuncs }, /* rop */
+	{ 0x0044, &nv04_graph_ofuncs }, /* pattern */
+	{ 0x004a, &nv04_graph_ofuncs }, /* gdi */
+	{ 0x0052, &nv04_graph_ofuncs }, /* swzsurf */
+	{ 0x005f, &nv04_graph_ofuncs }, /* blit */
+	{ 0x0062, &nv04_graph_ofuncs }, /* surf2d */
+	{ 0x0072, &nv04_graph_ofuncs }, /* beta4 */
+	{ 0x0089, &nv04_graph_ofuncs }, /* sifm */
+	{ 0x008a, &nv04_graph_ofuncs }, /* ifc */
+	{ 0x009f, &nv04_graph_ofuncs }, /* blit */
+	{ 0x0093, &nv04_graph_ofuncs }, /* surf3d */
+	{ 0x0094, &nv04_graph_ofuncs }, /* ttri */
+	{ 0x0095, &nv04_graph_ofuncs }, /* mtri */
+	{ 0x0056, &nv04_graph_ofuncs }, /* celcius */
+	{},
+};
+
+static struct nouveau_oclass
+nv15_graph_sclass[] = {
+	{ 0x0012, &nv04_graph_ofuncs }, /* beta1 */
+	{ 0x0019, &nv04_graph_ofuncs }, /* clip */
+	{ 0x0030, &nv04_graph_ofuncs }, /* null */
+	{ 0x0039, &nv04_graph_ofuncs }, /* m2mf */
+	{ 0x0043, &nv04_graph_ofuncs }, /* rop */
+	{ 0x0044, &nv04_graph_ofuncs }, /* pattern */
+	{ 0x004a, &nv04_graph_ofuncs }, /* gdi */
+	{ 0x0052, &nv04_graph_ofuncs }, /* swzsurf */
+	{ 0x005f, &nv04_graph_ofuncs }, /* blit */
+	{ 0x0062, &nv04_graph_ofuncs }, /* surf2d */
+	{ 0x0072, &nv04_graph_ofuncs }, /* beta4 */
+	{ 0x0089, &nv04_graph_ofuncs }, /* sifm */
+	{ 0x008a, &nv04_graph_ofuncs }, /* ifc */
+	{ 0x009f, &nv04_graph_ofuncs }, /* blit */
+	{ 0x0093, &nv04_graph_ofuncs }, /* surf3d */
+	{ 0x0094, &nv04_graph_ofuncs }, /* ttri */
+	{ 0x0095, &nv04_graph_ofuncs }, /* mtri */
+	{ 0x0096, &nv04_graph_ofuncs }, /* celcius */
+	{},
+};
+
+static int
+nv17_graph_mthd_lma_window(struct nouveau_object *object, u32 mthd,
+			   void *args, u32 size)
+{
+	struct nv10_graph_chan *chan = (void *)object->parent;
+	struct nv10_graph_priv *priv = nv10_graph_priv(chan);
+	struct pipe_state *pipe = &chan->pipe_state;
+	u32 pipe_0x0040[1], pipe_0x64c0[8], pipe_0x6a80[3], pipe_0x6ab0[3];
+	u32 xfmode0, xfmode1;
+	u32 data = *(u32 *)args;
+	int i;
+
+	chan->lma_window[(mthd - 0x1638) / 4] = data;
+
+	if (mthd != 0x1644)
+		return 0;
+
+	nv04_graph_idle(priv);
+
+	PIPE_SAVE(priv, pipe_0x0040, 0x0040);
+	PIPE_SAVE(priv, pipe->pipe_0x0200, 0x0200);
+
+	PIPE_RESTORE(priv, chan->lma_window, 0x6790);
+
+	nv04_graph_idle(priv);
+
+	xfmode0 = nv_rd32(priv, NV10_PGRAPH_XFMODE0);
+	xfmode1 = nv_rd32(priv, NV10_PGRAPH_XFMODE1);
+
+	PIPE_SAVE(priv, pipe->pipe_0x4400, 0x4400);
+	PIPE_SAVE(priv, pipe_0x64c0, 0x64c0);
+	PIPE_SAVE(priv, pipe_0x6ab0, 0x6ab0);
+	PIPE_SAVE(priv, pipe_0x6a80, 0x6a80);
+
+	nv04_graph_idle(priv);
+
+	nv_wr32(priv, NV10_PGRAPH_XFMODE0, 0x10000000);
+	nv_wr32(priv, NV10_PGRAPH_XFMODE1, 0x00000000);
+	nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, 0x000064c0);
+	for (i = 0; i < 4; i++)
+		nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x3f800000);
+	for (i = 0; i < 4; i++)
+		nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x00000000);
+
+	nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, 0x00006ab0);
+	for (i = 0; i < 3; i++)
+		nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x3f800000);
+
+	nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, 0x00006a80);
+	for (i = 0; i < 3; i++)
+		nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x00000000);
+
+	nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, 0x00000040);
+	nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x00000008);
+
+	PIPE_RESTORE(priv, pipe->pipe_0x0200, 0x0200);
+
+	nv04_graph_idle(priv);
+
+	PIPE_RESTORE(priv, pipe_0x0040, 0x0040);
+
+	nv_wr32(priv, NV10_PGRAPH_XFMODE0, xfmode0);
+	nv_wr32(priv, NV10_PGRAPH_XFMODE1, xfmode1);
+
+	PIPE_RESTORE(priv, pipe_0x64c0, 0x64c0);
+	PIPE_RESTORE(priv, pipe_0x6ab0, 0x6ab0);
+	PIPE_RESTORE(priv, pipe_0x6a80, 0x6a80);
+	PIPE_RESTORE(priv, pipe->pipe_0x4400, 0x4400);
+
+	nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, 0x000000c0);
+	nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x00000000);
+
+	nv04_graph_idle(priv);
+
+	return 0;
+}
+
+static int
+nv17_graph_mthd_lma_enable(struct nouveau_object *object, u32 mthd,
+			   void *args, u32 size)
+{
+	struct nv10_graph_chan *chan = (void *)object->parent;
+	struct nv10_graph_priv *priv = nv10_graph_priv(chan);
+
+	nv04_graph_idle(priv);
+
+	nv_mask(priv, NV10_PGRAPH_DEBUG_4, 0x00000100, 0x00000100);
+	nv_mask(priv, 0x4006b0, 0x08000000, 0x08000000);
+	return 0;
+}
+
+static struct nouveau_omthds
+nv17_celcius_omthds[] = {
+	{ 0x1638, 0x1638, nv17_graph_mthd_lma_window },
+	{ 0x163c, 0x163c, nv17_graph_mthd_lma_window },
+	{ 0x1640, 0x1640, nv17_graph_mthd_lma_window },
+	{ 0x1644, 0x1644, nv17_graph_mthd_lma_window },
+	{ 0x1658, 0x1658, nv17_graph_mthd_lma_enable },
+	{}
+};
+
+static struct nouveau_oclass
+nv17_graph_sclass[] = {
+	{ 0x0012, &nv04_graph_ofuncs }, /* beta1 */
+	{ 0x0019, &nv04_graph_ofuncs }, /* clip */
+	{ 0x0030, &nv04_graph_ofuncs }, /* null */
+	{ 0x0039, &nv04_graph_ofuncs }, /* m2mf */
+	{ 0x0043, &nv04_graph_ofuncs }, /* rop */
+	{ 0x0044, &nv04_graph_ofuncs }, /* pattern */
+	{ 0x004a, &nv04_graph_ofuncs }, /* gdi */
+	{ 0x0052, &nv04_graph_ofuncs }, /* swzsurf */
+	{ 0x005f, &nv04_graph_ofuncs }, /* blit */
+	{ 0x0062, &nv04_graph_ofuncs }, /* surf2d */
+	{ 0x0072, &nv04_graph_ofuncs }, /* beta4 */
+	{ 0x0089, &nv04_graph_ofuncs }, /* sifm */
+	{ 0x008a, &nv04_graph_ofuncs }, /* ifc */
+	{ 0x009f, &nv04_graph_ofuncs }, /* blit */
+	{ 0x0093, &nv04_graph_ofuncs }, /* surf3d */
+	{ 0x0094, &nv04_graph_ofuncs }, /* ttri */
+	{ 0x0095, &nv04_graph_ofuncs }, /* mtri */
+	{ 0x0099, &nv04_graph_ofuncs, nv17_celcius_omthds },
+	{},
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static struct nv10_graph_chan *
+nv10_graph_channel(struct nv10_graph_priv *priv)
+{
+	struct nv10_graph_chan *chan = NULL;
+	if (nv_rd32(priv, 0x400144) & 0x00010000) {
+		int chid = nv_rd32(priv, 0x400148) >> 24;
+		if (chid < ARRAY_SIZE(priv->chan))
+			chan = priv->chan[chid];
+	}
+	return chan;
+}
+
+static void
+nv10_graph_save_pipe(struct nv10_graph_chan *chan)
+{
+	struct nv10_graph_priv *priv = nv10_graph_priv(chan);
+	struct pipe_state *pipe = &chan->pipe_state;
+
+	PIPE_SAVE(priv, pipe->pipe_0x4400, 0x4400);
+	PIPE_SAVE(priv, pipe->pipe_0x0200, 0x0200);
+	PIPE_SAVE(priv, pipe->pipe_0x6400, 0x6400);
+	PIPE_SAVE(priv, pipe->pipe_0x6800, 0x6800);
+	PIPE_SAVE(priv, pipe->pipe_0x6c00, 0x6c00);
+	PIPE_SAVE(priv, pipe->pipe_0x7000, 0x7000);
+	PIPE_SAVE(priv, pipe->pipe_0x7400, 0x7400);
+	PIPE_SAVE(priv, pipe->pipe_0x7800, 0x7800);
+	PIPE_SAVE(priv, pipe->pipe_0x0040, 0x0040);
+	PIPE_SAVE(priv, pipe->pipe_0x0000, 0x0000);
+}
+
+static void
+nv10_graph_load_pipe(struct nv10_graph_chan *chan)
+{
+	struct nv10_graph_priv *priv = nv10_graph_priv(chan);
+	struct pipe_state *pipe = &chan->pipe_state;
+	u32 xfmode0, xfmode1;
+	int i;
+
+	nv04_graph_idle(priv);
+	/* XXX check haiku comments */
+	xfmode0 = nv_rd32(priv, NV10_PGRAPH_XFMODE0);
+	xfmode1 = nv_rd32(priv, NV10_PGRAPH_XFMODE1);
+	nv_wr32(priv, NV10_PGRAPH_XFMODE0, 0x10000000);
+	nv_wr32(priv, NV10_PGRAPH_XFMODE1, 0x00000000);
+	nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, 0x000064c0);
+	for (i = 0; i < 4; i++)
+		nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x3f800000);
+	for (i = 0; i < 4; i++)
+		nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x00000000);
+
+	nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, 0x00006ab0);
+	for (i = 0; i < 3; i++)
+		nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x3f800000);
+
+	nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, 0x00006a80);
+	for (i = 0; i < 3; i++)
+		nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x00000000);
+
+	nv_wr32(priv, NV10_PGRAPH_PIPE_ADDRESS, 0x00000040);
+	nv_wr32(priv, NV10_PGRAPH_PIPE_DATA, 0x00000008);
+
+
+	PIPE_RESTORE(priv, pipe->pipe_0x0200, 0x0200);
+	nv04_graph_idle(priv);
+
+	/* restore XFMODE */
+	nv_wr32(priv, NV10_PGRAPH_XFMODE0, xfmode0);
+	nv_wr32(priv, NV10_PGRAPH_XFMODE1, xfmode1);
+	PIPE_RESTORE(priv, pipe->pipe_0x6400, 0x6400);
+	PIPE_RESTORE(priv, pipe->pipe_0x6800, 0x6800);
+	PIPE_RESTORE(priv, pipe->pipe_0x6c00, 0x6c00);
+	PIPE_RESTORE(priv, pipe->pipe_0x7000, 0x7000);
+	PIPE_RESTORE(priv, pipe->pipe_0x7400, 0x7400);
+	PIPE_RESTORE(priv, pipe->pipe_0x7800, 0x7800);
+	PIPE_RESTORE(priv, pipe->pipe_0x4400, 0x4400);
+	PIPE_RESTORE(priv, pipe->pipe_0x0000, 0x0000);
+	PIPE_RESTORE(priv, pipe->pipe_0x0040, 0x0040);
+	nv04_graph_idle(priv);
+}
+
+static void
+nv10_graph_create_pipe(struct nv10_graph_chan *chan)
+{
+	struct nv10_graph_priv *priv = nv10_graph_priv(chan);
+	struct pipe_state *pipe_state = &chan->pipe_state;
+	u32 *pipe_state_addr;
+	int i;
+#define PIPE_INIT(addr) \
+	do { \
+		pipe_state_addr = pipe_state->pipe_##addr; \
+	} while (0)
+#define PIPE_INIT_END(addr) \
+	do { \
+		u32 *__end_addr = pipe_state->pipe_##addr + \
+				ARRAY_SIZE(pipe_state->pipe_##addr); \
+		if (pipe_state_addr != __end_addr) \
+			nv_error(priv, "incomplete pipe init for 0x%x :  %p/%p\n", \
+				addr, pipe_state_addr, __end_addr); \
+	} while (0)
+#define NV_WRITE_PIPE_INIT(value) *(pipe_state_addr++) = value
+
+	PIPE_INIT(0x0200);
+	for (i = 0; i < 48; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	PIPE_INIT_END(0x0200);
+
+	PIPE_INIT(0x6400);
+	for (i = 0; i < 211; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x3f800000);
+	NV_WRITE_PIPE_INIT(0x40000000);
+	NV_WRITE_PIPE_INIT(0x40000000);
+	NV_WRITE_PIPE_INIT(0x40000000);
+	NV_WRITE_PIPE_INIT(0x40000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x3f800000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x3f000000);
+	NV_WRITE_PIPE_INIT(0x3f000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x3f800000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x3f800000);
+	NV_WRITE_PIPE_INIT(0x3f800000);
+	NV_WRITE_PIPE_INIT(0x3f800000);
+	NV_WRITE_PIPE_INIT(0x3f800000);
+	PIPE_INIT_END(0x6400);
+
+	PIPE_INIT(0x6800);
+	for (i = 0; i < 162; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x3f800000);
+	for (i = 0; i < 25; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	PIPE_INIT_END(0x6800);
+
+	PIPE_INIT(0x6c00);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0xbf800000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	PIPE_INIT_END(0x6c00);
+
+	PIPE_INIT(0x7000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x7149f2ca);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x7149f2ca);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x7149f2ca);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x7149f2ca);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x7149f2ca);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x7149f2ca);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x7149f2ca);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x00000000);
+	NV_WRITE_PIPE_INIT(0x7149f2ca);
+	for (i = 0; i < 35; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	PIPE_INIT_END(0x7000);
+
+	PIPE_INIT(0x7400);
+	for (i = 0; i < 48; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	PIPE_INIT_END(0x7400);
+
+	PIPE_INIT(0x7800);
+	for (i = 0; i < 48; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	PIPE_INIT_END(0x7800);
+
+	PIPE_INIT(0x4400);
+	for (i = 0; i < 32; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	PIPE_INIT_END(0x4400);
+
+	PIPE_INIT(0x0000);
+	for (i = 0; i < 16; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	PIPE_INIT_END(0x0000);
+
+	PIPE_INIT(0x0040);
+	for (i = 0; i < 4; i++)
+		NV_WRITE_PIPE_INIT(0x00000000);
+	PIPE_INIT_END(0x0040);
+
+#undef PIPE_INIT
+#undef PIPE_INIT_END
+#undef NV_WRITE_PIPE_INIT
+}
+
+static int
+nv10_graph_ctx_regs_find_offset(struct nv10_graph_priv *priv, int reg)
+{
+	int i;
+	for (i = 0; i < ARRAY_SIZE(nv10_graph_ctx_regs); i++) {
+		if (nv10_graph_ctx_regs[i] == reg)
+			return i;
+	}
+	nv_error(priv, "unknow offset nv10_ctx_regs %d\n", reg);
+	return -1;
+}
+
+static int
+nv17_graph_ctx_regs_find_offset(struct nv10_graph_priv *priv, int reg)
+{
+	int i;
+	for (i = 0; i < ARRAY_SIZE(nv17_graph_ctx_regs); i++) {
+		if (nv17_graph_ctx_regs[i] == reg)
+			return i;
+	}
+	nv_error(priv, "unknow offset nv17_ctx_regs %d\n", reg);
+	return -1;
+}
+
+static void
+nv10_graph_load_dma_vtxbuf(struct nv10_graph_chan *chan, int chid, u32 inst)
+{
+	struct nv10_graph_priv *priv = nv10_graph_priv(chan);
+	u32 st2, st2_dl, st2_dh, fifo_ptr, fifo[0x60/4];
+	u32 ctx_user, ctx_switch[5];
+	int i, subchan = -1;
+
+	/* NV10TCL_DMA_VTXBUF (method 0x18c) modifies hidden state
+	 * that cannot be restored via MMIO. Do it through the FIFO
+	 * instead.
+	 */
+
+	/* Look for a celsius object */
+	for (i = 0; i < 8; i++) {
+		int class = nv_rd32(priv, NV10_PGRAPH_CTX_CACHE(i, 0)) & 0xfff;
+
+		if (class == 0x56 || class == 0x96 || class == 0x99) {
+			subchan = i;
+			break;
+		}
+	}
+
+	if (subchan < 0 || !inst)
+		return;
+
+	/* Save the current ctx object */
+	ctx_user = nv_rd32(priv, NV10_PGRAPH_CTX_USER);
+	for (i = 0; i < 5; i++)
+		ctx_switch[i] = nv_rd32(priv, NV10_PGRAPH_CTX_SWITCH(i));
+
+	/* Save the FIFO state */
+	st2 = nv_rd32(priv, NV10_PGRAPH_FFINTFC_ST2);
+	st2_dl = nv_rd32(priv, NV10_PGRAPH_FFINTFC_ST2_DL);
+	st2_dh = nv_rd32(priv, NV10_PGRAPH_FFINTFC_ST2_DH);
+	fifo_ptr = nv_rd32(priv, NV10_PGRAPH_FFINTFC_FIFO_PTR);
+
+	for (i = 0; i < ARRAY_SIZE(fifo); i++)
+		fifo[i] = nv_rd32(priv, 0x4007a0 + 4 * i);
+
+	/* Switch to the celsius subchannel */
+	for (i = 0; i < 5; i++)
+		nv_wr32(priv, NV10_PGRAPH_CTX_SWITCH(i),
+			nv_rd32(priv, NV10_PGRAPH_CTX_CACHE(subchan, i)));
+	nv_mask(priv, NV10_PGRAPH_CTX_USER, 0xe000, subchan << 13);
+
+	/* Inject NV10TCL_DMA_VTXBUF */
+	nv_wr32(priv, NV10_PGRAPH_FFINTFC_FIFO_PTR, 0);
+	nv_wr32(priv, NV10_PGRAPH_FFINTFC_ST2,
+		0x2c000000 | chid << 20 | subchan << 16 | 0x18c);
+	nv_wr32(priv, NV10_PGRAPH_FFINTFC_ST2_DL, inst);
+	nv_mask(priv, NV10_PGRAPH_CTX_CONTROL, 0, 0x10000);
+	nv_mask(priv, NV04_PGRAPH_FIFO, 0x00000001, 0x00000001);
+	nv_mask(priv, NV04_PGRAPH_FIFO, 0x00000001, 0x00000000);
+
+	/* Restore the FIFO state */
+	for (i = 0; i < ARRAY_SIZE(fifo); i++)
+		nv_wr32(priv, 0x4007a0 + 4 * i, fifo[i]);
+
+	nv_wr32(priv, NV10_PGRAPH_FFINTFC_FIFO_PTR, fifo_ptr);
+	nv_wr32(priv, NV10_PGRAPH_FFINTFC_ST2, st2);
+	nv_wr32(priv, NV10_PGRAPH_FFINTFC_ST2_DL, st2_dl);
+	nv_wr32(priv, NV10_PGRAPH_FFINTFC_ST2_DH, st2_dh);
+
+	/* Restore the current ctx object */
+	for (i = 0; i < 5; i++)
+		nv_wr32(priv, NV10_PGRAPH_CTX_SWITCH(i), ctx_switch[i]);
+	nv_wr32(priv, NV10_PGRAPH_CTX_USER, ctx_user);
+}
+
+static int
+nv10_graph_load_context(struct nv10_graph_chan *chan, int chid)
+{
+	struct nv10_graph_priv *priv = nv10_graph_priv(chan);
+	u32 inst;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(nv10_graph_ctx_regs); i++)
+		nv_wr32(priv, nv10_graph_ctx_regs[i], chan->nv10[i]);
+
+	if (nv_device(priv)->chipset >= 0x17) {
+		for (i = 0; i < ARRAY_SIZE(nv17_graph_ctx_regs); i++)
+			nv_wr32(priv, nv17_graph_ctx_regs[i], chan->nv17[i]);
+	}
+
+	nv10_graph_load_pipe(chan);
+
+	inst = nv_rd32(priv, NV10_PGRAPH_GLOBALSTATE1) & 0xffff;
+	nv10_graph_load_dma_vtxbuf(chan, chid, inst);
+
+	nv_wr32(priv, NV10_PGRAPH_CTX_CONTROL, 0x10010100);
+	nv_mask(priv, NV10_PGRAPH_CTX_USER, 0xff000000, chid << 24);
+	nv_mask(priv, NV10_PGRAPH_FFINTFC_ST2, 0x30000000, 0x00000000);
+	return 0;
+}
+
+static int
+nv10_graph_unload_context(struct nv10_graph_chan *chan)
+{
+	struct nv10_graph_priv *priv = nv10_graph_priv(chan);
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(nv10_graph_ctx_regs); i++)
+		chan->nv10[i] = nv_rd32(priv, nv10_graph_ctx_regs[i]);
+
+	if (nv_device(priv)->chipset >= 0x17) {
+		for (i = 0; i < ARRAY_SIZE(nv17_graph_ctx_regs); i++)
+			chan->nv17[i] = nv_rd32(priv, nv17_graph_ctx_regs[i]);
+	}
+
+	nv10_graph_save_pipe(chan);
+
+	nv_wr32(priv, NV10_PGRAPH_CTX_CONTROL, 0x10000000);
+	nv_mask(priv, NV10_PGRAPH_CTX_USER, 0xff000000, 0x1f000000);
+	return 0;
+}
+
+static void
+nv10_graph_context_switch(struct nv10_graph_priv *priv)
+{
+	struct nv10_graph_chan *prev = NULL;
+	struct nv10_graph_chan *next = NULL;
+	unsigned long flags;
+	int chid;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	nv04_graph_idle(priv);
+
+	/* If previous context is valid, we need to save it */
+	prev = nv10_graph_channel(priv);
+	if (prev)
+		nv10_graph_unload_context(prev);
+
+	/* load context for next channel */
+	chid = (nv_rd32(priv, NV04_PGRAPH_TRAPPED_ADDR) >> 20) & 0x1f;
+	next = priv->chan[chid];
+	if (next)
+		nv10_graph_load_context(next, chid);
+
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+#define NV_WRITE_CTX(reg, val) do { \
+	int offset = nv10_graph_ctx_regs_find_offset(priv, reg); \
+	if (offset > 0) \
+		chan->nv10[offset] = val; \
+	} while (0)
+
+#define NV17_WRITE_CTX(reg, val) do { \
+	int offset = nv17_graph_ctx_regs_find_offset(priv, reg); \
+	if (offset > 0) \
+		chan->nv17[offset] = val; \
+	} while (0)
+
+static int
+nv10_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nouveau_fifo_chan *fifo = (void *)parent;
+	struct nv10_graph_priv *priv = (void *)engine;
+	struct nv10_graph_chan *chan;
+	unsigned long flags;
+	int ret;
+
+	ret = nouveau_object_create(parent, engine, oclass, 0, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	if (priv->chan[fifo->chid]) {
+		*pobject = nv_object(priv->chan[fifo->chid]);
+		atomic_inc(&(*pobject)->refcount);
+		spin_unlock_irqrestore(&priv->lock, flags);
+		nouveau_object_destroy(&chan->base);
+		return 1;
+	}
+
+	NV_WRITE_CTX(0x00400e88, 0x08000000);
+	NV_WRITE_CTX(0x00400e9c, 0x4b7fffff);
+	NV_WRITE_CTX(NV03_PGRAPH_XY_LOGIC_MISC0, 0x0001ffff);
+	NV_WRITE_CTX(0x00400e10, 0x00001000);
+	NV_WRITE_CTX(0x00400e14, 0x00001000);
+	NV_WRITE_CTX(0x00400e30, 0x00080008);
+	NV_WRITE_CTX(0x00400e34, 0x00080008);
+	if (nv_device(priv)->chipset >= 0x17) {
+		/* is it really needed ??? */
+		NV17_WRITE_CTX(NV10_PGRAPH_DEBUG_4,
+					nv_rd32(priv, NV10_PGRAPH_DEBUG_4));
+		NV17_WRITE_CTX(0x004006b0, nv_rd32(priv, 0x004006b0));
+		NV17_WRITE_CTX(0x00400eac, 0x0fff0000);
+		NV17_WRITE_CTX(0x00400eb0, 0x0fff0000);
+		NV17_WRITE_CTX(0x00400ec0, 0x00000080);
+		NV17_WRITE_CTX(0x00400ed0, 0x00000080);
+	}
+	NV_WRITE_CTX(NV10_PGRAPH_CTX_USER, chan->chid << 24);
+
+	nv10_graph_create_pipe(chan);
+
+	priv->chan[fifo->chid] = chan;
+	chan->chid = fifo->chid;
+	spin_unlock_irqrestore(&priv->lock, flags);
+	return 0;
+}
+
+static void
+nv10_graph_context_dtor(struct nouveau_object *object)
+{
+	struct nv10_graph_priv *priv = (void *)object->engine;
+	struct nv10_graph_chan *chan = (void *)object;
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	priv->chan[chan->chid] = NULL;
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	nouveau_object_destroy(&chan->base);
+}
+
+static int
+nv10_graph_context_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv10_graph_priv *priv = (void *)object->engine;
+	struct nv10_graph_chan *chan = (void *)object;
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	nv_mask(priv, NV04_PGRAPH_FIFO, 0x00000001, 0x00000000);
+	if (nv10_graph_channel(priv) == chan)
+		nv10_graph_unload_context(chan);
+	nv_mask(priv, NV04_PGRAPH_FIFO, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	return nouveau_object_fini(&chan->base, suspend);
+}
+
+static struct nouveau_oclass
+nv10_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x10),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv10_graph_context_ctor,
+		.dtor = nv10_graph_context_dtor,
+		.init = nouveau_object_init,
+		.fini = nv10_graph_context_fini,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static void
+nv10_graph_tile_prog(struct nouveau_engine *engine, int i)
+{
+	struct nouveau_fb_tile *tile = &nouveau_fb(engine)->tile.region[i];
+	struct nouveau_fifo *pfifo = nouveau_fifo(engine);
+	struct nv10_graph_priv *priv = (void *)engine;
+	unsigned long flags;
+
+	pfifo->pause(pfifo, &flags);
+	nv04_graph_idle(priv);
+
+	nv_wr32(priv, NV10_PGRAPH_TLIMIT(i), tile->limit);
+	nv_wr32(priv, NV10_PGRAPH_TSIZE(i), tile->pitch);
+	nv_wr32(priv, NV10_PGRAPH_TILE(i), tile->addr);
+
+	pfifo->start(pfifo, &flags);
+}
+
+const struct nouveau_bitfield nv10_graph_intr_name[] = {
+	{ NV_PGRAPH_INTR_NOTIFY, "NOTIFY" },
+	{ NV_PGRAPH_INTR_ERROR,  "ERROR"  },
+	{}
+};
+
+const struct nouveau_bitfield nv10_graph_nstatus[] = {
+	{ NV10_PGRAPH_NSTATUS_STATE_IN_USE,       "STATE_IN_USE" },
+	{ NV10_PGRAPH_NSTATUS_INVALID_STATE,      "INVALID_STATE" },
+	{ NV10_PGRAPH_NSTATUS_BAD_ARGUMENT,       "BAD_ARGUMENT" },
+	{ NV10_PGRAPH_NSTATUS_PROTECTION_FAULT,   "PROTECTION_FAULT" },
+	{}
+};
+
+static void
+nv10_graph_intr(struct nouveau_subdev *subdev)
+{
+	struct nv10_graph_priv *priv = (void *)subdev;
+	struct nv10_graph_chan *chan = NULL;
+	struct nouveau_namedb *namedb = NULL;
+	struct nouveau_handle *handle = NULL;
+	u32 stat = nv_rd32(priv, NV03_PGRAPH_INTR);
+	u32 nsource = nv_rd32(priv, NV03_PGRAPH_NSOURCE);
+	u32 nstatus = nv_rd32(priv, NV03_PGRAPH_NSTATUS);
+	u32 addr = nv_rd32(priv, NV04_PGRAPH_TRAPPED_ADDR);
+	u32 chid = (addr & 0x01f00000) >> 20;
+	u32 subc = (addr & 0x00070000) >> 16;
+	u32 mthd = (addr & 0x00001ffc);
+	u32 data = nv_rd32(priv, NV04_PGRAPH_TRAPPED_DATA);
+	u32 class = nv_rd32(priv, 0x400160 + subc * 4) & 0xfff;
+	u32 show = stat;
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	chan = priv->chan[chid];
+	if (chan)
+		namedb = (void *)nv_pclass(nv_object(chan), NV_NAMEDB_CLASS);
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	if (stat & NV_PGRAPH_INTR_ERROR) {
+		if (chan && (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD)) {
+			handle = nouveau_namedb_get_class(namedb, class);
+			if (handle && !nv_call(handle->object, mthd, data))
+				show &= ~NV_PGRAPH_INTR_ERROR;
+		}
+	}
+
+	if (stat & NV_PGRAPH_INTR_CONTEXT_SWITCH) {
+		nv_wr32(priv, NV03_PGRAPH_INTR, NV_PGRAPH_INTR_CONTEXT_SWITCH);
+		stat &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+		show &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+		nv10_graph_context_switch(priv);
+	}
+
+	nv_wr32(priv, NV03_PGRAPH_INTR, stat);
+	nv_wr32(priv, NV04_PGRAPH_FIFO, 0x00000001);
+
+	if (show) {
+		nv_error(priv, "");
+		nouveau_bitfield_print(nv10_graph_intr_name, show);
+		printk(" nsource:");
+		nouveau_bitfield_print(nv04_graph_nsource, nsource);
+		printk(" nstatus:");
+		nouveau_bitfield_print(nv10_graph_nstatus, nstatus);
+		printk("\n");
+		nv_error(priv, "ch %d/%d class 0x%04x "
+			       "mthd 0x%04x data 0x%08x\n",
+			 chid, subc, class, mthd, data);
+	}
+
+	nouveau_namedb_put(handle);
+}
+
+static int
+nv10_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv10_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00001000;
+	nv_subdev(priv)->intr = nv10_graph_intr;
+	nv_engine(priv)->cclass = &nv10_graph_cclass;
+
+	if (nv_device(priv)->chipset <= 0x10)
+		nv_engine(priv)->sclass = nv10_graph_sclass;
+	else
+	if (nv_device(priv)->chipset <  0x17 ||
+	    nv_device(priv)->chipset == 0x1a)
+		nv_engine(priv)->sclass = nv15_graph_sclass;
+	else
+		nv_engine(priv)->sclass = nv17_graph_sclass;
+
+	nv_engine(priv)->tile_prog = nv10_graph_tile_prog;
+	spin_lock_init(&priv->lock);
+	return 0;
+}
+
+static void
+nv10_graph_dtor(struct nouveau_object *object)
+{
+	struct nv10_graph_priv *priv = (void *)object;
+	nouveau_graph_destroy(&priv->base);
+}
+
+static int
+nv10_graph_init(struct nouveau_object *object)
+{
+	struct nouveau_engine *engine = nv_engine(object);
+	struct nouveau_fb *pfb = nouveau_fb(object);
+	struct nv10_graph_priv *priv = (void *)engine;
+	int ret, i;
+
+	ret = nouveau_graph_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
+	nv_wr32(priv, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
+
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0xFFFFFFFF);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0x00000000);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_1, 0x00118700);
+	/* nv_wr32(priv, NV04_PGRAPH_DEBUG_2, 0x24E00810); */ /* 0x25f92ad9 */
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_2, 0x25f92ad9);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_3, 0x55DE0830 | (1 << 29) | (1 << 31));
+
+	if (nv_device(priv)->chipset >= 0x17) {
+		nv_wr32(priv, NV10_PGRAPH_DEBUG_4, 0x1f000000);
+		nv_wr32(priv, 0x400a10, 0x03ff3fb6);
+		nv_wr32(priv, 0x400838, 0x002f8684);
+		nv_wr32(priv, 0x40083c, 0x00115f3f);
+		nv_wr32(priv, 0x4006b0, 0x40000020);
+	} else {
+		nv_wr32(priv, NV10_PGRAPH_DEBUG_4, 0x00000000);
+	}
+
+	/* Turn all the tiling regions off. */
+	for (i = 0; i < pfb->tile.regions; i++)
+		engine->tile_prog(engine, i);
+
+	nv_wr32(priv, NV10_PGRAPH_CTX_SWITCH(0), 0x00000000);
+	nv_wr32(priv, NV10_PGRAPH_CTX_SWITCH(1), 0x00000000);
+	nv_wr32(priv, NV10_PGRAPH_CTX_SWITCH(2), 0x00000000);
+	nv_wr32(priv, NV10_PGRAPH_CTX_SWITCH(3), 0x00000000);
+	nv_wr32(priv, NV10_PGRAPH_CTX_SWITCH(4), 0x00000000);
+	nv_wr32(priv, NV10_PGRAPH_STATE, 0xFFFFFFFF);
+
+	nv_mask(priv, NV10_PGRAPH_CTX_USER, 0xff000000, 0x1f000000);
+	nv_wr32(priv, NV10_PGRAPH_CTX_CONTROL, 0x10000100);
+	nv_wr32(priv, NV10_PGRAPH_FFINTFC_ST2, 0x08000000);
+	return 0;
+}
+
+static int
+nv10_graph_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv10_graph_priv *priv = (void *)object;
+	return nouveau_graph_fini(&priv->base, suspend);
+}
+
+struct nouveau_oclass
+nv10_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x10),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv10_graph_ctor,
+		.dtor = nv10_graph_dtor,
+		.init = nv10_graph_init,
+		.fini = nv10_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv20.c b/nvkm/engine/graph/nv20.c
new file mode 100644
index 000000000..5b20401bf
--- /dev/null
+++ b/nvkm/engine/graph/nv20.c
@@ -0,0 +1,381 @@
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/handle.h>
+#include <core/enum.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+
+#include <engine/graph.h>
+#include <engine/fifo.h>
+
+#include "nv20.h"
+#include "regs.h"
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv20_graph_sclass[] = {
+	{ 0x0012, &nv04_graph_ofuncs, NULL }, /* beta1 */
+	{ 0x0019, &nv04_graph_ofuncs, NULL }, /* clip */
+	{ 0x0030, &nv04_graph_ofuncs, NULL }, /* null */
+	{ 0x0039, &nv04_graph_ofuncs, NULL }, /* m2mf */
+	{ 0x0043, &nv04_graph_ofuncs, NULL }, /* rop */
+	{ 0x0044, &nv04_graph_ofuncs, NULL }, /* patt */
+	{ 0x004a, &nv04_graph_ofuncs, NULL }, /* gdi */
+	{ 0x0062, &nv04_graph_ofuncs, NULL }, /* surf2d */
+	{ 0x0072, &nv04_graph_ofuncs, NULL }, /* beta4 */
+	{ 0x0089, &nv04_graph_ofuncs, NULL }, /* sifm */
+	{ 0x008a, &nv04_graph_ofuncs, NULL }, /* ifc */
+	{ 0x0096, &nv04_graph_ofuncs, NULL }, /* celcius */
+	{ 0x0097, &nv04_graph_ofuncs, NULL }, /* kelvin */
+	{ 0x009e, &nv04_graph_ofuncs, NULL }, /* swzsurf */
+	{ 0x009f, &nv04_graph_ofuncs, NULL }, /* imageblit */
+	{},
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static int
+nv20_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nv20_graph_chan *chan;
+	int ret, i;
+
+	ret = nouveau_graph_context_create(parent, engine, oclass, NULL,
+					   0x37f0, 16, NVOBJ_FLAG_ZERO_ALLOC,
+					   &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	chan->chid = nouveau_fifo_chan(parent)->chid;
+
+	nv_wo32(chan, 0x0000, 0x00000001 | (chan->chid << 24));
+	nv_wo32(chan, 0x033c, 0xffff0000);
+	nv_wo32(chan, 0x03a0, 0x0fff0000);
+	nv_wo32(chan, 0x03a4, 0x0fff0000);
+	nv_wo32(chan, 0x047c, 0x00000101);
+	nv_wo32(chan, 0x0490, 0x00000111);
+	nv_wo32(chan, 0x04a8, 0x44400000);
+	for (i = 0x04d4; i <= 0x04e0; i += 4)
+		nv_wo32(chan, i, 0x00030303);
+	for (i = 0x04f4; i <= 0x0500; i += 4)
+		nv_wo32(chan, i, 0x00080000);
+	for (i = 0x050c; i <= 0x0518; i += 4)
+		nv_wo32(chan, i, 0x01012000);
+	for (i = 0x051c; i <= 0x0528; i += 4)
+		nv_wo32(chan, i, 0x000105b8);
+	for (i = 0x052c; i <= 0x0538; i += 4)
+		nv_wo32(chan, i, 0x00080008);
+	for (i = 0x055c; i <= 0x0598; i += 4)
+		nv_wo32(chan, i, 0x07ff0000);
+	nv_wo32(chan, 0x05a4, 0x4b7fffff);
+	nv_wo32(chan, 0x05fc, 0x00000001);
+	nv_wo32(chan, 0x0604, 0x00004000);
+	nv_wo32(chan, 0x0610, 0x00000001);
+	nv_wo32(chan, 0x0618, 0x00040000);
+	nv_wo32(chan, 0x061c, 0x00010000);
+	for (i = 0x1c1c; i <= 0x248c; i += 16) {
+		nv_wo32(chan, (i + 0), 0x10700ff9);
+		nv_wo32(chan, (i + 4), 0x0436086c);
+		nv_wo32(chan, (i + 8), 0x000c001b);
+	}
+	nv_wo32(chan, 0x281c, 0x3f800000);
+	nv_wo32(chan, 0x2830, 0x3f800000);
+	nv_wo32(chan, 0x285c, 0x40000000);
+	nv_wo32(chan, 0x2860, 0x3f800000);
+	nv_wo32(chan, 0x2864, 0x3f000000);
+	nv_wo32(chan, 0x286c, 0x40000000);
+	nv_wo32(chan, 0x2870, 0x3f800000);
+	nv_wo32(chan, 0x2878, 0xbf800000);
+	nv_wo32(chan, 0x2880, 0xbf800000);
+	nv_wo32(chan, 0x34a4, 0x000fe000);
+	nv_wo32(chan, 0x3530, 0x000003f8);
+	nv_wo32(chan, 0x3540, 0x002fe000);
+	for (i = 0x355c; i <= 0x3578; i += 4)
+		nv_wo32(chan, i, 0x001c527c);
+	return 0;
+}
+
+int
+nv20_graph_context_init(struct nouveau_object *object)
+{
+	struct nv20_graph_priv *priv = (void *)object->engine;
+	struct nv20_graph_chan *chan = (void *)object;
+	int ret;
+
+	ret = nouveau_graph_context_init(&chan->base);
+	if (ret)
+		return ret;
+
+	nv_wo32(priv->ctxtab, chan->chid * 4, nv_gpuobj(chan)->addr >> 4);
+	return 0;
+}
+
+int
+nv20_graph_context_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv20_graph_priv *priv = (void *)object->engine;
+	struct nv20_graph_chan *chan = (void *)object;
+	int chid = -1;
+
+	nv_mask(priv, 0x400720, 0x00000001, 0x00000000);
+	if (nv_rd32(priv, 0x400144) & 0x00010000)
+		chid = (nv_rd32(priv, 0x400148) & 0x1f000000) >> 24;
+	if (chan->chid == chid) {
+		nv_wr32(priv, 0x400784, nv_gpuobj(chan)->addr >> 4);
+		nv_wr32(priv, 0x400788, 0x00000002);
+		nv_wait(priv, 0x400700, 0xffffffff, 0x00000000);
+		nv_wr32(priv, 0x400144, 0x10000000);
+		nv_mask(priv, 0x400148, 0xff000000, 0x1f000000);
+	}
+	nv_mask(priv, 0x400720, 0x00000001, 0x00000001);
+
+	nv_wo32(priv->ctxtab, chan->chid * 4, 0x00000000);
+	return nouveau_graph_context_fini(&chan->base, suspend);
+}
+
+static struct nouveau_oclass
+nv20_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x20),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv20_graph_context_ctor,
+		.dtor = _nouveau_graph_context_dtor,
+		.init = nv20_graph_context_init,
+		.fini = nv20_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+void
+nv20_graph_tile_prog(struct nouveau_engine *engine, int i)
+{
+	struct nouveau_fb_tile *tile = &nouveau_fb(engine)->tile.region[i];
+	struct nouveau_fifo *pfifo = nouveau_fifo(engine);
+	struct nv20_graph_priv *priv = (void *)engine;
+	unsigned long flags;
+
+	pfifo->pause(pfifo, &flags);
+	nv04_graph_idle(priv);
+
+	nv_wr32(priv, NV20_PGRAPH_TLIMIT(i), tile->limit);
+	nv_wr32(priv, NV20_PGRAPH_TSIZE(i), tile->pitch);
+	nv_wr32(priv, NV20_PGRAPH_TILE(i), tile->addr);
+
+	nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00EA0030 + 4 * i);
+	nv_wr32(priv, NV10_PGRAPH_RDI_DATA, tile->limit);
+	nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00EA0050 + 4 * i);
+	nv_wr32(priv, NV10_PGRAPH_RDI_DATA, tile->pitch);
+	nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00EA0010 + 4 * i);
+	nv_wr32(priv, NV10_PGRAPH_RDI_DATA, tile->addr);
+
+	if (nv_device(engine)->chipset != 0x34) {
+		nv_wr32(priv, NV20_PGRAPH_ZCOMP(i), tile->zcomp);
+		nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00ea0090 + 4 * i);
+		nv_wr32(priv, NV10_PGRAPH_RDI_DATA, tile->zcomp);
+	}
+
+	pfifo->start(pfifo, &flags);
+}
+
+void
+nv20_graph_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_engine *engine = nv_engine(subdev);
+	struct nouveau_object *engctx;
+	struct nouveau_handle *handle;
+	struct nv20_graph_priv *priv = (void *)subdev;
+	u32 stat = nv_rd32(priv, NV03_PGRAPH_INTR);
+	u32 nsource = nv_rd32(priv, NV03_PGRAPH_NSOURCE);
+	u32 nstatus = nv_rd32(priv, NV03_PGRAPH_NSTATUS);
+	u32 addr = nv_rd32(priv, NV04_PGRAPH_TRAPPED_ADDR);
+	u32 chid = (addr & 0x01f00000) >> 20;
+	u32 subc = (addr & 0x00070000) >> 16;
+	u32 mthd = (addr & 0x00001ffc);
+	u32 data = nv_rd32(priv, NV04_PGRAPH_TRAPPED_DATA);
+	u32 class = nv_rd32(priv, 0x400160 + subc * 4) & 0xfff;
+	u32 show = stat;
+
+	engctx = nouveau_engctx_get(engine, chid);
+	if (stat & NV_PGRAPH_INTR_ERROR) {
+		if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
+			handle = nouveau_handle_get_class(engctx, class);
+			if (handle && !nv_call(handle->object, mthd, data))
+				show &= ~NV_PGRAPH_INTR_ERROR;
+			nouveau_handle_put(handle);
+		}
+	}
+
+	nv_wr32(priv, NV03_PGRAPH_INTR, stat);
+	nv_wr32(priv, NV04_PGRAPH_FIFO, 0x00000001);
+
+	if (show) {
+		nv_error(priv, "");
+		nouveau_bitfield_print(nv10_graph_intr_name, show);
+		printk(" nsource:");
+		nouveau_bitfield_print(nv04_graph_nsource, nsource);
+		printk(" nstatus:");
+		nouveau_bitfield_print(nv10_graph_nstatus, nstatus);
+		printk("\n");
+		nv_error(priv, "ch %d/%d class 0x%04x mthd 0x%04x data 0x%08x\n",
+			chid, subc, class, mthd, data);
+	}
+
+	nouveau_engctx_put(engctx);
+}
+
+static int
+nv20_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv20_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 32 * 4, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &priv->ctxtab);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00001000;
+	nv_subdev(priv)->intr = nv20_graph_intr;
+	nv_engine(priv)->cclass = &nv20_graph_cclass;
+	nv_engine(priv)->sclass = nv20_graph_sclass;
+	nv_engine(priv)->tile_prog = nv20_graph_tile_prog;
+	return 0;
+}
+
+void
+nv20_graph_dtor(struct nouveau_object *object)
+{
+	struct nv20_graph_priv *priv = (void *)object;
+	nouveau_gpuobj_ref(NULL, &priv->ctxtab);
+	nouveau_graph_destroy(&priv->base);
+}
+
+int
+nv20_graph_init(struct nouveau_object *object)
+{
+	struct nouveau_engine *engine = nv_engine(object);
+	struct nv20_graph_priv *priv = (void *)engine;
+	struct nouveau_fb *pfb = nouveau_fb(object);
+	u32 tmp, vramsz;
+	int ret, i;
+
+	ret = nouveau_graph_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, NV20_PGRAPH_CHANNEL_CTX_TABLE, priv->ctxtab->addr >> 4);
+
+	if (nv_device(priv)->chipset == 0x20) {
+		nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x003d0000);
+		for (i = 0; i < 15; i++)
+			nv_wr32(priv, NV10_PGRAPH_RDI_DATA, 0x00000000);
+		nv_wait(priv, 0x400700, 0xffffffff, 0x00000000);
+	} else {
+		nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x02c80000);
+		for (i = 0; i < 32; i++)
+			nv_wr32(priv, NV10_PGRAPH_RDI_DATA, 0x00000000);
+		nv_wait(priv, 0x400700, 0xffffffff, 0x00000000);
+	}
+
+	nv_wr32(priv, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
+	nv_wr32(priv, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
+
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0xFFFFFFFF);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0x00000000);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_1, 0x00118700);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_3, 0xF3CE0475); /* 0x4 = auto ctx switch */
+	nv_wr32(priv, NV10_PGRAPH_DEBUG_4, 0x00000000);
+	nv_wr32(priv, 0x40009C           , 0x00000040);
+
+	if (nv_device(priv)->chipset >= 0x25) {
+		nv_wr32(priv, 0x400890, 0x00a8cfff);
+		nv_wr32(priv, 0x400610, 0x304B1FB6);
+		nv_wr32(priv, 0x400B80, 0x1cbd3883);
+		nv_wr32(priv, 0x400B84, 0x44000000);
+		nv_wr32(priv, 0x400098, 0x40000080);
+		nv_wr32(priv, 0x400B88, 0x000000ff);
+
+	} else {
+		nv_wr32(priv, 0x400880, 0x0008c7df);
+		nv_wr32(priv, 0x400094, 0x00000005);
+		nv_wr32(priv, 0x400B80, 0x45eae20e);
+		nv_wr32(priv, 0x400B84, 0x24000000);
+		nv_wr32(priv, 0x400098, 0x00000040);
+		nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00E00038);
+		nv_wr32(priv, NV10_PGRAPH_RDI_DATA , 0x00000030);
+		nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00E10038);
+		nv_wr32(priv, NV10_PGRAPH_RDI_DATA , 0x00000030);
+	}
+
+	/* Turn all the tiling regions off. */
+	for (i = 0; i < pfb->tile.regions; i++)
+		engine->tile_prog(engine, i);
+
+	nv_wr32(priv, 0x4009a0, nv_rd32(priv, 0x100324));
+	nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00EA000C);
+	nv_wr32(priv, NV10_PGRAPH_RDI_DATA, nv_rd32(priv, 0x100324));
+
+	nv_wr32(priv, NV10_PGRAPH_CTX_CONTROL, 0x10000100);
+	nv_wr32(priv, NV10_PGRAPH_STATE      , 0xFFFFFFFF);
+
+	tmp = nv_rd32(priv, NV10_PGRAPH_SURFACE) & 0x0007ff00;
+	nv_wr32(priv, NV10_PGRAPH_SURFACE, tmp);
+	tmp = nv_rd32(priv, NV10_PGRAPH_SURFACE) | 0x00020100;
+	nv_wr32(priv, NV10_PGRAPH_SURFACE, tmp);
+
+	/* begin RAM config */
+	vramsz = pci_resource_len(nv_device(priv)->pdev, 0) - 1;
+	nv_wr32(priv, 0x4009A4, nv_rd32(priv, 0x100200));
+	nv_wr32(priv, 0x4009A8, nv_rd32(priv, 0x100204));
+	nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00EA0000);
+	nv_wr32(priv, NV10_PGRAPH_RDI_DATA , nv_rd32(priv, 0x100200));
+	nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00EA0004);
+	nv_wr32(priv, NV10_PGRAPH_RDI_DATA , nv_rd32(priv, 0x100204));
+	nv_wr32(priv, 0x400820, 0);
+	nv_wr32(priv, 0x400824, 0);
+	nv_wr32(priv, 0x400864, vramsz - 1);
+	nv_wr32(priv, 0x400868, vramsz - 1);
+
+	/* interesting.. the below overwrites some of the tile setup above.. */
+	nv_wr32(priv, 0x400B20, 0x00000000);
+	nv_wr32(priv, 0x400B04, 0xFFFFFFFF);
+
+	nv_wr32(priv, NV03_PGRAPH_ABS_UCLIP_XMIN, 0);
+	nv_wr32(priv, NV03_PGRAPH_ABS_UCLIP_YMIN, 0);
+	nv_wr32(priv, NV03_PGRAPH_ABS_UCLIP_XMAX, 0x7fff);
+	nv_wr32(priv, NV03_PGRAPH_ABS_UCLIP_YMAX, 0x7fff);
+	return 0;
+}
+
+struct nouveau_oclass
+nv20_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x20),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv20_graph_ctor,
+		.dtor = nv20_graph_dtor,
+		.init = nv20_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv20.h b/nvkm/engine/graph/nv20.h
new file mode 100644
index 000000000..2bea7313e
--- /dev/null
+++ b/nvkm/engine/graph/nv20.h
@@ -0,0 +1,31 @@
+#ifndef __NV20_GRAPH_H__
+#define __NV20_GRAPH_H__
+
+#include <core/enum.h>
+
+#include <engine/graph.h>
+#include <engine/fifo.h>
+
+struct nv20_graph_priv {
+	struct nouveau_graph base;
+	struct nouveau_gpuobj *ctxtab;
+};
+
+struct nv20_graph_chan {
+	struct nouveau_graph_chan base;
+	int chid;
+};
+
+extern struct nouveau_oclass nv25_graph_sclass[];
+int  nv20_graph_context_init(struct nouveau_object *);
+int  nv20_graph_context_fini(struct nouveau_object *, bool);
+
+void nv20_graph_tile_prog(struct nouveau_engine *, int);
+void nv20_graph_intr(struct nouveau_subdev *);
+
+void nv20_graph_dtor(struct nouveau_object *);
+int  nv20_graph_init(struct nouveau_object *);
+
+int  nv30_graph_init(struct nouveau_object *);
+
+#endif
diff --git a/nvkm/engine/graph/nv25.c b/nvkm/engine/graph/nv25.c
new file mode 100644
index 000000000..b2b650dd8
--- /dev/null
+++ b/nvkm/engine/graph/nv25.c
@@ -0,0 +1,167 @@
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/enum.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+
+#include <engine/graph.h>
+
+#include "nv20.h"
+#include "regs.h"
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+struct nouveau_oclass
+nv25_graph_sclass[] = {
+	{ 0x0012, &nv04_graph_ofuncs, NULL }, /* beta1 */
+	{ 0x0019, &nv04_graph_ofuncs, NULL }, /* clip */
+	{ 0x0030, &nv04_graph_ofuncs, NULL }, /* null */
+	{ 0x0039, &nv04_graph_ofuncs, NULL }, /* m2mf */
+	{ 0x0043, &nv04_graph_ofuncs, NULL }, /* rop */
+	{ 0x0044, &nv04_graph_ofuncs, NULL }, /* patt */
+	{ 0x004a, &nv04_graph_ofuncs, NULL }, /* gdi */
+	{ 0x0062, &nv04_graph_ofuncs, NULL }, /* surf2d */
+	{ 0x0072, &nv04_graph_ofuncs, NULL }, /* beta4 */
+	{ 0x0089, &nv04_graph_ofuncs, NULL }, /* sifm */
+	{ 0x008a, &nv04_graph_ofuncs, NULL }, /* ifc */
+	{ 0x0096, &nv04_graph_ofuncs, NULL }, /* celcius */
+	{ 0x009e, &nv04_graph_ofuncs, NULL }, /* swzsurf */
+	{ 0x009f, &nv04_graph_ofuncs, NULL }, /* imageblit */
+	{ 0x0597, &nv04_graph_ofuncs, NULL }, /* kelvin */
+	{},
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static int
+nv25_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nv20_graph_chan *chan;
+	int ret, i;
+
+	ret = nouveau_graph_context_create(parent, engine, oclass, NULL, 0x3724,
+					   16, NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	chan->chid = nouveau_fifo_chan(parent)->chid;
+
+	nv_wo32(chan, 0x0028, 0x00000001 | (chan->chid << 24));
+	nv_wo32(chan, 0x035c, 0xffff0000);
+	nv_wo32(chan, 0x03c0, 0x0fff0000);
+	nv_wo32(chan, 0x03c4, 0x0fff0000);
+	nv_wo32(chan, 0x049c, 0x00000101);
+	nv_wo32(chan, 0x04b0, 0x00000111);
+	nv_wo32(chan, 0x04c8, 0x00000080);
+	nv_wo32(chan, 0x04cc, 0xffff0000);
+	nv_wo32(chan, 0x04d0, 0x00000001);
+	nv_wo32(chan, 0x04e4, 0x44400000);
+	nv_wo32(chan, 0x04fc, 0x4b800000);
+	for (i = 0x0510; i <= 0x051c; i += 4)
+		nv_wo32(chan, i, 0x00030303);
+	for (i = 0x0530; i <= 0x053c; i += 4)
+		nv_wo32(chan, i, 0x00080000);
+	for (i = 0x0548; i <= 0x0554; i += 4)
+		nv_wo32(chan, i, 0x01012000);
+	for (i = 0x0558; i <= 0x0564; i += 4)
+		nv_wo32(chan, i, 0x000105b8);
+	for (i = 0x0568; i <= 0x0574; i += 4)
+		nv_wo32(chan, i, 0x00080008);
+	for (i = 0x0598; i <= 0x05d4; i += 4)
+		nv_wo32(chan, i, 0x07ff0000);
+	nv_wo32(chan, 0x05e0, 0x4b7fffff);
+	nv_wo32(chan, 0x0620, 0x00000080);
+	nv_wo32(chan, 0x0624, 0x30201000);
+	nv_wo32(chan, 0x0628, 0x70605040);
+	nv_wo32(chan, 0x062c, 0xb0a09080);
+	nv_wo32(chan, 0x0630, 0xf0e0d0c0);
+	nv_wo32(chan, 0x0664, 0x00000001);
+	nv_wo32(chan, 0x066c, 0x00004000);
+	nv_wo32(chan, 0x0678, 0x00000001);
+	nv_wo32(chan, 0x0680, 0x00040000);
+	nv_wo32(chan, 0x0684, 0x00010000);
+	for (i = 0x1b04; i <= 0x2374; i += 16) {
+		nv_wo32(chan, (i + 0), 0x10700ff9);
+		nv_wo32(chan, (i + 4), 0x0436086c);
+		nv_wo32(chan, (i + 8), 0x000c001b);
+	}
+	nv_wo32(chan, 0x2704, 0x3f800000);
+	nv_wo32(chan, 0x2718, 0x3f800000);
+	nv_wo32(chan, 0x2744, 0x40000000);
+	nv_wo32(chan, 0x2748, 0x3f800000);
+	nv_wo32(chan, 0x274c, 0x3f000000);
+	nv_wo32(chan, 0x2754, 0x40000000);
+	nv_wo32(chan, 0x2758, 0x3f800000);
+	nv_wo32(chan, 0x2760, 0xbf800000);
+	nv_wo32(chan, 0x2768, 0xbf800000);
+	nv_wo32(chan, 0x308c, 0x000fe000);
+	nv_wo32(chan, 0x3108, 0x000003f8);
+	nv_wo32(chan, 0x3468, 0x002fe000);
+	for (i = 0x3484; i <= 0x34a0; i += 4)
+		nv_wo32(chan, i, 0x001c527c);
+	return 0;
+}
+
+static struct nouveau_oclass
+nv25_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x25),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv25_graph_context_ctor,
+		.dtor = _nouveau_graph_context_dtor,
+		.init = nv20_graph_context_init,
+		.fini = nv20_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv25_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv20_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 32 * 4, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &priv->ctxtab);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00001000;
+	nv_subdev(priv)->intr = nv20_graph_intr;
+	nv_engine(priv)->cclass = &nv25_graph_cclass;
+	nv_engine(priv)->sclass = nv25_graph_sclass;
+	nv_engine(priv)->tile_prog = nv20_graph_tile_prog;
+	return 0;
+}
+
+struct nouveau_oclass
+nv25_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x25),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv25_graph_ctor,
+		.dtor = nv20_graph_dtor,
+		.init = nv20_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv2a.c b/nvkm/engine/graph/nv2a.c
new file mode 100644
index 000000000..700462fa0
--- /dev/null
+++ b/nvkm/engine/graph/nv2a.c
@@ -0,0 +1,134 @@
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/enum.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+
+#include <engine/graph.h>
+
+#include "nv20.h"
+#include "regs.h"
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static int
+nv2a_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nv20_graph_chan *chan;
+	int ret, i;
+
+	ret = nouveau_graph_context_create(parent, engine, oclass, NULL, 0x36b0,
+					   16, NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	chan->chid = nouveau_fifo_chan(parent)->chid;
+
+	nv_wo32(chan, 0x0000, 0x00000001 | (chan->chid << 24));
+	nv_wo32(chan, 0x033c, 0xffff0000);
+	nv_wo32(chan, 0x03a0, 0x0fff0000);
+	nv_wo32(chan, 0x03a4, 0x0fff0000);
+	nv_wo32(chan, 0x047c, 0x00000101);
+	nv_wo32(chan, 0x0490, 0x00000111);
+	nv_wo32(chan, 0x04a8, 0x44400000);
+	for (i = 0x04d4; i <= 0x04e0; i += 4)
+		nv_wo32(chan, i, 0x00030303);
+	for (i = 0x04f4; i <= 0x0500; i += 4)
+		nv_wo32(chan, i, 0x00080000);
+	for (i = 0x050c; i <= 0x0518; i += 4)
+		nv_wo32(chan, i, 0x01012000);
+	for (i = 0x051c; i <= 0x0528; i += 4)
+		nv_wo32(chan, i, 0x000105b8);
+	for (i = 0x052c; i <= 0x0538; i += 4)
+		nv_wo32(chan, i, 0x00080008);
+	for (i = 0x055c; i <= 0x0598; i += 4)
+		nv_wo32(chan, i, 0x07ff0000);
+	nv_wo32(chan, 0x05a4, 0x4b7fffff);
+	nv_wo32(chan, 0x05fc, 0x00000001);
+	nv_wo32(chan, 0x0604, 0x00004000);
+	nv_wo32(chan, 0x0610, 0x00000001);
+	nv_wo32(chan, 0x0618, 0x00040000);
+	nv_wo32(chan, 0x061c, 0x00010000);
+	for (i = 0x1a9c; i <= 0x22fc; i += 16) { /*XXX: check!! */
+		nv_wo32(chan, (i + 0), 0x10700ff9);
+		nv_wo32(chan, (i + 4), 0x0436086c);
+		nv_wo32(chan, (i + 8), 0x000c001b);
+	}
+	nv_wo32(chan, 0x269c, 0x3f800000);
+	nv_wo32(chan, 0x26b0, 0x3f800000);
+	nv_wo32(chan, 0x26dc, 0x40000000);
+	nv_wo32(chan, 0x26e0, 0x3f800000);
+	nv_wo32(chan, 0x26e4, 0x3f000000);
+	nv_wo32(chan, 0x26ec, 0x40000000);
+	nv_wo32(chan, 0x26f0, 0x3f800000);
+	nv_wo32(chan, 0x26f8, 0xbf800000);
+	nv_wo32(chan, 0x2700, 0xbf800000);
+	nv_wo32(chan, 0x3024, 0x000fe000);
+	nv_wo32(chan, 0x30a0, 0x000003f8);
+	nv_wo32(chan, 0x33fc, 0x002fe000);
+	for (i = 0x341c; i <= 0x3438; i += 4)
+		nv_wo32(chan, i, 0x001c527c);
+	return 0;
+}
+
+static struct nouveau_oclass
+nv2a_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x2a),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv2a_graph_context_ctor,
+		.dtor = _nouveau_graph_context_dtor,
+		.init = nv20_graph_context_init,
+		.fini = nv20_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv2a_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv20_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 32 * 4, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &priv->ctxtab);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00001000;
+	nv_subdev(priv)->intr = nv20_graph_intr;
+	nv_engine(priv)->cclass = &nv2a_graph_cclass;
+	nv_engine(priv)->sclass = nv25_graph_sclass;
+	nv_engine(priv)->tile_prog = nv20_graph_tile_prog;
+	return 0;
+}
+
+struct nouveau_oclass
+nv2a_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x2a),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv2a_graph_ctor,
+		.dtor = nv20_graph_dtor,
+		.init = nv20_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv30.c b/nvkm/engine/graph/nv30.c
new file mode 100644
index 000000000..cedadaa92
--- /dev/null
+++ b/nvkm/engine/graph/nv30.c
@@ -0,0 +1,238 @@
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/enum.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+
+#include <engine/graph.h>
+
+#include "nv20.h"
+#include "regs.h"
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv30_graph_sclass[] = {
+	{ 0x0012, &nv04_graph_ofuncs, NULL }, /* beta1 */
+	{ 0x0019, &nv04_graph_ofuncs, NULL }, /* clip */
+	{ 0x0030, &nv04_graph_ofuncs, NULL }, /* null */
+	{ 0x0039, &nv04_graph_ofuncs, NULL }, /* m2mf */
+	{ 0x0043, &nv04_graph_ofuncs, NULL }, /* rop */
+	{ 0x0044, &nv04_graph_ofuncs, NULL }, /* patt */
+	{ 0x004a, &nv04_graph_ofuncs, NULL }, /* gdi */
+	{ 0x0062, &nv04_graph_ofuncs, NULL }, /* surf2d */
+	{ 0x0072, &nv04_graph_ofuncs, NULL }, /* beta4 */
+	{ 0x0089, &nv04_graph_ofuncs, NULL }, /* sifm */
+	{ 0x008a, &nv04_graph_ofuncs, NULL }, /* ifc */
+	{ 0x009f, &nv04_graph_ofuncs, NULL }, /* imageblit */
+	{ 0x0362, &nv04_graph_ofuncs, NULL }, /* surf2d (nv30) */
+	{ 0x0389, &nv04_graph_ofuncs, NULL }, /* sifm (nv30) */
+	{ 0x038a, &nv04_graph_ofuncs, NULL }, /* ifc (nv30) */
+	{ 0x039e, &nv04_graph_ofuncs, NULL }, /* swzsurf (nv30) */
+	{ 0x0397, &nv04_graph_ofuncs, NULL }, /* rankine */
+	{},
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static int
+nv30_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nv20_graph_chan *chan;
+	int ret, i;
+
+	ret = nouveau_graph_context_create(parent, engine, oclass, NULL, 0x5f48,
+					   16, NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	chan->chid = nouveau_fifo_chan(parent)->chid;
+
+	nv_wo32(chan, 0x0028, 0x00000001 | (chan->chid << 24));
+	nv_wo32(chan, 0x0410, 0x00000101);
+	nv_wo32(chan, 0x0424, 0x00000111);
+	nv_wo32(chan, 0x0428, 0x00000060);
+	nv_wo32(chan, 0x0444, 0x00000080);
+	nv_wo32(chan, 0x0448, 0xffff0000);
+	nv_wo32(chan, 0x044c, 0x00000001);
+	nv_wo32(chan, 0x0460, 0x44400000);
+	nv_wo32(chan, 0x048c, 0xffff0000);
+	for (i = 0x04e0; i < 0x04e8; i += 4)
+		nv_wo32(chan, i, 0x0fff0000);
+	nv_wo32(chan, 0x04ec, 0x00011100);
+	for (i = 0x0508; i < 0x0548; i += 4)
+		nv_wo32(chan, i, 0x07ff0000);
+	nv_wo32(chan, 0x0550, 0x4b7fffff);
+	nv_wo32(chan, 0x058c, 0x00000080);
+	nv_wo32(chan, 0x0590, 0x30201000);
+	nv_wo32(chan, 0x0594, 0x70605040);
+	nv_wo32(chan, 0x0598, 0xb8a89888);
+	nv_wo32(chan, 0x059c, 0xf8e8d8c8);
+	nv_wo32(chan, 0x05b0, 0xb0000000);
+	for (i = 0x0600; i < 0x0640; i += 4)
+		nv_wo32(chan, i, 0x00010588);
+	for (i = 0x0640; i < 0x0680; i += 4)
+		nv_wo32(chan, i, 0x00030303);
+	for (i = 0x06c0; i < 0x0700; i += 4)
+		nv_wo32(chan, i, 0x0008aae4);
+	for (i = 0x0700; i < 0x0740; i += 4)
+		nv_wo32(chan, i, 0x01012000);
+	for (i = 0x0740; i < 0x0780; i += 4)
+		nv_wo32(chan, i, 0x00080008);
+	nv_wo32(chan, 0x085c, 0x00040000);
+	nv_wo32(chan, 0x0860, 0x00010000);
+	for (i = 0x0864; i < 0x0874; i += 4)
+		nv_wo32(chan, i, 0x00040004);
+	for (i = 0x1f18; i <= 0x3088 ; i += 16) {
+		nv_wo32(chan, i + 0, 0x10700ff9);
+		nv_wo32(chan, i + 1, 0x0436086c);
+		nv_wo32(chan, i + 2, 0x000c001b);
+	}
+	for (i = 0x30b8; i < 0x30c8; i += 4)
+		nv_wo32(chan, i, 0x0000ffff);
+	nv_wo32(chan, 0x344c, 0x3f800000);
+	nv_wo32(chan, 0x3808, 0x3f800000);
+	nv_wo32(chan, 0x381c, 0x3f800000);
+	nv_wo32(chan, 0x3848, 0x40000000);
+	nv_wo32(chan, 0x384c, 0x3f800000);
+	nv_wo32(chan, 0x3850, 0x3f000000);
+	nv_wo32(chan, 0x3858, 0x40000000);
+	nv_wo32(chan, 0x385c, 0x3f800000);
+	nv_wo32(chan, 0x3864, 0xbf800000);
+	nv_wo32(chan, 0x386c, 0xbf800000);
+	return 0;
+}
+
+static struct nouveau_oclass
+nv30_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x30),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv30_graph_context_ctor,
+		.dtor = _nouveau_graph_context_dtor,
+		.init = nv20_graph_context_init,
+		.fini = nv20_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv30_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv20_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 32 * 4, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &priv->ctxtab);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00001000;
+	nv_subdev(priv)->intr = nv20_graph_intr;
+	nv_engine(priv)->cclass = &nv30_graph_cclass;
+	nv_engine(priv)->sclass = nv30_graph_sclass;
+	nv_engine(priv)->tile_prog = nv20_graph_tile_prog;
+	return 0;
+}
+
+int
+nv30_graph_init(struct nouveau_object *object)
+{
+	struct nouveau_engine *engine = nv_engine(object);
+	struct nv20_graph_priv *priv = (void *)engine;
+	struct nouveau_fb *pfb = nouveau_fb(object);
+	int ret, i;
+
+	ret = nouveau_graph_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, NV20_PGRAPH_CHANNEL_CTX_TABLE, priv->ctxtab->addr >> 4);
+
+	nv_wr32(priv, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
+	nv_wr32(priv, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
+
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0xFFFFFFFF);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0x00000000);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_1, 0x401287c0);
+	nv_wr32(priv, 0x400890, 0x01b463ff);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_3, 0xf2de0475);
+	nv_wr32(priv, NV10_PGRAPH_DEBUG_4, 0x00008000);
+	nv_wr32(priv, NV04_PGRAPH_LIMIT_VIOL_PIX, 0xf04bdff6);
+	nv_wr32(priv, 0x400B80, 0x1003d888);
+	nv_wr32(priv, 0x400B84, 0x0c000000);
+	nv_wr32(priv, 0x400098, 0x00000000);
+	nv_wr32(priv, 0x40009C, 0x0005ad00);
+	nv_wr32(priv, 0x400B88, 0x62ff00ff); /* suspiciously like PGRAPH_DEBUG_2 */
+	nv_wr32(priv, 0x4000a0, 0x00000000);
+	nv_wr32(priv, 0x4000a4, 0x00000008);
+	nv_wr32(priv, 0x4008a8, 0xb784a400);
+	nv_wr32(priv, 0x400ba0, 0x002f8685);
+	nv_wr32(priv, 0x400ba4, 0x00231f3f);
+	nv_wr32(priv, 0x4008a4, 0x40000020);
+
+	if (nv_device(priv)->chipset == 0x34) {
+		nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00EA0004);
+		nv_wr32(priv, NV10_PGRAPH_RDI_DATA , 0x00200201);
+		nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00EA0008);
+		nv_wr32(priv, NV10_PGRAPH_RDI_DATA , 0x00000008);
+		nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00EA0000);
+		nv_wr32(priv, NV10_PGRAPH_RDI_DATA , 0x00000032);
+		nv_wr32(priv, NV10_PGRAPH_RDI_INDEX, 0x00E00004);
+		nv_wr32(priv, NV10_PGRAPH_RDI_DATA , 0x00000002);
+	}
+
+	nv_wr32(priv, 0x4000c0, 0x00000016);
+
+	/* Turn all the tiling regions off. */
+	for (i = 0; i < pfb->tile.regions; i++)
+		engine->tile_prog(engine, i);
+
+	nv_wr32(priv, NV10_PGRAPH_CTX_CONTROL, 0x10000100);
+	nv_wr32(priv, NV10_PGRAPH_STATE      , 0xFFFFFFFF);
+	nv_wr32(priv, 0x0040075c             , 0x00000001);
+
+	/* begin RAM config */
+	/* vramsz = pci_resource_len(priv->dev->pdev, 0) - 1; */
+	nv_wr32(priv, 0x4009A4, nv_rd32(priv, 0x100200));
+	nv_wr32(priv, 0x4009A8, nv_rd32(priv, 0x100204));
+	if (nv_device(priv)->chipset != 0x34) {
+		nv_wr32(priv, 0x400750, 0x00EA0000);
+		nv_wr32(priv, 0x400754, nv_rd32(priv, 0x100200));
+		nv_wr32(priv, 0x400750, 0x00EA0004);
+		nv_wr32(priv, 0x400754, nv_rd32(priv, 0x100204));
+	}
+	return 0;
+}
+
+struct nouveau_oclass
+nv30_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x30),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv30_graph_ctor,
+		.dtor = nv20_graph_dtor,
+		.init = nv30_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv34.c b/nvkm/engine/graph/nv34.c
new file mode 100644
index 000000000..273f63200
--- /dev/null
+++ b/nvkm/engine/graph/nv34.c
@@ -0,0 +1,168 @@
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/enum.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+
+#include <engine/graph.h>
+
+#include "nv20.h"
+#include "regs.h"
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv34_graph_sclass[] = {
+	{ 0x0012, &nv04_graph_ofuncs, NULL }, /* beta1 */
+	{ 0x0019, &nv04_graph_ofuncs, NULL }, /* clip */
+	{ 0x0030, &nv04_graph_ofuncs, NULL }, /* null */
+	{ 0x0039, &nv04_graph_ofuncs, NULL }, /* m2mf */
+	{ 0x0043, &nv04_graph_ofuncs, NULL }, /* rop */
+	{ 0x0044, &nv04_graph_ofuncs, NULL }, /* patt */
+	{ 0x004a, &nv04_graph_ofuncs, NULL }, /* gdi */
+	{ 0x0062, &nv04_graph_ofuncs, NULL }, /* surf2d */
+	{ 0x0072, &nv04_graph_ofuncs, NULL }, /* beta4 */
+	{ 0x0089, &nv04_graph_ofuncs, NULL }, /* sifm */
+	{ 0x008a, &nv04_graph_ofuncs, NULL }, /* ifc */
+	{ 0x009f, &nv04_graph_ofuncs, NULL }, /* imageblit */
+	{ 0x0362, &nv04_graph_ofuncs, NULL }, /* surf2d (nv30) */
+	{ 0x0389, &nv04_graph_ofuncs, NULL }, /* sifm (nv30) */
+	{ 0x038a, &nv04_graph_ofuncs, NULL }, /* ifc (nv30) */
+	{ 0x039e, &nv04_graph_ofuncs, NULL }, /* swzsurf (nv30) */
+	{ 0x0697, &nv04_graph_ofuncs, NULL }, /* rankine */
+	{},
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static int
+nv34_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nv20_graph_chan *chan;
+	int ret, i;
+
+	ret = nouveau_graph_context_create(parent, engine, oclass, NULL, 0x46dc,
+					   16, NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	chan->chid = nouveau_fifo_chan(parent)->chid;
+
+	nv_wo32(chan, 0x0028, 0x00000001 | (chan->chid << 24));
+	nv_wo32(chan, 0x040c, 0x01000101);
+	nv_wo32(chan, 0x0420, 0x00000111);
+	nv_wo32(chan, 0x0424, 0x00000060);
+	nv_wo32(chan, 0x0440, 0x00000080);
+	nv_wo32(chan, 0x0444, 0xffff0000);
+	nv_wo32(chan, 0x0448, 0x00000001);
+	nv_wo32(chan, 0x045c, 0x44400000);
+	nv_wo32(chan, 0x0480, 0xffff0000);
+	for (i = 0x04d4; i < 0x04dc; i += 4)
+		nv_wo32(chan, i, 0x0fff0000);
+	nv_wo32(chan, 0x04e0, 0x00011100);
+	for (i = 0x04fc; i < 0x053c; i += 4)
+		nv_wo32(chan, i, 0x07ff0000);
+	nv_wo32(chan, 0x0544, 0x4b7fffff);
+	nv_wo32(chan, 0x057c, 0x00000080);
+	nv_wo32(chan, 0x0580, 0x30201000);
+	nv_wo32(chan, 0x0584, 0x70605040);
+	nv_wo32(chan, 0x0588, 0xb8a89888);
+	nv_wo32(chan, 0x058c, 0xf8e8d8c8);
+	nv_wo32(chan, 0x05a0, 0xb0000000);
+	for (i = 0x05f0; i < 0x0630; i += 4)
+		nv_wo32(chan, i, 0x00010588);
+	for (i = 0x0630; i < 0x0670; i += 4)
+		nv_wo32(chan, i, 0x00030303);
+	for (i = 0x06b0; i < 0x06f0; i += 4)
+		nv_wo32(chan, i, 0x0008aae4);
+	for (i = 0x06f0; i < 0x0730; i += 4)
+		nv_wo32(chan, i, 0x01012000);
+	for (i = 0x0730; i < 0x0770; i += 4)
+		nv_wo32(chan, i, 0x00080008);
+	nv_wo32(chan, 0x0850, 0x00040000);
+	nv_wo32(chan, 0x0854, 0x00010000);
+	for (i = 0x0858; i < 0x0868; i += 4)
+		nv_wo32(chan, i, 0x00040004);
+	for (i = 0x15ac; i <= 0x271c ; i += 16) {
+		nv_wo32(chan, i + 0, 0x10700ff9);
+		nv_wo32(chan, i + 1, 0x0436086c);
+		nv_wo32(chan, i + 2, 0x000c001b);
+	}
+	for (i = 0x274c; i < 0x275c; i += 4)
+		nv_wo32(chan, i, 0x0000ffff);
+	nv_wo32(chan, 0x2ae0, 0x3f800000);
+	nv_wo32(chan, 0x2e9c, 0x3f800000);
+	nv_wo32(chan, 0x2eb0, 0x3f800000);
+	nv_wo32(chan, 0x2edc, 0x40000000);
+	nv_wo32(chan, 0x2ee0, 0x3f800000);
+	nv_wo32(chan, 0x2ee4, 0x3f000000);
+	nv_wo32(chan, 0x2eec, 0x40000000);
+	nv_wo32(chan, 0x2ef0, 0x3f800000);
+	nv_wo32(chan, 0x2ef8, 0xbf800000);
+	nv_wo32(chan, 0x2f00, 0xbf800000);
+	return 0;
+}
+
+static struct nouveau_oclass
+nv34_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x34),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv34_graph_context_ctor,
+		.dtor = _nouveau_graph_context_dtor,
+		.init = nv20_graph_context_init,
+		.fini = nv20_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv34_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv20_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 32 * 4, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &priv->ctxtab);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00001000;
+	nv_subdev(priv)->intr = nv20_graph_intr;
+	nv_engine(priv)->cclass = &nv34_graph_cclass;
+	nv_engine(priv)->sclass = nv34_graph_sclass;
+	nv_engine(priv)->tile_prog = nv20_graph_tile_prog;
+	return 0;
+}
+
+struct nouveau_oclass
+nv34_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x34),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv34_graph_ctor,
+		.dtor = nv20_graph_dtor,
+		.init = nv30_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv35.c b/nvkm/engine/graph/nv35.c
new file mode 100644
index 000000000..f40ee2116
--- /dev/null
+++ b/nvkm/engine/graph/nv35.c
@@ -0,0 +1,166 @@
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/enum.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+
+#include "nv20.h"
+#include "regs.h"
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv35_graph_sclass[] = {
+	{ 0x0012, &nv04_graph_ofuncs, NULL }, /* beta1 */
+	{ 0x0019, &nv04_graph_ofuncs, NULL }, /* clip */
+	{ 0x0030, &nv04_graph_ofuncs, NULL }, /* null */
+	{ 0x0039, &nv04_graph_ofuncs, NULL }, /* m2mf */
+	{ 0x0043, &nv04_graph_ofuncs, NULL }, /* rop */
+	{ 0x0044, &nv04_graph_ofuncs, NULL }, /* patt */
+	{ 0x004a, &nv04_graph_ofuncs, NULL }, /* gdi */
+	{ 0x0062, &nv04_graph_ofuncs, NULL }, /* surf2d */
+	{ 0x0072, &nv04_graph_ofuncs, NULL }, /* beta4 */
+	{ 0x0089, &nv04_graph_ofuncs, NULL }, /* sifm */
+	{ 0x008a, &nv04_graph_ofuncs, NULL }, /* ifc */
+	{ 0x009f, &nv04_graph_ofuncs, NULL }, /* imageblit */
+	{ 0x0362, &nv04_graph_ofuncs, NULL }, /* surf2d (nv30) */
+	{ 0x0389, &nv04_graph_ofuncs, NULL }, /* sifm (nv30) */
+	{ 0x038a, &nv04_graph_ofuncs, NULL }, /* ifc (nv30) */
+	{ 0x039e, &nv04_graph_ofuncs, NULL }, /* swzsurf (nv30) */
+	{ 0x0497, &nv04_graph_ofuncs, NULL }, /* rankine */
+	{},
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static int
+nv35_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nv20_graph_chan *chan;
+	int ret, i;
+
+	ret = nouveau_graph_context_create(parent, engine, oclass, NULL, 0x577c,
+					   16, NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	chan->chid = nouveau_fifo_chan(parent)->chid;
+
+	nv_wo32(chan, 0x0028, 0x00000001 | (chan->chid << 24));
+	nv_wo32(chan, 0x040c, 0x00000101);
+	nv_wo32(chan, 0x0420, 0x00000111);
+	nv_wo32(chan, 0x0424, 0x00000060);
+	nv_wo32(chan, 0x0440, 0x00000080);
+	nv_wo32(chan, 0x0444, 0xffff0000);
+	nv_wo32(chan, 0x0448, 0x00000001);
+	nv_wo32(chan, 0x045c, 0x44400000);
+	nv_wo32(chan, 0x0488, 0xffff0000);
+	for (i = 0x04dc; i < 0x04e4; i += 4)
+		nv_wo32(chan, i, 0x0fff0000);
+	nv_wo32(chan, 0x04e8, 0x00011100);
+	for (i = 0x0504; i < 0x0544; i += 4)
+		nv_wo32(chan, i, 0x07ff0000);
+	nv_wo32(chan, 0x054c, 0x4b7fffff);
+	nv_wo32(chan, 0x0588, 0x00000080);
+	nv_wo32(chan, 0x058c, 0x30201000);
+	nv_wo32(chan, 0x0590, 0x70605040);
+	nv_wo32(chan, 0x0594, 0xb8a89888);
+	nv_wo32(chan, 0x0598, 0xf8e8d8c8);
+	nv_wo32(chan, 0x05ac, 0xb0000000);
+	for (i = 0x0604; i < 0x0644; i += 4)
+		nv_wo32(chan, i, 0x00010588);
+	for (i = 0x0644; i < 0x0684; i += 4)
+		nv_wo32(chan, i, 0x00030303);
+	for (i = 0x06c4; i < 0x0704; i += 4)
+		nv_wo32(chan, i, 0x0008aae4);
+	for (i = 0x0704; i < 0x0744; i += 4)
+		nv_wo32(chan, i, 0x01012000);
+	for (i = 0x0744; i < 0x0784; i += 4)
+		nv_wo32(chan, i, 0x00080008);
+	nv_wo32(chan, 0x0860, 0x00040000);
+	nv_wo32(chan, 0x0864, 0x00010000);
+	for (i = 0x0868; i < 0x0878; i += 4)
+		nv_wo32(chan, i, 0x00040004);
+	for (i = 0x1f1c; i <= 0x308c ; i += 16) {
+		nv_wo32(chan, i + 0, 0x10700ff9);
+		nv_wo32(chan, i + 4, 0x0436086c);
+		nv_wo32(chan, i + 8, 0x000c001b);
+	}
+	for (i = 0x30bc; i < 0x30cc; i += 4)
+		nv_wo32(chan, i, 0x0000ffff);
+	nv_wo32(chan, 0x3450, 0x3f800000);
+	nv_wo32(chan, 0x380c, 0x3f800000);
+	nv_wo32(chan, 0x3820, 0x3f800000);
+	nv_wo32(chan, 0x384c, 0x40000000);
+	nv_wo32(chan, 0x3850, 0x3f800000);
+	nv_wo32(chan, 0x3854, 0x3f000000);
+	nv_wo32(chan, 0x385c, 0x40000000);
+	nv_wo32(chan, 0x3860, 0x3f800000);
+	nv_wo32(chan, 0x3868, 0xbf800000);
+	nv_wo32(chan, 0x3870, 0xbf800000);
+	return 0;
+}
+
+static struct nouveau_oclass
+nv35_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x35),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv35_graph_context_ctor,
+		.dtor = _nouveau_graph_context_dtor,
+		.init = nv20_graph_context_init,
+		.fini = nv20_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv35_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv20_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 32 * 4, 16,
+				 NVOBJ_FLAG_ZERO_ALLOC, &priv->ctxtab);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00001000;
+	nv_subdev(priv)->intr = nv20_graph_intr;
+	nv_engine(priv)->cclass = &nv35_graph_cclass;
+	nv_engine(priv)->sclass = nv35_graph_sclass;
+	nv_engine(priv)->tile_prog = nv20_graph_tile_prog;
+	return 0;
+}
+
+struct nouveau_oclass
+nv35_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x35),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv35_graph_ctor,
+		.dtor = nv20_graph_dtor,
+		.init = nv30_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv40.c b/nvkm/engine/graph/nv40.c
new file mode 100644
index 000000000..0b36dd3de
--- /dev/null
+++ b/nvkm/engine/graph/nv40.c
@@ -0,0 +1,525 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/handle.h>
+#include <core/engctx.h>
+
+#include <subdev/fb.h>
+#include <subdev/timer.h>
+
+#include <engine/graph.h>
+#include <engine/fifo.h>
+
+#include "nv40.h"
+#include "regs.h"
+
+struct nv40_graph_priv {
+	struct nouveau_graph base;
+	u32 size;
+};
+
+struct nv40_graph_chan {
+	struct nouveau_graph_chan base;
+};
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+static int
+nv40_graph_object_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nouveau_gpuobj *obj;
+	int ret;
+
+	ret = nouveau_gpuobj_create(parent, engine, oclass, 0, parent,
+				    20, 16, 0, &obj);
+	*pobject = nv_object(obj);
+	if (ret)
+		return ret;
+
+	nv_wo32(obj, 0x00, nv_mclass(obj));
+	nv_wo32(obj, 0x04, 0x00000000);
+	nv_wo32(obj, 0x08, 0x00000000);
+#ifdef __BIG_ENDIAN
+	nv_mo32(obj, 0x08, 0x01000000, 0x01000000);
+#endif
+	nv_wo32(obj, 0x0c, 0x00000000);
+	nv_wo32(obj, 0x10, 0x00000000);
+	return 0;
+}
+
+static struct nouveau_ofuncs
+nv40_graph_ofuncs = {
+	.ctor = nv40_graph_object_ctor,
+	.dtor = _nouveau_gpuobj_dtor,
+	.init = _nouveau_gpuobj_init,
+	.fini = _nouveau_gpuobj_fini,
+	.rd32 = _nouveau_gpuobj_rd32,
+	.wr32 = _nouveau_gpuobj_wr32,
+};
+
+static struct nouveau_oclass
+nv40_graph_sclass[] = {
+	{ 0x0012, &nv40_graph_ofuncs, NULL }, /* beta1 */
+	{ 0x0019, &nv40_graph_ofuncs, NULL }, /* clip */
+	{ 0x0030, &nv40_graph_ofuncs, NULL }, /* null */
+	{ 0x0039, &nv40_graph_ofuncs, NULL }, /* m2mf */
+	{ 0x0043, &nv40_graph_ofuncs, NULL }, /* rop */
+	{ 0x0044, &nv40_graph_ofuncs, NULL }, /* patt */
+	{ 0x004a, &nv40_graph_ofuncs, NULL }, /* gdi */
+	{ 0x0062, &nv40_graph_ofuncs, NULL }, /* surf2d */
+	{ 0x0072, &nv40_graph_ofuncs, NULL }, /* beta4 */
+	{ 0x0089, &nv40_graph_ofuncs, NULL }, /* sifm */
+	{ 0x008a, &nv40_graph_ofuncs, NULL }, /* ifc */
+	{ 0x009f, &nv40_graph_ofuncs, NULL }, /* imageblit */
+	{ 0x3062, &nv40_graph_ofuncs, NULL }, /* surf2d (nv40) */
+	{ 0x3089, &nv40_graph_ofuncs, NULL }, /* sifm (nv40) */
+	{ 0x309e, &nv40_graph_ofuncs, NULL }, /* swzsurf (nv40) */
+	{ 0x4097, &nv40_graph_ofuncs, NULL }, /* curie */
+	{},
+};
+
+static struct nouveau_oclass
+nv44_graph_sclass[] = {
+	{ 0x0012, &nv40_graph_ofuncs, NULL }, /* beta1 */
+	{ 0x0019, &nv40_graph_ofuncs, NULL }, /* clip */
+	{ 0x0030, &nv40_graph_ofuncs, NULL }, /* null */
+	{ 0x0039, &nv40_graph_ofuncs, NULL }, /* m2mf */
+	{ 0x0043, &nv40_graph_ofuncs, NULL }, /* rop */
+	{ 0x0044, &nv40_graph_ofuncs, NULL }, /* patt */
+	{ 0x004a, &nv40_graph_ofuncs, NULL }, /* gdi */
+	{ 0x0062, &nv40_graph_ofuncs, NULL }, /* surf2d */
+	{ 0x0072, &nv40_graph_ofuncs, NULL }, /* beta4 */
+	{ 0x0089, &nv40_graph_ofuncs, NULL }, /* sifm */
+	{ 0x008a, &nv40_graph_ofuncs, NULL }, /* ifc */
+	{ 0x009f, &nv40_graph_ofuncs, NULL }, /* imageblit */
+	{ 0x3062, &nv40_graph_ofuncs, NULL }, /* surf2d (nv40) */
+	{ 0x3089, &nv40_graph_ofuncs, NULL }, /* sifm (nv40) */
+	{ 0x309e, &nv40_graph_ofuncs, NULL }, /* swzsurf (nv40) */
+	{ 0x4497, &nv40_graph_ofuncs, NULL }, /* curie */
+	{},
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static int
+nv40_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nv40_graph_priv *priv = (void *)engine;
+	struct nv40_graph_chan *chan;
+	int ret;
+
+	ret = nouveau_graph_context_create(parent, engine, oclass, NULL,
+					   priv->size, 16,
+					   NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv40_grctx_fill(nv_device(priv), nv_gpuobj(chan));
+	nv_wo32(chan, 0x00000, nv_gpuobj(chan)->addr >> 4);
+	return 0;
+}
+
+static int
+nv40_graph_context_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv40_graph_priv *priv = (void *)object->engine;
+	struct nv40_graph_chan *chan = (void *)object;
+	u32 inst = 0x01000000 | nv_gpuobj(chan)->addr >> 4;
+	int ret = 0;
+
+	nv_mask(priv, 0x400720, 0x00000001, 0x00000000);
+
+	if (nv_rd32(priv, 0x40032c) == inst) {
+		if (suspend) {
+			nv_wr32(priv, 0x400720, 0x00000000);
+			nv_wr32(priv, 0x400784, inst);
+			nv_mask(priv, 0x400310, 0x00000020, 0x00000020);
+			nv_mask(priv, 0x400304, 0x00000001, 0x00000001);
+			if (!nv_wait(priv, 0x400300, 0x00000001, 0x00000000)) {
+				u32 insn = nv_rd32(priv, 0x400308);
+				nv_warn(priv, "ctxprog timeout 0x%08x\n", insn);
+				ret = -EBUSY;
+			}
+		}
+
+		nv_mask(priv, 0x40032c, 0x01000000, 0x00000000);
+	}
+
+	if (nv_rd32(priv, 0x400330) == inst)
+		nv_mask(priv, 0x400330, 0x01000000, 0x00000000);
+
+	nv_mask(priv, 0x400720, 0x00000001, 0x00000001);
+	return ret;
+}
+
+static struct nouveau_oclass
+nv40_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv40_graph_context_ctor,
+		.dtor = _nouveau_graph_context_dtor,
+		.init = _nouveau_graph_context_init,
+		.fini = nv40_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static void
+nv40_graph_tile_prog(struct nouveau_engine *engine, int i)
+{
+	struct nouveau_fb_tile *tile = &nouveau_fb(engine)->tile.region[i];
+	struct nouveau_fifo *pfifo = nouveau_fifo(engine);
+	struct nv40_graph_priv *priv = (void *)engine;
+	unsigned long flags;
+
+	pfifo->pause(pfifo, &flags);
+	nv04_graph_idle(priv);
+
+	switch (nv_device(priv)->chipset) {
+	case 0x40:
+	case 0x41:
+	case 0x42:
+	case 0x43:
+	case 0x45:
+	case 0x4e:
+		nv_wr32(priv, NV20_PGRAPH_TSIZE(i), tile->pitch);
+		nv_wr32(priv, NV20_PGRAPH_TLIMIT(i), tile->limit);
+		nv_wr32(priv, NV20_PGRAPH_TILE(i), tile->addr);
+		nv_wr32(priv, NV40_PGRAPH_TSIZE1(i), tile->pitch);
+		nv_wr32(priv, NV40_PGRAPH_TLIMIT1(i), tile->limit);
+		nv_wr32(priv, NV40_PGRAPH_TILE1(i), tile->addr);
+		switch (nv_device(priv)->chipset) {
+		case 0x40:
+		case 0x45:
+			nv_wr32(priv, NV20_PGRAPH_ZCOMP(i), tile->zcomp);
+			nv_wr32(priv, NV40_PGRAPH_ZCOMP1(i), tile->zcomp);
+			break;
+		case 0x41:
+		case 0x42:
+		case 0x43:
+			nv_wr32(priv, NV41_PGRAPH_ZCOMP0(i), tile->zcomp);
+			nv_wr32(priv, NV41_PGRAPH_ZCOMP1(i), tile->zcomp);
+			break;
+		default:
+			break;
+		}
+		break;
+	case 0x44:
+	case 0x4a:
+		nv_wr32(priv, NV20_PGRAPH_TSIZE(i), tile->pitch);
+		nv_wr32(priv, NV20_PGRAPH_TLIMIT(i), tile->limit);
+		nv_wr32(priv, NV20_PGRAPH_TILE(i), tile->addr);
+		break;
+	case 0x46:
+	case 0x4c:
+	case 0x47:
+	case 0x49:
+	case 0x4b:
+	case 0x63:
+	case 0x67:
+	case 0x68:
+		nv_wr32(priv, NV47_PGRAPH_TSIZE(i), tile->pitch);
+		nv_wr32(priv, NV47_PGRAPH_TLIMIT(i), tile->limit);
+		nv_wr32(priv, NV47_PGRAPH_TILE(i), tile->addr);
+		nv_wr32(priv, NV40_PGRAPH_TSIZE1(i), tile->pitch);
+		nv_wr32(priv, NV40_PGRAPH_TLIMIT1(i), tile->limit);
+		nv_wr32(priv, NV40_PGRAPH_TILE1(i), tile->addr);
+		switch (nv_device(priv)->chipset) {
+		case 0x47:
+		case 0x49:
+		case 0x4b:
+			nv_wr32(priv, NV47_PGRAPH_ZCOMP0(i), tile->zcomp);
+			nv_wr32(priv, NV47_PGRAPH_ZCOMP1(i), tile->zcomp);
+			break;
+		default:
+			break;
+		}
+		break;
+	default:
+		break;
+	}
+
+	pfifo->start(pfifo, &flags);
+}
+
+static void
+nv40_graph_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_fifo *pfifo = nouveau_fifo(subdev);
+	struct nouveau_engine *engine = nv_engine(subdev);
+	struct nouveau_object *engctx;
+	struct nouveau_handle *handle = NULL;
+	struct nv40_graph_priv *priv = (void *)subdev;
+	u32 stat = nv_rd32(priv, NV03_PGRAPH_INTR);
+	u32 nsource = nv_rd32(priv, NV03_PGRAPH_NSOURCE);
+	u32 nstatus = nv_rd32(priv, NV03_PGRAPH_NSTATUS);
+	u32 inst = nv_rd32(priv, 0x40032c) & 0x000fffff;
+	u32 addr = nv_rd32(priv, NV04_PGRAPH_TRAPPED_ADDR);
+	u32 subc = (addr & 0x00070000) >> 16;
+	u32 mthd = (addr & 0x00001ffc);
+	u32 data = nv_rd32(priv, NV04_PGRAPH_TRAPPED_DATA);
+	u32 class = nv_rd32(priv, 0x400160 + subc * 4) & 0xffff;
+	u32 show = stat;
+	int chid;
+
+	engctx = nouveau_engctx_get(engine, inst);
+	chid   = pfifo->chid(pfifo, engctx);
+
+	if (stat & NV_PGRAPH_INTR_ERROR) {
+		if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
+			handle = nouveau_handle_get_class(engctx, class);
+			if (handle && !nv_call(handle->object, mthd, data))
+				show &= ~NV_PGRAPH_INTR_ERROR;
+			nouveau_handle_put(handle);
+		}
+
+		if (nsource & NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION) {
+			nv_mask(priv, 0x402000, 0, 0);
+		}
+	}
+
+	nv_wr32(priv, NV03_PGRAPH_INTR, stat);
+	nv_wr32(priv, NV04_PGRAPH_FIFO, 0x00000001);
+
+	if (show) {
+		nv_error(priv, "");
+		nouveau_bitfield_print(nv10_graph_intr_name, show);
+		printk(" nsource:");
+		nouveau_bitfield_print(nv04_graph_nsource, nsource);
+		printk(" nstatus:");
+		nouveau_bitfield_print(nv10_graph_nstatus, nstatus);
+		printk("\n");
+		nv_error(priv, "ch %d [0x%08x] subc %d class 0x%04x "
+			       "mthd 0x%04x data 0x%08x\n",
+			 chid, inst << 4, subc, class, mthd, data);
+	}
+
+	nouveau_engctx_put(engctx);
+}
+
+static int
+nv40_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv40_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00001000;
+	nv_subdev(priv)->intr = nv40_graph_intr;
+	nv_engine(priv)->cclass = &nv40_graph_cclass;
+	if (nv44_graph_class(priv))
+		nv_engine(priv)->sclass = nv44_graph_sclass;
+	else
+		nv_engine(priv)->sclass = nv40_graph_sclass;
+	nv_engine(priv)->tile_prog = nv40_graph_tile_prog;
+	return 0;
+}
+
+static int
+nv40_graph_init(struct nouveau_object *object)
+{
+	struct nouveau_engine *engine = nv_engine(object);
+	struct nouveau_fb *pfb = nouveau_fb(object);
+	struct nv40_graph_priv *priv = (void *)engine;
+	int ret, i, j;
+	u32 vramsz;
+
+	ret = nouveau_graph_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* generate and upload context program */
+	ret = nv40_grctx_init(nv_device(priv), &priv->size);
+	if (ret)
+		return ret;
+
+	/* No context present currently */
+	nv_wr32(priv, NV40_PGRAPH_CTXCTL_CUR, 0x00000000);
+
+	nv_wr32(priv, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
+	nv_wr32(priv, NV40_PGRAPH_INTR_EN, 0xFFFFFFFF);
+
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0xFFFFFFFF);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_0, 0x00000000);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_1, 0x401287c0);
+	nv_wr32(priv, NV04_PGRAPH_DEBUG_3, 0xe0de8055);
+	nv_wr32(priv, NV10_PGRAPH_DEBUG_4, 0x00008000);
+	nv_wr32(priv, NV04_PGRAPH_LIMIT_VIOL_PIX, 0x00be3c5f);
+
+	nv_wr32(priv, NV10_PGRAPH_CTX_CONTROL, 0x10010100);
+	nv_wr32(priv, NV10_PGRAPH_STATE      , 0xFFFFFFFF);
+
+	j = nv_rd32(priv, 0x1540) & 0xff;
+	if (j) {
+		for (i = 0; !(j & 1); j >>= 1, i++)
+			;
+		nv_wr32(priv, 0x405000, i);
+	}
+
+	if (nv_device(priv)->chipset == 0x40) {
+		nv_wr32(priv, 0x4009b0, 0x83280fff);
+		nv_wr32(priv, 0x4009b4, 0x000000a0);
+	} else {
+		nv_wr32(priv, 0x400820, 0x83280eff);
+		nv_wr32(priv, 0x400824, 0x000000a0);
+	}
+
+	switch (nv_device(priv)->chipset) {
+	case 0x40:
+	case 0x45:
+		nv_wr32(priv, 0x4009b8, 0x0078e366);
+		nv_wr32(priv, 0x4009bc, 0x0000014c);
+		break;
+	case 0x41:
+	case 0x42: /* pciid also 0x00Cx */
+	/* case 0x0120: XXX (pciid) */
+		nv_wr32(priv, 0x400828, 0x007596ff);
+		nv_wr32(priv, 0x40082c, 0x00000108);
+		break;
+	case 0x43:
+		nv_wr32(priv, 0x400828, 0x0072cb77);
+		nv_wr32(priv, 0x40082c, 0x00000108);
+		break;
+	case 0x44:
+	case 0x46: /* G72 */
+	case 0x4a:
+	case 0x4c: /* G7x-based C51 */
+	case 0x4e:
+		nv_wr32(priv, 0x400860, 0);
+		nv_wr32(priv, 0x400864, 0);
+		break;
+	case 0x47: /* G70 */
+	case 0x49: /* G71 */
+	case 0x4b: /* G73 */
+		nv_wr32(priv, 0x400828, 0x07830610);
+		nv_wr32(priv, 0x40082c, 0x0000016A);
+		break;
+	default:
+		break;
+	}
+
+	nv_wr32(priv, 0x400b38, 0x2ffff800);
+	nv_wr32(priv, 0x400b3c, 0x00006000);
+
+	/* Tiling related stuff. */
+	switch (nv_device(priv)->chipset) {
+	case 0x44:
+	case 0x4a:
+		nv_wr32(priv, 0x400bc4, 0x1003d888);
+		nv_wr32(priv, 0x400bbc, 0xb7a7b500);
+		break;
+	case 0x46:
+		nv_wr32(priv, 0x400bc4, 0x0000e024);
+		nv_wr32(priv, 0x400bbc, 0xb7a7b520);
+		break;
+	case 0x4c:
+	case 0x4e:
+	case 0x67:
+		nv_wr32(priv, 0x400bc4, 0x1003d888);
+		nv_wr32(priv, 0x400bbc, 0xb7a7b540);
+		break;
+	default:
+		break;
+	}
+
+	/* Turn all the tiling regions off. */
+	for (i = 0; i < pfb->tile.regions; i++)
+		engine->tile_prog(engine, i);
+
+	/* begin RAM config */
+	vramsz = pci_resource_len(nv_device(priv)->pdev, 0) - 1;
+	switch (nv_device(priv)->chipset) {
+	case 0x40:
+		nv_wr32(priv, 0x4009A4, nv_rd32(priv, 0x100200));
+		nv_wr32(priv, 0x4009A8, nv_rd32(priv, 0x100204));
+		nv_wr32(priv, 0x4069A4, nv_rd32(priv, 0x100200));
+		nv_wr32(priv, 0x4069A8, nv_rd32(priv, 0x100204));
+		nv_wr32(priv, 0x400820, 0);
+		nv_wr32(priv, 0x400824, 0);
+		nv_wr32(priv, 0x400864, vramsz);
+		nv_wr32(priv, 0x400868, vramsz);
+		break;
+	default:
+		switch (nv_device(priv)->chipset) {
+		case 0x41:
+		case 0x42:
+		case 0x43:
+		case 0x45:
+		case 0x4e:
+		case 0x44:
+		case 0x4a:
+			nv_wr32(priv, 0x4009F0, nv_rd32(priv, 0x100200));
+			nv_wr32(priv, 0x4009F4, nv_rd32(priv, 0x100204));
+			break;
+		default:
+			nv_wr32(priv, 0x400DF0, nv_rd32(priv, 0x100200));
+			nv_wr32(priv, 0x400DF4, nv_rd32(priv, 0x100204));
+			break;
+		}
+		nv_wr32(priv, 0x4069F0, nv_rd32(priv, 0x100200));
+		nv_wr32(priv, 0x4069F4, nv_rd32(priv, 0x100204));
+		nv_wr32(priv, 0x400840, 0);
+		nv_wr32(priv, 0x400844, 0);
+		nv_wr32(priv, 0x4008A0, vramsz);
+		nv_wr32(priv, 0x4008A4, vramsz);
+		break;
+	}
+
+	return 0;
+}
+
+struct nouveau_oclass
+nv40_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv40_graph_ctor,
+		.dtor = _nouveau_graph_dtor,
+		.init = nv40_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv40.h b/nvkm/engine/graph/nv40.h
new file mode 100644
index 000000000..7da35a4e7
--- /dev/null
+++ b/nvkm/engine/graph/nv40.h
@@ -0,0 +1,21 @@
+#ifndef __NV40_GRAPH_H__
+#define __NV40_GRAPH_H__
+
+/* returns 1 if device is one of the nv4x using the 0x4497 object class,
+ * helpful to determine a number of other hardware features
+ */
+static inline int
+nv44_graph_class(void *priv)
+{
+	struct nouveau_device *device = nv_device(priv);
+
+	if ((device->chipset & 0xf0) == 0x60)
+		return 1;
+
+	return !(0x0baf & (1 << (device->chipset & 0x0f)));
+}
+
+int  nv40_grctx_init(struct nouveau_device *, u32 *size);
+void nv40_grctx_fill(struct nouveau_device *, struct nouveau_gpuobj *);
+
+#endif
diff --git a/nvkm/engine/graph/nv50.c b/nvkm/engine/graph/nv50.c
new file mode 100644
index 000000000..b1c3d835b
--- /dev/null
+++ b/nvkm/engine/graph/nv50.c
@@ -0,0 +1,955 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/handle.h>
+#include <core/engctx.h>
+#include <core/enum.h>
+
+#include <subdev/fb.h>
+#include <subdev/vm.h>
+#include <subdev/timer.h>
+
+#include <engine/fifo.h>
+#include <engine/graph.h>
+
+#include "nv50.h"
+
+struct nv50_graph_priv {
+	struct nouveau_graph base;
+	spinlock_t lock;
+	u32 size;
+};
+
+struct nv50_graph_chan {
+	struct nouveau_graph_chan base;
+};
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+static int
+nv50_graph_object_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nouveau_gpuobj *obj;
+	int ret;
+
+	ret = nouveau_gpuobj_create(parent, engine, oclass, 0, parent,
+				    16, 16, 0, &obj);
+	*pobject = nv_object(obj);
+	if (ret)
+		return ret;
+
+	nv_wo32(obj, 0x00, nv_mclass(obj));
+	nv_wo32(obj, 0x04, 0x00000000);
+	nv_wo32(obj, 0x08, 0x00000000);
+	nv_wo32(obj, 0x0c, 0x00000000);
+	return 0;
+}
+
+static struct nouveau_ofuncs
+nv50_graph_ofuncs = {
+	.ctor = nv50_graph_object_ctor,
+	.dtor = _nouveau_gpuobj_dtor,
+	.init = _nouveau_gpuobj_init,
+	.fini = _nouveau_gpuobj_fini,
+	.rd32 = _nouveau_gpuobj_rd32,
+	.wr32 = _nouveau_gpuobj_wr32,
+};
+
+static struct nouveau_oclass
+nv50_graph_sclass[] = {
+	{ 0x0030, &nv50_graph_ofuncs },
+	{ 0x502d, &nv50_graph_ofuncs },
+	{ 0x5039, &nv50_graph_ofuncs },
+	{ 0x5097, &nv50_graph_ofuncs },
+	{ 0x50c0, &nv50_graph_ofuncs },
+	{}
+};
+
+static struct nouveau_oclass
+nv84_graph_sclass[] = {
+	{ 0x0030, &nv50_graph_ofuncs },
+	{ 0x502d, &nv50_graph_ofuncs },
+	{ 0x5039, &nv50_graph_ofuncs },
+	{ 0x50c0, &nv50_graph_ofuncs },
+	{ 0x8297, &nv50_graph_ofuncs },
+	{}
+};
+
+static struct nouveau_oclass
+nva0_graph_sclass[] = {
+	{ 0x0030, &nv50_graph_ofuncs },
+	{ 0x502d, &nv50_graph_ofuncs },
+	{ 0x5039, &nv50_graph_ofuncs },
+	{ 0x50c0, &nv50_graph_ofuncs },
+	{ 0x8397, &nv50_graph_ofuncs },
+	{}
+};
+
+static struct nouveau_oclass
+nva3_graph_sclass[] = {
+	{ 0x0030, &nv50_graph_ofuncs },
+	{ 0x502d, &nv50_graph_ofuncs },
+	{ 0x5039, &nv50_graph_ofuncs },
+	{ 0x50c0, &nv50_graph_ofuncs },
+	{ 0x8597, &nv50_graph_ofuncs },
+	{ 0x85c0, &nv50_graph_ofuncs },
+	{}
+};
+
+static struct nouveau_oclass
+nvaf_graph_sclass[] = {
+	{ 0x0030, &nv50_graph_ofuncs },
+	{ 0x502d, &nv50_graph_ofuncs },
+	{ 0x5039, &nv50_graph_ofuncs },
+	{ 0x50c0, &nv50_graph_ofuncs },
+	{ 0x85c0, &nv50_graph_ofuncs },
+	{ 0x8697, &nv50_graph_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static int
+nv50_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *data, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nv50_graph_priv *priv = (void *)engine;
+	struct nv50_graph_chan *chan;
+	int ret;
+
+	ret = nouveau_graph_context_create(parent, engine, oclass, NULL,
+					   priv->size, 0,
+					   NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv50_grctx_fill(nv_device(priv), nv_gpuobj(chan));
+	return 0;
+}
+
+static struct nouveau_oclass
+nv50_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_graph_context_ctor,
+		.dtor = _nouveau_graph_context_dtor,
+		.init = _nouveau_graph_context_init,
+		.fini = _nouveau_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv50_graph_tlb_flush(struct nouveau_engine *engine)
+{
+	nv50_vm_flush_engine(&engine->base, 0x00);
+	return 0;
+}
+
+static const struct nouveau_bitfield nv50_pgraph_status[] = {
+	{ 0x00000001, "BUSY" }, /* set when any bit is set */
+	{ 0x00000002, "DISPATCH" },
+	{ 0x00000004, "UNK2" },
+	{ 0x00000008, "UNK3" },
+	{ 0x00000010, "UNK4" },
+	{ 0x00000020, "UNK5" },
+	{ 0x00000040, "M2MF" },
+	{ 0x00000080, "UNK7" },
+	{ 0x00000100, "CTXPROG" },
+	{ 0x00000200, "VFETCH" },
+	{ 0x00000400, "CCACHE_UNK4" },
+	{ 0x00000800, "STRMOUT_GSCHED_UNK5" },
+	{ 0x00001000, "UNK14XX" },
+	{ 0x00002000, "UNK24XX_CSCHED" },
+	{ 0x00004000, "UNK1CXX" },
+	{ 0x00008000, "CLIPID" },
+	{ 0x00010000, "ZCULL" },
+	{ 0x00020000, "ENG2D" },
+	{ 0x00040000, "UNK34XX" },
+	{ 0x00080000, "TPRAST" },
+	{ 0x00100000, "TPROP" },
+	{ 0x00200000, "TEX" },
+	{ 0x00400000, "TPVP" },
+	{ 0x00800000, "MP" },
+	{ 0x01000000, "ROP" },
+	{}
+};
+
+static const char *const nv50_pgraph_vstatus_0[] = {
+	"VFETCH", "CCACHE", "UNK4", "UNK5", "GSCHED", "STRMOUT", "UNK14XX", NULL
+};
+
+static const char *const nv50_pgraph_vstatus_1[] = {
+	"TPRAST", "TPROP", "TEXTURE", "TPVP", "MP", NULL
+};
+
+static const char *const nv50_pgraph_vstatus_2[] = {
+	"UNK24XX", "CSCHED", "UNK1CXX", "CLIPID", "ZCULL", "ENG2D", "UNK34XX",
+	"ROP", NULL
+};
+
+static void nouveau_pgraph_vstatus_print(struct nv50_graph_priv *priv, int r,
+		const char *const units[], u32 status)
+{
+	int i;
+
+	nv_error(priv, "PGRAPH_VSTATUS%d: 0x%08x", r, status);
+
+	for (i = 0; units[i] && status; i++) {
+		if ((status & 7) == 1)
+			pr_cont(" %s", units[i]);
+		status >>= 3;
+	}
+	if (status)
+		pr_cont(" (invalid: 0x%x)", status);
+	pr_cont("\n");
+}
+
+static int
+nv84_graph_tlb_flush(struct nouveau_engine *engine)
+{
+	struct nouveau_timer *ptimer = nouveau_timer(engine);
+	struct nv50_graph_priv *priv = (void *)engine;
+	bool idle, timeout = false;
+	unsigned long flags;
+	u64 start;
+	u32 tmp;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	nv_mask(priv, 0x400500, 0x00000001, 0x00000000);
+
+	start = ptimer->read(ptimer);
+	do {
+		idle = true;
+
+		for (tmp = nv_rd32(priv, 0x400380); tmp && idle; tmp >>= 3) {
+			if ((tmp & 7) == 1)
+				idle = false;
+		}
+
+		for (tmp = nv_rd32(priv, 0x400384); tmp && idle; tmp >>= 3) {
+			if ((tmp & 7) == 1)
+				idle = false;
+		}
+
+		for (tmp = nv_rd32(priv, 0x400388); tmp && idle; tmp >>= 3) {
+			if ((tmp & 7) == 1)
+				idle = false;
+		}
+	} while (!idle &&
+		 !(timeout = ptimer->read(ptimer) - start > 2000000000));
+
+	if (timeout) {
+		nv_error(priv, "PGRAPH TLB flush idle timeout fail\n");
+
+		tmp = nv_rd32(priv, 0x400700);
+		nv_error(priv, "PGRAPH_STATUS  : 0x%08x", tmp);
+		nouveau_bitfield_print(nv50_pgraph_status, tmp);
+		pr_cont("\n");
+
+		nouveau_pgraph_vstatus_print(priv, 0, nv50_pgraph_vstatus_0,
+				nv_rd32(priv, 0x400380));
+		nouveau_pgraph_vstatus_print(priv, 1, nv50_pgraph_vstatus_1,
+				nv_rd32(priv, 0x400384));
+		nouveau_pgraph_vstatus_print(priv, 2, nv50_pgraph_vstatus_2,
+				nv_rd32(priv, 0x400388));
+	}
+
+	nv50_vm_flush_engine(&engine->base, 0x00);
+
+	nv_mask(priv, 0x400500, 0x00000001, 0x00000001);
+	spin_unlock_irqrestore(&priv->lock, flags);
+	return timeout ? -EBUSY : 0;
+}
+
+static const struct nouveau_enum nv50_mp_exec_error_names[] = {
+	{ 3, "STACK_UNDERFLOW", NULL },
+	{ 4, "QUADON_ACTIVE", NULL },
+	{ 8, "TIMEOUT", NULL },
+	{ 0x10, "INVALID_OPCODE", NULL },
+	{ 0x40, "BREAKPOINT", NULL },
+	{}
+};
+
+static const struct nouveau_bitfield nv50_graph_trap_m2mf[] = {
+	{ 0x00000001, "NOTIFY" },
+	{ 0x00000002, "IN" },
+	{ 0x00000004, "OUT" },
+	{}
+};
+
+static const struct nouveau_bitfield nv50_graph_trap_vfetch[] = {
+	{ 0x00000001, "FAULT" },
+	{}
+};
+
+static const struct nouveau_bitfield nv50_graph_trap_strmout[] = {
+	{ 0x00000001, "FAULT" },
+	{}
+};
+
+static const struct nouveau_bitfield nv50_graph_trap_ccache[] = {
+	{ 0x00000001, "FAULT" },
+	{}
+};
+
+/* There must be a *lot* of these. Will take some time to gather them up. */
+const struct nouveau_enum nv50_data_error_names[] = {
+	{ 0x00000003, "INVALID_OPERATION", NULL },
+	{ 0x00000004, "INVALID_VALUE", NULL },
+	{ 0x00000005, "INVALID_ENUM", NULL },
+	{ 0x00000008, "INVALID_OBJECT", NULL },
+	{ 0x00000009, "READ_ONLY_OBJECT", NULL },
+	{ 0x0000000a, "SUPERVISOR_OBJECT", NULL },
+	{ 0x0000000b, "INVALID_ADDRESS_ALIGNMENT", NULL },
+	{ 0x0000000c, "INVALID_BITFIELD", NULL },
+	{ 0x0000000d, "BEGIN_END_ACTIVE", NULL },
+	{ 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT", NULL },
+	{ 0x0000000f, "VIEWPORT_ID_NEEDS_GP", NULL },
+	{ 0x00000010, "RT_DOUBLE_BIND", NULL },
+	{ 0x00000011, "RT_TYPES_MISMATCH", NULL },
+	{ 0x00000012, "RT_LINEAR_WITH_ZETA", NULL },
+	{ 0x00000015, "FP_TOO_FEW_REGS", NULL },
+	{ 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH", NULL },
+	{ 0x00000017, "RT_LINEAR_WITH_MSAA", NULL },
+	{ 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT", NULL },
+	{ 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT", NULL },
+	{ 0x0000001a, "RT_INVALID_ALIGNMENT", NULL },
+	{ 0x0000001b, "SAMPLER_OVER_LIMIT", NULL },
+	{ 0x0000001c, "TEXTURE_OVER_LIMIT", NULL },
+	{ 0x0000001e, "GP_TOO_MANY_OUTPUTS", NULL },
+	{ 0x0000001f, "RT_BPP128_WITH_MS8", NULL },
+	{ 0x00000021, "Z_OUT_OF_BOUNDS", NULL },
+	{ 0x00000023, "XY_OUT_OF_BOUNDS", NULL },
+	{ 0x00000024, "VP_ZERO_INPUTS", NULL },
+	{ 0x00000027, "CP_MORE_PARAMS_THAN_SHARED", NULL },
+	{ 0x00000028, "CP_NO_REG_SPACE_STRIPED", NULL },
+	{ 0x00000029, "CP_NO_REG_SPACE_PACKED", NULL },
+	{ 0x0000002a, "CP_NOT_ENOUGH_WARPS", NULL },
+	{ 0x0000002b, "CP_BLOCK_SIZE_MISMATCH", NULL },
+	{ 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS", NULL },
+	{ 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS", NULL },
+	{ 0x0000002e, "CP_NO_BLOCKDIM_LATCH", NULL },
+	{ 0x00000031, "ENG2D_FORMAT_MISMATCH", NULL },
+	{ 0x0000003f, "PRIMITIVE_ID_NEEDS_GP", NULL },
+	{ 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT", NULL },
+	{ 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT", NULL },
+	{ 0x00000046, "LAYER_ID_NEEDS_GP", NULL },
+	{ 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT", NULL },
+	{ 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT", NULL },
+	{}
+};
+
+static const struct nouveau_bitfield nv50_graph_intr_name[] = {
+	{ 0x00000001, "NOTIFY" },
+	{ 0x00000002, "COMPUTE_QUERY" },
+	{ 0x00000010, "ILLEGAL_MTHD" },
+	{ 0x00000020, "ILLEGAL_CLASS" },
+	{ 0x00000040, "DOUBLE_NOTIFY" },
+	{ 0x00001000, "CONTEXT_SWITCH" },
+	{ 0x00010000, "BUFFER_NOTIFY" },
+	{ 0x00100000, "DATA_ERROR" },
+	{ 0x00200000, "TRAP" },
+	{ 0x01000000, "SINGLE_STEP" },
+	{}
+};
+
+static void
+nv50_priv_mp_trap(struct nv50_graph_priv *priv, int tpid, int display)
+{
+	u32 units = nv_rd32(priv, 0x1540);
+	u32 addr, mp10, status, pc, oplow, ophigh;
+	int i;
+	int mps = 0;
+	for (i = 0; i < 4; i++) {
+		if (!(units & 1 << (i+24)))
+			continue;
+		if (nv_device(priv)->chipset < 0xa0)
+			addr = 0x408200 + (tpid << 12) + (i << 7);
+		else
+			addr = 0x408100 + (tpid << 11) + (i << 7);
+		mp10 = nv_rd32(priv, addr + 0x10);
+		status = nv_rd32(priv, addr + 0x14);
+		if (!status)
+			continue;
+		if (display) {
+			nv_rd32(priv, addr + 0x20);
+			pc = nv_rd32(priv, addr + 0x24);
+			oplow = nv_rd32(priv, addr + 0x70);
+			ophigh = nv_rd32(priv, addr + 0x74);
+			nv_error(priv, "TRAP_MP_EXEC - "
+					"TP %d MP %d: ", tpid, i);
+			nouveau_enum_print(nv50_mp_exec_error_names, status);
+			printk(" at %06x warp %d, opcode %08x %08x\n",
+					pc&0xffffff, pc >> 24,
+					oplow, ophigh);
+		}
+		nv_wr32(priv, addr + 0x10, mp10);
+		nv_wr32(priv, addr + 0x14, 0);
+		mps++;
+	}
+	if (!mps && display)
+		nv_error(priv, "TRAP_MP_EXEC - TP %d: "
+				"No MPs claiming errors?\n", tpid);
+}
+
+static void
+nv50_priv_tp_trap(struct nv50_graph_priv *priv, int type, u32 ustatus_old,
+		u32 ustatus_new, int display, const char *name)
+{
+	int tps = 0;
+	u32 units = nv_rd32(priv, 0x1540);
+	int i, r;
+	u32 ustatus_addr, ustatus;
+	for (i = 0; i < 16; i++) {
+		if (!(units & (1 << i)))
+			continue;
+		if (nv_device(priv)->chipset < 0xa0)
+			ustatus_addr = ustatus_old + (i << 12);
+		else
+			ustatus_addr = ustatus_new + (i << 11);
+		ustatus = nv_rd32(priv, ustatus_addr) & 0x7fffffff;
+		if (!ustatus)
+			continue;
+		tps++;
+		switch (type) {
+		case 6: /* texture error... unknown for now */
+			if (display) {
+				nv_error(priv, "magic set %d:\n", i);
+				for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4)
+					nv_error(priv, "\t0x%08x: 0x%08x\n", r,
+						nv_rd32(priv, r));
+			}
+			break;
+		case 7: /* MP error */
+			if (ustatus & 0x04030000) {
+				nv50_priv_mp_trap(priv, i, display);
+				ustatus &= ~0x04030000;
+			}
+			break;
+		case 8: /* TPDMA error */
+			{
+			u32 e0c = nv_rd32(priv, ustatus_addr + 4);
+			u32 e10 = nv_rd32(priv, ustatus_addr + 8);
+			u32 e14 = nv_rd32(priv, ustatus_addr + 0xc);
+			u32 e18 = nv_rd32(priv, ustatus_addr + 0x10);
+			u32 e1c = nv_rd32(priv, ustatus_addr + 0x14);
+			u32 e20 = nv_rd32(priv, ustatus_addr + 0x18);
+			u32 e24 = nv_rd32(priv, ustatus_addr + 0x1c);
+			/* 2d engine destination */
+			if (ustatus & 0x00000010) {
+				if (display) {
+					nv_error(priv, "TRAP_TPDMA_2D - TP %d - Unknown fault at address %02x%08x\n",
+							i, e14, e10);
+					nv_error(priv, "TRAP_TPDMA_2D - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
+							i, e0c, e18, e1c, e20, e24);
+				}
+				ustatus &= ~0x00000010;
+			}
+			/* Render target */
+			if (ustatus & 0x00000040) {
+				if (display) {
+					nv_error(priv, "TRAP_TPDMA_RT - TP %d - Unknown fault at address %02x%08x\n",
+							i, e14, e10);
+					nv_error(priv, "TRAP_TPDMA_RT - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
+							i, e0c, e18, e1c, e20, e24);
+				}
+				ustatus &= ~0x00000040;
+			}
+			/* CUDA memory: l[], g[] or stack. */
+			if (ustatus & 0x00000080) {
+				if (display) {
+					if (e18 & 0x80000000) {
+						/* g[] read fault? */
+						nv_error(priv, "TRAP_TPDMA - TP %d - Global read fault at address %02x%08x\n",
+								i, e14, e10 | ((e18 >> 24) & 0x1f));
+						e18 &= ~0x1f000000;
+					} else if (e18 & 0xc) {
+						/* g[] write fault? */
+						nv_error(priv, "TRAP_TPDMA - TP %d - Global write fault at address %02x%08x\n",
+								i, e14, e10 | ((e18 >> 7) & 0x1f));
+						e18 &= ~0x00000f80;
+					} else {
+						nv_error(priv, "TRAP_TPDMA - TP %d - Unknown CUDA fault at address %02x%08x\n",
+								i, e14, e10);
+					}
+					nv_error(priv, "TRAP_TPDMA - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
+							i, e0c, e18, e1c, e20, e24);
+				}
+				ustatus &= ~0x00000080;
+			}
+			}
+			break;
+		}
+		if (ustatus) {
+			if (display)
+				nv_error(priv, "%s - TP%d: Unhandled ustatus 0x%08x\n", name, i, ustatus);
+		}
+		nv_wr32(priv, ustatus_addr, 0xc0000000);
+	}
+
+	if (!tps && display)
+		nv_warn(priv, "%s - No TPs claiming errors?\n", name);
+}
+
+static int
+nv50_graph_trap_handler(struct nv50_graph_priv *priv, u32 display,
+			int chid, u64 inst)
+{
+	u32 status = nv_rd32(priv, 0x400108);
+	u32 ustatus;
+
+	if (!status && display) {
+		nv_error(priv, "TRAP: no units reporting traps?\n");
+		return 1;
+	}
+
+	/* DISPATCH: Relays commands to other units and handles NOTIFY,
+	 * COND, QUERY. If you get a trap from it, the command is still stuck
+	 * in DISPATCH and you need to do something about it. */
+	if (status & 0x001) {
+		ustatus = nv_rd32(priv, 0x400804) & 0x7fffffff;
+		if (!ustatus && display) {
+			nv_error(priv, "TRAP_DISPATCH - no ustatus?\n");
+		}
+
+		nv_wr32(priv, 0x400500, 0x00000000);
+
+		/* Known to be triggered by screwed up NOTIFY and COND... */
+		if (ustatus & 0x00000001) {
+			u32 addr = nv_rd32(priv, 0x400808);
+			u32 subc = (addr & 0x00070000) >> 16;
+			u32 mthd = (addr & 0x00001ffc);
+			u32 datal = nv_rd32(priv, 0x40080c);
+			u32 datah = nv_rd32(priv, 0x400810);
+			u32 class = nv_rd32(priv, 0x400814);
+			u32 r848 = nv_rd32(priv, 0x400848);
+
+			nv_error(priv, "TRAP DISPATCH_FAULT\n");
+			if (display && (addr & 0x80000000)) {
+				nv_error(priv, "ch %d [0x%010llx] "
+					     "subc %d class 0x%04x mthd 0x%04x "
+					     "data 0x%08x%08x "
+					     "400808 0x%08x 400848 0x%08x\n",
+					chid, inst, subc, class, mthd, datah,
+					datal, addr, r848);
+			} else
+			if (display) {
+				nv_error(priv, "no stuck command?\n");
+			}
+
+			nv_wr32(priv, 0x400808, 0);
+			nv_wr32(priv, 0x4008e8, nv_rd32(priv, 0x4008e8) & 3);
+			nv_wr32(priv, 0x400848, 0);
+			ustatus &= ~0x00000001;
+		}
+
+		if (ustatus & 0x00000002) {
+			u32 addr = nv_rd32(priv, 0x40084c);
+			u32 subc = (addr & 0x00070000) >> 16;
+			u32 mthd = (addr & 0x00001ffc);
+			u32 data = nv_rd32(priv, 0x40085c);
+			u32 class = nv_rd32(priv, 0x400814);
+
+			nv_error(priv, "TRAP DISPATCH_QUERY\n");
+			if (display && (addr & 0x80000000)) {
+				nv_error(priv, "ch %d [0x%010llx] "
+					     "subc %d class 0x%04x mthd 0x%04x "
+					     "data 0x%08x 40084c 0x%08x\n",
+					chid, inst, subc, class, mthd,
+					data, addr);
+			} else
+			if (display) {
+				nv_error(priv, "no stuck command?\n");
+			}
+
+			nv_wr32(priv, 0x40084c, 0);
+			ustatus &= ~0x00000002;
+		}
+
+		if (ustatus && display) {
+			nv_error(priv, "TRAP_DISPATCH (unknown "
+				      "0x%08x)\n", ustatus);
+		}
+
+		nv_wr32(priv, 0x400804, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x001);
+		status &= ~0x001;
+		if (!status)
+			return 0;
+	}
+
+	/* M2MF: Memory to memory copy engine. */
+	if (status & 0x002) {
+		u32 ustatus = nv_rd32(priv, 0x406800) & 0x7fffffff;
+		if (display) {
+			nv_error(priv, "TRAP_M2MF");
+			nouveau_bitfield_print(nv50_graph_trap_m2mf, ustatus);
+			printk("\n");
+			nv_error(priv, "TRAP_M2MF %08x %08x %08x %08x\n",
+				nv_rd32(priv, 0x406804), nv_rd32(priv, 0x406808),
+				nv_rd32(priv, 0x40680c), nv_rd32(priv, 0x406810));
+
+		}
+
+		/* No sane way found yet -- just reset the bugger. */
+		nv_wr32(priv, 0x400040, 2);
+		nv_wr32(priv, 0x400040, 0);
+		nv_wr32(priv, 0x406800, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x002);
+		status &= ~0x002;
+	}
+
+	/* VFETCH: Fetches data from vertex buffers. */
+	if (status & 0x004) {
+		u32 ustatus = nv_rd32(priv, 0x400c04) & 0x7fffffff;
+		if (display) {
+			nv_error(priv, "TRAP_VFETCH");
+			nouveau_bitfield_print(nv50_graph_trap_vfetch, ustatus);
+			printk("\n");
+			nv_error(priv, "TRAP_VFETCH %08x %08x %08x %08x\n",
+				nv_rd32(priv, 0x400c00), nv_rd32(priv, 0x400c08),
+				nv_rd32(priv, 0x400c0c), nv_rd32(priv, 0x400c10));
+		}
+
+		nv_wr32(priv, 0x400c04, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x004);
+		status &= ~0x004;
+	}
+
+	/* STRMOUT: DirectX streamout / OpenGL transform feedback. */
+	if (status & 0x008) {
+		ustatus = nv_rd32(priv, 0x401800) & 0x7fffffff;
+		if (display) {
+			nv_error(priv, "TRAP_STRMOUT");
+			nouveau_bitfield_print(nv50_graph_trap_strmout, ustatus);
+			printk("\n");
+			nv_error(priv, "TRAP_STRMOUT %08x %08x %08x %08x\n",
+				nv_rd32(priv, 0x401804), nv_rd32(priv, 0x401808),
+				nv_rd32(priv, 0x40180c), nv_rd32(priv, 0x401810));
+
+		}
+
+		/* No sane way found yet -- just reset the bugger. */
+		nv_wr32(priv, 0x400040, 0x80);
+		nv_wr32(priv, 0x400040, 0);
+		nv_wr32(priv, 0x401800, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x008);
+		status &= ~0x008;
+	}
+
+	/* CCACHE: Handles code and c[] caches and fills them. */
+	if (status & 0x010) {
+		ustatus = nv_rd32(priv, 0x405018) & 0x7fffffff;
+		if (display) {
+			nv_error(priv, "TRAP_CCACHE");
+			nouveau_bitfield_print(nv50_graph_trap_ccache, ustatus);
+			printk("\n");
+			nv_error(priv, "TRAP_CCACHE %08x %08x %08x %08x"
+				     " %08x %08x %08x\n",
+				nv_rd32(priv, 0x405000), nv_rd32(priv, 0x405004),
+				nv_rd32(priv, 0x405008), nv_rd32(priv, 0x40500c),
+				nv_rd32(priv, 0x405010), nv_rd32(priv, 0x405014),
+				nv_rd32(priv, 0x40501c));
+
+		}
+
+		nv_wr32(priv, 0x405018, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x010);
+		status &= ~0x010;
+	}
+
+	/* Unknown, not seen yet... 0x402000 is the only trap status reg
+	 * remaining, so try to handle it anyway. Perhaps related to that
+	 * unknown DMA slot on tesla? */
+	if (status & 0x20) {
+		ustatus = nv_rd32(priv, 0x402000) & 0x7fffffff;
+		if (display)
+			nv_error(priv, "TRAP_UNKC04 0x%08x\n", ustatus);
+		nv_wr32(priv, 0x402000, 0xc0000000);
+		/* no status modifiction on purpose */
+	}
+
+	/* TEXTURE: CUDA texturing units */
+	if (status & 0x040) {
+		nv50_priv_tp_trap(priv, 6, 0x408900, 0x408600, display,
+				    "TRAP_TEXTURE");
+		nv_wr32(priv, 0x400108, 0x040);
+		status &= ~0x040;
+	}
+
+	/* MP: CUDA execution engines. */
+	if (status & 0x080) {
+		nv50_priv_tp_trap(priv, 7, 0x408314, 0x40831c, display,
+				    "TRAP_MP");
+		nv_wr32(priv, 0x400108, 0x080);
+		status &= ~0x080;
+	}
+
+	/* TPDMA:  Handles TP-initiated uncached memory accesses:
+	 * l[], g[], stack, 2d surfaces, render targets. */
+	if (status & 0x100) {
+		nv50_priv_tp_trap(priv, 8, 0x408e08, 0x408708, display,
+				    "TRAP_TPDMA");
+		nv_wr32(priv, 0x400108, 0x100);
+		status &= ~0x100;
+	}
+
+	if (status) {
+		if (display)
+			nv_error(priv, "TRAP: unknown 0x%08x\n", status);
+		nv_wr32(priv, 0x400108, status);
+	}
+
+	return 1;
+}
+
+static void
+nv50_graph_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_fifo *pfifo = nouveau_fifo(subdev);
+	struct nouveau_engine *engine = nv_engine(subdev);
+	struct nouveau_object *engctx;
+	struct nouveau_handle *handle = NULL;
+	struct nv50_graph_priv *priv = (void *)subdev;
+	u32 stat = nv_rd32(priv, 0x400100);
+	u32 inst = nv_rd32(priv, 0x40032c) & 0x0fffffff;
+	u32 addr = nv_rd32(priv, 0x400704);
+	u32 subc = (addr & 0x00070000) >> 16;
+	u32 mthd = (addr & 0x00001ffc);
+	u32 data = nv_rd32(priv, 0x400708);
+	u32 class = nv_rd32(priv, 0x400814);
+	u32 show = stat;
+	int chid;
+
+	engctx = nouveau_engctx_get(engine, inst);
+	chid   = pfifo->chid(pfifo, engctx);
+
+	if (stat & 0x00000010) {
+		handle = nouveau_handle_get_class(engctx, class);
+		if (handle && !nv_call(handle->object, mthd, data))
+			show &= ~0x00000010;
+		nouveau_handle_put(handle);
+	}
+
+	if (show & 0x00100000) {
+		u32 ecode = nv_rd32(priv, 0x400110);
+		nv_error(priv, "DATA_ERROR ");
+		nouveau_enum_print(nv50_data_error_names, ecode);
+		printk("\n");
+	}
+
+	if (stat & 0x00200000) {
+		if (!nv50_graph_trap_handler(priv, show, chid, (u64)inst << 12))
+			show &= ~0x00200000;
+	}
+
+	nv_wr32(priv, 0x400100, stat);
+	nv_wr32(priv, 0x400500, 0x00010001);
+
+	if (show) {
+		nv_error(priv, "");
+		nouveau_bitfield_print(nv50_graph_intr_name, show);
+		printk("\n");
+		nv_error(priv, "ch %d [0x%010llx] subc %d class 0x%04x "
+			       "mthd 0x%04x data 0x%08x\n",
+			 chid, (u64)inst << 12, subc, class, mthd, data);
+	}
+
+	if (nv_rd32(priv, 0x400824) & (1 << 31))
+		nv_wr32(priv, 0x400824, nv_rd32(priv, 0x400824) & ~(1 << 31));
+
+	nouveau_engctx_put(engctx);
+}
+
+static int
+nv50_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_graph_priv *priv;
+	int ret;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00201000;
+	nv_subdev(priv)->intr = nv50_graph_intr;
+	nv_engine(priv)->cclass = &nv50_graph_cclass;
+
+	switch (nv_device(priv)->chipset) {
+	case 0x50:
+		nv_engine(priv)->sclass = nv50_graph_sclass;
+		break;
+	case 0x84:
+	case 0x86:
+	case 0x92:
+	case 0x94:
+	case 0x96:
+	case 0x98:
+		nv_engine(priv)->sclass = nv84_graph_sclass;
+		break;
+	case 0xa0:
+	case 0xaa:
+	case 0xac:
+		nv_engine(priv)->sclass = nva0_graph_sclass;
+		break;
+	case 0xa3:
+	case 0xa5:
+	case 0xa8:
+		nv_engine(priv)->sclass = nva3_graph_sclass;
+		break;
+	case 0xaf:
+		nv_engine(priv)->sclass = nvaf_graph_sclass;
+		break;
+
+	};
+
+	if (nv_device(priv)->chipset == 0x50 ||
+	    nv_device(priv)->chipset == 0xac)
+		nv_engine(priv)->tlb_flush = nv50_graph_tlb_flush;
+	else
+		nv_engine(priv)->tlb_flush = nv84_graph_tlb_flush;
+
+	spin_lock_init(&priv->lock);
+	return 0;
+}
+
+static int
+nv50_graph_init(struct nouveau_object *object)
+{
+	struct nv50_graph_priv *priv = (void *)object;
+	int ret, units, i;
+
+	ret = nouveau_graph_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* NV_PGRAPH_DEBUG_3_HW_CTX_SWITCH_ENABLED */
+	nv_wr32(priv, 0x40008c, 0x00000004);
+
+	/* reset/enable traps and interrupts */
+	nv_wr32(priv, 0x400804, 0xc0000000);
+	nv_wr32(priv, 0x406800, 0xc0000000);
+	nv_wr32(priv, 0x400c04, 0xc0000000);
+	nv_wr32(priv, 0x401800, 0xc0000000);
+	nv_wr32(priv, 0x405018, 0xc0000000);
+	nv_wr32(priv, 0x402000, 0xc0000000);
+
+	units = nv_rd32(priv, 0x001540);
+	for (i = 0; i < 16; i++) {
+		if (!(units & (1 << i)))
+			continue;
+
+		if (nv_device(priv)->chipset < 0xa0) {
+			nv_wr32(priv, 0x408900 + (i << 12), 0xc0000000);
+			nv_wr32(priv, 0x408e08 + (i << 12), 0xc0000000);
+			nv_wr32(priv, 0x408314 + (i << 12), 0xc0000000);
+		} else {
+			nv_wr32(priv, 0x408600 + (i << 11), 0xc0000000);
+			nv_wr32(priv, 0x408708 + (i << 11), 0xc0000000);
+			nv_wr32(priv, 0x40831c + (i << 11), 0xc0000000);
+		}
+	}
+
+	nv_wr32(priv, 0x400108, 0xffffffff);
+	nv_wr32(priv, 0x400138, 0xffffffff);
+	nv_wr32(priv, 0x400100, 0xffffffff);
+	nv_wr32(priv, 0x40013c, 0xffffffff);
+	nv_wr32(priv, 0x400500, 0x00010001);
+
+	/* upload context program, initialise ctxctl defaults */
+	ret = nv50_grctx_init(nv_device(priv), &priv->size);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x400824, 0x00000000);
+	nv_wr32(priv, 0x400828, 0x00000000);
+	nv_wr32(priv, 0x40082c, 0x00000000);
+	nv_wr32(priv, 0x400830, 0x00000000);
+	nv_wr32(priv, 0x400724, 0x00000000);
+	nv_wr32(priv, 0x40032c, 0x00000000);
+	nv_wr32(priv, 0x400320, 4);	/* CTXCTL_CMD = NEWCTXDMA */
+
+	/* some unknown zcull magic */
+	switch (nv_device(priv)->chipset & 0xf0) {
+	case 0x50:
+	case 0x80:
+	case 0x90:
+		nv_wr32(priv, 0x402ca8, 0x00000800);
+		break;
+	case 0xa0:
+	default:
+		nv_wr32(priv, 0x402cc0, 0x00000000);
+		if (nv_device(priv)->chipset == 0xa0 ||
+		    nv_device(priv)->chipset == 0xaa ||
+		    nv_device(priv)->chipset == 0xac) {
+			nv_wr32(priv, 0x402ca8, 0x00000802);
+		} else {
+			nv_wr32(priv, 0x402cc0, 0x00000000);
+			nv_wr32(priv, 0x402ca8, 0x00000002);
+		}
+
+		break;
+	}
+
+	/* zero out zcull regions */
+	for (i = 0; i < 8; i++) {
+		nv_wr32(priv, 0x402c20 + (i * 8), 0x00000000);
+		nv_wr32(priv, 0x402c24 + (i * 8), 0x00000000);
+		nv_wr32(priv, 0x402c28 + (i * 8), 0x00000000);
+		nv_wr32(priv, 0x402c2c + (i * 8), 0x00000000);
+	}
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_graph_ctor,
+		.dtor = _nouveau_graph_dtor,
+		.init = nv50_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nv50.h b/nvkm/engine/graph/nv50.h
new file mode 100644
index 000000000..0505fb419
--- /dev/null
+++ b/nvkm/engine/graph/nv50.h
@@ -0,0 +1,7 @@
+#ifndef __NV50_GRAPH_H__
+#define __NV50_GRAPH_H__
+
+int  nv50_grctx_init(struct nouveau_device *, u32 *size);
+void nv50_grctx_fill(struct nouveau_device *, struct nouveau_gpuobj *);
+
+#endif
diff --git a/nvkm/engine/graph/nvc0.c b/nvkm/engine/graph/nvc0.c
new file mode 100644
index 000000000..45aff5f50
--- /dev/null
+++ b/nvkm/engine/graph/nvc0.c
@@ -0,0 +1,946 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "nvc0.h"
+#include "fuc/hubnvc0.fuc.h"
+#include "fuc/gpcnvc0.fuc.h"
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nvc0_graph_sclass[] = {
+	{ 0x902d, &nouveau_object_ofuncs },
+	{ 0x9039, &nouveau_object_ofuncs },
+	{ 0x9097, &nouveau_object_ofuncs },
+	{ 0x90c0, &nouveau_object_ofuncs },
+	{}
+};
+
+static struct nouveau_oclass
+nvc1_graph_sclass[] = {
+	{ 0x902d, &nouveau_object_ofuncs },
+	{ 0x9039, &nouveau_object_ofuncs },
+	{ 0x9097, &nouveau_object_ofuncs },
+	{ 0x90c0, &nouveau_object_ofuncs },
+	{ 0x9197, &nouveau_object_ofuncs },
+	{}
+};
+
+static struct nouveau_oclass
+nvc8_graph_sclass[] = {
+	{ 0x902d, &nouveau_object_ofuncs },
+	{ 0x9039, &nouveau_object_ofuncs },
+	{ 0x9097, &nouveau_object_ofuncs },
+	{ 0x90c0, &nouveau_object_ofuncs },
+	{ 0x9197, &nouveau_object_ofuncs },
+	{ 0x9297, &nouveau_object_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+int
+nvc0_graph_context_ctor(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass, void *args, u32 size,
+			struct nouveau_object **pobject)
+{
+	struct nouveau_vm *vm = nouveau_client(parent)->vm;
+	struct nvc0_graph_priv *priv = (void *)engine;
+	struct nvc0_graph_data *data = priv->mmio_data;
+	struct nvc0_graph_mmio *mmio = priv->mmio_list;
+	struct nvc0_graph_chan *chan;
+	int ret, i;
+
+	/* allocate memory for context, and fill with default values */
+	ret = nouveau_graph_context_create(parent, engine, oclass, NULL,
+					   priv->size, 0x100,
+					   NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	/* allocate memory for a "mmio list" buffer that's used by the HUB
+	 * fuc to modify some per-context register settings on first load
+	 * of the context.
+	 */
+	ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 0x100, 0, &chan->mmio);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_map_vm(nv_gpuobj(chan->mmio), vm,
+				    NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS,
+				    &chan->mmio_vma);
+	if (ret)
+		return ret;
+
+	/* allocate buffers referenced by mmio list */
+	for (i = 0; data->size && i < ARRAY_SIZE(priv->mmio_data); i++) {
+		ret = nouveau_gpuobj_new(parent, NULL, data->size, data->align,
+					 0, &chan->data[i].mem);
+		if (ret)
+			return ret;
+
+		ret = nouveau_gpuobj_map_vm(chan->data[i].mem, vm, data->access,
+					   &chan->data[i].vma);
+		if (ret)
+			return ret;
+
+		data++;
+	}
+
+	/* finally, fill in the mmio list and point the context at it */
+	for (i = 0; mmio->addr && i < ARRAY_SIZE(priv->mmio_list); i++) {
+		u32 addr = mmio->addr;
+		u32 data = mmio->data;
+
+		if (mmio->shift) {
+			u64 info = chan->data[mmio->buffer].vma.offset;
+			data |= info >> mmio->shift;
+		}
+
+		nv_wo32(chan->mmio, chan->mmio_nr++ * 4, addr);
+		nv_wo32(chan->mmio, chan->mmio_nr++ * 4, data);
+		mmio++;
+	}
+
+	for (i = 0; i < priv->size; i += 4)
+		nv_wo32(chan, i, priv->data[i / 4]);
+
+	if (!priv->firmware) {
+		nv_wo32(chan, 0x00, chan->mmio_nr / 2);
+		nv_wo32(chan, 0x04, chan->mmio_vma.offset >> 8);
+	} else {
+		nv_wo32(chan, 0xf4, 0);
+		nv_wo32(chan, 0xf8, 0);
+		nv_wo32(chan, 0x10, chan->mmio_nr / 2);
+		nv_wo32(chan, 0x14, lower_32_bits(chan->mmio_vma.offset));
+		nv_wo32(chan, 0x18, upper_32_bits(chan->mmio_vma.offset));
+		nv_wo32(chan, 0x1c, 1);
+		nv_wo32(chan, 0x20, 0);
+		nv_wo32(chan, 0x28, 0);
+		nv_wo32(chan, 0x2c, 0);
+	}
+
+	return 0;
+}
+
+void
+nvc0_graph_context_dtor(struct nouveau_object *object)
+{
+	struct nvc0_graph_chan *chan = (void *)object;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(chan->data); i++) {
+		nouveau_gpuobj_unmap(&chan->data[i].vma);
+		nouveau_gpuobj_ref(NULL, &chan->data[i].mem);
+	}
+
+	nouveau_gpuobj_unmap(&chan->mmio_vma);
+	nouveau_gpuobj_ref(NULL, &chan->mmio);
+
+	nouveau_graph_context_destroy(&chan->base);
+}
+
+static struct nouveau_oclass
+nvc0_graph_cclass = {
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_graph_context_ctor,
+		.dtor = nvc0_graph_context_dtor,
+		.init = _nouveau_graph_context_init,
+		.fini = _nouveau_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static void
+nvc0_graph_ctxctl_debug_unit(struct nvc0_graph_priv *priv, u32 base)
+{
+	nv_error(priv, "%06x - done 0x%08x\n", base,
+		 nv_rd32(priv, base + 0x400));
+	nv_error(priv, "%06x - stat 0x%08x 0x%08x 0x%08x 0x%08x\n", base,
+		 nv_rd32(priv, base + 0x800), nv_rd32(priv, base + 0x804),
+		 nv_rd32(priv, base + 0x808), nv_rd32(priv, base + 0x80c));
+	nv_error(priv, "%06x - stat 0x%08x 0x%08x 0x%08x 0x%08x\n", base,
+		 nv_rd32(priv, base + 0x810), nv_rd32(priv, base + 0x814),
+		 nv_rd32(priv, base + 0x818), nv_rd32(priv, base + 0x81c));
+}
+
+void
+nvc0_graph_ctxctl_debug(struct nvc0_graph_priv *priv)
+{
+	u32 gpcnr = nv_rd32(priv, 0x409604) & 0xffff;
+	u32 gpc;
+
+	nvc0_graph_ctxctl_debug_unit(priv, 0x409000);
+	for (gpc = 0; gpc < gpcnr; gpc++)
+		nvc0_graph_ctxctl_debug_unit(priv, 0x502000 + (gpc * 0x8000));
+}
+
+static void
+nvc0_graph_ctxctl_isr(struct nvc0_graph_priv *priv)
+{
+	u32 ustat = nv_rd32(priv, 0x409c18);
+
+	if (ustat & 0x00000001)
+		nv_error(priv, "CTXCTRL ucode error\n");
+	if (ustat & 0x00080000)
+		nv_error(priv, "CTXCTRL watchdog timeout\n");
+	if (ustat & ~0x00080001)
+		nv_error(priv, "CTXCTRL 0x%08x\n", ustat);
+
+	nvc0_graph_ctxctl_debug(priv);
+	nv_wr32(priv, 0x409c20, ustat);
+}
+
+static void
+nvc0_graph_trap_tpc(struct nvc0_graph_priv *priv, int gpc, int tpc)
+{
+	u32 stat = nv_rd32(priv, TPC_UNIT(gpc, tpc, 0x0508));
+
+	if (stat & 0x00000001) {
+		u32 trap = nv_rd32(priv, TPC_UNIT(gpc, tpc, 0x0224));
+		nv_error(priv, "GPC%d/TPC%d/TEX: 0x%08x\n", gpc, tpc, trap);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0224), 0xc0000000);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0508), 0x00000001);
+		stat &= ~0x00000001;
+	}
+
+	if (stat & 0x00000002) {
+		u32 trap0 = nv_rd32(priv, TPC_UNIT(gpc, tpc, 0x0644));
+		u32 trap1 = nv_rd32(priv, TPC_UNIT(gpc, tpc, 0x064c));
+		nv_error(priv, "GPC%d/TPC%d/MP: 0x%08x 0x%08x\n",
+			       gpc, tpc, trap0, trap1);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0644), 0x001ffffe);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x064c), 0x0000000f);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0508), 0x00000002);
+		stat &= ~0x00000002;
+	}
+
+	if (stat & 0x00000004) {
+		u32 trap = nv_rd32(priv, TPC_UNIT(gpc, tpc, 0x0084));
+		nv_error(priv, "GPC%d/TPC%d/POLY: 0x%08x\n", gpc, tpc, trap);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0084), 0xc0000000);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0508), 0x00000004);
+		stat &= ~0x00000004;
+	}
+
+	if (stat & 0x00000008) {
+		u32 trap = nv_rd32(priv, TPC_UNIT(gpc, tpc, 0x048c));
+		nv_error(priv, "GPC%d/TPC%d/L1C: 0x%08x\n", gpc, tpc, trap);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x048c), 0xc0000000);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0508), 0x00000008);
+		stat &= ~0x00000008;
+	}
+
+	if (stat) {
+		nv_error(priv, "GPC%d/TPC%d/0x%08x: unknown\n", gpc, tpc, stat);
+		nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x0508), stat);
+	}
+}
+
+static void
+nvc0_graph_trap_gpc(struct nvc0_graph_priv *priv, int gpc)
+{
+	u32 stat = nv_rd32(priv, GPC_UNIT(gpc, 0x2c90));
+	int tpc;
+
+	if (stat & 0x00000001) {
+		u32 trap = nv_rd32(priv, GPC_UNIT(gpc, 0x0420));
+		nv_error(priv, "GPC%d/PROP: 0x%08x\n", gpc, trap);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0420), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x2c90), 0x00000001);
+		stat &= ~0x00000001;
+	}
+
+	if (stat & 0x00000002) {
+		u32 trap = nv_rd32(priv, GPC_UNIT(gpc, 0x0900));
+		nv_error(priv, "GPC%d/ZCULL: 0x%08x\n", gpc, trap);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0900), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x2c90), 0x00000002);
+		stat &= ~0x00000002;
+	}
+
+	if (stat & 0x00000004) {
+		u32 trap = nv_rd32(priv, GPC_UNIT(gpc, 0x1028));
+		nv_error(priv, "GPC%d/CCACHE: 0x%08x\n", gpc, trap);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x1028), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x2c90), 0x00000004);
+		stat &= ~0x00000004;
+	}
+
+	if (stat & 0x00000008) {
+		u32 trap = nv_rd32(priv, GPC_UNIT(gpc, 0x0824));
+		nv_error(priv, "GPC%d/ESETUP: 0x%08x\n", gpc, trap);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0824), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x2c90), 0x00000008);
+		stat &= ~0x00000009;
+	}
+
+	for (tpc = 0; tpc < priv->tpc_nr[gpc]; tpc++) {
+		u32 mask = 0x00010000 << tpc;
+		if (stat & mask) {
+			nvc0_graph_trap_tpc(priv, gpc, tpc);
+			nv_wr32(priv, GPC_UNIT(gpc, 0x2c90), mask);
+			stat &= ~mask;
+		}
+	}
+
+	if (stat) {
+		nv_error(priv, "GPC%d/0x%08x: unknown\n", gpc, stat);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x2c90), stat);
+	}
+}
+
+static void
+nvc0_graph_trap_intr(struct nvc0_graph_priv *priv)
+{
+	u32 trap = nv_rd32(priv, 0x400108);
+	int rop, gpc;
+
+	if (trap & 0x00000001) {
+		u32 stat = nv_rd32(priv, 0x404000);
+		nv_error(priv, "DISPATCH 0x%08x\n", stat);
+		nv_wr32(priv, 0x404000, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x00000001);
+		trap &= ~0x00000001;
+	}
+
+	if (trap & 0x00000002) {
+		u32 stat = nv_rd32(priv, 0x404600);
+		nv_error(priv, "M2MF 0x%08x\n", stat);
+		nv_wr32(priv, 0x404600, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x00000002);
+		trap &= ~0x00000002;
+	}
+
+	if (trap & 0x00000008) {
+		u32 stat = nv_rd32(priv, 0x408030);
+		nv_error(priv, "CCACHE 0x%08x\n", stat);
+		nv_wr32(priv, 0x408030, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x00000008);
+		trap &= ~0x00000008;
+	}
+
+	if (trap & 0x00000010) {
+		u32 stat = nv_rd32(priv, 0x405840);
+		nv_error(priv, "SHADER 0x%08x\n", stat);
+		nv_wr32(priv, 0x405840, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x00000010);
+		trap &= ~0x00000010;
+	}
+
+	if (trap & 0x00000040) {
+		u32 stat = nv_rd32(priv, 0x40601c);
+		nv_error(priv, "UNK6 0x%08x\n", stat);
+		nv_wr32(priv, 0x40601c, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x00000040);
+		trap &= ~0x00000040;
+	}
+
+	if (trap & 0x00000080) {
+		u32 stat = nv_rd32(priv, 0x404490);
+		nv_error(priv, "MACRO 0x%08x\n", stat);
+		nv_wr32(priv, 0x404490, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x00000080);
+		trap &= ~0x00000080;
+	}
+
+	if (trap & 0x01000000) {
+		u32 stat = nv_rd32(priv, 0x400118);
+		for (gpc = 0; stat && gpc < priv->gpc_nr; gpc++) {
+			u32 mask = 0x00000001 << gpc;
+			if (stat & mask) {
+				nvc0_graph_trap_gpc(priv, gpc);
+				nv_wr32(priv, 0x400118, mask);
+				stat &= ~mask;
+			}
+		}
+		nv_wr32(priv, 0x400108, 0x01000000);
+		trap &= ~0x01000000;
+	}
+
+	if (trap & 0x02000000) {
+		for (rop = 0; rop < priv->rop_nr; rop++) {
+			u32 statz = nv_rd32(priv, ROP_UNIT(rop, 0x070));
+			u32 statc = nv_rd32(priv, ROP_UNIT(rop, 0x144));
+			nv_error(priv, "ROP%d 0x%08x 0x%08x\n",
+				 rop, statz, statc);
+			nv_wr32(priv, ROP_UNIT(rop, 0x070), 0xc0000000);
+			nv_wr32(priv, ROP_UNIT(rop, 0x144), 0xc0000000);
+		}
+		nv_wr32(priv, 0x400108, 0x02000000);
+		trap &= ~0x02000000;
+	}
+
+	if (trap) {
+		nv_error(priv, "TRAP UNHANDLED 0x%08x\n", trap);
+		nv_wr32(priv, 0x400108, trap);
+	}
+}
+
+static void
+nvc0_graph_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_fifo *pfifo = nouveau_fifo(subdev);
+	struct nouveau_engine *engine = nv_engine(subdev);
+	struct nouveau_object *engctx;
+	struct nouveau_handle *handle;
+	struct nvc0_graph_priv *priv = (void *)subdev;
+	u64 inst = nv_rd32(priv, 0x409b00) & 0x0fffffff;
+	u32 stat = nv_rd32(priv, 0x400100);
+	u32 addr = nv_rd32(priv, 0x400704);
+	u32 mthd = (addr & 0x00003ffc);
+	u32 subc = (addr & 0x00070000) >> 16;
+	u32 data = nv_rd32(priv, 0x400708);
+	u32 code = nv_rd32(priv, 0x400110);
+	u32 class = nv_rd32(priv, 0x404200 + (subc * 4));
+	int chid;
+
+	engctx = nouveau_engctx_get(engine, inst);
+	chid   = pfifo->chid(pfifo, engctx);
+
+	if (stat & 0x00000010) {
+		handle = nouveau_handle_get_class(engctx, class);
+		if (!handle || nv_call(handle->object, mthd, data)) {
+			nv_error(priv, "ILLEGAL_MTHD ch %d [0x%010llx] "
+				     "subc %d class 0x%04x mthd 0x%04x "
+				     "data 0x%08x\n",
+				 chid, inst << 12, subc, class, mthd, data);
+		}
+		nouveau_handle_put(handle);
+		nv_wr32(priv, 0x400100, 0x00000010);
+		stat &= ~0x00000010;
+	}
+
+	if (stat & 0x00000020) {
+		nv_error(priv, "ILLEGAL_CLASS ch %d [0x%010llx] subc %d "
+			     "class 0x%04x mthd 0x%04x data 0x%08x\n",
+			chid, inst << 12, subc, class, mthd, data);
+		nv_wr32(priv, 0x400100, 0x00000020);
+		stat &= ~0x00000020;
+	}
+
+	if (stat & 0x00100000) {
+		nv_error(priv, "DATA_ERROR [");
+		nouveau_enum_print(nv50_data_error_names, code);
+		printk("] ch %d [0x%010llx] subc %d class 0x%04x "
+		       "mthd 0x%04x data 0x%08x\n",
+		       chid, inst << 12, subc, class, mthd, data);
+		nv_wr32(priv, 0x400100, 0x00100000);
+		stat &= ~0x00100000;
+	}
+
+	if (stat & 0x00200000) {
+		nv_error(priv, "TRAP ch %d [0x%010llx]\n", chid, inst << 12);
+		nvc0_graph_trap_intr(priv);
+		nv_wr32(priv, 0x400100, 0x00200000);
+		stat &= ~0x00200000;
+	}
+
+	if (stat & 0x00080000) {
+		nvc0_graph_ctxctl_isr(priv);
+		nv_wr32(priv, 0x400100, 0x00080000);
+		stat &= ~0x00080000;
+	}
+
+	if (stat) {
+		nv_error(priv, "unknown stat 0x%08x\n", stat);
+		nv_wr32(priv, 0x400100, stat);
+	}
+
+	nv_wr32(priv, 0x400500, 0x00010001);
+	nouveau_engctx_put(engctx);
+}
+
+int
+nvc0_graph_ctor_fw(struct nvc0_graph_priv *priv, const char *fwname,
+		   struct nvc0_graph_fuc *fuc)
+{
+	struct nouveau_device *device = nv_device(priv);
+	const struct firmware *fw;
+	char f[32];
+	int ret;
+
+	snprintf(f, sizeof(f), "nouveau/nv%02x_%s", device->chipset, fwname);
+	ret = request_firmware(&fw, f, &device->pdev->dev);
+	if (ret) {
+		snprintf(f, sizeof(f), "nouveau/%s", fwname);
+		ret = request_firmware(&fw, f, &device->pdev->dev);
+		if (ret) {
+			nv_error(priv, "failed to load %s\n", fwname);
+			return ret;
+		}
+	}
+
+	fuc->size = fw->size;
+	fuc->data = kmemdup(fw->data, fuc->size, GFP_KERNEL);
+	release_firmware(fw);
+	return (fuc->data != NULL) ? 0 : -ENOMEM;
+}
+
+static int
+nvc0_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nvc0_graph_priv *priv;
+	int ret, i;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x18001000;
+	nv_subdev(priv)->intr = nvc0_graph_intr;
+	nv_engine(priv)->cclass = &nvc0_graph_cclass;
+
+	if (nouveau_boolopt(device->cfgopt, "NvGrUseFW", false)) {
+		nv_info(priv, "using external firmware\n");
+		if (nvc0_graph_ctor_fw(priv, "fuc409c", &priv->fuc409c) ||
+		    nvc0_graph_ctor_fw(priv, "fuc409d", &priv->fuc409d) ||
+		    nvc0_graph_ctor_fw(priv, "fuc41ac", &priv->fuc41ac) ||
+		    nvc0_graph_ctor_fw(priv, "fuc41ad", &priv->fuc41ad))
+			return -EINVAL;
+		priv->firmware = true;
+	}
+
+	switch (nvc0_graph_class(priv)) {
+	case 0x9097:
+		nv_engine(priv)->sclass = nvc0_graph_sclass;
+		break;
+	case 0x9197:
+		nv_engine(priv)->sclass = nvc1_graph_sclass;
+		break;
+	case 0x9297:
+		nv_engine(priv)->sclass = nvc8_graph_sclass;
+		break;
+	}
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 256, 0, &priv->unk4188b4);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 256, 0, &priv->unk4188b8);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < 0x1000; i += 4) {
+		nv_wo32(priv->unk4188b4, i, 0x00000010);
+		nv_wo32(priv->unk4188b8, i, 0x00000010);
+	}
+
+	priv->rop_nr = (nv_rd32(priv, 0x409604) & 0x001f0000) >> 16;
+	priv->gpc_nr =  nv_rd32(priv, 0x409604) & 0x0000001f;
+	for (i = 0; i < priv->gpc_nr; i++) {
+		priv->tpc_nr[i]  = nv_rd32(priv, GPC_UNIT(i, 0x2608));
+		priv->tpc_total += priv->tpc_nr[i];
+	}
+
+	/*XXX: these need figuring out... though it might not even matter */
+	switch (nv_device(priv)->chipset) {
+	case 0xc0:
+		if (priv->tpc_total == 11) { /* 465, 3/4/4/0, 4 */
+			priv->magic_not_rop_nr = 0x07;
+		} else
+		if (priv->tpc_total == 14) { /* 470, 3/3/4/4, 5 */
+			priv->magic_not_rop_nr = 0x05;
+		} else
+		if (priv->tpc_total == 15) { /* 480, 3/4/4/4, 6 */
+			priv->magic_not_rop_nr = 0x06;
+		}
+		break;
+	case 0xc3: /* 450, 4/0/0/0, 2 */
+		priv->magic_not_rop_nr = 0x03;
+		break;
+	case 0xc4: /* 460, 3/4/0/0, 4 */
+		priv->magic_not_rop_nr = 0x01;
+		break;
+	case 0xc1: /* 2/0/0/0, 1 */
+		priv->magic_not_rop_nr = 0x01;
+		break;
+	case 0xc8: /* 4/4/3/4, 5 */
+		priv->magic_not_rop_nr = 0x06;
+		break;
+	case 0xce: /* 4/4/0/0, 4 */
+		priv->magic_not_rop_nr = 0x03;
+		break;
+	case 0xcf: /* 4/0/0/0, 3 */
+		priv->magic_not_rop_nr = 0x03;
+		break;
+	case 0xd9: /* 1/0/0/0, 1 */
+		priv->magic_not_rop_nr = 0x01;
+		break;
+	}
+
+	return 0;
+}
+
+static void
+nvc0_graph_dtor_fw(struct nvc0_graph_fuc *fuc)
+{
+	if (fuc->data) {
+		kfree(fuc->data);
+		fuc->data = NULL;
+	}
+}
+
+void
+nvc0_graph_dtor(struct nouveau_object *object)
+{
+	struct nvc0_graph_priv *priv = (void *)object;
+
+	if (priv->data)
+		kfree(priv->data);
+
+	nvc0_graph_dtor_fw(&priv->fuc409c);
+	nvc0_graph_dtor_fw(&priv->fuc409d);
+	nvc0_graph_dtor_fw(&priv->fuc41ac);
+	nvc0_graph_dtor_fw(&priv->fuc41ad);
+
+	nouveau_gpuobj_ref(NULL, &priv->unk4188b8);
+	nouveau_gpuobj_ref(NULL, &priv->unk4188b4);
+
+	nouveau_graph_destroy(&priv->base);
+}
+
+static void
+nvc0_graph_init_obj418880(struct nvc0_graph_priv *priv)
+{
+	int i;
+
+	nv_wr32(priv, GPC_BCAST(0x0880), 0x00000000);
+	nv_wr32(priv, GPC_BCAST(0x08a4), 0x00000000);
+	for (i = 0; i < 4; i++)
+		nv_wr32(priv, GPC_BCAST(0x0888) + (i * 4), 0x00000000);
+	nv_wr32(priv, GPC_BCAST(0x08b4), priv->unk4188b4->addr >> 8);
+	nv_wr32(priv, GPC_BCAST(0x08b8), priv->unk4188b8->addr >> 8);
+}
+
+static void
+nvc0_graph_init_regs(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x400080, 0x003083c2);
+	nv_wr32(priv, 0x400088, 0x00006fe7);
+	nv_wr32(priv, 0x40008c, 0x00000000);
+	nv_wr32(priv, 0x400090, 0x00000030);
+	nv_wr32(priv, 0x40013c, 0x013901f7);
+	nv_wr32(priv, 0x400140, 0x00000100);
+	nv_wr32(priv, 0x400144, 0x00000000);
+	nv_wr32(priv, 0x400148, 0x00000110);
+	nv_wr32(priv, 0x400138, 0x00000000);
+	nv_wr32(priv, 0x400130, 0x00000000);
+	nv_wr32(priv, 0x400134, 0x00000000);
+	nv_wr32(priv, 0x400124, 0x00000002);
+}
+
+static void
+nvc0_graph_init_gpc_0(struct nvc0_graph_priv *priv)
+{
+	const u32 magicgpc918 = DIV_ROUND_UP(0x00800000, priv->tpc_total);
+	u32 data[TPC_MAX / 8];
+	u8  tpcnr[GPC_MAX];
+	int i, gpc, tpc;
+
+	nv_wr32(priv, TPC_UNIT(0, 0, 0x5c), 1); /* affects TFB offset queries */
+
+	/*
+	 *      TP      ROP UNKVAL(magic_not_rop_nr)
+	 * 450: 4/0/0/0 2        3
+	 * 460: 3/4/0/0 4        1
+	 * 465: 3/4/4/0 4        7
+	 * 470: 3/3/4/4 5        5
+	 * 480: 3/4/4/4 6        6
+	 */
+
+	memset(data, 0x00, sizeof(data));
+	memcpy(tpcnr, priv->tpc_nr, sizeof(priv->tpc_nr));
+	for (i = 0, gpc = -1; i < priv->tpc_total; i++) {
+		do {
+			gpc = (gpc + 1) % priv->gpc_nr;
+		} while (!tpcnr[gpc]);
+		tpc = priv->tpc_nr[gpc] - tpcnr[gpc]--;
+
+		data[i / 8] |= tpc << ((i % 8) * 4);
+	}
+
+	nv_wr32(priv, GPC_BCAST(0x0980), data[0]);
+	nv_wr32(priv, GPC_BCAST(0x0984), data[1]);
+	nv_wr32(priv, GPC_BCAST(0x0988), data[2]);
+	nv_wr32(priv, GPC_BCAST(0x098c), data[3]);
+
+	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0914), priv->magic_not_rop_nr << 8 |
+						  priv->tpc_nr[gpc]);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0910), 0x00040000 | priv->tpc_total);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0918), magicgpc918);
+	}
+
+	nv_wr32(priv, GPC_BCAST(0x1bd4), magicgpc918);
+	nv_wr32(priv, GPC_BCAST(0x08ac), nv_rd32(priv, 0x100800));
+}
+
+static void
+nvc0_graph_init_units(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x409c24, 0x000f0000);
+	nv_wr32(priv, 0x404000, 0xc0000000); /* DISPATCH */
+	nv_wr32(priv, 0x404600, 0xc0000000); /* M2MF */
+	nv_wr32(priv, 0x408030, 0xc0000000);
+	nv_wr32(priv, 0x40601c, 0xc0000000);
+	nv_wr32(priv, 0x404490, 0xc0000000); /* MACRO */
+	nv_wr32(priv, 0x406018, 0xc0000000);
+	nv_wr32(priv, 0x405840, 0xc0000000);
+	nv_wr32(priv, 0x405844, 0x00ffffff);
+	nv_mask(priv, 0x419cc0, 0x00000008, 0x00000008);
+	nv_mask(priv, 0x419eb4, 0x00001000, 0x00001000);
+}
+
+static void
+nvc0_graph_init_gpc_1(struct nvc0_graph_priv *priv)
+{
+	int gpc, tpc;
+
+	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0420), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0900), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x1028), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0824), 0xc0000000);
+		for (tpc = 0; tpc < priv->tpc_nr[gpc]; tpc++) {
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x508), 0xffffffff);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x50c), 0xffffffff);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x224), 0xc0000000);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x48c), 0xc0000000);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x084), 0xc0000000);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x644), 0x001ffffe);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x64c), 0x0000000f);
+		}
+		nv_wr32(priv, GPC_UNIT(gpc, 0x2c90), 0xffffffff);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x2c94), 0xffffffff);
+	}
+}
+
+static void
+nvc0_graph_init_rop(struct nvc0_graph_priv *priv)
+{
+	int rop;
+
+	for (rop = 0; rop < priv->rop_nr; rop++) {
+		nv_wr32(priv, ROP_UNIT(rop, 0x144), 0xc0000000);
+		nv_wr32(priv, ROP_UNIT(rop, 0x070), 0xc0000000);
+		nv_wr32(priv, ROP_UNIT(rop, 0x204), 0xffffffff);
+		nv_wr32(priv, ROP_UNIT(rop, 0x208), 0xffffffff);
+	}
+}
+
+void
+nvc0_graph_init_fw(struct nvc0_graph_priv *priv, u32 fuc_base,
+		   struct nvc0_graph_fuc *code, struct nvc0_graph_fuc *data)
+{
+	int i;
+
+	nv_wr32(priv, fuc_base + 0x01c0, 0x01000000);
+	for (i = 0; i < data->size / 4; i++)
+		nv_wr32(priv, fuc_base + 0x01c4, data->data[i]);
+
+	nv_wr32(priv, fuc_base + 0x0180, 0x01000000);
+	for (i = 0; i < code->size / 4; i++) {
+		if ((i & 0x3f) == 0)
+			nv_wr32(priv, fuc_base + 0x0188, i >> 6);
+		nv_wr32(priv, fuc_base + 0x0184, code->data[i]);
+	}
+}
+
+static int
+nvc0_graph_init_ctxctl(struct nvc0_graph_priv *priv)
+{
+	u32 r000260;
+	int i;
+
+	if (priv->firmware) {
+		/* load fuc microcode */
+		r000260 = nv_mask(priv, 0x000260, 0x00000001, 0x00000000);
+		nvc0_graph_init_fw(priv, 0x409000, &priv->fuc409c,
+						   &priv->fuc409d);
+		nvc0_graph_init_fw(priv, 0x41a000, &priv->fuc41ac,
+						   &priv->fuc41ad);
+		nv_wr32(priv, 0x000260, r000260);
+
+		/* start both of them running */
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x41a10c, 0x00000000);
+		nv_wr32(priv, 0x40910c, 0x00000000);
+		nv_wr32(priv, 0x41a100, 0x00000002);
+		nv_wr32(priv, 0x409100, 0x00000002);
+		if (!nv_wait(priv, 0x409800, 0x00000001, 0x00000001))
+			nv_warn(priv, "0x409800 wait failed\n");
+
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x409500, 0x7fffffff);
+		nv_wr32(priv, 0x409504, 0x00000021);
+
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x409500, 0x00000000);
+		nv_wr32(priv, 0x409504, 0x00000010);
+		if (!nv_wait_ne(priv, 0x409800, 0xffffffff, 0x00000000)) {
+			nv_error(priv, "fuc09 req 0x10 timeout\n");
+			return -EBUSY;
+		}
+		priv->size = nv_rd32(priv, 0x409800);
+
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x409500, 0x00000000);
+		nv_wr32(priv, 0x409504, 0x00000016);
+		if (!nv_wait_ne(priv, 0x409800, 0xffffffff, 0x00000000)) {
+			nv_error(priv, "fuc09 req 0x16 timeout\n");
+			return -EBUSY;
+		}
+
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x409500, 0x00000000);
+		nv_wr32(priv, 0x409504, 0x00000025);
+		if (!nv_wait_ne(priv, 0x409800, 0xffffffff, 0x00000000)) {
+			nv_error(priv, "fuc09 req 0x25 timeout\n");
+			return -EBUSY;
+		}
+
+		if (priv->data == NULL) {
+			int ret = nvc0_grctx_generate(priv);
+			if (ret) {
+				nv_error(priv, "failed to construct context\n");
+				return ret;
+			}
+		}
+
+		return 0;
+	}
+
+	/* load HUB microcode */
+	r000260 = nv_mask(priv, 0x000260, 0x00000001, 0x00000000);
+	nv_wr32(priv, 0x4091c0, 0x01000000);
+	for (i = 0; i < sizeof(nvc0_grhub_data) / 4; i++)
+		nv_wr32(priv, 0x4091c4, nvc0_grhub_data[i]);
+
+	nv_wr32(priv, 0x409180, 0x01000000);
+	for (i = 0; i < sizeof(nvc0_grhub_code) / 4; i++) {
+		if ((i & 0x3f) == 0)
+			nv_wr32(priv, 0x409188, i >> 6);
+		nv_wr32(priv, 0x409184, nvc0_grhub_code[i]);
+	}
+
+	/* load GPC microcode */
+	nv_wr32(priv, 0x41a1c0, 0x01000000);
+	for (i = 0; i < sizeof(nvc0_grgpc_data) / 4; i++)
+		nv_wr32(priv, 0x41a1c4, nvc0_grgpc_data[i]);
+
+	nv_wr32(priv, 0x41a180, 0x01000000);
+	for (i = 0; i < sizeof(nvc0_grgpc_code) / 4; i++) {
+		if ((i & 0x3f) == 0)
+			nv_wr32(priv, 0x41a188, i >> 6);
+		nv_wr32(priv, 0x41a184, nvc0_grgpc_code[i]);
+	}
+	nv_wr32(priv, 0x000260, r000260);
+
+	/* start HUB ucode running, it'll init the GPCs */
+	nv_wr32(priv, 0x409800, nv_device(priv)->chipset);
+	nv_wr32(priv, 0x40910c, 0x00000000);
+	nv_wr32(priv, 0x409100, 0x00000002);
+	if (!nv_wait(priv, 0x409800, 0x80000000, 0x80000000)) {
+		nv_error(priv, "HUB_INIT timed out\n");
+		nvc0_graph_ctxctl_debug(priv);
+		return -EBUSY;
+	}
+
+	priv->size = nv_rd32(priv, 0x409804);
+	if (priv->data == NULL) {
+		int ret = nvc0_grctx_generate(priv);
+		if (ret) {
+			nv_error(priv, "failed to construct context\n");
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+static int
+nvc0_graph_init(struct nouveau_object *object)
+{
+	struct nvc0_graph_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_graph_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nvc0_graph_init_obj418880(priv);
+	nvc0_graph_init_regs(priv);
+	/*nvc0_graph_init_unitplemented_magics(priv);*/
+	nvc0_graph_init_gpc_0(priv);
+	/*nvc0_graph_init_unitplemented_c242(priv);*/
+
+	nv_wr32(priv, 0x400500, 0x00010001);
+	nv_wr32(priv, 0x400100, 0xffffffff);
+	nv_wr32(priv, 0x40013c, 0xffffffff);
+
+	nvc0_graph_init_units(priv);
+	nvc0_graph_init_gpc_1(priv);
+	nvc0_graph_init_rop(priv);
+
+	nv_wr32(priv, 0x400108, 0xffffffff);
+	nv_wr32(priv, 0x400138, 0xffffffff);
+	nv_wr32(priv, 0x400118, 0xffffffff);
+	nv_wr32(priv, 0x400130, 0xffffffff);
+	nv_wr32(priv, 0x40011c, 0xffffffff);
+	nv_wr32(priv, 0x400134, 0xffffffff);
+	nv_wr32(priv, 0x400054, 0x34ce3464);
+
+	ret = nvc0_graph_init_ctxctl(priv);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_graph_ctor,
+		.dtor = nvc0_graph_dtor,
+		.init = nvc0_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/nvc0.h b/nvkm/engine/graph/nvc0.h
new file mode 100644
index 000000000..a1e78de46
--- /dev/null
+++ b/nvkm/engine/graph/nvc0.h
@@ -0,0 +1,172 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#ifndef __NVC0_GRAPH_H__
+#define __NVC0_GRAPH_H__
+
+#include <core/client.h>
+#include <core/handle.h>
+#include <core/gpuobj.h>
+#include <core/option.h>
+
+#include <subdev/fb.h>
+#include <subdev/vm.h>
+#include <subdev/bar.h>
+#include <subdev/timer.h>
+
+#include <engine/fifo.h>
+#include <engine/graph.h>
+
+#define GPC_MAX 4
+#define TPC_MAX 32
+
+#define ROP_BCAST(r)      (0x408800 + (r))
+#define ROP_UNIT(u, r)    (0x410000 + (u) * 0x400 + (r))
+#define GPC_BCAST(r)      (0x418000 + (r))
+#define GPC_UNIT(t, r)    (0x500000 + (t) * 0x8000 + (r))
+#define TPC_UNIT(t, m, r) (0x504000 + (t) * 0x8000 + (m) * 0x800 + (r))
+
+struct nvc0_graph_data {
+	u32 size;
+	u32 align;
+	u32 access;
+};
+
+struct nvc0_graph_mmio {
+	u32 addr;
+	u32 data;
+	u32 shift;
+	u32 buffer;
+};
+
+struct nvc0_graph_fuc {
+	u32 *data;
+	u32  size;
+};
+
+struct nvc0_graph_priv {
+	struct nouveau_graph base;
+
+	struct nvc0_graph_fuc fuc409c;
+	struct nvc0_graph_fuc fuc409d;
+	struct nvc0_graph_fuc fuc41ac;
+	struct nvc0_graph_fuc fuc41ad;
+	bool firmware;
+
+	u8 rop_nr;
+	u8 gpc_nr;
+	u8 tpc_nr[GPC_MAX];
+	u8 tpc_total;
+
+	struct nouveau_gpuobj *unk4188b4;
+	struct nouveau_gpuobj *unk4188b8;
+
+	struct nvc0_graph_data mmio_data[4];
+	struct nvc0_graph_mmio mmio_list[4096/8];
+	u32  size;
+	u32 *data;
+
+	u8 magic_not_rop_nr;
+};
+
+struct nvc0_graph_chan {
+	struct nouveau_graph_chan base;
+
+	struct nouveau_gpuobj *mmio;
+	struct nouveau_vma mmio_vma;
+	int mmio_nr;
+	struct {
+		struct nouveau_gpuobj *mem;
+		struct nouveau_vma vma;
+	} data[4];
+};
+
+static inline u32
+nvc0_graph_class(void *obj)
+{
+	struct nouveau_device *device = nv_device(obj);
+
+	switch (device->chipset) {
+	case 0xc0:
+	case 0xc3:
+	case 0xc4:
+	case 0xce: /* guess, mmio trace shows only 0x9097 state */
+	case 0xcf: /* guess, mmio trace shows only 0x9097 state */
+		return 0x9097;
+	case 0xc1:
+		return 0x9197;
+	case 0xc8:
+	case 0xd9:
+		return 0x9297;
+	case 0xe4:
+	case 0xe7:
+	case 0xe6:
+		return 0xa097;
+	default:
+		return 0;
+	}
+}
+
+void nv_icmd(struct nvc0_graph_priv *priv, u32 icmd, u32 data);
+
+static inline void
+nv_mthd(struct nvc0_graph_priv *priv, u32 class, u32 mthd, u32 data)
+{
+	nv_wr32(priv, 0x40448c, data);
+	nv_wr32(priv, 0x404488, 0x80000000 | (mthd << 14) | class);
+}
+
+struct nvc0_grctx {
+	struct nvc0_graph_priv *priv;
+	struct nvc0_graph_data *data;
+	struct nvc0_graph_mmio *mmio;
+	struct nouveau_gpuobj *chan;
+	int buffer_nr;
+	u64 buffer[4];
+	u64 addr;
+};
+
+int  nvc0_grctx_generate(struct nvc0_graph_priv *);
+int  nvc0_grctx_init(struct nvc0_graph_priv *, struct nvc0_grctx *);
+void nvc0_grctx_data(struct nvc0_grctx *, u32, u32, u32);
+void nvc0_grctx_mmio(struct nvc0_grctx *, u32, u32, u32, u32);
+int  nvc0_grctx_fini(struct nvc0_grctx *);
+
+int  nve0_grctx_generate(struct nvc0_graph_priv *);
+
+#define mmio_data(s,a,p) nvc0_grctx_data(&info, (s), (a), (p))
+#define mmio_list(r,d,s,b) nvc0_grctx_mmio(&info, (r), (d), (s), (b))
+
+void nvc0_graph_ctxctl_debug(struct nvc0_graph_priv *);
+int  nvc0_graph_ctor_fw(struct nvc0_graph_priv *, const char *,
+			struct nvc0_graph_fuc *);
+void nvc0_graph_dtor(struct nouveau_object *);
+void nvc0_graph_init_fw(struct nvc0_graph_priv *, u32 base,
+			struct nvc0_graph_fuc *, struct nvc0_graph_fuc *);
+int  nvc0_graph_context_ctor(struct nouveau_object *, struct nouveau_object *,
+			     struct nouveau_oclass *, void *, u32,
+			     struct nouveau_object **);
+void nvc0_graph_context_dtor(struct nouveau_object *);
+
+#endif
diff --git a/nvkm/engine/graph/nve0.c b/nvkm/engine/graph/nve0.c
new file mode 100644
index 000000000..9f82e9702
--- /dev/null
+++ b/nvkm/engine/graph/nve0.c
@@ -0,0 +1,577 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "nvc0.h"
+#include "fuc/hubnve0.fuc.h"
+#include "fuc/gpcnve0.fuc.h"
+
+/*******************************************************************************
+ * Graphics object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nve0_graph_sclass[] = {
+	{ 0x902d, &nouveau_object_ofuncs },
+	{ 0xa040, &nouveau_object_ofuncs },
+	{ 0xa097, &nouveau_object_ofuncs },
+	{ 0xa0c0, &nouveau_object_ofuncs },
+	{ 0xa0b5, &nouveau_object_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * PGRAPH context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nve0_graph_cclass = {
+	.handle = NV_ENGCTX(GR, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_graph_context_ctor,
+		.dtor = nvc0_graph_context_dtor,
+		.init = _nouveau_graph_context_init,
+		.fini = _nouveau_graph_context_fini,
+		.rd32 = _nouveau_graph_context_rd32,
+		.wr32 = _nouveau_graph_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PGRAPH engine/subdev functions
+ ******************************************************************************/
+
+static void
+nve0_graph_ctxctl_isr(struct nvc0_graph_priv *priv)
+{
+	u32 ustat = nv_rd32(priv, 0x409c18);
+
+	if (ustat & 0x00000001)
+		nv_error(priv, "CTXCTRL ucode error\n");
+	if (ustat & 0x00080000)
+		nv_error(priv, "CTXCTRL watchdog timeout\n");
+	if (ustat & ~0x00080001)
+		nv_error(priv, "CTXCTRL 0x%08x\n", ustat);
+
+	nvc0_graph_ctxctl_debug(priv);
+	nv_wr32(priv, 0x409c20, ustat);
+}
+
+static void
+nve0_graph_trap_isr(struct nvc0_graph_priv *priv, int chid, u64 inst)
+{
+	u32 trap = nv_rd32(priv, 0x400108);
+	int rop;
+
+	if (trap & 0x00000001) {
+		u32 stat = nv_rd32(priv, 0x404000);
+		nv_error(priv, "DISPATCH ch %d [0x%010llx] 0x%08x\n",
+			 chid, inst, stat);
+		nv_wr32(priv, 0x404000, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x00000001);
+		trap &= ~0x00000001;
+	}
+
+	if (trap & 0x00000010) {
+		u32 stat = nv_rd32(priv, 0x405840);
+		nv_error(priv, "SHADER ch %d [0x%010llx] 0x%08x\n",
+			 chid, inst, stat);
+		nv_wr32(priv, 0x405840, 0xc0000000);
+		nv_wr32(priv, 0x400108, 0x00000010);
+		trap &= ~0x00000010;
+	}
+
+	if (trap & 0x02000000) {
+		for (rop = 0; rop < priv->rop_nr; rop++) {
+			u32 statz = nv_rd32(priv, ROP_UNIT(rop, 0x070));
+			u32 statc = nv_rd32(priv, ROP_UNIT(rop, 0x144));
+			nv_error(priv, "ROP%d ch %d [0x%010llx] 0x%08x 0x%08x\n",
+				 rop, chid, inst, statz, statc);
+			nv_wr32(priv, ROP_UNIT(rop, 0x070), 0xc0000000);
+			nv_wr32(priv, ROP_UNIT(rop, 0x144), 0xc0000000);
+		}
+		nv_wr32(priv, 0x400108, 0x02000000);
+		trap &= ~0x02000000;
+	}
+
+	if (trap) {
+		nv_error(priv, "TRAP ch %d [0x%010llx] 0x%08x\n",
+			 chid, inst, trap);
+		nv_wr32(priv, 0x400108, trap);
+	}
+}
+
+static void
+nve0_graph_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_fifo *pfifo = nouveau_fifo(subdev);
+	struct nouveau_engine *engine = nv_engine(subdev);
+	struct nouveau_object *engctx;
+	struct nouveau_handle *handle;
+	struct nvc0_graph_priv *priv = (void *)subdev;
+	u64 inst = nv_rd32(priv, 0x409b00) & 0x0fffffff;
+	u32 stat = nv_rd32(priv, 0x400100);
+	u32 addr = nv_rd32(priv, 0x400704);
+	u32 mthd = (addr & 0x00003ffc);
+	u32 subc = (addr & 0x00070000) >> 16;
+	u32 data = nv_rd32(priv, 0x400708);
+	u32 code = nv_rd32(priv, 0x400110);
+	u32 class = nv_rd32(priv, 0x404200 + (subc * 4));
+	int chid;
+
+	engctx = nouveau_engctx_get(engine, inst);
+	chid   = pfifo->chid(pfifo, engctx);
+
+	if (stat & 0x00000010) {
+		handle = nouveau_handle_get_class(engctx, class);
+		if (!handle || nv_call(handle->object, mthd, data)) {
+			nv_error(priv, "ILLEGAL_MTHD ch %d [0x%010llx] "
+				     "subc %d class 0x%04x mthd 0x%04x "
+				     "data 0x%08x\n",
+				 chid, inst, subc, class, mthd, data);
+		}
+		nouveau_handle_put(handle);
+		nv_wr32(priv, 0x400100, 0x00000010);
+		stat &= ~0x00000010;
+	}
+
+	if (stat & 0x00000020) {
+		nv_error(priv, "ILLEGAL_CLASS ch %d [0x%010llx] subc %d "
+			     "class 0x%04x mthd 0x%04x data 0x%08x\n",
+			 chid, inst, subc, class, mthd, data);
+		nv_wr32(priv, 0x400100, 0x00000020);
+		stat &= ~0x00000020;
+	}
+
+	if (stat & 0x00100000) {
+		nv_error(priv, "DATA_ERROR [");
+		nouveau_enum_print(nv50_data_error_names, code);
+		printk("] ch %d [0x%010llx] subc %d class 0x%04x "
+		       "mthd 0x%04x data 0x%08x\n",
+		       chid, inst, subc, class, mthd, data);
+		nv_wr32(priv, 0x400100, 0x00100000);
+		stat &= ~0x00100000;
+	}
+
+	if (stat & 0x00200000) {
+		nve0_graph_trap_isr(priv, chid, inst);
+		nv_wr32(priv, 0x400100, 0x00200000);
+		stat &= ~0x00200000;
+	}
+
+	if (stat & 0x00080000) {
+		nve0_graph_ctxctl_isr(priv);
+		nv_wr32(priv, 0x400100, 0x00080000);
+		stat &= ~0x00080000;
+	}
+
+	if (stat) {
+		nv_error(priv, "unknown stat 0x%08x\n", stat);
+		nv_wr32(priv, 0x400100, stat);
+	}
+
+	nv_wr32(priv, 0x400500, 0x00010001);
+	nouveau_engctx_put(engctx);
+}
+
+static int
+nve0_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nvc0_graph_priv *priv;
+	int ret, i;
+
+	ret = nouveau_graph_create(parent, engine, oclass, true, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x18001000;
+	nv_subdev(priv)->intr = nve0_graph_intr;
+	nv_engine(priv)->cclass = &nve0_graph_cclass;
+	nv_engine(priv)->sclass = nve0_graph_sclass;
+
+	if (nouveau_boolopt(device->cfgopt, "NvGrUseFW", false)) {
+		nv_info(priv, "using external firmware\n");
+		if (nvc0_graph_ctor_fw(priv, "fuc409c", &priv->fuc409c) ||
+		    nvc0_graph_ctor_fw(priv, "fuc409d", &priv->fuc409d) ||
+		    nvc0_graph_ctor_fw(priv, "fuc41ac", &priv->fuc41ac) ||
+		    nvc0_graph_ctor_fw(priv, "fuc41ad", &priv->fuc41ad))
+			return -EINVAL;
+		priv->firmware = true;
+	}
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 256, 0, &priv->unk4188b4);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 256, 0, &priv->unk4188b8);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < 0x1000; i += 4) {
+		nv_wo32(priv->unk4188b4, i, 0x00000010);
+		nv_wo32(priv->unk4188b8, i, 0x00000010);
+	}
+
+	priv->gpc_nr =  nv_rd32(priv, 0x409604) & 0x0000001f;
+	priv->rop_nr = (nv_rd32(priv, 0x409604) & 0x001f0000) >> 16;
+	for (i = 0; i < priv->gpc_nr; i++) {
+		priv->tpc_nr[i] = nv_rd32(priv, GPC_UNIT(i, 0x2608));
+		priv->tpc_total += priv->tpc_nr[i];
+	}
+
+	switch (nv_device(priv)->chipset) {
+	case 0xe4:
+		if (priv->tpc_total == 8)
+			priv->magic_not_rop_nr = 3;
+		else
+		if (priv->tpc_total == 7)
+			priv->magic_not_rop_nr = 1;
+		break;
+	case 0xe7:
+	case 0xe6:
+		priv->magic_not_rop_nr = 1;
+		break;
+	default:
+		break;
+	}
+
+	return 0;
+}
+
+static void
+nve0_graph_init_obj418880(struct nvc0_graph_priv *priv)
+{
+	int i;
+
+	nv_wr32(priv, GPC_BCAST(0x0880), 0x00000000);
+	nv_wr32(priv, GPC_BCAST(0x08a4), 0x00000000);
+	for (i = 0; i < 4; i++)
+		nv_wr32(priv, GPC_BCAST(0x0888) + (i * 4), 0x00000000);
+	nv_wr32(priv, GPC_BCAST(0x08b4), priv->unk4188b4->addr >> 8);
+	nv_wr32(priv, GPC_BCAST(0x08b8), priv->unk4188b8->addr >> 8);
+}
+
+static void
+nve0_graph_init_regs(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x400080, 0x003083c2);
+	nv_wr32(priv, 0x400088, 0x0001ffe7);
+	nv_wr32(priv, 0x40008c, 0x00000000);
+	nv_wr32(priv, 0x400090, 0x00000030);
+	nv_wr32(priv, 0x40013c, 0x003901f7);
+	nv_wr32(priv, 0x400140, 0x00000100);
+	nv_wr32(priv, 0x400144, 0x00000000);
+	nv_wr32(priv, 0x400148, 0x00000110);
+	nv_wr32(priv, 0x400138, 0x00000000);
+	nv_wr32(priv, 0x400130, 0x00000000);
+	nv_wr32(priv, 0x400134, 0x00000000);
+	nv_wr32(priv, 0x400124, 0x00000002);
+}
+
+static void
+nve0_graph_init_units(struct nvc0_graph_priv *priv)
+{
+	nv_wr32(priv, 0x409ffc, 0x00000000);
+	nv_wr32(priv, 0x409c14, 0x00003e3e);
+	nv_wr32(priv, 0x409c24, 0x000f0000);
+
+	nv_wr32(priv, 0x404000, 0xc0000000);
+	nv_wr32(priv, 0x404600, 0xc0000000);
+	nv_wr32(priv, 0x408030, 0xc0000000);
+	nv_wr32(priv, 0x404490, 0xc0000000);
+	nv_wr32(priv, 0x406018, 0xc0000000);
+	nv_wr32(priv, 0x407020, 0xc0000000);
+	nv_wr32(priv, 0x405840, 0xc0000000);
+	nv_wr32(priv, 0x405844, 0x00ffffff);
+
+	nv_mask(priv, 0x419cc0, 0x00000008, 0x00000008);
+	nv_mask(priv, 0x419eb4, 0x00001000, 0x00001000);
+
+}
+
+static void
+nve0_graph_init_gpc_0(struct nvc0_graph_priv *priv)
+{
+	const u32 magicgpc918 = DIV_ROUND_UP(0x00800000, priv->tpc_total);
+	u32 data[TPC_MAX / 8];
+	u8  tpcnr[GPC_MAX];
+	int i, gpc, tpc;
+
+	nv_wr32(priv, GPC_UNIT(0, 0x3018), 0x00000001);
+
+	memset(data, 0x00, sizeof(data));
+	memcpy(tpcnr, priv->tpc_nr, sizeof(priv->tpc_nr));
+	for (i = 0, gpc = -1; i < priv->tpc_total; i++) {
+		do {
+			gpc = (gpc + 1) % priv->gpc_nr;
+		} while (!tpcnr[gpc]);
+		tpc = priv->tpc_nr[gpc] - tpcnr[gpc]--;
+
+		data[i / 8] |= tpc << ((i % 8) * 4);
+	}
+
+	nv_wr32(priv, GPC_BCAST(0x0980), data[0]);
+	nv_wr32(priv, GPC_BCAST(0x0984), data[1]);
+	nv_wr32(priv, GPC_BCAST(0x0988), data[2]);
+	nv_wr32(priv, GPC_BCAST(0x098c), data[3]);
+
+	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0914), priv->magic_not_rop_nr << 8 |
+						  priv->tpc_nr[gpc]);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0910), 0x00040000 | priv->tpc_total);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0918), magicgpc918);
+	}
+
+	nv_wr32(priv, GPC_BCAST(0x1bd4), magicgpc918);
+	nv_wr32(priv, GPC_BCAST(0x08ac), nv_rd32(priv, 0x100800));
+}
+
+static void
+nve0_graph_init_gpc_1(struct nvc0_graph_priv *priv)
+{
+	int gpc, tpc;
+
+	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
+		nv_wr32(priv, GPC_UNIT(gpc, 0x3038), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0420), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0900), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x1028), 0xc0000000);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x0824), 0xc0000000);
+		for (tpc = 0; tpc < priv->tpc_nr[gpc]; tpc++) {
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x508), 0xffffffff);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x50c), 0xffffffff);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x224), 0xc0000000);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x48c), 0xc0000000);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x084), 0xc0000000);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x644), 0x001ffffe);
+			nv_wr32(priv, TPC_UNIT(gpc, tpc, 0x64c), 0x0000000f);
+		}
+		nv_wr32(priv, GPC_UNIT(gpc, 0x2c90), 0xffffffff);
+		nv_wr32(priv, GPC_UNIT(gpc, 0x2c94), 0xffffffff);
+	}
+}
+
+static void
+nve0_graph_init_rop(struct nvc0_graph_priv *priv)
+{
+	int rop;
+
+	for (rop = 0; rop < priv->rop_nr; rop++) {
+		nv_wr32(priv, ROP_UNIT(rop, 0x144), 0xc0000000);
+		nv_wr32(priv, ROP_UNIT(rop, 0x070), 0xc0000000);
+		nv_wr32(priv, ROP_UNIT(rop, 0x204), 0xffffffff);
+		nv_wr32(priv, ROP_UNIT(rop, 0x208), 0xffffffff);
+	}
+}
+
+static int
+nve0_graph_init_ctxctl(struct nvc0_graph_priv *priv)
+{
+	u32 r000260;
+	int i;
+
+	if (priv->firmware) {
+		/* load fuc microcode */
+		r000260 = nv_mask(priv, 0x000260, 0x00000001, 0x00000000);
+		nvc0_graph_init_fw(priv, 0x409000, &priv->fuc409c, &priv->fuc409d);
+		nvc0_graph_init_fw(priv, 0x41a000, &priv->fuc41ac, &priv->fuc41ad);
+		nv_wr32(priv, 0x000260, r000260);
+
+		/* start both of them running */
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x41a10c, 0x00000000);
+		nv_wr32(priv, 0x40910c, 0x00000000);
+		nv_wr32(priv, 0x41a100, 0x00000002);
+		nv_wr32(priv, 0x409100, 0x00000002);
+		if (!nv_wait(priv, 0x409800, 0x00000001, 0x00000001))
+			nv_error(priv, "0x409800 wait failed\n");
+
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x409500, 0x7fffffff);
+		nv_wr32(priv, 0x409504, 0x00000021);
+
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x409500, 0x00000000);
+		nv_wr32(priv, 0x409504, 0x00000010);
+		if (!nv_wait_ne(priv, 0x409800, 0xffffffff, 0x00000000)) {
+			nv_error(priv, "fuc09 req 0x10 timeout\n");
+			return -EBUSY;
+		}
+		priv->size = nv_rd32(priv, 0x409800);
+
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x409500, 0x00000000);
+		nv_wr32(priv, 0x409504, 0x00000016);
+		if (!nv_wait_ne(priv, 0x409800, 0xffffffff, 0x00000000)) {
+			nv_error(priv, "fuc09 req 0x16 timeout\n");
+			return -EBUSY;
+		}
+
+		nv_wr32(priv, 0x409840, 0xffffffff);
+		nv_wr32(priv, 0x409500, 0x00000000);
+		nv_wr32(priv, 0x409504, 0x00000025);
+		if (!nv_wait_ne(priv, 0x409800, 0xffffffff, 0x00000000)) {
+			nv_error(priv, "fuc09 req 0x25 timeout\n");
+			return -EBUSY;
+		}
+
+		nv_wr32(priv, 0x409800, 0x00000000);
+		nv_wr32(priv, 0x409500, 0x00000001);
+		nv_wr32(priv, 0x409504, 0x00000030);
+		if (!nv_wait_ne(priv, 0x409800, 0xffffffff, 0x00000000)) {
+			nv_error(priv, "fuc09 req 0x30 timeout\n");
+			return -EBUSY;
+		}
+
+		nv_wr32(priv, 0x409810, 0xb00095c8);
+		nv_wr32(priv, 0x409800, 0x00000000);
+		nv_wr32(priv, 0x409500, 0x00000001);
+		nv_wr32(priv, 0x409504, 0x00000031);
+		if (!nv_wait_ne(priv, 0x409800, 0xffffffff, 0x00000000)) {
+			nv_error(priv, "fuc09 req 0x31 timeout\n");
+			return -EBUSY;
+		}
+
+		nv_wr32(priv, 0x409810, 0x00080420);
+		nv_wr32(priv, 0x409800, 0x00000000);
+		nv_wr32(priv, 0x409500, 0x00000001);
+		nv_wr32(priv, 0x409504, 0x00000032);
+		if (!nv_wait_ne(priv, 0x409800, 0xffffffff, 0x00000000)) {
+			nv_error(priv, "fuc09 req 0x32 timeout\n");
+			return -EBUSY;
+		}
+
+		nv_wr32(priv, 0x409614, 0x00000070);
+		nv_wr32(priv, 0x409614, 0x00000770);
+		nv_wr32(priv, 0x40802c, 0x00000001);
+
+		if (priv->data == NULL) {
+			int ret = nve0_grctx_generate(priv);
+			if (ret) {
+				nv_error(priv, "failed to construct context\n");
+				return ret;
+			}
+		}
+
+		return 0;
+	}
+
+	/* load HUB microcode */
+	r000260 = nv_mask(priv, 0x000260, 0x00000001, 0x00000000);
+	nv_wr32(priv, 0x4091c0, 0x01000000);
+	for (i = 0; i < sizeof(nve0_grhub_data) / 4; i++)
+		nv_wr32(priv, 0x4091c4, nve0_grhub_data[i]);
+
+	nv_wr32(priv, 0x409180, 0x01000000);
+	for (i = 0; i < sizeof(nve0_grhub_code) / 4; i++) {
+		if ((i & 0x3f) == 0)
+			nv_wr32(priv, 0x409188, i >> 6);
+		nv_wr32(priv, 0x409184, nve0_grhub_code[i]);
+	}
+
+	/* load GPC microcode */
+	nv_wr32(priv, 0x41a1c0, 0x01000000);
+	for (i = 0; i < sizeof(nve0_grgpc_data) / 4; i++)
+		nv_wr32(priv, 0x41a1c4, nve0_grgpc_data[i]);
+
+	nv_wr32(priv, 0x41a180, 0x01000000);
+	for (i = 0; i < sizeof(nve0_grgpc_code) / 4; i++) {
+		if ((i & 0x3f) == 0)
+			nv_wr32(priv, 0x41a188, i >> 6);
+		nv_wr32(priv, 0x41a184, nve0_grgpc_code[i]);
+	}
+	nv_wr32(priv, 0x000260, r000260);
+
+	/* start HUB ucode running, it'll init the GPCs */
+	nv_wr32(priv, 0x409800, nv_device(priv)->chipset);
+	nv_wr32(priv, 0x40910c, 0x00000000);
+	nv_wr32(priv, 0x409100, 0x00000002);
+	if (!nv_wait(priv, 0x409800, 0x80000000, 0x80000000)) {
+		nv_error(priv, "HUB_INIT timed out\n");
+		nvc0_graph_ctxctl_debug(priv);
+		return -EBUSY;
+	}
+
+	priv->size = nv_rd32(priv, 0x409804);
+	if (priv->data == NULL) {
+		int ret = nve0_grctx_generate(priv);
+		if (ret) {
+			nv_error(priv, "failed to construct context\n");
+			return ret;
+		}
+	}
+
+	return 0;
+}
+
+static int
+nve0_graph_init(struct nouveau_object *object)
+{
+	struct nvc0_graph_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_graph_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nve0_graph_init_obj418880(priv);
+	nve0_graph_init_regs(priv);
+	nve0_graph_init_gpc_0(priv);
+
+	nv_wr32(priv, 0x400500, 0x00010001);
+	nv_wr32(priv, 0x400100, 0xffffffff);
+	nv_wr32(priv, 0x40013c, 0xffffffff);
+
+	nve0_graph_init_units(priv);
+	nve0_graph_init_gpc_1(priv);
+	nve0_graph_init_rop(priv);
+
+	nv_wr32(priv, 0x400108, 0xffffffff);
+	nv_wr32(priv, 0x400138, 0xffffffff);
+	nv_wr32(priv, 0x400118, 0xffffffff);
+	nv_wr32(priv, 0x400130, 0xffffffff);
+	nv_wr32(priv, 0x40011c, 0xffffffff);
+	nv_wr32(priv, 0x400134, 0xffffffff);
+	nv_wr32(priv, 0x400054, 0x34ce3464);
+
+	ret = nve0_graph_init_ctxctl(priv);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+struct nouveau_oclass
+nve0_graph_oclass = {
+	.handle = NV_ENGINE(GR, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nve0_graph_ctor,
+		.dtor = nvc0_graph_dtor,
+		.init = nve0_graph_init,
+		.fini = _nouveau_graph_fini,
+	},
+};
diff --git a/nvkm/engine/graph/regs.h b/nvkm/engine/graph/regs.h
new file mode 100644
index 000000000..fde8e2441
--- /dev/null
+++ b/nvkm/engine/graph/regs.h
@@ -0,0 +1,274 @@
+#ifndef __NOUVEAU_GRAPH_REGS_H__
+#define __NOUVEAU_GRAPH_REGS_H__
+
+#define NV04_PGRAPH_DEBUG_0                                0x00400080
+#define NV04_PGRAPH_DEBUG_1                                0x00400084
+#define NV04_PGRAPH_DEBUG_2                                0x00400088
+#define NV04_PGRAPH_DEBUG_3                                0x0040008c
+#define NV10_PGRAPH_DEBUG_4                                0x00400090
+#define NV03_PGRAPH_INTR                                   0x00400100
+#define NV03_PGRAPH_NSTATUS                                0x00400104
+#    define NV04_PGRAPH_NSTATUS_STATE_IN_USE                  (1<<11)
+#    define NV04_PGRAPH_NSTATUS_INVALID_STATE                 (1<<12)
+#    define NV04_PGRAPH_NSTATUS_BAD_ARGUMENT                  (1<<13)
+#    define NV04_PGRAPH_NSTATUS_PROTECTION_FAULT              (1<<14)
+#    define NV10_PGRAPH_NSTATUS_STATE_IN_USE                  (1<<23)
+#    define NV10_PGRAPH_NSTATUS_INVALID_STATE                 (1<<24)
+#    define NV10_PGRAPH_NSTATUS_BAD_ARGUMENT                  (1<<25)
+#    define NV10_PGRAPH_NSTATUS_PROTECTION_FAULT              (1<<26)
+#define NV03_PGRAPH_NSOURCE                                0x00400108
+#    define NV03_PGRAPH_NSOURCE_NOTIFICATION                   (1<<0)
+#    define NV03_PGRAPH_NSOURCE_DATA_ERROR                     (1<<1)
+#    define NV03_PGRAPH_NSOURCE_PROTECTION_ERROR               (1<<2)
+#    define NV03_PGRAPH_NSOURCE_RANGE_EXCEPTION                (1<<3)
+#    define NV03_PGRAPH_NSOURCE_LIMIT_COLOR                    (1<<4)
+#    define NV03_PGRAPH_NSOURCE_LIMIT_ZETA                     (1<<5)
+#    define NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD                   (1<<6)
+#    define NV03_PGRAPH_NSOURCE_DMA_R_PROTECTION               (1<<7)
+#    define NV03_PGRAPH_NSOURCE_DMA_W_PROTECTION               (1<<8)
+#    define NV03_PGRAPH_NSOURCE_FORMAT_EXCEPTION               (1<<9)
+#    define NV03_PGRAPH_NSOURCE_PATCH_EXCEPTION               (1<<10)
+#    define NV03_PGRAPH_NSOURCE_STATE_INVALID                 (1<<11)
+#    define NV03_PGRAPH_NSOURCE_DOUBLE_NOTIFY                 (1<<12)
+#    define NV03_PGRAPH_NSOURCE_NOTIFY_IN_USE                 (1<<13)
+#    define NV03_PGRAPH_NSOURCE_METHOD_CNT                    (1<<14)
+#    define NV03_PGRAPH_NSOURCE_BFR_NOTIFICATION              (1<<15)
+#    define NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION            (1<<16)
+#    define NV03_PGRAPH_NSOURCE_DMA_WIDTH_A                   (1<<17)
+#    define NV03_PGRAPH_NSOURCE_DMA_WIDTH_B                   (1<<18)
+#define NV03_PGRAPH_INTR_EN                                0x00400140
+#define NV40_PGRAPH_INTR_EN                                0x0040013C
+#    define NV_PGRAPH_INTR_NOTIFY                              (1<<0)
+#    define NV_PGRAPH_INTR_MISSING_HW                          (1<<4)
+#    define NV_PGRAPH_INTR_CONTEXT_SWITCH                     (1<<12)
+#    define NV_PGRAPH_INTR_BUFFER_NOTIFY                      (1<<16)
+#    define NV_PGRAPH_INTR_ERROR                              (1<<20)
+#define NV10_PGRAPH_CTX_CONTROL                            0x00400144
+#define NV10_PGRAPH_CTX_USER                               0x00400148
+#define NV10_PGRAPH_CTX_SWITCH(i)                         (0x0040014C + 0x4*(i))
+#define NV04_PGRAPH_CTX_SWITCH1                            0x00400160
+#define NV10_PGRAPH_CTX_CACHE(i, j)                       (0x00400160	\
+							   + 0x4*(i) + 0x20*(j))
+#define NV04_PGRAPH_CTX_SWITCH2                            0x00400164
+#define NV04_PGRAPH_CTX_SWITCH3                            0x00400168
+#define NV04_PGRAPH_CTX_SWITCH4                            0x0040016C
+#define NV04_PGRAPH_CTX_CONTROL                            0x00400170
+#define NV04_PGRAPH_CTX_USER                               0x00400174
+#define NV04_PGRAPH_CTX_CACHE1                             0x00400180
+#define NV03_PGRAPH_CTX_CONTROL                            0x00400190
+#define NV03_PGRAPH_CTX_USER                               0x00400194
+#define NV04_PGRAPH_CTX_CACHE2                             0x004001A0
+#define NV04_PGRAPH_CTX_CACHE3                             0x004001C0
+#define NV04_PGRAPH_CTX_CACHE4                             0x004001E0
+#define NV40_PGRAPH_CTXCTL_0304                            0x00400304
+#define NV40_PGRAPH_CTXCTL_0304_XFER_CTX                   0x00000001
+#define NV40_PGRAPH_CTXCTL_UCODE_STAT                      0x00400308
+#define NV40_PGRAPH_CTXCTL_UCODE_STAT_IP_MASK              0xff000000
+#define NV40_PGRAPH_CTXCTL_UCODE_STAT_IP_SHIFT                     24
+#define NV40_PGRAPH_CTXCTL_UCODE_STAT_OP_MASK              0x00ffffff
+#define NV40_PGRAPH_CTXCTL_0310                            0x00400310
+#define NV40_PGRAPH_CTXCTL_0310_XFER_SAVE                  0x00000020
+#define NV40_PGRAPH_CTXCTL_0310_XFER_LOAD                  0x00000040
+#define NV40_PGRAPH_CTXCTL_030C                            0x0040030c
+#define NV40_PGRAPH_CTXCTL_UCODE_INDEX                     0x00400324
+#define NV40_PGRAPH_CTXCTL_UCODE_DATA                      0x00400328
+#define NV40_PGRAPH_CTXCTL_CUR                             0x0040032c
+#define NV40_PGRAPH_CTXCTL_CUR_LOADED                      0x01000000
+#define NV40_PGRAPH_CTXCTL_CUR_INSTANCE                    0x000FFFFF
+#define NV40_PGRAPH_CTXCTL_NEXT                            0x00400330
+#define NV40_PGRAPH_CTXCTL_NEXT_INSTANCE                   0x000fffff
+#define NV50_PGRAPH_CTXCTL_CUR                             0x0040032c
+#define NV50_PGRAPH_CTXCTL_CUR_LOADED                      0x80000000
+#define NV50_PGRAPH_CTXCTL_CUR_INSTANCE                    0x00ffffff
+#define NV50_PGRAPH_CTXCTL_NEXT                            0x00400330
+#define NV50_PGRAPH_CTXCTL_NEXT_INSTANCE                   0x00ffffff
+#define NV03_PGRAPH_ABS_X_RAM                              0x00400400
+#define NV03_PGRAPH_ABS_Y_RAM                              0x00400480
+#define NV03_PGRAPH_X_MISC                                 0x00400500
+#define NV03_PGRAPH_Y_MISC                                 0x00400504
+#define NV04_PGRAPH_VALID1                                 0x00400508
+#define NV04_PGRAPH_SOURCE_COLOR                           0x0040050C
+#define NV04_PGRAPH_MISC24_0                               0x00400510
+#define NV03_PGRAPH_XY_LOGIC_MISC0                         0x00400514
+#define NV03_PGRAPH_XY_LOGIC_MISC1                         0x00400518
+#define NV03_PGRAPH_XY_LOGIC_MISC2                         0x0040051C
+#define NV03_PGRAPH_XY_LOGIC_MISC3                         0x00400520
+#define NV03_PGRAPH_CLIPX_0                                0x00400524
+#define NV03_PGRAPH_CLIPX_1                                0x00400528
+#define NV03_PGRAPH_CLIPY_0                                0x0040052C
+#define NV03_PGRAPH_CLIPY_1                                0x00400530
+#define NV03_PGRAPH_ABS_ICLIP_XMAX                         0x00400534
+#define NV03_PGRAPH_ABS_ICLIP_YMAX                         0x00400538
+#define NV03_PGRAPH_ABS_UCLIP_XMIN                         0x0040053C
+#define NV03_PGRAPH_ABS_UCLIP_YMIN                         0x00400540
+#define NV03_PGRAPH_ABS_UCLIP_XMAX                         0x00400544
+#define NV03_PGRAPH_ABS_UCLIP_YMAX                         0x00400548
+#define NV03_PGRAPH_ABS_UCLIPA_XMIN                        0x00400560
+#define NV03_PGRAPH_ABS_UCLIPA_YMIN                        0x00400564
+#define NV03_PGRAPH_ABS_UCLIPA_XMAX                        0x00400568
+#define NV03_PGRAPH_ABS_UCLIPA_YMAX                        0x0040056C
+#define NV04_PGRAPH_MISC24_1                               0x00400570
+#define NV04_PGRAPH_MISC24_2                               0x00400574
+#define NV04_PGRAPH_VALID2                                 0x00400578
+#define NV04_PGRAPH_PASSTHRU_0                             0x0040057C
+#define NV04_PGRAPH_PASSTHRU_1                             0x00400580
+#define NV04_PGRAPH_PASSTHRU_2                             0x00400584
+#define NV10_PGRAPH_DIMX_TEXTURE                           0x00400588
+#define NV10_PGRAPH_WDIMX_TEXTURE                          0x0040058C
+#define NV04_PGRAPH_COMBINE_0_ALPHA                        0x00400590
+#define NV04_PGRAPH_COMBINE_0_COLOR                        0x00400594
+#define NV04_PGRAPH_COMBINE_1_ALPHA                        0x00400598
+#define NV04_PGRAPH_COMBINE_1_COLOR                        0x0040059C
+#define NV04_PGRAPH_FORMAT_0                               0x004005A8
+#define NV04_PGRAPH_FORMAT_1                               0x004005AC
+#define NV04_PGRAPH_FILTER_0                               0x004005B0
+#define NV04_PGRAPH_FILTER_1                               0x004005B4
+#define NV03_PGRAPH_MONO_COLOR0                            0x00400600
+#define NV04_PGRAPH_ROP3                                   0x00400604
+#define NV04_PGRAPH_BETA_AND                               0x00400608
+#define NV04_PGRAPH_BETA_PREMULT                           0x0040060C
+#define NV04_PGRAPH_LIMIT_VIOL_PIX                         0x00400610
+#define NV04_PGRAPH_FORMATS                                0x00400618
+#define NV10_PGRAPH_DEBUG_2                                0x00400620
+#define NV04_PGRAPH_BOFFSET0                               0x00400640
+#define NV04_PGRAPH_BOFFSET1                               0x00400644
+#define NV04_PGRAPH_BOFFSET2                               0x00400648
+#define NV04_PGRAPH_BOFFSET3                               0x0040064C
+#define NV04_PGRAPH_BOFFSET4                               0x00400650
+#define NV04_PGRAPH_BOFFSET5                               0x00400654
+#define NV04_PGRAPH_BBASE0                                 0x00400658
+#define NV04_PGRAPH_BBASE1                                 0x0040065C
+#define NV04_PGRAPH_BBASE2                                 0x00400660
+#define NV04_PGRAPH_BBASE3                                 0x00400664
+#define NV04_PGRAPH_BBASE4                                 0x00400668
+#define NV04_PGRAPH_BBASE5                                 0x0040066C
+#define NV04_PGRAPH_BPITCH0                                0x00400670
+#define NV04_PGRAPH_BPITCH1                                0x00400674
+#define NV04_PGRAPH_BPITCH2                                0x00400678
+#define NV04_PGRAPH_BPITCH3                                0x0040067C
+#define NV04_PGRAPH_BPITCH4                                0x00400680
+#define NV04_PGRAPH_BLIMIT0                                0x00400684
+#define NV04_PGRAPH_BLIMIT1                                0x00400688
+#define NV04_PGRAPH_BLIMIT2                                0x0040068C
+#define NV04_PGRAPH_BLIMIT3                                0x00400690
+#define NV04_PGRAPH_BLIMIT4                                0x00400694
+#define NV04_PGRAPH_BLIMIT5                                0x00400698
+#define NV04_PGRAPH_BSWIZZLE2                              0x0040069C
+#define NV04_PGRAPH_BSWIZZLE5                              0x004006A0
+#define NV03_PGRAPH_STATUS                                 0x004006B0
+#define NV04_PGRAPH_STATUS                                 0x00400700
+#    define NV40_PGRAPH_STATUS_SYNC_STALL                  0x00004000
+#define NV04_PGRAPH_TRAPPED_ADDR                           0x00400704
+#define NV04_PGRAPH_TRAPPED_DATA                           0x00400708
+#define NV04_PGRAPH_SURFACE                                0x0040070C
+#define NV10_PGRAPH_TRAPPED_DATA_HIGH                      0x0040070C
+#define NV04_PGRAPH_STATE                                  0x00400710
+#define NV10_PGRAPH_SURFACE                                0x00400710
+#define NV04_PGRAPH_NOTIFY                                 0x00400714
+#define NV10_PGRAPH_STATE                                  0x00400714
+#define NV10_PGRAPH_NOTIFY                                 0x00400718
+
+#define NV04_PGRAPH_FIFO                                   0x00400720
+
+#define NV04_PGRAPH_BPIXEL                                 0x00400724
+#define NV10_PGRAPH_RDI_INDEX                              0x00400750
+#define NV04_PGRAPH_FFINTFC_ST2                            0x00400754
+#define NV10_PGRAPH_RDI_DATA                               0x00400754
+#define NV04_PGRAPH_DMA_PITCH                              0x00400760
+#define NV10_PGRAPH_FFINTFC_FIFO_PTR                       0x00400760
+#define NV04_PGRAPH_DVD_COLORFMT                           0x00400764
+#define NV10_PGRAPH_FFINTFC_ST2                            0x00400764
+#define NV04_PGRAPH_SCALED_FORMAT                          0x00400768
+#define NV10_PGRAPH_FFINTFC_ST2_DL                         0x00400768
+#define NV10_PGRAPH_FFINTFC_ST2_DH                         0x0040076c
+#define NV10_PGRAPH_DMA_PITCH                              0x00400770
+#define NV10_PGRAPH_DVD_COLORFMT                           0x00400774
+#define NV10_PGRAPH_SCALED_FORMAT                          0x00400778
+#define NV20_PGRAPH_CHANNEL_CTX_TABLE                      0x00400780
+#define NV20_PGRAPH_CHANNEL_CTX_POINTER                    0x00400784
+#define NV20_PGRAPH_CHANNEL_CTX_XFER                       0x00400788
+#define NV20_PGRAPH_CHANNEL_CTX_XFER_LOAD                  0x00000001
+#define NV20_PGRAPH_CHANNEL_CTX_XFER_SAVE                  0x00000002
+#define NV04_PGRAPH_PATT_COLOR0                            0x00400800
+#define NV04_PGRAPH_PATT_COLOR1                            0x00400804
+#define NV04_PGRAPH_PATTERN                                0x00400808
+#define NV04_PGRAPH_PATTERN_SHAPE                          0x00400810
+#define NV04_PGRAPH_CHROMA                                 0x00400814
+#define NV04_PGRAPH_CONTROL0                               0x00400818
+#define NV04_PGRAPH_CONTROL1                               0x0040081C
+#define NV04_PGRAPH_CONTROL2                               0x00400820
+#define NV04_PGRAPH_BLEND                                  0x00400824
+#define NV04_PGRAPH_STORED_FMT                             0x00400830
+#define NV04_PGRAPH_PATT_COLORRAM                          0x00400900
+#define NV20_PGRAPH_TILE(i)                                (0x00400900 + (i*16))
+#define NV20_PGRAPH_TLIMIT(i)                              (0x00400904 + (i*16))
+#define NV20_PGRAPH_TSIZE(i)                               (0x00400908 + (i*16))
+#define NV20_PGRAPH_TSTATUS(i)                             (0x0040090C + (i*16))
+#define NV20_PGRAPH_ZCOMP(i)                               (0x00400980 + 4*(i))
+#define NV41_PGRAPH_ZCOMP0(i)                              (0x004009c0 + 4*(i))
+#define NV10_PGRAPH_TILE(i)                                (0x00400B00 + (i*16))
+#define NV10_PGRAPH_TLIMIT(i)                              (0x00400B04 + (i*16))
+#define NV10_PGRAPH_TSIZE(i)                               (0x00400B08 + (i*16))
+#define NV10_PGRAPH_TSTATUS(i)                             (0x00400B0C + (i*16))
+#define NV04_PGRAPH_U_RAM                                  0x00400D00
+#define NV47_PGRAPH_TILE(i)                                (0x00400D00 + (i*16))
+#define NV47_PGRAPH_TLIMIT(i)                              (0x00400D04 + (i*16))
+#define NV47_PGRAPH_TSIZE(i)                               (0x00400D08 + (i*16))
+#define NV47_PGRAPH_TSTATUS(i)                             (0x00400D0C + (i*16))
+#define NV04_PGRAPH_V_RAM                                  0x00400D40
+#define NV04_PGRAPH_W_RAM                                  0x00400D80
+#define NV47_PGRAPH_ZCOMP0(i)                              (0x00400e00 + 4*(i))
+#define NV10_PGRAPH_COMBINER0_IN_ALPHA                     0x00400E40
+#define NV10_PGRAPH_COMBINER1_IN_ALPHA                     0x00400E44
+#define NV10_PGRAPH_COMBINER0_IN_RGB                       0x00400E48
+#define NV10_PGRAPH_COMBINER1_IN_RGB                       0x00400E4C
+#define NV10_PGRAPH_COMBINER_COLOR0                        0x00400E50
+#define NV10_PGRAPH_COMBINER_COLOR1                        0x00400E54
+#define NV10_PGRAPH_COMBINER0_OUT_ALPHA                    0x00400E58
+#define NV10_PGRAPH_COMBINER1_OUT_ALPHA                    0x00400E5C
+#define NV10_PGRAPH_COMBINER0_OUT_RGB                      0x00400E60
+#define NV10_PGRAPH_COMBINER1_OUT_RGB                      0x00400E64
+#define NV10_PGRAPH_COMBINER_FINAL0                        0x00400E68
+#define NV10_PGRAPH_COMBINER_FINAL1                        0x00400E6C
+#define NV10_PGRAPH_WINDOWCLIP_HORIZONTAL                  0x00400F00
+#define NV10_PGRAPH_WINDOWCLIP_VERTICAL                    0x00400F20
+#define NV10_PGRAPH_XFMODE0                                0x00400F40
+#define NV10_PGRAPH_XFMODE1                                0x00400F44
+#define NV10_PGRAPH_GLOBALSTATE0                           0x00400F48
+#define NV10_PGRAPH_GLOBALSTATE1                           0x00400F4C
+#define NV10_PGRAPH_PIPE_ADDRESS                           0x00400F50
+#define NV10_PGRAPH_PIPE_DATA                              0x00400F54
+#define NV04_PGRAPH_DMA_START_0                            0x00401000
+#define NV04_PGRAPH_DMA_START_1                            0x00401004
+#define NV04_PGRAPH_DMA_LENGTH                             0x00401008
+#define NV04_PGRAPH_DMA_MISC                               0x0040100C
+#define NV04_PGRAPH_DMA_DATA_0                             0x00401020
+#define NV04_PGRAPH_DMA_DATA_1                             0x00401024
+#define NV04_PGRAPH_DMA_RM                                 0x00401030
+#define NV04_PGRAPH_DMA_A_XLATE_INST                       0x00401040
+#define NV04_PGRAPH_DMA_A_CONTROL                          0x00401044
+#define NV04_PGRAPH_DMA_A_LIMIT                            0x00401048
+#define NV04_PGRAPH_DMA_A_TLB_PTE                          0x0040104C
+#define NV04_PGRAPH_DMA_A_TLB_TAG                          0x00401050
+#define NV04_PGRAPH_DMA_A_ADJ_OFFSET                       0x00401054
+#define NV04_PGRAPH_DMA_A_OFFSET                           0x00401058
+#define NV04_PGRAPH_DMA_A_SIZE                             0x0040105C
+#define NV04_PGRAPH_DMA_A_Y_SIZE                           0x00401060
+#define NV04_PGRAPH_DMA_B_XLATE_INST                       0x00401080
+#define NV04_PGRAPH_DMA_B_CONTROL                          0x00401084
+#define NV04_PGRAPH_DMA_B_LIMIT                            0x00401088
+#define NV04_PGRAPH_DMA_B_TLB_PTE                          0x0040108C
+#define NV04_PGRAPH_DMA_B_TLB_TAG                          0x00401090
+#define NV04_PGRAPH_DMA_B_ADJ_OFFSET                       0x00401094
+#define NV04_PGRAPH_DMA_B_OFFSET                           0x00401098
+#define NV04_PGRAPH_DMA_B_SIZE                             0x0040109C
+#define NV04_PGRAPH_DMA_B_Y_SIZE                           0x004010A0
+#define NV47_PGRAPH_ZCOMP1(i)                              (0x004068c0 + 4*(i))
+#define NV40_PGRAPH_TILE1(i)                               (0x00406900 + (i*16))
+#define NV40_PGRAPH_TLIMIT1(i)                             (0x00406904 + (i*16))
+#define NV40_PGRAPH_TSIZE1(i)                              (0x00406908 + (i*16))
+#define NV40_PGRAPH_TSTATUS1(i)                            (0x0040690C + (i*16))
+#define NV40_PGRAPH_ZCOMP1(i)                              (0x00406980 + 4*(i))
+#define NV41_PGRAPH_ZCOMP1(i)                              (0x004069c0 + 4*(i))
+
+#endif
diff --git a/nvkm/engine/mpeg/Makefile.am b/nvkm/engine/mpeg/Makefile.am
new file mode 100644
index 000000000..1b0d90096
--- /dev/null
+++ b/nvkm/engine/mpeg/Makefile.am
@@ -0,0 +1,8 @@
+noinst_LTLIBRARIES = libmpeg.la
+
+libmpeg_la_SOURCES = nv31.c \
+		     nv40.c \
+		     nv50.c \
+		     nv84.c
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/mpeg/nv31.c b/nvkm/engine/mpeg/nv31.c
new file mode 100644
index 000000000..9fd86375f
--- /dev/null
+++ b/nvkm/engine/mpeg/nv31.c
@@ -0,0 +1,308 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/handle.h>
+
+#include <subdev/fb.h>
+#include <subdev/timer.h>
+#include <subdev/instmem.h>
+
+#include <engine/fifo.h>
+#include <engine/mpeg.h>
+#include <engine/graph/nv40.h>
+
+struct nv31_mpeg_priv {
+	struct nouveau_mpeg base;
+	atomic_t refcount;
+};
+
+struct nv31_mpeg_chan {
+	struct nouveau_object base;
+};
+
+/*******************************************************************************
+ * MPEG object classes
+ ******************************************************************************/
+
+static int
+nv31_mpeg_object_ctor(struct nouveau_object *parent,
+		      struct nouveau_object *engine,
+		      struct nouveau_oclass *oclass, void *data, u32 size,
+		      struct nouveau_object **pobject)
+{
+	struct nouveau_gpuobj *obj;
+	int ret;
+
+	ret = nouveau_gpuobj_create(parent, engine, oclass, 0, parent,
+				    20, 16, 0, &obj);
+	*pobject = nv_object(obj);
+	if (ret)
+		return ret;
+
+	nv_wo32(obj, 0x00, nv_mclass(obj));
+	nv_wo32(obj, 0x04, 0x00000000);
+	nv_wo32(obj, 0x08, 0x00000000);
+	nv_wo32(obj, 0x0c, 0x00000000);
+	return 0;
+}
+
+static int
+nv31_mpeg_mthd_dma(struct nouveau_object *object, u32 mthd, void *arg, u32 len)
+{
+	struct nouveau_instmem *imem = nouveau_instmem(object);
+	struct nv31_mpeg_priv *priv = (void *)object->engine;
+	u32 inst = *(u32 *)arg << 4;
+	u32 dma0 = nv_ro32(imem, inst + 0);
+	u32 dma1 = nv_ro32(imem, inst + 4);
+	u32 dma2 = nv_ro32(imem, inst + 8);
+	u32 base = (dma2 & 0xfffff000) | (dma0 >> 20);
+	u32 size = dma1 + 1;
+
+	/* only allow linear DMA objects */
+	if (!(dma0 & 0x00002000))
+		return -EINVAL;
+
+	if (mthd == 0x0190) {
+		/* DMA_CMD */
+		nv_mask(priv, 0x00b300, 0x00030000, (dma0 & 0x00030000));
+		nv_wr32(priv, 0x00b334, base);
+		nv_wr32(priv, 0x00b324, size);
+	} else
+	if (mthd == 0x01a0) {
+		/* DMA_DATA */
+		nv_mask(priv, 0x00b300, 0x000c0000, (dma0 & 0x00030000) << 2);
+		nv_wr32(priv, 0x00b360, base);
+		nv_wr32(priv, 0x00b364, size);
+	} else {
+		/* DMA_IMAGE, VRAM only */
+		if (dma0 & 0x000c0000)
+			return -EINVAL;
+
+		nv_wr32(priv, 0x00b370, base);
+		nv_wr32(priv, 0x00b374, size);
+	}
+
+	return 0;
+}
+
+static struct nouveau_ofuncs
+nv31_mpeg_ofuncs = {
+	.ctor = nv31_mpeg_object_ctor,
+	.dtor = _nouveau_gpuobj_dtor,
+	.init = _nouveau_gpuobj_init,
+	.fini = _nouveau_gpuobj_fini,
+	.rd32 = _nouveau_gpuobj_rd32,
+	.wr32 = _nouveau_gpuobj_wr32,
+};
+
+static struct nouveau_omthds
+nv31_mpeg_omthds[] = {
+	{ 0x0190, 0x0190, nv31_mpeg_mthd_dma },
+	{ 0x01a0, 0x01a0, nv31_mpeg_mthd_dma },
+	{ 0x01b0, 0x01b0, nv31_mpeg_mthd_dma },
+	{}
+};
+
+struct nouveau_oclass
+nv31_mpeg_sclass[] = {
+	{ 0x3174, &nv31_mpeg_ofuncs, nv31_mpeg_omthds },
+	{}
+};
+
+/*******************************************************************************
+ * PMPEG context
+ ******************************************************************************/
+
+static int
+nv31_mpeg_context_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nv31_mpeg_priv *priv = (void *)engine;
+	struct nv31_mpeg_chan *chan;
+	int ret;
+
+	if (!atomic_add_unless(&priv->refcount, 1, 1))
+		return -EBUSY;
+
+	ret = nouveau_object_create(parent, engine, oclass, 0, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static void
+nv31_mpeg_context_dtor(struct nouveau_object *object)
+{
+	struct nv31_mpeg_priv *priv = (void *)object->engine;
+	struct nv31_mpeg_chan *chan = (void *)object;
+	atomic_dec(&priv->refcount);
+	nouveau_object_destroy(&chan->base);
+}
+
+static struct nouveau_oclass
+nv31_mpeg_cclass = {
+	.handle = NV_ENGCTX(MPEG, 0x31),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv31_mpeg_context_ctor,
+		.dtor = nv31_mpeg_context_dtor,
+		.init = nouveau_object_init,
+		.fini = nouveau_object_fini,
+	},
+};
+
+/*******************************************************************************
+ * PMPEG engine/subdev functions
+ ******************************************************************************/
+
+void
+nv31_mpeg_tile_prog(struct nouveau_engine *engine, int i)
+{
+	struct nouveau_fb_tile *tile = &nouveau_fb(engine)->tile.region[i];
+	struct nv31_mpeg_priv *priv = (void *)engine;
+
+	nv_wr32(priv, 0x00b008 + (i * 0x10), tile->pitch);
+	nv_wr32(priv, 0x00b004 + (i * 0x10), tile->limit);
+	nv_wr32(priv, 0x00b000 + (i * 0x10), tile->addr);
+}
+
+void
+nv31_mpeg_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_fifo *pfifo = nouveau_fifo(subdev);
+	struct nouveau_engine *engine = nv_engine(subdev);
+	struct nouveau_object *engctx;
+	struct nouveau_handle *handle;
+	struct nv31_mpeg_priv *priv = (void *)subdev;
+	u32 inst = nv_rd32(priv, 0x00b318) & 0x000fffff;
+	u32 stat = nv_rd32(priv, 0x00b100);
+	u32 type = nv_rd32(priv, 0x00b230);
+	u32 mthd = nv_rd32(priv, 0x00b234);
+	u32 data = nv_rd32(priv, 0x00b238);
+	u32 show = stat;
+	int chid;
+
+	engctx = nouveau_engctx_get(engine, inst);
+	chid   = pfifo->chid(pfifo, engctx);
+
+	if (stat & 0x01000000) {
+		/* happens on initial binding of the object */
+		if (type == 0x00000020 && mthd == 0x0000) {
+			nv_mask(priv, 0x00b308, 0x00000000, 0x00000000);
+			show &= ~0x01000000;
+		}
+
+		if (type == 0x00000010) {
+			handle = nouveau_handle_get_class(engctx, 0x3174);
+			if (handle && !nv_call(handle->object, mthd, data))
+				show &= ~0x01000000;
+			nouveau_handle_put(handle);
+		}
+	}
+
+	nv_wr32(priv, 0x00b100, stat);
+	nv_wr32(priv, 0x00b230, 0x00000001);
+
+	if (show) {
+		nv_error(priv, "ch %d [0x%08x] 0x%08x 0x%08x 0x%08x 0x%08x\n",
+			 chid, inst << 4, stat, type, mthd, data);
+	}
+
+	nouveau_engctx_put(engctx);
+}
+
+static int
+nv31_mpeg_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv31_mpeg_priv *priv;
+	int ret;
+
+	ret = nouveau_mpeg_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000002;
+	nv_subdev(priv)->intr = nv31_mpeg_intr;
+	nv_engine(priv)->cclass = &nv31_mpeg_cclass;
+	nv_engine(priv)->sclass = nv31_mpeg_sclass;
+	nv_engine(priv)->tile_prog = nv31_mpeg_tile_prog;
+	return 0;
+}
+
+int
+nv31_mpeg_init(struct nouveau_object *object)
+{
+	struct nouveau_engine *engine = nv_engine(object->engine);
+	struct nv31_mpeg_priv *priv = (void *)engine;
+	struct nouveau_fb *pfb = nouveau_fb(object);
+	int ret, i;
+
+	ret = nouveau_mpeg_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* VPE init */
+	nv_wr32(priv, 0x00b0e0, 0x00000020); /* nvidia: rd 0x01, wr 0x20 */
+	nv_wr32(priv, 0x00b0e8, 0x00000020); /* nvidia: rd 0x01, wr 0x20 */
+
+	for (i = 0; i < pfb->tile.regions; i++)
+		engine->tile_prog(engine, i);
+
+	/* PMPEG init */
+	nv_wr32(priv, 0x00b32c, 0x00000000);
+	nv_wr32(priv, 0x00b314, 0x00000100);
+	nv_wr32(priv, 0x00b220, nv44_graph_class(priv) ? 0x00000044 : 0x00000031);
+	nv_wr32(priv, 0x00b300, 0x02001ec1);
+	nv_mask(priv, 0x00b32c, 0x00000001, 0x00000001);
+
+	nv_wr32(priv, 0x00b100, 0xffffffff);
+	nv_wr32(priv, 0x00b140, 0xffffffff);
+
+	if (!nv_wait(priv, 0x00b200, 0x00000001, 0x00000000)) {
+		nv_error(priv, "timeout 0x%08x\n", nv_rd32(priv, 0x00b200));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+struct nouveau_oclass
+nv31_mpeg_oclass = {
+	.handle = NV_ENGINE(MPEG, 0x31),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv31_mpeg_ctor,
+		.dtor = _nouveau_mpeg_dtor,
+		.init = nv31_mpeg_init,
+		.fini = _nouveau_mpeg_fini,
+	},
+};
diff --git a/nvkm/engine/mpeg/nv40.c b/nvkm/engine/mpeg/nv40.c
new file mode 100644
index 000000000..f7c581ad1
--- /dev/null
+++ b/nvkm/engine/mpeg/nv40.c
@@ -0,0 +1,144 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+
+#include <subdev/fb.h>
+#include <subdev/timer.h>
+#include <subdev/instmem.h>
+
+#include <engine/mpeg.h>
+#include <engine/graph/nv40.h>
+
+struct nv40_mpeg_priv {
+	struct nouveau_mpeg base;
+};
+
+struct nv40_mpeg_chan {
+	struct nouveau_mpeg_chan base;
+};
+
+/*******************************************************************************
+ * PMPEG context
+ ******************************************************************************/
+
+static int
+nv40_mpeg_context_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nv40_mpeg_chan *chan;
+	int ret;
+
+	ret = nouveau_mpeg_context_create(parent, engine, oclass, NULL,
+					  264 * 4, 16,
+					  NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int
+nv40_mpeg_context_fini(struct nouveau_object *object, bool suspend)
+{
+
+	struct nv40_mpeg_priv *priv = (void *)object->engine;
+	struct nv40_mpeg_chan *chan = (void *)object;
+	u32 inst = 0x80000000 | nv_gpuobj(chan)->addr >> 4;
+
+	nv_mask(priv, 0x00b32c, 0x00000001, 0x00000000);
+	if (nv_rd32(priv, 0x00b318) == inst)
+		nv_mask(priv, 0x00b318, 0x80000000, 0x00000000);
+	nv_mask(priv, 0x00b32c, 0x00000001, 0x00000001);
+	return 0;
+}
+
+static struct nouveau_oclass
+nv40_mpeg_cclass = {
+	.handle = NV_ENGCTX(MPEG, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv40_mpeg_context_ctor,
+		.dtor = _nouveau_mpeg_context_dtor,
+		.init = _nouveau_mpeg_context_init,
+		.fini = nv40_mpeg_context_fini,
+		.rd32 = _nouveau_mpeg_context_rd32,
+		.wr32 = _nouveau_mpeg_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PMPEG engine/subdev functions
+ ******************************************************************************/
+
+static void
+nv40_mpeg_intr(struct nouveau_subdev *subdev)
+{
+	struct nv40_mpeg_priv *priv = (void *)subdev;
+	u32 stat;
+
+	if ((stat = nv_rd32(priv, 0x00b100)))
+		nv31_mpeg_intr(subdev);
+
+	if ((stat = nv_rd32(priv, 0x00b800))) {
+		nv_error(priv, "PMSRCH 0x%08x\n", stat);
+		nv_wr32(priv, 0x00b800, stat);
+	}
+}
+
+static int
+nv40_mpeg_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv40_mpeg_priv *priv;
+	int ret;
+
+	ret = nouveau_mpeg_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000002;
+	nv_subdev(priv)->intr = nv40_mpeg_intr;
+	nv_engine(priv)->cclass = &nv40_mpeg_cclass;
+	nv_engine(priv)->sclass = nv31_mpeg_sclass;
+	nv_engine(priv)->tile_prog = nv31_mpeg_tile_prog;
+	return 0;
+}
+
+struct nouveau_oclass
+nv40_mpeg_oclass = {
+	.handle = NV_ENGINE(MPEG, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv40_mpeg_ctor,
+		.dtor = _nouveau_mpeg_dtor,
+		.init = nv31_mpeg_init,
+		.fini = _nouveau_mpeg_fini,
+	},
+};
diff --git a/nvkm/engine/mpeg/nv50.c b/nvkm/engine/mpeg/nv50.c
new file mode 100644
index 000000000..bc7d12b30
--- /dev/null
+++ b/nvkm/engine/mpeg/nv50.c
@@ -0,0 +1,239 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+
+#include <subdev/vm.h>
+#include <subdev/bar.h>
+#include <subdev/timer.h>
+
+#include <engine/mpeg.h>
+
+struct nv50_mpeg_priv {
+	struct nouveau_mpeg base;
+};
+
+struct nv50_mpeg_chan {
+	struct nouveau_mpeg_chan base;
+};
+
+/*******************************************************************************
+ * MPEG object classes
+ ******************************************************************************/
+
+static int
+nv50_mpeg_object_ctor(struct nouveau_object *parent,
+		      struct nouveau_object *engine,
+		      struct nouveau_oclass *oclass, void *data, u32 size,
+		      struct nouveau_object **pobject)
+{
+	struct nouveau_gpuobj *obj;
+	int ret;
+
+	ret = nouveau_gpuobj_create(parent, engine, oclass, 0, parent,
+				    16, 16, 0, &obj);
+	*pobject = nv_object(obj);
+	if (ret)
+		return ret;
+
+	nv_wo32(obj, 0x00, nv_mclass(obj));
+	nv_wo32(obj, 0x04, 0x00000000);
+	nv_wo32(obj, 0x08, 0x00000000);
+	nv_wo32(obj, 0x0c, 0x00000000);
+	return 0;
+}
+
+struct nouveau_ofuncs
+nv50_mpeg_ofuncs = {
+	.ctor = nv50_mpeg_object_ctor,
+	.dtor = _nouveau_gpuobj_dtor,
+	.init = _nouveau_gpuobj_init,
+	.fini = _nouveau_gpuobj_fini,
+	.rd32 = _nouveau_gpuobj_rd32,
+	.wr32 = _nouveau_gpuobj_wr32,
+};
+
+static struct nouveau_oclass
+nv50_mpeg_sclass[] = {
+	{ 0x3174, &nv50_mpeg_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * PMPEG context
+ ******************************************************************************/
+
+int
+nv50_mpeg_context_ctor(struct nouveau_object *parent,
+		       struct nouveau_object *engine,
+		       struct nouveau_oclass *oclass, void *data, u32 size,
+		       struct nouveau_object **pobject)
+{
+	struct nouveau_bar *bar = nouveau_bar(parent);
+	struct nv50_mpeg_chan *chan;
+	int ret;
+
+	ret = nouveau_mpeg_context_create(parent, engine, oclass, NULL, 128 * 4,
+					  0, NVOBJ_FLAG_ZERO_ALLOC, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	nv_wo32(chan, 0x0070, 0x00801ec1);
+	nv_wo32(chan, 0x007c, 0x0000037c);
+	bar->flush(bar);
+	return 0;
+}
+
+static struct nouveau_oclass
+nv50_mpeg_cclass = {
+	.handle = NV_ENGCTX(MPEG, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_mpeg_context_ctor,
+		.dtor = _nouveau_mpeg_context_dtor,
+		.init = _nouveau_mpeg_context_init,
+		.fini = _nouveau_mpeg_context_fini,
+		.rd32 = _nouveau_mpeg_context_rd32,
+		.wr32 = _nouveau_mpeg_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PMPEG engine/subdev functions
+ ******************************************************************************/
+
+int
+nv50_mpeg_tlb_flush(struct nouveau_engine *engine)
+{
+	nv50_vm_flush_engine(&engine->base, 0x08);
+	return 0;
+}
+
+void
+nv50_mpeg_intr(struct nouveau_subdev *subdev)
+{
+	struct nv50_mpeg_priv *priv = (void *)subdev;
+	u32 stat = nv_rd32(priv, 0x00b100);
+	u32 type = nv_rd32(priv, 0x00b230);
+	u32 mthd = nv_rd32(priv, 0x00b234);
+	u32 data = nv_rd32(priv, 0x00b238);
+	u32 show = stat;
+
+	if (stat & 0x01000000) {
+		/* happens on initial binding of the object */
+		if (type == 0x00000020 && mthd == 0x0000) {
+			nv_wr32(priv, 0x00b308, 0x00000100);
+			show &= ~0x01000000;
+		}
+	}
+
+	if (show) {
+		nv_info(priv, "0x%08x 0x%08x 0x%08x 0x%08x\n",
+			stat, type, mthd, data);
+	}
+
+	nv_wr32(priv, 0x00b100, stat);
+	nv_wr32(priv, 0x00b230, 0x00000001);
+}
+
+static void
+nv50_vpe_intr(struct nouveau_subdev *subdev)
+{
+	struct nv50_mpeg_priv *priv = (void *)subdev;
+
+	if (nv_rd32(priv, 0x00b100))
+		nv50_mpeg_intr(subdev);
+
+	if (nv_rd32(priv, 0x00b800)) {
+		u32 stat = nv_rd32(priv, 0x00b800);
+		nv_info(priv, "PMSRCH: 0x%08x\n", stat);
+		nv_wr32(priv, 0xb800, stat);
+	}
+}
+
+static int
+nv50_mpeg_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_mpeg_priv *priv;
+	int ret;
+
+	ret = nouveau_mpeg_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00400002;
+	nv_subdev(priv)->intr = nv50_vpe_intr;
+	nv_engine(priv)->cclass = &nv50_mpeg_cclass;
+	nv_engine(priv)->sclass = nv50_mpeg_sclass;
+	nv_engine(priv)->tlb_flush = nv50_mpeg_tlb_flush;
+	return 0;
+}
+
+int
+nv50_mpeg_init(struct nouveau_object *object)
+{
+	struct nv50_mpeg_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_mpeg_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x00b32c, 0x00000000);
+	nv_wr32(priv, 0x00b314, 0x00000100);
+	nv_wr32(priv, 0x00b0e0, 0x0000001a);
+
+	nv_wr32(priv, 0x00b220, 0x00000044);
+	nv_wr32(priv, 0x00b300, 0x00801ec1);
+	nv_wr32(priv, 0x00b390, 0x00000000);
+	nv_wr32(priv, 0x00b394, 0x00000000);
+	nv_wr32(priv, 0x00b398, 0x00000000);
+	nv_mask(priv, 0x00b32c, 0x00000001, 0x00000001);
+
+	nv_wr32(priv, 0x00b100, 0xffffffff);
+	nv_wr32(priv, 0x00b140, 0xffffffff);
+
+	if (!nv_wait(priv, 0x00b200, 0x00000001, 0x00000000)) {
+		nv_error(priv, "timeout 0x%08x\n", nv_rd32(priv, 0x00b200));
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_mpeg_oclass = {
+	.handle = NV_ENGINE(MPEG, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_mpeg_ctor,
+		.dtor = _nouveau_mpeg_dtor,
+		.init = nv50_mpeg_init,
+		.fini = _nouveau_mpeg_fini,
+	},
+};
diff --git a/nvkm/engine/mpeg/nv84.c b/nvkm/engine/mpeg/nv84.c
new file mode 100644
index 000000000..8f805b44d
--- /dev/null
+++ b/nvkm/engine/mpeg/nv84.c
@@ -0,0 +1,104 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+
+#include <subdev/vm.h>
+#include <subdev/bar.h>
+#include <subdev/timer.h>
+
+#include <engine/mpeg.h>
+
+struct nv84_mpeg_priv {
+	struct nouveau_mpeg base;
+};
+
+struct nv84_mpeg_chan {
+	struct nouveau_mpeg_chan base;
+};
+
+/*******************************************************************************
+ * MPEG object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv84_mpeg_sclass[] = {
+	{ 0x8274, &nv50_mpeg_ofuncs },
+	{}
+};
+
+/*******************************************************************************
+ * PMPEG context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv84_mpeg_cclass = {
+	.handle = NV_ENGCTX(MPEG, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_mpeg_context_ctor,
+		.dtor = _nouveau_mpeg_context_dtor,
+		.init = _nouveau_mpeg_context_init,
+		.fini = _nouveau_mpeg_context_fini,
+		.rd32 = _nouveau_mpeg_context_rd32,
+		.wr32 = _nouveau_mpeg_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PMPEG engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv84_mpeg_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv84_mpeg_priv *priv;
+	int ret;
+
+	ret = nouveau_mpeg_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000002;
+	nv_subdev(priv)->intr = nv50_mpeg_intr;
+	nv_engine(priv)->cclass = &nv84_mpeg_cclass;
+	nv_engine(priv)->sclass = nv84_mpeg_sclass;
+	nv_engine(priv)->tlb_flush = nv50_mpeg_tlb_flush;
+	return 0;
+}
+
+struct nouveau_oclass
+nv84_mpeg_oclass = {
+	.handle = NV_ENGINE(MPEG, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv84_mpeg_ctor,
+		.dtor = _nouveau_mpeg_dtor,
+		.init = nv50_mpeg_init,
+		.fini = _nouveau_mpeg_fini,
+	},
+};
diff --git a/nvkm/engine/ppp/Makefile.am b/nvkm/engine/ppp/Makefile.am
new file mode 100644
index 000000000..f9a1d5a8d
--- /dev/null
+++ b/nvkm/engine/ppp/Makefile.am
@@ -0,0 +1,6 @@
+noinst_LTLIBRARIES = libppp.la
+
+libppp_la_SOURCES = nv98.c \
+		    nvc0.c
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/ppp/nv98.c b/nvkm/engine/ppp/nv98.c
new file mode 100644
index 000000000..5a5b2a773
--- /dev/null
+++ b/nvkm/engine/ppp/nv98.c
@@ -0,0 +1,98 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/engine.h>
+#include <core/engctx.h>
+#include <core/class.h>
+
+#include <engine/ppp.h>
+
+struct nv98_ppp_priv {
+	struct nouveau_engine base;
+};
+
+struct nv98_ppp_chan {
+	struct nouveau_engctx base;
+};
+
+/*******************************************************************************
+ * PPP object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv98_ppp_sclass[] = {
+	{},
+};
+
+/*******************************************************************************
+ * PPPP context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv98_ppp_cclass = {
+	.handle = NV_ENGCTX(PPP, 0x98),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_engctx_ctor,
+		.dtor = _nouveau_engctx_dtor,
+		.init = _nouveau_engctx_init,
+		.fini = _nouveau_engctx_fini,
+		.rd32 = _nouveau_engctx_rd32,
+		.wr32 = _nouveau_engctx_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PPPP engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv98_ppp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nv98_ppp_priv *priv;
+	int ret;
+
+	ret = nouveau_engine_create(parent, engine, oclass, true,
+				    "PPPP", "ppp", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00400002;
+	nv_engine(priv)->cclass = &nv98_ppp_cclass;
+	nv_engine(priv)->sclass = nv98_ppp_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nv98_ppp_oclass = {
+	.handle = NV_ENGINE(PPP, 0x98),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv98_ppp_ctor,
+		.dtor = _nouveau_engine_dtor,
+		.init = _nouveau_engine_init,
+		.fini = _nouveau_engine_fini,
+	},
+};
diff --git a/nvkm/engine/ppp/nvc0.c b/nvkm/engine/ppp/nvc0.c
new file mode 100644
index 000000000..ebf0d860e
--- /dev/null
+++ b/nvkm/engine/ppp/nvc0.c
@@ -0,0 +1,110 @@
+/*
+ * Copyright 2012 Maarten Lankhorst
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Maarten Lankhorst
+ */
+
+#include <core/falcon.h>
+
+#include <engine/ppp.h>
+
+struct nvc0_ppp_priv {
+	struct nouveau_falcon base;
+};
+
+/*******************************************************************************
+ * PPP object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nvc0_ppp_sclass[] = {
+	{ 0x90b3, &nouveau_object_ofuncs },
+	{},
+};
+
+/*******************************************************************************
+ * PPPP context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nvc0_ppp_cclass = {
+	.handle = NV_ENGCTX(PPP, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_falcon_context_ctor,
+		.dtor = _nouveau_falcon_context_dtor,
+		.init = _nouveau_falcon_context_init,
+		.fini = _nouveau_falcon_context_fini,
+		.rd32 = _nouveau_falcon_context_rd32,
+		.wr32 = _nouveau_falcon_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PPPP engine/subdev functions
+ ******************************************************************************/
+
+static int
+nvc0_ppp_init(struct nouveau_object *object)
+{
+	struct nvc0_ppp_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_falcon_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x086010, 0x0000fff2);
+	nv_wr32(priv, 0x08601c, 0x0000fff2);
+	return 0;
+}
+
+static int
+nvc0_ppp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nvc0_ppp_priv *priv;
+	int ret;
+
+	ret = nouveau_falcon_create(parent, engine, oclass, 0x086000, true,
+				    "PPPP", "ppp", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00000002;
+	nv_engine(priv)->cclass = &nvc0_ppp_cclass;
+	nv_engine(priv)->sclass = nvc0_ppp_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_ppp_oclass = {
+	.handle = NV_ENGINE(PPP, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_ppp_ctor,
+		.dtor = _nouveau_falcon_dtor,
+		.init = nvc0_ppp_init,
+		.fini = _nouveau_falcon_fini,
+		.rd32 = _nouveau_falcon_rd32,
+		.wr32 = _nouveau_falcon_wr32,
+	},
+};
diff --git a/nvkm/engine/software/Makefile.am b/nvkm/engine/software/Makefile.am
new file mode 100644
index 000000000..e17df3a2a
--- /dev/null
+++ b/nvkm/engine/software/Makefile.am
@@ -0,0 +1,8 @@
+noinst_LTLIBRARIES = libsoftware.la
+
+libsoftware_la_SOURCES = nv04.c \
+			 nv10.c \
+			 nv50.c \
+			 nvc0.c
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/software/nv04.c b/nvkm/engine/software/nv04.c
new file mode 100644
index 000000000..2a859a31c
--- /dev/null
+++ b/nvkm/engine/software/nv04.c
@@ -0,0 +1,147 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+
+#include <engine/software.h>
+#include <engine/fifo.h>
+
+struct nv04_software_priv {
+	struct nouveau_software base;
+};
+
+struct nv04_software_chan {
+	struct nouveau_software_chan base;
+};
+
+/*******************************************************************************
+ * software object classes
+ ******************************************************************************/
+
+static int
+nv04_software_set_ref(struct nouveau_object *object, u32 mthd,
+		      void *data, u32 size)
+{
+	struct nouveau_object *channel = (void *)nv_engctx(object->parent);
+	struct nouveau_fifo_chan *fifo = (void *)channel->parent;
+	atomic_set(&fifo->refcnt, *(u32*)data);
+	return 0;
+}
+
+static int
+nv04_software_flip(struct nouveau_object *object, u32 mthd,
+		   void *args, u32 size)
+{
+	struct nv04_software_chan *chan = (void *)nv_engctx(object->parent);
+	if (chan->base.flip)
+		return chan->base.flip(chan->base.flip_data);
+	return -EINVAL;
+}
+
+static struct nouveau_omthds
+nv04_software_omthds[] = {
+	{ 0x0150, 0x0150, nv04_software_set_ref },
+	{ 0x0500, 0x0500, nv04_software_flip },
+	{}
+};
+
+static struct nouveau_oclass
+nv04_software_sclass[] = {
+	{ 0x006e, &nouveau_object_ofuncs, nv04_software_omthds },
+	{}
+};
+
+/*******************************************************************************
+ * software context
+ ******************************************************************************/
+
+static int
+nv04_software_context_ctor(struct nouveau_object *parent,
+		      struct nouveau_object *engine,
+		      struct nouveau_oclass *oclass, void *data, u32 size,
+		      struct nouveau_object **pobject)
+{
+	struct nv04_software_chan *chan;
+	int ret;
+
+	ret = nouveau_software_context_create(parent, engine, oclass, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static struct nouveau_oclass
+nv04_software_cclass = {
+	.handle = NV_ENGCTX(SW, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_software_context_ctor,
+		.dtor = _nouveau_software_context_dtor,
+		.init = _nouveau_software_context_init,
+		.fini = _nouveau_software_context_fini,
+	},
+};
+
+/*******************************************************************************
+ * software engine/subdev functions
+ ******************************************************************************/
+
+void
+nv04_software_intr(struct nouveau_subdev *subdev)
+{
+	nv_mask(subdev, 0x000100, 0x80000000, 0x00000000);
+}
+
+static int
+nv04_software_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nv04_software_priv *priv;
+	int ret;
+
+	ret = nouveau_software_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->cclass = &nv04_software_cclass;
+	nv_engine(priv)->sclass = nv04_software_sclass;
+	nv_subdev(priv)->intr = nv04_software_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nv04_software_oclass = {
+	.handle = NV_ENGINE(SW, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_software_ctor,
+		.dtor = _nouveau_software_dtor,
+		.init = _nouveau_software_init,
+		.fini = _nouveau_software_fini,
+	},
+};
diff --git a/nvkm/engine/software/nv10.c b/nvkm/engine/software/nv10.c
new file mode 100644
index 000000000..a019364b1
--- /dev/null
+++ b/nvkm/engine/software/nv10.c
@@ -0,0 +1,129 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+
+#include <engine/software.h>
+
+struct nv10_software_priv {
+	struct nouveau_software base;
+};
+
+struct nv10_software_chan {
+	struct nouveau_software_chan base;
+};
+
+/*******************************************************************************
+ * software object classes
+ ******************************************************************************/
+
+static int
+nv10_software_flip(struct nouveau_object *object, u32 mthd,
+		   void *args, u32 size)
+{
+	struct nv10_software_chan *chan = (void *)nv_engctx(object->parent);
+	if (chan->base.flip)
+		return chan->base.flip(chan->base.flip_data);
+	return -EINVAL;
+}
+
+static struct nouveau_omthds
+nv10_software_omthds[] = {
+	{ 0x0500, 0x0500, nv10_software_flip },
+	{}
+};
+
+static struct nouveau_oclass
+nv10_software_sclass[] = {
+	{ 0x016e, &nouveau_object_ofuncs, nv10_software_omthds },
+	{}
+};
+
+/*******************************************************************************
+ * software context
+ ******************************************************************************/
+
+static int
+nv10_software_context_ctor(struct nouveau_object *parent,
+		      struct nouveau_object *engine,
+		      struct nouveau_oclass *oclass, void *data, u32 size,
+		      struct nouveau_object **pobject)
+{
+	struct nv10_software_chan *chan;
+	int ret;
+
+	ret = nouveau_software_context_create(parent, engine, oclass, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static struct nouveau_oclass
+nv10_software_cclass = {
+	.handle = NV_ENGCTX(SW, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv10_software_context_ctor,
+		.dtor = _nouveau_software_context_dtor,
+		.init = _nouveau_software_context_init,
+		.fini = _nouveau_software_context_fini,
+	},
+};
+
+/*******************************************************************************
+ * software engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv10_software_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nv10_software_priv *priv;
+	int ret;
+
+	ret = nouveau_software_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->cclass = &nv10_software_cclass;
+	nv_engine(priv)->sclass = nv10_software_sclass;
+	nv_subdev(priv)->intr = nv04_software_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nv10_software_oclass = {
+	.handle = NV_ENGINE(SW, 0x10),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv10_software_ctor,
+		.dtor = _nouveau_software_dtor,
+		.init = _nouveau_software_init,
+		.fini = _nouveau_software_fini,
+	},
+};
diff --git a/nvkm/engine/software/nv50.c b/nvkm/engine/software/nv50.c
new file mode 100644
index 000000000..b0e7e1c01
--- /dev/null
+++ b/nvkm/engine/software/nv50.c
@@ -0,0 +1,199 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+#include <core/namedb.h>
+#include <core/handle.h>
+#include <core/gpuobj.h>
+
+#include <engine/software.h>
+#include <engine/disp.h>
+
+struct nv50_software_priv {
+	struct nouveau_software base;
+};
+
+struct nv50_software_chan {
+	struct nouveau_software_chan base;
+};
+
+/*******************************************************************************
+ * software object classes
+ ******************************************************************************/
+
+static int
+nv50_software_mthd_dma_vblsem(struct nouveau_object *object, u32 mthd,
+			      void *args, u32 size)
+{
+	struct nv50_software_chan *chan = (void *)nv_engctx(object->parent);
+	struct nouveau_fifo_chan *fifo = (void *)nv_object(chan)->parent;
+	struct nouveau_handle *handle;
+	int ret = -EINVAL;
+
+	handle = nouveau_namedb_get(nv_namedb(fifo), *(u32 *)args);
+	if (!handle)
+		return -ENOENT;
+
+	if (nv_iclass(handle->object, NV_GPUOBJ_CLASS)) {
+		struct nouveau_gpuobj *gpuobj = nv_gpuobj(handle->object);
+		chan->base.vblank.ctxdma = gpuobj->node->offset >> 4;
+		ret = 0;
+	}
+	nouveau_namedb_put(handle);
+	return ret;
+}
+
+static int
+nv50_software_mthd_vblsem_offset(struct nouveau_object *object, u32 mthd,
+				 void *args, u32 size)
+{
+	struct nv50_software_chan *chan = (void *)nv_engctx(object->parent);
+	chan->base.vblank.offset = *(u32 *)args;
+	return 0;
+}
+
+static int
+nv50_software_mthd_vblsem_value(struct nouveau_object *object, u32 mthd,
+				void *args, u32 size)
+{
+	struct nv50_software_chan *chan = (void *)nv_engctx(object->parent);
+	chan->base.vblank.value = *(u32 *)args;
+	return 0;
+}
+
+static int
+nv50_software_mthd_vblsem_release(struct nouveau_object *object, u32 mthd,
+				  void *args, u32 size)
+{
+	struct nv50_software_chan *chan = (void *)nv_engctx(object->parent);
+	struct nouveau_disp *disp = nouveau_disp(object);
+	unsigned long flags;
+	u32 crtc = *(u32 *)args;
+
+	if (crtc > 1)
+		return -EINVAL;
+
+	disp->vblank.get(disp->vblank.data, crtc);
+
+	spin_lock_irqsave(&disp->vblank.lock, flags);
+	list_add(&chan->base.vblank.head, &disp->vblank.list);
+	chan->base.vblank.crtc = crtc;
+	spin_unlock_irqrestore(&disp->vblank.lock, flags);
+	return 0;
+}
+
+static int
+nv50_software_mthd_flip(struct nouveau_object *object, u32 mthd,
+			void *args, u32 size)
+{
+	struct nv50_software_chan *chan = (void *)nv_engctx(object->parent);
+	if (chan->base.flip)
+		return chan->base.flip(chan->base.flip_data);
+	return -EINVAL;
+}
+
+static struct nouveau_omthds
+nv50_software_omthds[] = {
+	{ 0x018c, 0x018c, nv50_software_mthd_dma_vblsem },
+	{ 0x0400, 0x0400, nv50_software_mthd_vblsem_offset },
+	{ 0x0404, 0x0404, nv50_software_mthd_vblsem_value },
+	{ 0x0408, 0x0408, nv50_software_mthd_vblsem_release },
+	{ 0x0500, 0x0500, nv50_software_mthd_flip },
+	{}
+};
+
+static struct nouveau_oclass
+nv50_software_sclass[] = {
+	{ 0x506e, &nouveau_object_ofuncs, nv50_software_omthds },
+	{}
+};
+
+/*******************************************************************************
+ * software context
+ ******************************************************************************/
+
+static int
+nv50_software_context_ctor(struct nouveau_object *parent,
+			   struct nouveau_object *engine,
+			   struct nouveau_oclass *oclass, void *data, u32 size,
+			   struct nouveau_object **pobject)
+{
+	struct nv50_software_chan *chan;
+	int ret;
+
+	ret = nouveau_software_context_create(parent, engine, oclass, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	chan->base.vblank.channel = nv_gpuobj(parent->parent)->addr >> 12;
+	return 0;
+}
+
+static struct nouveau_oclass
+nv50_software_cclass = {
+	.handle = NV_ENGCTX(SW, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_software_context_ctor,
+		.dtor = _nouveau_software_context_dtor,
+		.init = _nouveau_software_context_init,
+		.fini = _nouveau_software_context_fini,
+	},
+};
+
+/*******************************************************************************
+ * software engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv50_software_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nv50_software_priv *priv;
+	int ret;
+
+	ret = nouveau_software_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->cclass = &nv50_software_cclass;
+	nv_engine(priv)->sclass = nv50_software_sclass;
+	nv_subdev(priv)->intr = nv04_software_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_software_oclass = {
+	.handle = NV_ENGINE(SW, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_software_ctor,
+		.dtor = _nouveau_software_dtor,
+		.init = _nouveau_software_init,
+		.fini = _nouveau_software_fini,
+	},
+};
diff --git a/nvkm/engine/software/nvc0.c b/nvkm/engine/software/nvc0.c
new file mode 100644
index 000000000..282a1cd1b
--- /dev/null
+++ b/nvkm/engine/software/nvc0.c
@@ -0,0 +1,181 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/os.h>
+#include <core/class.h>
+#include <core/engctx.h>
+
+#include <engine/software.h>
+#include <engine/disp.h>
+
+struct nvc0_software_priv {
+	struct nouveau_software base;
+};
+
+struct nvc0_software_chan {
+	struct nouveau_software_chan base;
+};
+
+/*******************************************************************************
+ * software object classes
+ ******************************************************************************/
+
+static int
+nvc0_software_mthd_vblsem_offset(struct nouveau_object *object, u32 mthd,
+				 void *args, u32 size)
+{
+	struct nvc0_software_chan *chan = (void *)nv_engctx(object->parent);
+	u64 data = *(u32 *)args;
+	if (mthd == 0x0400) {
+		chan->base.vblank.offset &= 0x00ffffffffULL;
+		chan->base.vblank.offset |= data << 32;
+	} else {
+		chan->base.vblank.offset &= 0xff00000000ULL;
+		chan->base.vblank.offset |= data;
+	}
+	return 0;
+}
+
+static int
+nvc0_software_mthd_vblsem_value(struct nouveau_object *object, u32 mthd,
+				void *args, u32 size)
+{
+	struct nvc0_software_chan *chan = (void *)nv_engctx(object->parent);
+	chan->base.vblank.value = *(u32 *)args;
+	return 0;
+}
+
+static int
+nvc0_software_mthd_vblsem_release(struct nouveau_object *object, u32 mthd,
+				  void *args, u32 size)
+{
+	struct nvc0_software_chan *chan = (void *)nv_engctx(object->parent);
+	struct nouveau_disp *disp = nouveau_disp(object);
+	unsigned long flags;
+	u32 crtc = *(u32 *)args;
+
+	if ((nv_device(object)->card_type < NV_E0 && crtc > 1) || crtc > 3)
+		return -EINVAL;
+
+	disp->vblank.get(disp->vblank.data, crtc);
+
+	spin_lock_irqsave(&disp->vblank.lock, flags);
+	list_add(&chan->base.vblank.head, &disp->vblank.list);
+	chan->base.vblank.crtc = crtc;
+	spin_unlock_irqrestore(&disp->vblank.lock, flags);
+	return 0;
+}
+
+static int
+nvc0_software_mthd_flip(struct nouveau_object *object, u32 mthd,
+			void *args, u32 size)
+{
+	struct nvc0_software_chan *chan = (void *)nv_engctx(object->parent);
+	if (chan->base.flip)
+		return chan->base.flip(chan->base.flip_data);
+	return -EINVAL;
+}
+
+static struct nouveau_omthds
+nvc0_software_omthds[] = {
+	{ 0x0400, 0x0400, nvc0_software_mthd_vblsem_offset },
+	{ 0x0404, 0x0404, nvc0_software_mthd_vblsem_offset },
+	{ 0x0408, 0x0408, nvc0_software_mthd_vblsem_value },
+	{ 0x040c, 0x040c, nvc0_software_mthd_vblsem_release },
+	{ 0x0500, 0x0500, nvc0_software_mthd_flip },
+	{}
+};
+
+static struct nouveau_oclass
+nvc0_software_sclass[] = {
+	{ 0x906e, &nouveau_object_ofuncs, nvc0_software_omthds },
+	{}
+};
+
+/*******************************************************************************
+ * software context
+ ******************************************************************************/
+
+static int
+nvc0_software_context_ctor(struct nouveau_object *parent,
+			   struct nouveau_object *engine,
+			   struct nouveau_oclass *oclass, void *data, u32 size,
+			   struct nouveau_object **pobject)
+{
+	struct nvc0_software_chan *chan;
+	int ret;
+
+	ret = nouveau_software_context_create(parent, engine, oclass, &chan);
+	*pobject = nv_object(chan);
+	if (ret)
+		return ret;
+
+	chan->base.vblank.channel = nv_gpuobj(parent->parent)->addr >> 12;
+	return 0;
+}
+
+static struct nouveau_oclass
+nvc0_software_cclass = {
+	.handle = NV_ENGCTX(SW, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_software_context_ctor,
+		.dtor = _nouveau_software_context_dtor,
+		.init = _nouveau_software_context_init,
+		.fini = _nouveau_software_context_fini,
+	},
+};
+
+/*******************************************************************************
+ * software engine/subdev functions
+ ******************************************************************************/
+
+static int
+nvc0_software_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		   struct nouveau_oclass *oclass, void *data, u32 size,
+		   struct nouveau_object **pobject)
+{
+	struct nvc0_software_priv *priv;
+	int ret;
+
+	ret = nouveau_software_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_engine(priv)->cclass = &nvc0_software_cclass;
+	nv_engine(priv)->sclass = nvc0_software_sclass;
+	nv_subdev(priv)->intr = nv04_software_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_software_oclass = {
+	.handle = NV_ENGINE(SW, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_software_ctor,
+		.dtor = _nouveau_software_dtor,
+		.init = _nouveau_software_init,
+		.fini = _nouveau_software_fini,
+	},
+};
diff --git a/nvkm/engine/vp/Makefile.am b/nvkm/engine/vp/Makefile.am
new file mode 100644
index 000000000..dda0f9a7c
--- /dev/null
+++ b/nvkm/engine/vp/Makefile.am
@@ -0,0 +1,7 @@
+noinst_LTLIBRARIES = libvp.la
+
+libvp_la_SOURCES = nv84.c \
+		   nvc0.c \
+		   nve0.c
+
+include $(srcdir)/../Makefile.engine
diff --git a/nvkm/engine/vp/nv84.c b/nvkm/engine/vp/nv84.c
new file mode 100644
index 000000000..261cd96e6
--- /dev/null
+++ b/nvkm/engine/vp/nv84.c
@@ -0,0 +1,93 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/engctx.h>
+#include <core/class.h>
+
+#include <engine/vp.h>
+
+struct nv84_vp_priv {
+	struct nouveau_engine base;
+};
+
+/*******************************************************************************
+ * VP object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv84_vp_sclass[] = {
+	{},
+};
+
+/*******************************************************************************
+ * PVP context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nv84_vp_cclass = {
+	.handle = NV_ENGCTX(VP, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_engctx_ctor,
+		.dtor = _nouveau_engctx_dtor,
+		.init = _nouveau_engctx_init,
+		.fini = _nouveau_engctx_fini,
+		.rd32 = _nouveau_engctx_rd32,
+		.wr32 = _nouveau_engctx_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PVP engine/subdev functions
+ ******************************************************************************/
+
+static int
+nv84_vp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv84_vp_priv *priv;
+	int ret;
+
+	ret = nouveau_engine_create(parent, engine, oclass, true,
+				    "PVP", "vp", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x01020000;
+	nv_engine(priv)->cclass = &nv84_vp_cclass;
+	nv_engine(priv)->sclass = nv84_vp_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nv84_vp_oclass = {
+	.handle = NV_ENGINE(VP, 0x84),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv84_vp_ctor,
+		.dtor = _nouveau_engine_dtor,
+		.init = _nouveau_engine_init,
+		.fini = _nouveau_engine_fini,
+	},
+};
diff --git a/nvkm/engine/vp/nvc0.c b/nvkm/engine/vp/nvc0.c
new file mode 100644
index 000000000..f761949d7
--- /dev/null
+++ b/nvkm/engine/vp/nvc0.c
@@ -0,0 +1,110 @@
+/*
+ * Copyright 2012 Maarten Lankhorst
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Maarten Lankhorst
+ */
+
+#include <core/falcon.h>
+
+#include <engine/vp.h>
+
+struct nvc0_vp_priv {
+	struct nouveau_falcon base;
+};
+
+/*******************************************************************************
+ * VP object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nvc0_vp_sclass[] = {
+	{ 0x90b2, &nouveau_object_ofuncs },
+	{},
+};
+
+/*******************************************************************************
+ * PVP context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nvc0_vp_cclass = {
+	.handle = NV_ENGCTX(VP, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_falcon_context_ctor,
+		.dtor = _nouveau_falcon_context_dtor,
+		.init = _nouveau_falcon_context_init,
+		.fini = _nouveau_falcon_context_fini,
+		.rd32 = _nouveau_falcon_context_rd32,
+		.wr32 = _nouveau_falcon_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PVP engine/subdev functions
+ ******************************************************************************/
+
+static int
+nvc0_vp_init(struct nouveau_object *object)
+{
+	struct nvc0_vp_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_falcon_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x085010, 0x0000fff2);
+	nv_wr32(priv, 0x08501c, 0x0000fff2);
+	return 0;
+}
+
+static int
+nvc0_vp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nvc0_vp_priv *priv;
+	int ret;
+
+	ret = nouveau_falcon_create(parent, engine, oclass, 0x085000, true,
+				    "PVP", "vp", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00020000;
+	nv_engine(priv)->cclass = &nvc0_vp_cclass;
+	nv_engine(priv)->sclass = nvc0_vp_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_vp_oclass = {
+	.handle = NV_ENGINE(VP, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_vp_ctor,
+		.dtor = _nouveau_falcon_dtor,
+		.init = nvc0_vp_init,
+		.fini = _nouveau_falcon_fini,
+		.rd32 = _nouveau_falcon_rd32,
+		.wr32 = _nouveau_falcon_wr32,
+	},
+};
diff --git a/nvkm/engine/vp/nve0.c b/nvkm/engine/vp/nve0.c
new file mode 100644
index 000000000..2384ce5db
--- /dev/null
+++ b/nvkm/engine/vp/nve0.c
@@ -0,0 +1,110 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/falcon.h>
+
+#include <engine/vp.h>
+
+struct nve0_vp_priv {
+	struct nouveau_falcon base;
+};
+
+/*******************************************************************************
+ * VP object classes
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nve0_vp_sclass[] = {
+	{ 0x95b2, &nouveau_object_ofuncs },
+	{},
+};
+
+/*******************************************************************************
+ * PVP context
+ ******************************************************************************/
+
+static struct nouveau_oclass
+nve0_vp_cclass = {
+	.handle = NV_ENGCTX(VP, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = _nouveau_falcon_context_ctor,
+		.dtor = _nouveau_falcon_context_dtor,
+		.init = _nouveau_falcon_context_init,
+		.fini = _nouveau_falcon_context_fini,
+		.rd32 = _nouveau_falcon_context_rd32,
+		.wr32 = _nouveau_falcon_context_wr32,
+	},
+};
+
+/*******************************************************************************
+ * PVP engine/subdev functions
+ ******************************************************************************/
+
+static int
+nve0_vp_init(struct nouveau_object *object)
+{
+	struct nve0_vp_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_falcon_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x085010, 0x0000fff2);
+	nv_wr32(priv, 0x08501c, 0x0000fff2);
+	return 0;
+}
+
+static int
+nve0_vp_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nve0_vp_priv *priv;
+	int ret;
+
+	ret = nouveau_falcon_create(parent, engine, oclass, 0x085000, true,
+				    "PVP", "vp", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->unit = 0x00020000;
+	nv_engine(priv)->cclass = &nve0_vp_cclass;
+	nv_engine(priv)->sclass = nve0_vp_sclass;
+	return 0;
+}
+
+struct nouveau_oclass
+nve0_vp_oclass = {
+	.handle = NV_ENGINE(VP, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nve0_vp_ctor,
+		.dtor = _nouveau_falcon_dtor,
+		.init = nve0_vp_init,
+		.fini = _nouveau_falcon_fini,
+		.rd32 = _nouveau_falcon_rd32,
+		.wr32 = _nouveau_falcon_wr32,
+	},
+};
diff --git a/nvkm/include/core/class.h b/nvkm/include/core/class.h
new file mode 100644
index 000000000..47c4b3a5b
--- /dev/null
+++ b/nvkm/include/core/class.h
@@ -0,0 +1,343 @@
+#ifndef __NOUVEAU_CLASS_H__
+#define __NOUVEAU_CLASS_H__
+
+/* Device class
+ *
+ * 0080: NV_DEVICE
+ */
+#define NV_DEVICE_CLASS                                              0x00000080
+
+#define NV_DEVICE_DISABLE_IDENTIFY                        0x0000000000000001ULL
+#define NV_DEVICE_DISABLE_MMIO                            0x0000000000000002ULL
+#define NV_DEVICE_DISABLE_VBIOS                           0x0000000000000004ULL
+#define NV_DEVICE_DISABLE_CORE                            0x0000000000000008ULL
+#define NV_DEVICE_DISABLE_DISP                            0x0000000000010000ULL
+#define NV_DEVICE_DISABLE_FIFO                            0x0000000000020000ULL
+#define NV_DEVICE_DISABLE_GRAPH                           0x0000000100000000ULL
+#define NV_DEVICE_DISABLE_MPEG                            0x0000000200000000ULL
+#define NV_DEVICE_DISABLE_ME                              0x0000000400000000ULL
+#define NV_DEVICE_DISABLE_VP                              0x0000000800000000ULL
+#define NV_DEVICE_DISABLE_CRYPT                           0x0000001000000000ULL
+#define NV_DEVICE_DISABLE_BSP                             0x0000002000000000ULL
+#define NV_DEVICE_DISABLE_PPP                             0x0000004000000000ULL
+#define NV_DEVICE_DISABLE_COPY0                           0x0000008000000000ULL
+#define NV_DEVICE_DISABLE_COPY1                           0x0000010000000000ULL
+#define NV_DEVICE_DISABLE_UNK1C1                          0x0000020000000000ULL
+#define NV_DEVICE_DISABLE_VENC                            0x0000040000000000ULL
+
+struct nv_device_class {
+	u64 device;	/* device identifier, ~0 for client default */
+	u64 disable;	/* disable particular subsystems */
+	u64 debug0;	/* as above, but *internal* ids, and *NOT* ABI */
+};
+
+/* DMA object classes
+ *
+ * 0002: NV_DMA_FROM_MEMORY
+ * 0003: NV_DMA_TO_MEMORY
+ * 003d: NV_DMA_IN_MEMORY
+ */
+#define NV_DMA_FROM_MEMORY_CLASS                                     0x00000002
+#define NV_DMA_TO_MEMORY_CLASS                                       0x00000003
+#define NV_DMA_IN_MEMORY_CLASS                                       0x0000003d
+
+#define NV_DMA_TARGET_MASK                                           0x000000ff
+#define NV_DMA_TARGET_VM                                             0x00000000
+#define NV_DMA_TARGET_VRAM                                           0x00000001
+#define NV_DMA_TARGET_PCI                                            0x00000002
+#define NV_DMA_TARGET_PCI_US                                         0x00000003
+#define NV_DMA_TARGET_AGP                                            0x00000004
+#define NV_DMA_ACCESS_MASK                                           0x00000f00
+#define NV_DMA_ACCESS_VM                                             0x00000000
+#define NV_DMA_ACCESS_RD                                             0x00000100
+#define NV_DMA_ACCESS_WR                                             0x00000200
+#define NV_DMA_ACCESS_RDWR                                           0x00000300
+
+/* NV50:NVC0 */
+#define NV50_DMA_CONF0_ENABLE                                        0x80000000
+#define NV50_DMA_CONF0_PRIV                                          0x00300000
+#define NV50_DMA_CONF0_PRIV_VM                                       0x00000000
+#define NV50_DMA_CONF0_PRIV_US                                       0x00100000
+#define NV50_DMA_CONF0_PRIV__S                                       0x00200000
+#define NV50_DMA_CONF0_PART                                          0x00030000
+#define NV50_DMA_CONF0_PART_VM                                       0x00000000
+#define NV50_DMA_CONF0_PART_256                                      0x00010000
+#define NV50_DMA_CONF0_PART_1KB                                      0x00020000
+#define NV50_DMA_CONF0_COMP                                          0x00000180
+#define NV50_DMA_CONF0_COMP_NONE                                     0x00000000
+#define NV50_DMA_CONF0_COMP_VM                                       0x00000180
+#define NV50_DMA_CONF0_TYPE                                          0x0000007f
+#define NV50_DMA_CONF0_TYPE_LINEAR                                   0x00000000
+#define NV50_DMA_CONF0_TYPE_VM                                       0x0000007f
+
+/* NVC0:NVD9 */
+#define NVC0_DMA_CONF0_ENABLE                                        0x80000000
+#define NVC0_DMA_CONF0_PRIV                                          0x00300000
+#define NVC0_DMA_CONF0_PRIV_VM                                       0x00000000
+#define NVC0_DMA_CONF0_PRIV_US                                       0x00100000
+#define NVC0_DMA_CONF0_PRIV__S                                       0x00200000
+#define NVC0_DMA_CONF0_UNKN /* PART? */                              0x00030000
+#define NVC0_DMA_CONF0_TYPE                                          0x000000ff
+#define NVC0_DMA_CONF0_TYPE_LINEAR                                   0x00000000
+#define NVC0_DMA_CONF0_TYPE_VM                                       0x000000ff
+
+/* NVD9- */
+#define NVD0_DMA_CONF0_ENABLE                                        0x80000000
+#define NVD0_DMA_CONF0_PAGE                                          0x00000400
+#define NVD0_DMA_CONF0_PAGE_LP                                       0x00000000
+#define NVD0_DMA_CONF0_PAGE_SP                                       0x00000400
+#define NVD0_DMA_CONF0_TYPE                                          0x000000ff
+#define NVD0_DMA_CONF0_TYPE_LINEAR                                   0x00000000
+#define NVD0_DMA_CONF0_TYPE_VM                                       0x000000ff
+
+struct nv_dma_class {
+	u32 flags;
+	u32 pad0;
+	u64 start;
+	u64 limit;
+	u32 conf0;
+};
+
+/* DMA FIFO channel classes
+ *
+ * 006b: NV03_CHANNEL_DMA
+ * 006e: NV10_CHANNEL_DMA
+ * 176e: NV17_CHANNEL_DMA
+ * 406e: NV40_CHANNEL_DMA
+ * 506e: NV50_CHANNEL_DMA
+ * 826e: NV84_CHANNEL_DMA
+ */
+#define NV03_CHANNEL_DMA_CLASS                                       0x0000006b
+#define NV10_CHANNEL_DMA_CLASS                                       0x0000006e
+#define NV17_CHANNEL_DMA_CLASS                                       0x0000176e
+#define NV40_CHANNEL_DMA_CLASS                                       0x0000406e
+#define NV50_CHANNEL_DMA_CLASS                                       0x0000506e
+#define NV84_CHANNEL_DMA_CLASS                                       0x0000826e
+
+struct nv03_channel_dma_class {
+	u32 pushbuf;
+	u32 pad0;
+	u64 offset;
+};
+
+/* Indirect FIFO channel classes
+ *
+ * 506f: NV50_CHANNEL_IND
+ * 826f: NV84_CHANNEL_IND
+ * 906f: NVC0_CHANNEL_IND
+ * a06f: NVE0_CHANNEL_IND
+ */
+
+#define NV50_CHANNEL_IND_CLASS                                       0x0000506f
+#define NV84_CHANNEL_IND_CLASS                                       0x0000826f
+#define NVC0_CHANNEL_IND_CLASS                                       0x0000906f
+#define NVE0_CHANNEL_IND_CLASS                                       0x0000a06f
+
+struct nv50_channel_ind_class {
+	u32 pushbuf;
+	u32 ilength;
+	u64 ioffset;
+};
+
+#define NVE0_CHANNEL_IND_ENGINE_GR                                   0x00000001
+#define NVE0_CHANNEL_IND_ENGINE_VP                                   0x00000002
+#define NVE0_CHANNEL_IND_ENGINE_PPP                                  0x00000004
+#define NVE0_CHANNEL_IND_ENGINE_BSP                                  0x00000008
+#define NVE0_CHANNEL_IND_ENGINE_CE0                                  0x00000010
+#define NVE0_CHANNEL_IND_ENGINE_CE1                                  0x00000020
+#define NVE0_CHANNEL_IND_ENGINE_ENC                                  0x00000040
+
+struct nve0_channel_ind_class {
+	u32 pushbuf;
+	u32 ilength;
+	u64 ioffset;
+	u32 engine;
+};
+
+/* 5070: NV50_DISP
+ * 8270: NV84_DISP
+ * 8370: NVA0_DISP
+ * 8870: NV94_DISP
+ * 8570: NVA3_DISP
+ * 9070: NVD0_DISP
+ * 9170: NVE0_DISP
+ */
+
+#define NV50_DISP_CLASS                                              0x00005070
+#define NV84_DISP_CLASS                                              0x00008270
+#define NVA0_DISP_CLASS                                              0x00008370
+#define NV94_DISP_CLASS                                              0x00008870
+#define NVA3_DISP_CLASS                                              0x00008570
+#define NVD0_DISP_CLASS                                              0x00009070
+#define NVE0_DISP_CLASS                                              0x00009170
+
+#define NV50_DISP_SOR_MTHD                                           0x00010000
+#define NV50_DISP_SOR_MTHD_TYPE                                      0x0000f000
+#define NV50_DISP_SOR_MTHD_HEAD                                      0x00000018
+#define NV50_DISP_SOR_MTHD_LINK                                      0x00000004
+#define NV50_DISP_SOR_MTHD_OR                                        0x00000003
+
+#define NV50_DISP_SOR_PWR                                            0x00010000
+#define NV50_DISP_SOR_PWR_STATE                                      0x00000001
+#define NV50_DISP_SOR_PWR_STATE_ON                                   0x00000001
+#define NV50_DISP_SOR_PWR_STATE_OFF                                  0x00000000
+#define NVA3_DISP_SOR_HDA_ELD                                        0x00010100
+#define NV84_DISP_SOR_HDMI_PWR                                       0x00012000
+#define NV84_DISP_SOR_HDMI_PWR_STATE                                 0x40000000
+#define NV84_DISP_SOR_HDMI_PWR_STATE_OFF                             0x00000000
+#define NV84_DISP_SOR_HDMI_PWR_STATE_ON                              0x40000000
+#define NV84_DISP_SOR_HDMI_PWR_MAX_AC_PACKET                         0x001f0000
+#define NV84_DISP_SOR_HDMI_PWR_REKEY                                 0x0000007f
+#define NV50_DISP_SOR_LVDS_SCRIPT                                    0x00013000
+#define NV50_DISP_SOR_LVDS_SCRIPT_ID                                 0x0000ffff
+#define NV94_DISP_SOR_DP_TRAIN                                       0x00016000
+#define NV94_DISP_SOR_DP_TRAIN_OP                                    0xf0000000
+#define NV94_DISP_SOR_DP_TRAIN_OP_PATTERN                            0x00000000
+#define NV94_DISP_SOR_DP_TRAIN_OP_INIT                               0x10000000
+#define NV94_DISP_SOR_DP_TRAIN_OP_FINI                               0x20000000
+#define NV94_DISP_SOR_DP_TRAIN_INIT_SPREAD                           0x00000001
+#define NV94_DISP_SOR_DP_TRAIN_INIT_SPREAD_OFF                       0x00000000
+#define NV94_DISP_SOR_DP_TRAIN_INIT_SPREAD_ON                        0x00000001
+#define NV94_DISP_SOR_DP_TRAIN_PATTERN                               0x00000003
+#define NV94_DISP_SOR_DP_TRAIN_PATTERN_DISABLED                      0x00000000
+#define NV94_DISP_SOR_DP_LNKCTL                                      0x00016040
+#define NV94_DISP_SOR_DP_LNKCTL_FRAME                                0x80000000
+#define NV94_DISP_SOR_DP_LNKCTL_FRAME_STD                            0x00000000
+#define NV94_DISP_SOR_DP_LNKCTL_FRAME_ENH                            0x80000000
+#define NV94_DISP_SOR_DP_LNKCTL_WIDTH                                0x00001f00
+#define NV94_DISP_SOR_DP_LNKCTL_COUNT                                0x00000007
+#define NV94_DISP_SOR_DP_DRVCTL(l)                     ((l) * 0x40 + 0x00016100)
+#define NV94_DISP_SOR_DP_DRVCTL_VS                                   0x00000300
+#define NV94_DISP_SOR_DP_DRVCTL_PE                                   0x00000003
+
+#define NV50_DISP_DAC_MTHD                                           0x00020000
+#define NV50_DISP_DAC_MTHD_TYPE                                      0x0000f000
+#define NV50_DISP_DAC_MTHD_OR                                        0x00000003
+
+#define NV50_DISP_DAC_PWR                                            0x00020000
+#define NV50_DISP_DAC_PWR_HSYNC                                      0x00000001
+#define NV50_DISP_DAC_PWR_HSYNC_ON                                   0x00000000
+#define NV50_DISP_DAC_PWR_HSYNC_LO                                   0x00000001
+#define NV50_DISP_DAC_PWR_VSYNC                                      0x00000004
+#define NV50_DISP_DAC_PWR_VSYNC_ON                                   0x00000000
+#define NV50_DISP_DAC_PWR_VSYNC_LO                                   0x00000004
+#define NV50_DISP_DAC_PWR_DATA                                       0x00000010
+#define NV50_DISP_DAC_PWR_DATA_ON                                    0x00000000
+#define NV50_DISP_DAC_PWR_DATA_LO                                    0x00000010
+#define NV50_DISP_DAC_PWR_STATE                                      0x00000040
+#define NV50_DISP_DAC_PWR_STATE_ON                                   0x00000000
+#define NV50_DISP_DAC_PWR_STATE_OFF                                  0x00000040
+#define NV50_DISP_DAC_LOAD                                           0x0002000c
+#define NV50_DISP_DAC_LOAD_VALUE                                     0x00000007
+
+struct nv50_display_class {
+};
+
+/* 507a: NV50_DISP_CURS
+ * 827a: NV84_DISP_CURS
+ * 837a: NVA0_DISP_CURS
+ * 887a: NV94_DISP_CURS
+ * 857a: NVA3_DISP_CURS
+ * 907a: NVD0_DISP_CURS
+ * 917a: NVE0_DISP_CURS
+ */
+
+#define NV50_DISP_CURS_CLASS                                         0x0000507a
+#define NV84_DISP_CURS_CLASS                                         0x0000827a
+#define NVA0_DISP_CURS_CLASS                                         0x0000837a
+#define NV94_DISP_CURS_CLASS                                         0x0000887a
+#define NVA3_DISP_CURS_CLASS                                         0x0000857a
+#define NVD0_DISP_CURS_CLASS                                         0x0000907a
+#define NVE0_DISP_CURS_CLASS                                         0x0000917a
+
+struct nv50_display_curs_class {
+	u32 head;
+};
+
+/* 507b: NV50_DISP_OIMM
+ * 827b: NV84_DISP_OIMM
+ * 837b: NVA0_DISP_OIMM
+ * 887b: NV94_DISP_OIMM
+ * 857b: NVA3_DISP_OIMM
+ * 907b: NVD0_DISP_OIMM
+ * 917b: NVE0_DISP_OIMM
+ */
+
+#define NV50_DISP_OIMM_CLASS                                         0x0000507b
+#define NV84_DISP_OIMM_CLASS                                         0x0000827b
+#define NVA0_DISP_OIMM_CLASS                                         0x0000837b
+#define NV94_DISP_OIMM_CLASS                                         0x0000887b
+#define NVA3_DISP_OIMM_CLASS                                         0x0000857b
+#define NVD0_DISP_OIMM_CLASS                                         0x0000907b
+#define NVE0_DISP_OIMM_CLASS                                         0x0000917b
+
+struct nv50_display_oimm_class {
+	u32 head;
+};
+
+/* 507c: NV50_DISP_SYNC
+ * 827c: NV84_DISP_SYNC
+ * 837c: NVA0_DISP_SYNC
+ * 887c: NV94_DISP_SYNC
+ * 857c: NVA3_DISP_SYNC
+ * 907c: NVD0_DISP_SYNC
+ * 917c: NVE0_DISP_SYNC
+ */
+
+#define NV50_DISP_SYNC_CLASS                                         0x0000507c
+#define NV84_DISP_SYNC_CLASS                                         0x0000827c
+#define NVA0_DISP_SYNC_CLASS                                         0x0000837c
+#define NV94_DISP_SYNC_CLASS                                         0x0000887c
+#define NVA3_DISP_SYNC_CLASS                                         0x0000857c
+#define NVD0_DISP_SYNC_CLASS                                         0x0000907c
+#define NVE0_DISP_SYNC_CLASS                                         0x0000917c
+
+struct nv50_display_sync_class {
+	u32 pushbuf;
+	u32 head;
+};
+
+/* 507d: NV50_DISP_MAST
+ * 827d: NV84_DISP_MAST
+ * 837d: NVA0_DISP_MAST
+ * 887d: NV94_DISP_MAST
+ * 857d: NVA3_DISP_MAST
+ * 907d: NVD0_DISP_MAST
+ * 917d: NVE0_DISP_MAST
+ */
+
+#define NV50_DISP_MAST_CLASS                                         0x0000507d
+#define NV84_DISP_MAST_CLASS                                         0x0000827d
+#define NVA0_DISP_MAST_CLASS                                         0x0000837d
+#define NV94_DISP_MAST_CLASS                                         0x0000887d
+#define NVA3_DISP_MAST_CLASS                                         0x0000857d
+#define NVD0_DISP_MAST_CLASS                                         0x0000907d
+#define NVE0_DISP_MAST_CLASS                                         0x0000917d
+
+struct nv50_display_mast_class {
+	u32 pushbuf;
+};
+
+/* 507e: NV50_DISP_OVLY
+ * 827e: NV84_DISP_OVLY
+ * 837e: NVA0_DISP_OVLY
+ * 887e: NV94_DISP_OVLY
+ * 857e: NVA3_DISP_OVLY
+ * 907e: NVD0_DISP_OVLY
+ * 917e: NVE0_DISP_OVLY
+ */
+
+#define NV50_DISP_OVLY_CLASS                                         0x0000507e
+#define NV84_DISP_OVLY_CLASS                                         0x0000827e
+#define NVA0_DISP_OVLY_CLASS                                         0x0000837e
+#define NV94_DISP_OVLY_CLASS                                         0x0000887e
+#define NVA3_DISP_OVLY_CLASS                                         0x0000857e
+#define NVD0_DISP_OVLY_CLASS                                         0x0000907e
+#define NVE0_DISP_OVLY_CLASS                                         0x0000917e
+
+struct nv50_display_ovly_class {
+	u32 pushbuf;
+	u32 head;
+};
+
+#endif
diff --git a/nvkm/include/core/client.h b/nvkm/include/core/client.h
new file mode 100644
index 000000000..0193532ce
--- /dev/null
+++ b/nvkm/include/core/client.h
@@ -0,0 +1,42 @@
+#ifndef __NOUVEAU_CLIENT_H__
+#define __NOUVEAU_CLIENT_H__
+
+#include <core/namedb.h>
+
+struct nouveau_client {
+	struct nouveau_namedb base;
+	struct nouveau_handle *root;
+	struct nouveau_object *device;
+	char name[16];
+	u32 debug;
+	struct nouveau_vm *vm;
+};
+
+static inline struct nouveau_client *
+nv_client(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!nv_iclass(obj, NV_CLIENT_CLASS)))
+		nv_assert("BAD CAST -> NvClient, %08x", nv_hclass(obj));
+#endif
+	return obj;
+}
+
+static inline struct nouveau_client *
+nouveau_client(void *obj)
+{
+	struct nouveau_object *client = nv_object(obj);
+	while (client && !(nv_iclass(client, NV_CLIENT_CLASS)))
+		client = client->parent;
+	return (void *)client;
+}
+
+#define nouveau_client_create(n,c,oc,od,d)                                     \
+	nouveau_client_create_((n), (c), (oc), (od), sizeof(**d), (void **)d)
+
+int  nouveau_client_create_(const char *name, u64 device, const char *cfg,
+			    const char *dbg, int, void **);
+int  nouveau_client_init(struct nouveau_client *);
+int  nouveau_client_fini(struct nouveau_client *, bool suspend);
+
+#endif
diff --git a/nvkm/include/core/debug.h b/nvkm/include/core/debug.h
new file mode 100644
index 000000000..9ea18dfcb
--- /dev/null
+++ b/nvkm/include/core/debug.h
@@ -0,0 +1,13 @@
+#ifndef __NOUVEAU_DEBUG_H__
+#define __NOUVEAU_DEBUG_H__
+
+#define NV_DBG_FATAL    0
+#define NV_DBG_ERROR    1
+#define NV_DBG_WARN     2
+#define NV_DBG_INFO     3
+#define NV_DBG_DEBUG    4
+#define NV_DBG_TRACE    5
+#define NV_DBG_PARANOIA 6
+#define NV_DBG_SPAM     7
+
+#endif
diff --git a/nvkm/include/core/device.h b/nvkm/include/core/device.h
new file mode 100644
index 000000000..e58b6f098
--- /dev/null
+++ b/nvkm/include/core/device.h
@@ -0,0 +1,136 @@
+#ifndef __NOUVEAU_DEVICE_H__
+#define __NOUVEAU_DEVICE_H__
+
+#include <core/object.h>
+#include <core/subdev.h>
+#include <core/engine.h>
+
+enum nv_subdev_type {
+	NVDEV_SUBDEV_DEVICE,
+	NVDEV_SUBDEV_VBIOS,
+
+	/* All subdevs from DEVINIT to DEVINIT_LAST will be created before
+	 * *any* of them are initialised.  This subdev category is used
+	 * for any subdevs that the VBIOS init table parsing may call out
+	 * to during POST.
+	 */
+	NVDEV_SUBDEV_DEVINIT,
+	NVDEV_SUBDEV_GPIO,
+	NVDEV_SUBDEV_I2C,
+	NVDEV_SUBDEV_CLOCK,
+	NVDEV_SUBDEV_DEVINIT_LAST = NVDEV_SUBDEV_CLOCK,
+
+	/* This grouping of subdevs are initialised right after they've
+	 * been created, and are allowed to assume any subdevs in the
+	 * list above them exist and have been initialised.
+	 */
+	NVDEV_SUBDEV_MXM,
+	NVDEV_SUBDEV_MC,
+	NVDEV_SUBDEV_TIMER,
+	NVDEV_SUBDEV_FB,
+	NVDEV_SUBDEV_LTCG,
+	NVDEV_SUBDEV_IBUS,
+	NVDEV_SUBDEV_INSTMEM,
+	NVDEV_SUBDEV_VM,
+	NVDEV_SUBDEV_BAR,
+	NVDEV_SUBDEV_VOLT,
+	NVDEV_SUBDEV_THERM,
+
+	NVDEV_ENGINE_DMAOBJ,
+	NVDEV_ENGINE_FIFO,
+	NVDEV_ENGINE_SW,
+	NVDEV_ENGINE_GR,
+	NVDEV_ENGINE_MPEG,
+	NVDEV_ENGINE_ME,
+	NVDEV_ENGINE_VP,
+	NVDEV_ENGINE_CRYPT,
+	NVDEV_ENGINE_BSP,
+	NVDEV_ENGINE_PPP,
+	NVDEV_ENGINE_COPY0,
+	NVDEV_ENGINE_COPY1,
+	NVDEV_ENGINE_UNK1C1,
+	NVDEV_ENGINE_VENC,
+	NVDEV_ENGINE_DISP,
+
+	NVDEV_SUBDEV_NR,
+};
+
+struct nouveau_device {
+	struct nouveau_subdev base;
+	struct list_head head;
+
+	struct pci_dev *pdev;
+	u64 handle;
+
+	const char *cfgopt;
+	const char *dbgopt;
+	const char *name;
+	const char *cname;
+
+	enum {
+		NV_04    = 0x04,
+		NV_10    = 0x10,
+		NV_20    = 0x20,
+		NV_30    = 0x30,
+		NV_40    = 0x40,
+		NV_50    = 0x50,
+		NV_C0    = 0xc0,
+		NV_D0    = 0xd0,
+		NV_E0    = 0xe0,
+	} card_type;
+	u32 chipset;
+	u32 crystal;
+
+	struct nouveau_oclass *oclass[NVDEV_SUBDEV_NR];
+	struct nouveau_object *subdev[NVDEV_SUBDEV_NR];
+};
+
+static inline struct nouveau_device *
+nv_device(void *obj)
+{
+	struct nouveau_object *object = nv_object(obj);
+	struct nouveau_object *device = object;
+
+	if (device->engine)
+		device = device->engine;
+	if (device->parent)
+		device = device->parent;
+
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!nv_iclass(device, NV_SUBDEV_CLASS) ||
+		     (nv_hclass(device) & 0xff) != NVDEV_SUBDEV_DEVICE)) {
+		nv_assert("BAD CAST -> NvDevice, 0x%08x 0x%08x",
+			  nv_hclass(object), nv_hclass(device));
+	}
+#endif
+
+	return (void *)device;
+}
+
+static inline struct nouveau_subdev *
+nouveau_subdev(void *obj, int sub)
+{
+	if (nv_device(obj)->subdev[sub])
+		return nv_subdev(nv_device(obj)->subdev[sub]);
+	return NULL;
+}
+
+static inline struct nouveau_engine *
+nouveau_engine(void *obj, int sub)
+{
+	struct nouveau_subdev *subdev = nouveau_subdev(obj, sub);
+	if (subdev && nv_iclass(subdev, NV_ENGINE_CLASS))
+		return nv_engine(subdev);
+	return NULL;
+}
+
+static inline bool
+nv_device_match(struct nouveau_object *object, u16 dev, u16 ven, u16 sub)
+{
+	struct nouveau_device *device = nv_device(object);
+	return device->pdev->device == dev &&
+	       device->pdev->subsystem_vendor == ven &&
+	       device->pdev->subsystem_device == sub;
+}
+
+#endif
diff --git a/nvkm/include/core/engctx.h b/nvkm/include/core/engctx.h
new file mode 100644
index 000000000..2fd48b564
--- /dev/null
+++ b/nvkm/include/core/engctx.h
@@ -0,0 +1,54 @@
+#ifndef __NOUVEAU_ENGCTX_H__
+#define __NOUVEAU_ENGCTX_H__
+
+#include <core/object.h>
+#include <core/gpuobj.h>
+
+#include <subdev/vm.h>
+
+#define NV_ENGCTX_(eng,var) (NV_ENGCTX_CLASS | ((var) << 8) | (eng))
+#define NV_ENGCTX(name,var)  NV_ENGCTX_(NVDEV_ENGINE_##name, (var))
+
+struct nouveau_engctx {
+	struct nouveau_gpuobj base;
+	struct nouveau_vma vma;
+	struct list_head head;
+	unsigned long save;
+	u64 addr;
+};
+
+static inline struct nouveau_engctx *
+nv_engctx(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!nv_iclass(obj, NV_ENGCTX_CLASS)))
+		nv_assert("BAD CAST -> NvEngCtx, %08x", nv_hclass(obj));
+#endif
+	return obj;
+}
+
+#define nouveau_engctx_create(p,e,c,g,s,a,f,d)                                 \
+	nouveau_engctx_create_((p), (e), (c), (g), (s), (a), (f),              \
+			       sizeof(**d), (void **)d)
+
+int  nouveau_engctx_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, struct nouveau_object *,
+			    u32 size, u32 align, u32 flags,
+			    int length, void **data);
+void nouveau_engctx_destroy(struct nouveau_engctx *);
+int  nouveau_engctx_init(struct nouveau_engctx *);
+int  nouveau_engctx_fini(struct nouveau_engctx *, bool suspend);
+
+int  _nouveau_engctx_ctor(struct nouveau_object *, struct nouveau_object *,
+			  struct nouveau_oclass *, void *, u32,
+			  struct nouveau_object **);
+void _nouveau_engctx_dtor(struct nouveau_object *);
+int  _nouveau_engctx_init(struct nouveau_object *);
+int  _nouveau_engctx_fini(struct nouveau_object *, bool suspend);
+#define _nouveau_engctx_rd32 _nouveau_gpuobj_rd32
+#define _nouveau_engctx_wr32 _nouveau_gpuobj_wr32
+
+struct nouveau_object *nouveau_engctx_get(struct nouveau_engine *, u64 addr);
+void nouveau_engctx_put(struct nouveau_object *);
+
+#endif
diff --git a/nvkm/include/core/engine.h b/nvkm/include/core/engine.h
new file mode 100644
index 000000000..666d06de7
--- /dev/null
+++ b/nvkm/include/core/engine.h
@@ -0,0 +1,57 @@
+#ifndef __NOUVEAU_ENGINE_H__
+#define __NOUVEAU_ENGINE_H__
+
+#include <core/object.h>
+#include <core/subdev.h>
+
+#define NV_ENGINE_(eng,var) (NV_ENGINE_CLASS | ((var) << 8) | (eng))
+#define NV_ENGINE(name,var)  NV_ENGINE_(NVDEV_ENGINE_##name, (var))
+
+struct nouveau_engine {
+	struct nouveau_subdev base;
+	struct nouveau_oclass *cclass;
+	struct nouveau_oclass *sclass;
+
+	struct list_head contexts;
+	spinlock_t lock;
+
+	void (*tile_prog)(struct nouveau_engine *, int region);
+	int  (*tlb_flush)(struct nouveau_engine *);
+};
+
+static inline struct nouveau_engine *
+nv_engine(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!nv_iclass(obj, NV_ENGINE_CLASS)))
+		nv_assert("BAD CAST -> NvEngine, %08x", nv_hclass(obj));
+#endif
+	return obj;
+}
+
+static inline int
+nv_engidx(struct nouveau_object *object)
+{
+	return nv_subidx(object);
+}
+
+#define nouveau_engine_create(p,e,c,d,i,f,r)                                   \
+	nouveau_engine_create_((p), (e), (c), (d), (i), (f),                   \
+			       sizeof(**r),(void **)r)
+
+#define nouveau_engine_destroy(p)                                              \
+	nouveau_subdev_destroy(&(p)->base)
+#define nouveau_engine_init(p)                                                 \
+	nouveau_subdev_init(&(p)->base)
+#define nouveau_engine_fini(p,s)                                               \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+int nouveau_engine_create_(struct nouveau_object *, struct nouveau_object *,
+			   struct nouveau_oclass *, bool, const char *,
+			   const char *, int, void **);
+
+#define _nouveau_engine_dtor _nouveau_subdev_dtor
+#define _nouveau_engine_init _nouveau_subdev_init
+#define _nouveau_engine_fini _nouveau_subdev_fini
+
+#endif
diff --git a/nvkm/include/core/enum.h b/nvkm/include/core/enum.h
new file mode 100644
index 000000000..e7b1e1819
--- /dev/null
+++ b/nvkm/include/core/enum.h
@@ -0,0 +1,23 @@
+#ifndef __NOUVEAU_ENUM_H__
+#define __NOUVEAU_ENUM_H__
+
+struct nouveau_enum {
+	u32 value;
+	const char *name;
+	const void *data;
+};
+
+const struct nouveau_enum *
+nouveau_enum_find(const struct nouveau_enum *, u32 value);
+
+void
+nouveau_enum_print(const struct nouveau_enum *en, u32 value);
+
+struct nouveau_bitfield {
+	u32 mask;
+	const char *name;
+};
+
+void nouveau_bitfield_print(const struct nouveau_bitfield *, u32 value);
+
+#endif
diff --git a/nvkm/include/core/falcon.h b/nvkm/include/core/falcon.h
new file mode 100644
index 000000000..1edec386a
--- /dev/null
+++ b/nvkm/include/core/falcon.h
@@ -0,0 +1,81 @@
+#ifndef __NOUVEAU_FALCON_H__
+#define __NOUVEAU_FALCON_H__
+
+#include <core/engine.h>
+#include <core/engctx.h>
+#include <core/gpuobj.h>
+
+struct nouveau_falcon_chan {
+	struct nouveau_engctx base;
+};
+
+#define nouveau_falcon_context_create(p,e,c,g,s,a,f,d)                         \
+	nouveau_engctx_create((p), (e), (c), (g), (s), (a), (f), (d))
+#define nouveau_falcon_context_destroy(d)                                      \
+	nouveau_engctx_destroy(&(d)->base)
+#define nouveau_falcon_context_init(d)                                         \
+	nouveau_engctx_init(&(d)->base)
+#define nouveau_falcon_context_fini(d,s)                                       \
+	nouveau_engctx_fini(&(d)->base, (s))
+
+#define _nouveau_falcon_context_ctor _nouveau_engctx_ctor
+#define _nouveau_falcon_context_dtor _nouveau_engctx_dtor
+#define _nouveau_falcon_context_init _nouveau_engctx_init
+#define _nouveau_falcon_context_fini _nouveau_engctx_fini
+#define _nouveau_falcon_context_rd32 _nouveau_engctx_rd32
+#define _nouveau_falcon_context_wr32 _nouveau_engctx_wr32
+
+struct nouveau_falcon_data {
+	bool external;
+};
+
+struct nouveau_falcon {
+	struct nouveau_engine base;
+
+	u32 addr;
+	u8  version;
+	u8  secret;
+
+	struct nouveau_gpuobj *core;
+	bool external;
+
+	struct {
+		u32 limit;
+		u32 *data;
+		u32  size;
+	} code;
+
+	struct {
+		u32 limit;
+		u32 *data;
+		u32  size;
+	} data;
+};
+
+#define nv_falcon(priv) (&(priv)->base)
+
+#define nouveau_falcon_create(p,e,c,b,d,i,f,r)                                 \
+	nouveau_falcon_create_((p), (e), (c), (b), (d), (i), (f),              \
+			       sizeof(**r),(void **)r)
+#define nouveau_falcon_destroy(p)                                              \
+	nouveau_engine_destroy(&(p)->base)
+#define nouveau_falcon_init(p) ({                                              \
+	struct nouveau_falcon *falcon = (p);                                   \
+	_nouveau_falcon_init(nv_object(falcon));                               \
+})
+#define nouveau_falcon_fini(p,s) ({                                            \
+	struct nouveau_falcon *falcon = (p);                                   \
+	_nouveau_falcon_fini(nv_object(falcon), (s));                          \
+})
+
+int nouveau_falcon_create_(struct nouveau_object *, struct nouveau_object *,
+			   struct nouveau_oclass *, u32, bool, const char *,
+			   const char *, int, void **);
+
+#define _nouveau_falcon_dtor _nouveau_engine_dtor
+int  _nouveau_falcon_init(struct nouveau_object *);
+int  _nouveau_falcon_fini(struct nouveau_object *, bool);
+u32  _nouveau_falcon_rd32(struct nouveau_object *, u64);
+void _nouveau_falcon_wr32(struct nouveau_object *, u64, u32);
+
+#endif
diff --git a/nvkm/include/core/gpuobj.h b/nvkm/include/core/gpuobj.h
new file mode 100644
index 000000000..b3b9ce4e9
--- /dev/null
+++ b/nvkm/include/core/gpuobj.h
@@ -0,0 +1,71 @@
+#ifndef __NOUVEAU_GPUOBJ_H__
+#define __NOUVEAU_GPUOBJ_H__
+
+#include <core/object.h>
+#include <core/device.h>
+#include <core/parent.h>
+#include <core/mm.h>
+
+struct nouveau_vma;
+struct nouveau_vm;
+
+#define NVOBJ_FLAG_ZERO_ALLOC 0x00000001
+#define NVOBJ_FLAG_ZERO_FREE  0x00000002
+#define NVOBJ_FLAG_HEAP       0x00000004
+
+struct nouveau_gpuobj {
+	struct nouveau_object base;
+	struct nouveau_object *parent;
+	struct nouveau_mm_node *node;
+	struct nouveau_mm heap;
+
+	u32 flags;
+	u64 addr;
+	u32 size;
+};
+
+static inline struct nouveau_gpuobj *
+nv_gpuobj(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!nv_iclass(obj, NV_GPUOBJ_CLASS)))
+		nv_assert("BAD CAST -> NvGpuObj, %08x", nv_hclass(obj));
+#endif
+	return obj;
+}
+
+#define nouveau_gpuobj_create(p,e,c,v,g,s,a,f,d)                               \
+	nouveau_gpuobj_create_((p), (e), (c), (v), (g), (s), (a), (f),         \
+			       sizeof(**d), (void **)d)
+#define nouveau_gpuobj_init(p) nouveau_object_init(&(p)->base)
+#define nouveau_gpuobj_fini(p,s) nouveau_object_fini(&(p)->base, (s))
+int  nouveau_gpuobj_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, u32 pclass,
+			    struct nouveau_object *, u32 size, u32 align,
+			    u32 flags, int length, void **);
+void nouveau_gpuobj_destroy(struct nouveau_gpuobj *);
+
+int nouveau_gpuobj_new(struct nouveau_object *, struct nouveau_object *,
+		       u32 size, u32 align, u32 flags,
+		       struct nouveau_gpuobj **);
+int nouveau_gpuobj_dup(struct nouveau_object *, struct nouveau_gpuobj *,
+		       struct nouveau_gpuobj **);
+
+int nouveau_gpuobj_map(struct nouveau_gpuobj *, u32 acc, struct nouveau_vma *);
+int nouveau_gpuobj_map_vm(struct nouveau_gpuobj *, struct nouveau_vm *,
+			  u32 access, struct nouveau_vma *);
+void nouveau_gpuobj_unmap(struct nouveau_vma *);
+
+static inline void
+nouveau_gpuobj_ref(struct nouveau_gpuobj *obj, struct nouveau_gpuobj **ref)
+{
+	nouveau_object_ref(&obj->base, (struct nouveau_object **)ref);
+}
+
+void _nouveau_gpuobj_dtor(struct nouveau_object *);
+int  _nouveau_gpuobj_init(struct nouveau_object *);
+int  _nouveau_gpuobj_fini(struct nouveau_object *, bool);
+u32  _nouveau_gpuobj_rd32(struct nouveau_object *, u64);
+void _nouveau_gpuobj_wr32(struct nouveau_object *, u64, u32);
+
+#endif
diff --git a/nvkm/include/core/handle.h b/nvkm/include/core/handle.h
new file mode 100644
index 000000000..363674cdf
--- /dev/null
+++ b/nvkm/include/core/handle.h
@@ -0,0 +1,31 @@
+#ifndef __NOUVEAU_HANDLE_H__
+#define __NOUVEAU_HANDLE_H__
+
+struct nouveau_handle {
+	struct nouveau_namedb *namedb;
+	struct list_head node;
+
+	struct list_head head;
+	struct list_head tree;
+	u32 name;
+	u32 priv;
+
+	struct nouveau_handle *parent;
+	struct nouveau_object *object;
+};
+
+int  nouveau_handle_create(struct nouveau_object *, u32 parent, u32 handle,
+			   struct nouveau_object *, struct nouveau_handle **);
+void nouveau_handle_destroy(struct nouveau_handle *);
+int  nouveau_handle_init(struct nouveau_handle *);
+int  nouveau_handle_fini(struct nouveau_handle *, bool suspend);
+
+struct nouveau_object *
+nouveau_handle_ref(struct nouveau_object *, u32 name);
+
+struct nouveau_handle *nouveau_handle_get_class(struct nouveau_object *, u16);
+struct nouveau_handle *nouveau_handle_get_vinst(struct nouveau_object *, u64);
+struct nouveau_handle *nouveau_handle_get_cinst(struct nouveau_object *, u32);
+void nouveau_handle_put(struct nouveau_handle *);
+
+#endif
diff --git a/nvkm/include/core/intr.h b/nvkm/include/core/intr.h
new file mode 100644
index 000000000..d2599746e
--- /dev/null
+++ b/nvkm/include/core/intr.h
@@ -0,0 +1,8 @@
+#ifndef __NOUVEAU_INTR_H__
+#define __NOUVEAU_INTR_H__
+
+struct nouveau_device;
+
+void nouveau_intr(struct nouveau_device *);
+
+#endif
diff --git a/nvkm/include/core/math.h b/nvkm/include/core/math.h
new file mode 100644
index 000000000..f808131c5
--- /dev/null
+++ b/nvkm/include/core/math.h
@@ -0,0 +1,16 @@
+#ifndef __NOUVEAU_MATH_H__
+#define __NOUVEAU_MATH_H__
+
+static inline int
+log2i(u64 base)
+{
+	u64 temp = base >> 1;
+	int log2;
+
+	for (log2 = 0; temp; log2++, temp >>= 1) {
+	}
+
+	return (base & (base - 1)) ? log2 + 1: log2;
+}
+
+#endif
diff --git a/nvkm/include/core/memobj.h b/nvkm/include/core/memobj.h
new file mode 100644
index 000000000..7ca7d9320
--- /dev/null
+++ b/nvkm/include/core/memobj.h
@@ -0,0 +1,32 @@
+#ifndef __NOUVEAU_MEMOBJ_H__
+#define __NOUVEAU_MEMOBJ_H__
+
+#include <core/object.h>
+#include <core/mm.h>
+
+struct nouveau_memobj {
+	struct nouveau_object base;
+	struct nouveau_mm heap;
+	u32 *suspend;
+	u64 addr;
+};
+
+static inline struct nouveau_memobj *
+nv_memobj(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!nv_iclass(obj, NV_MEMOBJ_CLASS)))
+		nv_assert("BAD CAST -> NvMemObj, %08x", nv_hclass(obj));
+#endif
+	return obj;
+}
+
+#define nouveau_memobj_create(p,e,c,d)                                         \
+	nouveau_object_create_((p), (e), (c), NV_MEMOBJ_CLASS,                 \
+			       sizeof(**d), (void **)d)
+#define nouveau_memobj_destroy(p) nouveau_object_destroy(&(p)->base)
+
+int nouveau_memobj_init(struct nouveau_memobj *);
+int nouveau_memobj_fini(struct nouveau_memobj *, bool suspend);
+
+#endif
diff --git a/nvkm/include/core/mm.h b/nvkm/include/core/mm.h
new file mode 100644
index 000000000..2514e81ad
--- /dev/null
+++ b/nvkm/include/core/mm.h
@@ -0,0 +1,38 @@
+#ifndef __NOUVEAU_MM_H__
+#define __NOUVEAU_MM_H__
+
+struct nouveau_mm_node {
+	struct list_head nl_entry;
+	struct list_head fl_entry;
+	struct list_head rl_entry;
+
+	u8  type;
+	u32 offset;
+	u32 length;
+};
+
+struct nouveau_mm {
+	struct list_head nodes;
+	struct list_head free;
+
+	struct mutex mutex;
+
+	u32 block_size;
+	int heap_nodes;
+};
+
+static inline bool
+nouveau_mm_initialised(struct nouveau_mm *mm)
+{
+	return mm->block_size != 0;
+}
+
+int  nouveau_mm_init(struct nouveau_mm *, u32 offset, u32 length, u32 block);
+int  nouveau_mm_fini(struct nouveau_mm *);
+int  nouveau_mm_head(struct nouveau_mm *, u8 type, u32 size_max, u32 size_min,
+		     u32 align, struct nouveau_mm_node **);
+int  nouveau_mm_tail(struct nouveau_mm *, u8 type, u32 size_max, u32 size_min,
+		     u32 align, struct nouveau_mm_node **);
+void nouveau_mm_free(struct nouveau_mm *, struct nouveau_mm_node **);
+
+#endif
diff --git a/nvkm/include/core/namedb.h b/nvkm/include/core/namedb.h
new file mode 100644
index 000000000..8897e0886
--- /dev/null
+++ b/nvkm/include/core/namedb.h
@@ -0,0 +1,56 @@
+#ifndef __NOUVEAU_NAMEDB_H__
+#define __NOUVEAU_NAMEDB_H__
+
+#include <core/parent.h>
+
+struct nouveau_handle;
+
+struct nouveau_namedb {
+	struct nouveau_parent base;
+	rwlock_t lock;
+	struct list_head list;
+};
+
+static inline struct nouveau_namedb *
+nv_namedb(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!nv_iclass(obj, NV_NAMEDB_CLASS)))
+		nv_assert("BAD CAST -> NvNameDB, %08x", nv_hclass(obj));
+#endif
+	return obj;
+}
+
+#define nouveau_namedb_create(p,e,c,v,s,m,d)                                   \
+	nouveau_namedb_create_((p), (e), (c), (v), (s), (m),                   \
+			       sizeof(**d), (void **)d)
+#define nouveau_namedb_init(p)                                                 \
+	nouveau_parent_init(&(p)->base)
+#define nouveau_namedb_fini(p,s)                                               \
+	nouveau_parent_fini(&(p)->base, (s))
+#define nouveau_namedb_destroy(p)                                              \
+	nouveau_parent_destroy(&(p)->base)
+
+int  nouveau_namedb_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, u32 pclass,
+			    struct nouveau_oclass *, u32 engcls,
+			    int size, void **);
+
+int  _nouveau_namedb_ctor(struct nouveau_object *, struct nouveau_object *,
+			  struct nouveau_oclass *, void *, u32,
+			  struct nouveau_object **);
+#define _nouveau_namedb_dtor _nouveau_parent_dtor
+#define _nouveau_namedb_init _nouveau_parent_init
+#define _nouveau_namedb_fini _nouveau_parent_fini
+
+int  nouveau_namedb_insert(struct nouveau_namedb *, u32 name,
+			   struct nouveau_object *, struct nouveau_handle *);
+void nouveau_namedb_remove(struct nouveau_handle *);
+
+struct nouveau_handle *nouveau_namedb_get(struct nouveau_namedb *, u32);
+struct nouveau_handle *nouveau_namedb_get_class(struct nouveau_namedb *, u16);
+struct nouveau_handle *nouveau_namedb_get_vinst(struct nouveau_namedb *, u64);
+struct nouveau_handle *nouveau_namedb_get_cinst(struct nouveau_namedb *, u32);
+void nouveau_namedb_put(struct nouveau_handle *);
+
+#endif
diff --git a/nvkm/include/core/object.h b/nvkm/include/core/object.h
new file mode 100644
index 000000000..5982935ee
--- /dev/null
+++ b/nvkm/include/core/object.h
@@ -0,0 +1,199 @@
+#ifndef __NOUVEAU_OBJECT_H__
+#define __NOUVEAU_OBJECT_H__
+
+#include <core/os.h>
+#include <core/printk.h>
+
+#define NV_PARENT_CLASS 0x80000000
+#define NV_NAMEDB_CLASS 0x40000000
+#define NV_CLIENT_CLASS 0x20000000
+#define NV_SUBDEV_CLASS 0x10000000
+#define NV_ENGINE_CLASS 0x08000000
+#define NV_MEMOBJ_CLASS 0x04000000
+#define NV_GPUOBJ_CLASS 0x02000000
+#define NV_ENGCTX_CLASS 0x01000000
+#define NV_OBJECT_CLASS 0x0000ffff
+
+struct nouveau_object {
+	struct nouveau_oclass *oclass;
+	struct nouveau_object *parent;
+	struct nouveau_object *engine;
+	atomic_t refcount;
+	atomic_t usecount;
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+#define NOUVEAU_OBJECT_MAGIC 0x75ef0bad
+	struct list_head list;
+	u32 _magic;
+#endif
+};
+
+static inline struct nouveau_object *
+nv_object(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (likely(obj)) {
+		struct nouveau_object *object = obj;
+		if (unlikely(object->_magic != NOUVEAU_OBJECT_MAGIC))
+			nv_assert("BAD CAST -> NvObject, invalid magic");
+	}
+#endif
+	return obj;
+}
+
+#define nouveau_object_create(p,e,c,s,d)                                       \
+	nouveau_object_create_((p), (e), (c), (s), sizeof(**d), (void **)d)
+int  nouveau_object_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, u32, int size, void **);
+void nouveau_object_destroy(struct nouveau_object *);
+int  nouveau_object_init(struct nouveau_object *);
+int  nouveau_object_fini(struct nouveau_object *, bool suspend);
+
+extern struct nouveau_ofuncs nouveau_object_ofuncs;
+
+struct nouveau_oclass {
+	u32 handle;
+	struct nouveau_ofuncs *ofuncs;
+	struct nouveau_omthds *omthds;
+};
+
+#define nv_oclass(o)    nv_object(o)->oclass
+#define nv_hclass(o)    nv_oclass(o)->handle
+#define nv_iclass(o,i) (nv_hclass(o) & (i))
+#define nv_mclass(o)    nv_iclass(o, NV_OBJECT_CLASS)
+
+static inline struct nouveau_object *
+nv_pclass(struct nouveau_object *parent, u32 oclass)
+{
+	while (parent && !nv_iclass(parent, oclass))
+		parent = parent->parent;
+	return parent;
+}
+
+struct nouveau_omthds {
+	u32 start;
+	u32 limit;
+	int (*call)(struct nouveau_object *, u32, void *, u32);
+};
+
+struct nouveau_ofuncs {
+	int  (*ctor)(struct nouveau_object *, struct nouveau_object *,
+		     struct nouveau_oclass *, void *data, u32 size,
+		     struct nouveau_object **);
+	void (*dtor)(struct nouveau_object *);
+	int  (*init)(struct nouveau_object *);
+	int  (*fini)(struct nouveau_object *, bool suspend);
+	u8   (*rd08)(struct nouveau_object *, u64 offset);
+	u16  (*rd16)(struct nouveau_object *, u64 offset);
+	u32  (*rd32)(struct nouveau_object *, u64 offset);
+	void (*wr08)(struct nouveau_object *, u64 offset, u8 data);
+	void (*wr16)(struct nouveau_object *, u64 offset, u16 data);
+	void (*wr32)(struct nouveau_object *, u64 offset, u32 data);
+};
+
+static inline struct nouveau_ofuncs *
+nv_ofuncs(void *obj)
+{
+	return nv_oclass(obj)->ofuncs;
+}
+
+int  nouveau_object_ctor(struct nouveau_object *, struct nouveau_object *,
+			 struct nouveau_oclass *, void *, u32,
+			 struct nouveau_object **);
+void nouveau_object_ref(struct nouveau_object *, struct nouveau_object **);
+int nouveau_object_inc(struct nouveau_object *);
+int nouveau_object_dec(struct nouveau_object *, bool suspend);
+
+int nouveau_object_new(struct nouveau_object *, u32 parent, u32 handle,
+		       u16 oclass, void *data, u32 size,
+		       struct nouveau_object **);
+int nouveau_object_del(struct nouveau_object *, u32 parent, u32 handle);
+void nouveau_object_debug(void);
+
+static inline int
+nv_exec(void *obj, u32 mthd, void *data, u32 size)
+{
+	struct nouveau_omthds *method = nv_oclass(obj)->omthds;
+
+	while (method && method->call) {
+		if (mthd >= method->start && mthd <= method->limit)
+			return method->call(obj, mthd, data, size);
+		method++;
+	}
+
+	return -EINVAL;
+}
+
+static inline int
+nv_call(void *obj, u32 mthd, u32 data)
+{
+	return nv_exec(obj, mthd, &data, sizeof(data));
+}
+
+static inline u8
+nv_ro08(void *obj, u64 addr)
+{
+	u8 data = nv_ofuncs(obj)->rd08(obj, addr);
+	nv_spam(obj, "nv_ro08 0x%08x 0x%02x\n", addr, data);
+	return data;
+}
+
+static inline u16
+nv_ro16(void *obj, u64 addr)
+{
+	u16 data = nv_ofuncs(obj)->rd16(obj, addr);
+	nv_spam(obj, "nv_ro16 0x%08x 0x%04x\n", addr, data);
+	return data;
+}
+
+static inline u32
+nv_ro32(void *obj, u64 addr)
+{
+	u32 data = nv_ofuncs(obj)->rd32(obj, addr);
+	nv_spam(obj, "nv_ro32 0x%08x 0x%08x\n", addr, data);
+	return data;
+}
+
+static inline void
+nv_wo08(void *obj, u64 addr, u8 data)
+{
+	nv_spam(obj, "nv_wo08 0x%08x 0x%02x\n", addr, data);
+	nv_ofuncs(obj)->wr08(obj, addr, data);
+}
+
+static inline void
+nv_wo16(void *obj, u64 addr, u16 data)
+{
+	nv_spam(obj, "nv_wo16 0x%08x 0x%04x\n", addr, data);
+	nv_ofuncs(obj)->wr16(obj, addr, data);
+}
+
+static inline void
+nv_wo32(void *obj, u64 addr, u32 data)
+{
+	nv_spam(obj, "nv_wo32 0x%08x 0x%08x\n", addr, data);
+	nv_ofuncs(obj)->wr32(obj, addr, data);
+}
+
+static inline u32
+nv_mo32(void *obj, u64 addr, u32 mask, u32 data)
+{
+	u32 temp = nv_ro32(obj, addr);
+	nv_wo32(obj, addr, (temp & ~mask) | data);
+	return temp;
+}
+
+static inline int
+nv_memcmp(void *obj, u32 addr, const char *str, u32 len)
+{
+	unsigned char c1, c2;
+
+	while (len--) {
+		c1 = nv_ro08(obj, addr++);
+		c2 = *(str++);
+		if (c1 != c2)
+			return c1 - c2;
+	}
+	return 0;
+}
+
+#endif
diff --git a/nvkm/include/core/option.h b/nvkm/include/core/option.h
new file mode 100644
index 000000000..27074957f
--- /dev/null
+++ b/nvkm/include/core/option.h
@@ -0,0 +1,11 @@
+#ifndef __NOUVEAU_OPTION_H__
+#define __NOUVEAU_OPTION_H__
+
+#include <core/os.h>
+
+const char *nouveau_stropt(const char *optstr, const char *opt, int *len);
+bool nouveau_boolopt(const char *optstr, const char *opt, bool value);
+
+int nouveau_dbgopt(const char *optstr, const char *sub);
+
+#endif
diff --git a/nvkm/include/core/parent.h b/nvkm/include/core/parent.h
new file mode 100644
index 000000000..31cd852c9
--- /dev/null
+++ b/nvkm/include/core/parent.h
@@ -0,0 +1,61 @@
+#ifndef __NOUVEAU_PARENT_H__
+#define __NOUVEAU_PARENT_H__
+
+#include <core/device.h>
+#include <core/object.h>
+
+struct nouveau_sclass {
+	struct nouveau_sclass *sclass;
+	struct nouveau_engine *engine;
+	struct nouveau_oclass *oclass;
+};
+
+struct nouveau_parent {
+	struct nouveau_object base;
+
+	struct nouveau_sclass *sclass;
+	u64 engine;
+
+	int  (*context_attach)(struct nouveau_object *,
+			       struct nouveau_object *);
+	int  (*context_detach)(struct nouveau_object *, bool suspend,
+			       struct nouveau_object *);
+
+	int  (*object_attach)(struct nouveau_object *parent,
+			      struct nouveau_object *object, u32 name);
+	void (*object_detach)(struct nouveau_object *parent, int cookie);
+};
+
+static inline struct nouveau_parent *
+nv_parent(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!(nv_iclass(obj, NV_PARENT_CLASS))))
+		nv_assert("BAD CAST -> NvParent, %08x", nv_hclass(obj));
+#endif
+	return obj;
+}
+
+#define nouveau_parent_create(p,e,c,v,s,m,d)                                   \
+	nouveau_parent_create_((p), (e), (c), (v), (s), (m),                   \
+			       sizeof(**d), (void **)d)
+#define nouveau_parent_init(p)                                                 \
+	nouveau_object_init(&(p)->base)
+#define nouveau_parent_fini(p,s)                                               \
+	nouveau_object_fini(&(p)->base, (s))
+
+int  nouveau_parent_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, u32 pclass,
+			    struct nouveau_oclass *, u64 engcls,
+			    int size, void **);
+void nouveau_parent_destroy(struct nouveau_parent *);
+
+void _nouveau_parent_dtor(struct nouveau_object *);
+#define _nouveau_parent_init _nouveau_object_init
+#define _nouveau_parent_fini _nouveau_object_fini
+
+int nouveau_parent_sclass(struct nouveau_object *, u16 handle,
+			  struct nouveau_object **pengine,
+			  struct nouveau_oclass **poclass);
+
+#endif
diff --git a/nvkm/include/core/printk.h b/nvkm/include/core/printk.h
new file mode 100644
index 000000000..1d629664f
--- /dev/null
+++ b/nvkm/include/core/printk.h
@@ -0,0 +1,39 @@
+#ifndef __NOUVEAU_PRINTK_H__
+#define __NOUVEAU_PRINTK_H__
+
+#include <core/os.h>
+#include <core/debug.h>
+
+struct nouveau_object;
+
+#define NV_PRINTK_FATAL    KERN_CRIT
+#define NV_PRINTK_ERROR    KERN_ERR
+#define NV_PRINTK_WARN     KERN_WARNING
+#define NV_PRINTK_INFO     KERN_INFO
+#define NV_PRINTK_DEBUG    KERN_DEBUG
+#define NV_PRINTK_PARANOIA KERN_DEBUG
+#define NV_PRINTK_TRACE    KERN_DEBUG
+#define NV_PRINTK_SPAM     KERN_DEBUG
+
+void nv_printk_(struct nouveau_object *, const char *, int, const char *, ...);
+
+#define nv_printk(o,l,f,a...) do {                                             \
+	if (NV_DBG_##l <= CONFIG_NOUVEAU_DEBUG)                                \
+		nv_printk_(nv_object(o), NV_PRINTK_##l, NV_DBG_##l, f, ##a);   \
+} while(0)
+
+#define nv_fatal(o,f,a...) nv_printk((o), FATAL, f, ##a)
+#define nv_error(o,f,a...) nv_printk((o), ERROR, f, ##a)
+#define nv_warn(o,f,a...) nv_printk((o), WARN, f, ##a)
+#define nv_info(o,f,a...) nv_printk((o), INFO, f, ##a)
+#define nv_debug(o,f,a...) nv_printk((o), DEBUG, f, ##a)
+#define nv_trace(o,f,a...) nv_printk((o), TRACE, f, ##a)
+#define nv_spam(o,f,a...) nv_printk((o), SPAM, f, ##a)
+
+#define nv_assert(f,a...) do {                                                 \
+	if (NV_DBG_FATAL <= CONFIG_NOUVEAU_DEBUG)                              \
+		nv_printk_(NULL, NV_PRINTK_FATAL, NV_DBG_FATAL, f "\n", ##a);  \
+	BUG_ON(1);                                                             \
+} while(0)
+
+#endif
diff --git a/nvkm/include/core/ramht.h b/nvkm/include/core/ramht.h
new file mode 100644
index 000000000..47e4cacbc
--- /dev/null
+++ b/nvkm/include/core/ramht.h
@@ -0,0 +1,23 @@
+#ifndef __NOUVEAU_RAMHT_H__
+#define __NOUVEAU_RAMHT_H__
+
+#include <core/gpuobj.h>
+
+struct nouveau_ramht {
+	struct nouveau_gpuobj base;
+	int bits;
+};
+
+int  nouveau_ramht_insert(struct nouveau_ramht *, int chid,
+			  u32 handle, u32 context);
+void nouveau_ramht_remove(struct nouveau_ramht *, int cookie);
+int  nouveau_ramht_new(struct nouveau_object *, struct nouveau_object *,
+		       u32 size, u32 align, struct nouveau_ramht **);
+
+static inline void
+nouveau_ramht_ref(struct nouveau_ramht *obj, struct nouveau_ramht **ref)
+{
+	nouveau_gpuobj_ref(&obj->base, (struct nouveau_gpuobj **)ref);
+}
+
+#endif
diff --git a/nvkm/include/core/subdev.h b/nvkm/include/core/subdev.h
new file mode 100644
index 000000000..e9632e931
--- /dev/null
+++ b/nvkm/include/core/subdev.h
@@ -0,0 +1,118 @@
+#ifndef __NOUVEAU_SUBDEV_H__
+#define __NOUVEAU_SUBDEV_H__
+
+#include <core/object.h>
+
+#define NV_SUBDEV_(sub,var) (NV_SUBDEV_CLASS | ((var) << 8) | (sub))
+#define NV_SUBDEV(name,var)  NV_SUBDEV_(NVDEV_SUBDEV_##name, (var))
+
+struct nouveau_subdev {
+	struct nouveau_object base;
+	struct mutex mutex;
+	const char *name;
+	void __iomem *mmio;
+	u32 debug;
+	u32 unit;
+
+	void (*intr)(struct nouveau_subdev *);
+};
+
+static inline struct nouveau_subdev *
+nv_subdev(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!nv_iclass(obj, NV_SUBDEV_CLASS)))
+		nv_assert("BAD CAST -> NvSubDev, %08x", nv_hclass(obj));
+#endif
+	return obj;
+}
+
+static inline int
+nv_subidx(struct nouveau_object *object)
+{
+	return nv_hclass(nv_subdev(object)) & 0xff;
+}
+
+#define nouveau_subdev_create(p,e,o,v,s,f,d)                                   \
+	nouveau_subdev_create_((p), (e), (o), (v), (s), (f),                   \
+			       sizeof(**d),(void **)d)
+
+int  nouveau_subdev_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, u32 pclass,
+			    const char *sname, const char *fname,
+			    int size, void **);
+void nouveau_subdev_destroy(struct nouveau_subdev *);
+int  nouveau_subdev_init(struct nouveau_subdev *);
+int  nouveau_subdev_fini(struct nouveau_subdev *, bool suspend);
+void nouveau_subdev_reset(struct nouveau_object *);
+
+void _nouveau_subdev_dtor(struct nouveau_object *);
+int  _nouveau_subdev_init(struct nouveau_object *);
+int  _nouveau_subdev_fini(struct nouveau_object *, bool suspend);
+
+#define s_printk(s,l,f,a...) do {                                              \
+	if ((s)->debug >= OS_DBG_##l) {                                        \
+		nv_printk((s)->base.parent, (s)->name, l, f, ##a);             \
+	}                                                                      \
+} while(0)
+
+static inline u8
+nv_rd08(void *obj, u32 addr)
+{
+	struct nouveau_subdev *subdev = nv_subdev(obj);
+	u8 data = ioread8(subdev->mmio + addr);
+	nv_spam(subdev, "nv_rd08 0x%06x 0x%02x\n", addr, data);
+	return data;
+}
+
+static inline u16
+nv_rd16(void *obj, u32 addr)
+{
+	struct nouveau_subdev *subdev = nv_subdev(obj);
+	u16 data = ioread16_native(subdev->mmio + addr);
+	nv_spam(subdev, "nv_rd16 0x%06x 0x%04x\n", addr, data);
+	return data;
+}
+
+static inline u32
+nv_rd32(void *obj, u32 addr)
+{
+	struct nouveau_subdev *subdev = nv_subdev(obj);
+	u32 data = ioread32_native(subdev->mmio + addr);
+	nv_spam(subdev, "nv_rd32 0x%06x 0x%08x\n", addr, data);
+	return data;
+}
+
+static inline void
+nv_wr08(void *obj, u32 addr, u8 data)
+{
+	struct nouveau_subdev *subdev = nv_subdev(obj);
+	nv_spam(subdev, "nv_wr08 0x%06x 0x%02x\n", addr, data);
+	iowrite8(data, subdev->mmio + addr);
+}
+
+static inline void
+nv_wr16(void *obj, u32 addr, u16 data)
+{
+	struct nouveau_subdev *subdev = nv_subdev(obj);
+	nv_spam(subdev, "nv_wr16 0x%06x 0x%04x\n", addr, data);
+	iowrite16_native(data, subdev->mmio + addr);
+}
+
+static inline void
+nv_wr32(void *obj, u32 addr, u32 data)
+{
+	struct nouveau_subdev *subdev = nv_subdev(obj);
+	nv_spam(subdev, "nv_wr32 0x%06x 0x%08x\n", addr, data);
+	iowrite32_native(data, subdev->mmio + addr);
+}
+
+static inline u32
+nv_mask(void *obj, u32 addr, u32 mask, u32 data)
+{
+	u32 temp = nv_rd32(obj, addr);
+	nv_wr32(obj, addr, (temp & ~mask) | data);
+	return temp;
+}
+
+#endif
diff --git a/nvkm/include/engine/bsp.h b/nvkm/include/engine/bsp.h
new file mode 100644
index 000000000..13ccdf54d
--- /dev/null
+++ b/nvkm/include/engine/bsp.h
@@ -0,0 +1,8 @@
+#ifndef __NOUVEAU_BSP_H__
+#define __NOUVEAU_BSP_H__
+
+extern struct nouveau_oclass nv84_bsp_oclass;
+extern struct nouveau_oclass nvc0_bsp_oclass;
+extern struct nouveau_oclass nve0_bsp_oclass;
+
+#endif
diff --git a/nvkm/include/engine/copy.h b/nvkm/include/engine/copy.h
new file mode 100644
index 000000000..8cad2cf28
--- /dev/null
+++ b/nvkm/include/engine/copy.h
@@ -0,0 +1,12 @@
+#ifndef __NOUVEAU_COPY_H__
+#define __NOUVEAU_COPY_H__
+
+void nva3_copy_intr(struct nouveau_subdev *);
+
+extern struct nouveau_oclass nva3_copy_oclass;
+extern struct nouveau_oclass nvc0_copy0_oclass;
+extern struct nouveau_oclass nvc0_copy1_oclass;
+extern struct nouveau_oclass nve0_copy0_oclass;
+extern struct nouveau_oclass nve0_copy1_oclass;
+
+#endif
diff --git a/nvkm/include/engine/crypt.h b/nvkm/include/engine/crypt.h
new file mode 100644
index 000000000..db975618e
--- /dev/null
+++ b/nvkm/include/engine/crypt.h
@@ -0,0 +1,7 @@
+#ifndef __NOUVEAU_CRYPT_H__
+#define __NOUVEAU_CRYPT_H__
+
+extern struct nouveau_oclass nv84_crypt_oclass;
+extern struct nouveau_oclass nv98_crypt_oclass;
+
+#endif
diff --git a/nvkm/include/engine/disp.h b/nvkm/include/engine/disp.h
new file mode 100644
index 000000000..46948285f
--- /dev/null
+++ b/nvkm/include/engine/disp.h
@@ -0,0 +1,49 @@
+#ifndef __NOUVEAU_DISP_H__
+#define __NOUVEAU_DISP_H__
+
+#include <core/object.h>
+#include <core/engine.h>
+#include <core/device.h>
+
+struct nouveau_disp {
+	struct nouveau_engine base;
+
+	struct {
+		struct list_head list;
+		spinlock_t lock;
+		void (*notify)(void *, int);
+		void (*get)(void *, int);
+		void (*put)(void *, int);
+		void *data;
+	} vblank;
+};
+
+static inline struct nouveau_disp *
+nouveau_disp(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_ENGINE_DISP];
+}
+
+#define nouveau_disp_create(p,e,c,i,x,d)                                       \
+	nouveau_engine_create((p), (e), (c), true, (i), (x), (d))
+#define nouveau_disp_destroy(d)                                                \
+	nouveau_engine_destroy(&(d)->base)
+#define nouveau_disp_init(d)                                                   \
+	nouveau_engine_init(&(d)->base)
+#define nouveau_disp_fini(d,s)                                                 \
+	nouveau_engine_fini(&(d)->base, (s))
+
+#define _nouveau_disp_dtor _nouveau_engine_dtor
+#define _nouveau_disp_init _nouveau_engine_init
+#define _nouveau_disp_fini _nouveau_engine_fini
+
+extern struct nouveau_oclass nv04_disp_oclass;
+extern struct nouveau_oclass nv50_disp_oclass;
+extern struct nouveau_oclass nv84_disp_oclass;
+extern struct nouveau_oclass nva0_disp_oclass;
+extern struct nouveau_oclass nv94_disp_oclass;
+extern struct nouveau_oclass nva3_disp_oclass;
+extern struct nouveau_oclass nvd0_disp_oclass;
+extern struct nouveau_oclass nve0_disp_oclass;
+
+#endif
diff --git a/nvkm/include/engine/dmaobj.h b/nvkm/include/engine/dmaobj.h
new file mode 100644
index 000000000..b28914ed1
--- /dev/null
+++ b/nvkm/include/engine/dmaobj.h
@@ -0,0 +1,48 @@
+#ifndef __NOUVEAU_DMAOBJ_H__
+#define __NOUVEAU_DMAOBJ_H__
+
+#include <core/object.h>
+#include <core/engine.h>
+
+struct nouveau_gpuobj;
+
+struct nouveau_dmaobj {
+	struct nouveau_object base;
+	u32 target;
+	u32 access;
+	u64 start;
+	u64 limit;
+	u32 conf0;
+};
+
+struct nouveau_dmaeng {
+	struct nouveau_engine base;
+
+	/* creates a "physical" dma object from a struct nouveau_dmaobj */
+	int (*bind)(struct nouveau_dmaeng *dmaeng,
+		    struct nouveau_object *parent,
+		    struct nouveau_dmaobj *dmaobj,
+		    struct nouveau_gpuobj **);
+};
+
+#define nouveau_dmaeng_create(p,e,c,d)                                         \
+	nouveau_engine_create((p), (e), (c), true, "DMAOBJ", "dmaobj", (d))
+#define nouveau_dmaeng_destroy(p)                                              \
+	nouveau_engine_destroy(&(p)->base)
+#define nouveau_dmaeng_init(p)                                                 \
+	nouveau_engine_init(&(p)->base)
+#define nouveau_dmaeng_fini(p,s)                                               \
+	nouveau_engine_fini(&(p)->base, (s))
+
+#define _nouveau_dmaeng_dtor _nouveau_engine_dtor
+#define _nouveau_dmaeng_init _nouveau_engine_init
+#define _nouveau_dmaeng_fini _nouveau_engine_fini
+
+extern struct nouveau_oclass nv04_dmaeng_oclass;
+extern struct nouveau_oclass nv50_dmaeng_oclass;
+extern struct nouveau_oclass nvc0_dmaeng_oclass;
+extern struct nouveau_oclass nvd0_dmaeng_oclass;
+
+extern struct nouveau_oclass nouveau_dmaobj_sclass[];
+
+#endif
diff --git a/nvkm/include/engine/fifo.h b/nvkm/include/engine/fifo.h
new file mode 100644
index 000000000..f18846c8c
--- /dev/null
+++ b/nvkm/include/engine/fifo.h
@@ -0,0 +1,111 @@
+#ifndef __NOUVEAU_FIFO_H__
+#define __NOUVEAU_FIFO_H__
+
+#include <core/namedb.h>
+#include <core/gpuobj.h>
+#include <core/engine.h>
+
+struct nouveau_fifo_chan {
+	struct nouveau_namedb base;
+	struct nouveau_dmaobj *pushdma;
+	struct nouveau_gpuobj *pushgpu;
+	void __iomem *user;
+	u32 size;
+	u16 chid;
+	atomic_t refcnt; /* NV04_NVSW_SET_REF */
+};
+
+static inline struct nouveau_fifo_chan *
+nouveau_fifo_chan(void *obj)
+{
+	return (void *)nv_namedb(obj);
+}
+
+#define nouveau_fifo_channel_create(p,e,c,b,a,s,n,m,d)                         \
+	nouveau_fifo_channel_create_((p), (e), (c), (b), (a), (s), (n),        \
+				     (m), sizeof(**d), (void **)d)
+#define nouveau_fifo_channel_init(p)                                           \
+	nouveau_namedb_init(&(p)->base)
+#define nouveau_fifo_channel_fini(p,s)                                         \
+	nouveau_namedb_fini(&(p)->base, (s))
+
+int  nouveau_fifo_channel_create_(struct nouveau_object *,
+				  struct nouveau_object *,
+				  struct nouveau_oclass *,
+				  int bar, u32 addr, u32 size, u32 push,
+				  u64 engmask, int len, void **);
+void nouveau_fifo_channel_destroy(struct nouveau_fifo_chan *);
+
+#define _nouveau_fifo_channel_init _nouveau_namedb_init
+#define _nouveau_fifo_channel_fini _nouveau_namedb_fini
+
+void _nouveau_fifo_channel_dtor(struct nouveau_object *);
+u32  _nouveau_fifo_channel_rd32(struct nouveau_object *, u64);
+void _nouveau_fifo_channel_wr32(struct nouveau_object *, u64, u32);
+
+struct nouveau_fifo_base {
+	struct nouveau_gpuobj base;
+};
+
+#define nouveau_fifo_context_create(p,e,c,g,s,a,f,d)                           \
+	nouveau_gpuobj_create((p), (e), (c), 0, (g), (s), (a), (f), (d))
+#define nouveau_fifo_context_destroy(p)                                        \
+	nouveau_gpuobj_destroy(&(p)->base)
+#define nouveau_fifo_context_init(p)                                           \
+	nouveau_gpuobj_init(&(p)->base)
+#define nouveau_fifo_context_fini(p,s)                                         \
+	nouveau_gpuobj_fini(&(p)->base, (s))
+
+#define _nouveau_fifo_context_dtor _nouveau_gpuobj_dtor
+#define _nouveau_fifo_context_init _nouveau_gpuobj_init
+#define _nouveau_fifo_context_fini _nouveau_gpuobj_fini
+#define _nouveau_fifo_context_rd32 _nouveau_gpuobj_rd32
+#define _nouveau_fifo_context_wr32 _nouveau_gpuobj_wr32
+
+struct nouveau_fifo {
+	struct nouveau_engine base;
+
+	struct nouveau_object **channel;
+	spinlock_t lock;
+	u16 min;
+	u16 max;
+
+	int  (*chid)(struct nouveau_fifo *, struct nouveau_object *);
+	void (*pause)(struct nouveau_fifo *, unsigned long *);
+	void (*start)(struct nouveau_fifo *, unsigned long *);
+};
+
+static inline struct nouveau_fifo *
+nouveau_fifo(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_ENGINE_FIFO];
+}
+
+#define nouveau_fifo_create(o,e,c,fc,lc,d)                                     \
+	nouveau_fifo_create_((o), (e), (c), (fc), (lc), sizeof(**d), (void **)d)
+#define nouveau_fifo_init(p)                                                   \
+	nouveau_engine_init(&(p)->base)
+#define nouveau_fifo_fini(p,s)                                                 \
+	nouveau_engine_fini(&(p)->base, (s))
+
+int nouveau_fifo_create_(struct nouveau_object *, struct nouveau_object *,
+			 struct nouveau_oclass *, int min, int max,
+			 int size, void **);
+void nouveau_fifo_destroy(struct nouveau_fifo *);
+
+#define _nouveau_fifo_init _nouveau_engine_init
+#define _nouveau_fifo_fini _nouveau_engine_fini
+
+extern struct nouveau_oclass nv04_fifo_oclass;
+extern struct nouveau_oclass nv10_fifo_oclass;
+extern struct nouveau_oclass nv17_fifo_oclass;
+extern struct nouveau_oclass nv40_fifo_oclass;
+extern struct nouveau_oclass nv50_fifo_oclass;
+extern struct nouveau_oclass nv84_fifo_oclass;
+extern struct nouveau_oclass nvc0_fifo_oclass;
+extern struct nouveau_oclass nve0_fifo_oclass;
+
+void nv04_fifo_intr(struct nouveau_subdev *);
+int  nv04_fifo_context_attach(struct nouveau_object *, struct nouveau_object *);
+
+#endif
diff --git a/nvkm/include/engine/graph.h b/nvkm/include/engine/graph.h
new file mode 100644
index 000000000..6943b40d0
--- /dev/null
+++ b/nvkm/include/engine/graph.h
@@ -0,0 +1,72 @@
+#ifndef __NOUVEAU_GRAPH_H__
+#define __NOUVEAU_GRAPH_H__
+
+#include <core/engine.h>
+#include <core/engctx.h>
+#include <core/enum.h>
+
+struct nouveau_graph_chan {
+	struct nouveau_engctx base;
+};
+
+#define nouveau_graph_context_create(p,e,c,g,s,a,f,d)                          \
+	nouveau_engctx_create((p), (e), (c), (g), (s), (a), (f), (d))
+#define nouveau_graph_context_destroy(d)                                       \
+	nouveau_engctx_destroy(&(d)->base)
+#define nouveau_graph_context_init(d)                                          \
+	nouveau_engctx_init(&(d)->base)
+#define nouveau_graph_context_fini(d,s)                                        \
+	nouveau_engctx_fini(&(d)->base, (s))
+
+#define _nouveau_graph_context_dtor _nouveau_engctx_dtor
+#define _nouveau_graph_context_init _nouveau_engctx_init
+#define _nouveau_graph_context_fini _nouveau_engctx_fini
+#define _nouveau_graph_context_rd32 _nouveau_engctx_rd32
+#define _nouveau_graph_context_wr32 _nouveau_engctx_wr32
+
+struct nouveau_graph {
+	struct nouveau_engine base;
+};
+
+static inline struct nouveau_graph *
+nouveau_graph(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_ENGINE_GR];
+}
+
+#define nouveau_graph_create(p,e,c,y,d)                                        \
+	nouveau_engine_create((p), (e), (c), (y), "PGRAPH", "graphics", (d))
+#define nouveau_graph_destroy(d)                                               \
+	nouveau_engine_destroy(&(d)->base)
+#define nouveau_graph_init(d)                                                  \
+	nouveau_engine_init(&(d)->base)
+#define nouveau_graph_fini(d,s)                                                \
+	nouveau_engine_fini(&(d)->base, (s))
+
+#define _nouveau_graph_dtor _nouveau_engine_dtor
+#define _nouveau_graph_init _nouveau_engine_init
+#define _nouveau_graph_fini _nouveau_engine_fini
+
+extern struct nouveau_oclass nv04_graph_oclass;
+extern struct nouveau_oclass nv10_graph_oclass;
+extern struct nouveau_oclass nv20_graph_oclass;
+extern struct nouveau_oclass nv25_graph_oclass;
+extern struct nouveau_oclass nv2a_graph_oclass;
+extern struct nouveau_oclass nv30_graph_oclass;
+extern struct nouveau_oclass nv34_graph_oclass;
+extern struct nouveau_oclass nv35_graph_oclass;
+extern struct nouveau_oclass nv40_graph_oclass;
+extern struct nouveau_oclass nv50_graph_oclass;
+extern struct nouveau_oclass nvc0_graph_oclass;
+extern struct nouveau_oclass nve0_graph_oclass;
+
+extern const struct nouveau_bitfield nv04_graph_nsource[];
+extern struct nouveau_ofuncs nv04_graph_ofuncs;
+bool nv04_graph_idle(void *obj);
+
+extern const struct nouveau_bitfield nv10_graph_intr_name[];
+extern const struct nouveau_bitfield nv10_graph_nstatus[];
+
+extern const struct nouveau_enum nv50_data_error_names[];
+
+#endif
diff --git a/nvkm/include/engine/mpeg.h b/nvkm/include/engine/mpeg.h
new file mode 100644
index 000000000..bbf0d4a5b
--- /dev/null
+++ b/nvkm/include/engine/mpeg.h
@@ -0,0 +1,61 @@
+#ifndef __NOUVEAU_MPEG_H__
+#define __NOUVEAU_MPEG_H__
+
+#include <core/engine.h>
+#include <core/engctx.h>
+
+struct nouveau_mpeg_chan {
+	struct nouveau_engctx base;
+};
+
+#define nouveau_mpeg_context_create(p,e,c,g,s,a,f,d)                           \
+	nouveau_engctx_create((p), (e), (c), (g), (s), (a), (f), (d))
+#define nouveau_mpeg_context_destroy(d)                                        \
+	nouveau_engctx_destroy(&(d)->base)
+#define nouveau_mpeg_context_init(d)                                           \
+	nouveau_engctx_init(&(d)->base)
+#define nouveau_mpeg_context_fini(d,s)                                         \
+	nouveau_engctx_fini(&(d)->base, (s))
+
+#define _nouveau_mpeg_context_dtor _nouveau_engctx_dtor
+#define _nouveau_mpeg_context_init _nouveau_engctx_init
+#define _nouveau_mpeg_context_fini _nouveau_engctx_fini
+#define _nouveau_mpeg_context_rd32 _nouveau_engctx_rd32
+#define _nouveau_mpeg_context_wr32 _nouveau_engctx_wr32
+
+struct nouveau_mpeg {
+	struct nouveau_engine base;
+};
+
+#define nouveau_mpeg_create(p,e,c,d)                                           \
+	nouveau_engine_create((p), (e), (c), true, "PMPEG", "mpeg", (d))
+#define nouveau_mpeg_destroy(d)                                                \
+	nouveau_engine_destroy(&(d)->base)
+#define nouveau_mpeg_init(d)                                                   \
+	nouveau_engine_init(&(d)->base)
+#define nouveau_mpeg_fini(d,s)                                                 \
+	nouveau_engine_fini(&(d)->base, (s))
+
+#define _nouveau_mpeg_dtor _nouveau_engine_dtor
+#define _nouveau_mpeg_init _nouveau_engine_init
+#define _nouveau_mpeg_fini _nouveau_engine_fini
+
+extern struct nouveau_oclass nv31_mpeg_oclass;
+extern struct nouveau_oclass nv40_mpeg_oclass;
+extern struct nouveau_oclass nv50_mpeg_oclass;
+extern struct nouveau_oclass nv84_mpeg_oclass;
+
+extern struct nouveau_oclass nv31_mpeg_sclass[];
+void nv31_mpeg_intr(struct nouveau_subdev *);
+void nv31_mpeg_tile_prog(struct nouveau_engine *, int);
+int  nv31_mpeg_init(struct nouveau_object *);
+
+extern struct nouveau_ofuncs nv50_mpeg_ofuncs;
+int  nv50_mpeg_context_ctor(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, void *, u32,
+			    struct nouveau_object **);
+int  nv50_mpeg_tlb_flush(struct nouveau_engine *);
+void nv50_mpeg_intr(struct nouveau_subdev *);
+int  nv50_mpeg_init(struct nouveau_object *);
+
+#endif
diff --git a/nvkm/include/engine/ppp.h b/nvkm/include/engine/ppp.h
new file mode 100644
index 000000000..0a66781e8
--- /dev/null
+++ b/nvkm/include/engine/ppp.h
@@ -0,0 +1,7 @@
+#ifndef __NOUVEAU_PPP_H__
+#define __NOUVEAU_PPP_H__
+
+extern struct nouveau_oclass nv98_ppp_oclass;
+extern struct nouveau_oclass nvc0_ppp_oclass;
+
+#endif
diff --git a/nvkm/include/engine/software.h b/nvkm/include/engine/software.h
new file mode 100644
index 000000000..c945691c8
--- /dev/null
+++ b/nvkm/include/engine/software.h
@@ -0,0 +1,60 @@
+#ifndef __NOUVEAU_SOFTWARE_H__
+#define __NOUVEAU_SOFTWARE_H__
+
+#include <core/engine.h>
+#include <core/engctx.h>
+
+struct nouveau_software_chan {
+	struct nouveau_engctx base;
+
+	struct {
+		struct list_head head;
+		u32 channel;
+		u32 ctxdma;
+		u64 offset;
+		u32 value;
+		u32 crtc;
+	} vblank;
+
+	int (*flip)(void *);
+	void *flip_data;
+};
+
+#define nouveau_software_context_create(p,e,c,d)                               \
+	nouveau_engctx_create((p), (e), (c), (p), 0, 0, 0, (d))
+#define nouveau_software_context_destroy(d)                                    \
+	nouveau_engctx_destroy(&(d)->base)
+#define nouveau_software_context_init(d)                                       \
+	nouveau_engctx_init(&(d)->base)
+#define nouveau_software_context_fini(d,s)                                     \
+	nouveau_engctx_fini(&(d)->base, (s))
+
+#define _nouveau_software_context_dtor _nouveau_engctx_dtor
+#define _nouveau_software_context_init _nouveau_engctx_init
+#define _nouveau_software_context_fini _nouveau_engctx_fini
+
+struct nouveau_software {
+	struct nouveau_engine base;
+};
+
+#define nouveau_software_create(p,e,c,d)                                       \
+	nouveau_engine_create((p), (e), (c), true, "SW", "software", (d))
+#define nouveau_software_destroy(d)                                            \
+	nouveau_engine_destroy(&(d)->base)
+#define nouveau_software_init(d)                                               \
+	nouveau_engine_init(&(d)->base)
+#define nouveau_software_fini(d,s)                                             \
+	nouveau_engine_fini(&(d)->base, (s))
+
+#define _nouveau_software_dtor _nouveau_engine_dtor
+#define _nouveau_software_init _nouveau_engine_init
+#define _nouveau_software_fini _nouveau_engine_fini
+
+extern struct nouveau_oclass nv04_software_oclass;
+extern struct nouveau_oclass nv10_software_oclass;
+extern struct nouveau_oclass nv50_software_oclass;
+extern struct nouveau_oclass nvc0_software_oclass;
+
+void nv04_software_intr(struct nouveau_subdev *);
+
+#endif
diff --git a/nvkm/include/engine/vp.h b/nvkm/include/engine/vp.h
new file mode 100644
index 000000000..d7b287b11
--- /dev/null
+++ b/nvkm/include/engine/vp.h
@@ -0,0 +1,8 @@
+#ifndef __NOUVEAU_VP_H__
+#define __NOUVEAU_VP_H__
+
+extern struct nouveau_oclass nv84_vp_oclass;
+extern struct nouveau_oclass nvc0_vp_oclass;
+extern struct nouveau_oclass nve0_vp_oclass;
+
+#endif
diff --git a/nvkm/include/subdev/bar.h b/nvkm/include/subdev/bar.h
new file mode 100644
index 000000000..4f4ff4502
--- /dev/null
+++ b/nvkm/include/subdev/bar.h
@@ -0,0 +1,55 @@
+#ifndef __NOUVEAU_BAR_H__
+#define __NOUVEAU_BAR_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+#include <subdev/fb.h>
+
+struct nouveau_vma;
+
+struct nouveau_bar {
+	struct nouveau_subdev base;
+
+	int (*alloc)(struct nouveau_bar *, struct nouveau_object *,
+		     struct nouveau_mem *, struct nouveau_object **);
+	void __iomem *iomem;
+
+	int (*kmap)(struct nouveau_bar *, struct nouveau_mem *,
+		    u32 flags, struct nouveau_vma *);
+	int (*umap)(struct nouveau_bar *, struct nouveau_mem *,
+		    u32 flags, struct nouveau_vma *);
+	void (*unmap)(struct nouveau_bar *, struct nouveau_vma *);
+	void (*flush)(struct nouveau_bar *);
+};
+
+static inline struct nouveau_bar *
+nouveau_bar(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_BAR];
+}
+
+#define nouveau_bar_create(p,e,o,d)                                            \
+	nouveau_bar_create_((p), (e), (o), sizeof(**d), (void **)d)
+#define nouveau_bar_init(p)                                                    \
+	nouveau_subdev_init(&(p)->base)
+#define nouveau_bar_fini(p,s)                                                  \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+int nouveau_bar_create_(struct nouveau_object *, struct nouveau_object *,
+			struct nouveau_oclass *, int, void **);
+void nouveau_bar_destroy(struct nouveau_bar *);
+
+void _nouveau_bar_dtor(struct nouveau_object *);
+#define _nouveau_bar_init _nouveau_subdev_init
+#define _nouveau_bar_fini _nouveau_subdev_fini
+
+extern struct nouveau_oclass nv50_bar_oclass;
+extern struct nouveau_oclass nvc0_bar_oclass;
+
+int nouveau_bar_alloc(struct nouveau_bar *, struct nouveau_object *,
+		      struct nouveau_mem *, struct nouveau_object **);
+
+void nv84_bar_flush(struct nouveau_bar *);
+
+#endif
diff --git a/nvkm/include/subdev/bios.h b/nvkm/include/subdev/bios.h
new file mode 100644
index 000000000..5bd1ca8cd
--- /dev/null
+++ b/nvkm/include/subdev/bios.h
@@ -0,0 +1,35 @@
+#ifndef __NOUVEAU_BIOS_H__
+#define __NOUVEAU_BIOS_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+struct nouveau_bios {
+	struct nouveau_subdev base;
+	u32 size;
+	u8 *data;
+
+	u32 bmp_offset;
+	u32 bit_offset;
+
+	struct {
+		u8 major;
+		u8 chip;
+		u8 minor;
+		u8 micro;
+		u8 patch;
+	} version;
+};
+
+static inline struct nouveau_bios *
+nouveau_bios(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_VBIOS];
+}
+
+u8  nvbios_checksum(const u8 *data, int size);
+u16 nvbios_findstr(const u8 *data, int size, const char *str, int len);
+
+extern struct nouveau_oclass nouveau_bios_oclass;
+
+#endif
diff --git a/nvkm/include/subdev/bios/bit.h b/nvkm/include/subdev/bios/bit.h
new file mode 100644
index 000000000..73f060b07
--- /dev/null
+++ b/nvkm/include/subdev/bios/bit.h
@@ -0,0 +1,13 @@
+#ifndef __NVBIOS_BIT_H__
+#define __NVBIOS_BIT_H__
+
+struct bit_entry {
+	u8  id;
+	u8  version;
+	u16 length;
+	u16 offset;
+};
+
+int bit_entry(struct nouveau_bios *, u8 id, struct bit_entry *);
+
+#endif
diff --git a/nvkm/include/subdev/bios/bmp.h b/nvkm/include/subdev/bios/bmp.h
new file mode 100644
index 000000000..10e4dbca6
--- /dev/null
+++ b/nvkm/include/subdev/bios/bmp.h
@@ -0,0 +1,39 @@
+#ifndef __NVBIOS_BMP_H__
+#define __NVBIOS_BMP_H__
+
+static inline u16
+bmp_version(struct nouveau_bios *bios)
+{
+	if (bios->bmp_offset) {
+		return nv_ro08(bios, bios->bmp_offset + 5) << 8 |
+		       nv_ro08(bios, bios->bmp_offset + 6);
+	}
+
+	return 0x0000;
+}
+
+static inline u16
+bmp_mem_init_table(struct nouveau_bios *bios)
+{
+	if (bmp_version(bios) >= 0x0300)
+		return nv_ro16(bios, bios->bmp_offset + 24);
+	return 0x0000;
+}
+
+static inline u16
+bmp_sdr_seq_table(struct nouveau_bios *bios)
+{
+	if (bmp_version(bios) >= 0x0300)
+		return nv_ro16(bios, bios->bmp_offset + 26);
+	return 0x0000;
+}
+
+static inline u16
+bmp_ddr_seq_table(struct nouveau_bios *bios)
+{
+	if (bmp_version(bios) >= 0x0300)
+		return nv_ro16(bios, bios->bmp_offset + 28);
+	return 0x0000;
+}
+
+#endif
diff --git a/nvkm/include/subdev/bios/conn.h b/nvkm/include/subdev/bios/conn.h
new file mode 100644
index 000000000..c1270548f
--- /dev/null
+++ b/nvkm/include/subdev/bios/conn.h
@@ -0,0 +1,27 @@
+#ifndef __NVBIOS_CONN_H__
+#define __NVBIOS_CONN_H__
+
+enum dcb_connector_type {
+	DCB_CONNECTOR_VGA = 0x00,
+	DCB_CONNECTOR_TV_0 = 0x10,
+	DCB_CONNECTOR_TV_1 = 0x11,
+	DCB_CONNECTOR_TV_3 = 0x13,
+	DCB_CONNECTOR_DVI_I = 0x30,
+	DCB_CONNECTOR_DVI_D = 0x31,
+	DCB_CONNECTOR_DMS59_0 = 0x38,
+	DCB_CONNECTOR_DMS59_1 = 0x39,
+	DCB_CONNECTOR_LVDS = 0x40,
+	DCB_CONNECTOR_LVDS_SPWG = 0x41,
+	DCB_CONNECTOR_DP = 0x46,
+	DCB_CONNECTOR_eDP = 0x47,
+	DCB_CONNECTOR_HDMI_0 = 0x60,
+	DCB_CONNECTOR_HDMI_1 = 0x61,
+	DCB_CONNECTOR_DMS59_DP0 = 0x64,
+	DCB_CONNECTOR_DMS59_DP1 = 0x65,
+	DCB_CONNECTOR_NONE = 0xff
+};
+
+u16 dcb_conntab(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len);
+u16 dcb_conn(struct nouveau_bios *bios, u8 idx, u8 *ver, u8 *len);
+
+#endif
diff --git a/nvkm/include/subdev/bios/dcb.h b/nvkm/include/subdev/bios/dcb.h
new file mode 100644
index 000000000..b79025da5
--- /dev/null
+++ b/nvkm/include/subdev/bios/dcb.h
@@ -0,0 +1,66 @@
+#ifndef __NVBIOS_DCB_H__
+#define __NVBIOS_DCB_H__
+
+struct nouveau_bios;
+
+enum dcb_output_type {
+	DCB_OUTPUT_ANALOG	= 0x0,
+	DCB_OUTPUT_TV		= 0x1,
+	DCB_OUTPUT_TMDS		= 0x2,
+	DCB_OUTPUT_LVDS		= 0x3,
+	DCB_OUTPUT_DP		= 0x6,
+	DCB_OUTPUT_EOL		= 0xe,
+	DCB_OUTPUT_UNUSED	= 0xf,
+	DCB_OUTPUT_ANY = -1,
+};
+
+struct dcb_output {
+	int index;	/* may not be raw dcb index if merging has happened */
+	enum dcb_output_type type;
+	uint8_t i2c_index;
+	uint8_t heads;
+	uint8_t connector;
+	uint8_t bus;
+	uint8_t location;
+	uint8_t or;
+	uint8_t link;
+	bool duallink_possible;
+	union {
+		struct sor_conf {
+			int link;
+		} sorconf;
+		struct {
+			int maxfreq;
+		} crtconf;
+		struct {
+			struct sor_conf sor;
+			bool use_straps_for_mode;
+			bool use_acpi_for_edid;
+			bool use_power_scripts;
+		} lvdsconf;
+		struct {
+			bool has_component_output;
+		} tvconf;
+		struct {
+			struct sor_conf sor;
+			int link_nr;
+			int link_bw;
+		} dpconf;
+		struct {
+			struct sor_conf sor;
+			int slave_addr;
+		} tmdsconf;
+	};
+	bool i2c_upper_default;
+};
+
+u16 dcb_table(struct nouveau_bios *, u8 *ver, u8 *hdr, u8 *ent, u8 *len);
+u16 dcb_outp(struct nouveau_bios *, u8 idx, u8 *ver, u8 *len);
+u16 dcb_outp_parse(struct nouveau_bios *, u8 idx, u8 *, u8 *,
+		   struct dcb_output *);
+u16 dcb_outp_match(struct nouveau_bios *, u16 type, u16 mask, u8 *, u8 *,
+		   struct dcb_output *);
+int dcb_outp_foreach(struct nouveau_bios *, void *data, int (*exec)
+		     (struct nouveau_bios *, void *, int index, u16 entry));
+
+#endif
diff --git a/nvkm/include/subdev/bios/dfp.h b/nvkm/include/subdev/bios/dfp.h
new file mode 100644
index 000000000..50cb52c11
--- /dev/null
+++ b/nvkm/include/subdev/bios/dfp.h
@@ -0,0 +1,4 @@
+#ifndef __NVBIOS_DFP_H__
+#define __NVBIOS_DFP_H__
+
+#endif
diff --git a/nvkm/include/subdev/bios/disp.h b/nvkm/include/subdev/bios/disp.h
new file mode 100644
index 000000000..c35937e2f
--- /dev/null
+++ b/nvkm/include/subdev/bios/disp.h
@@ -0,0 +1,48 @@
+#ifndef __NVBIOS_DISP_H__
+#define __NVBIOS_DISP_H__
+
+u16 nvbios_disp_table(struct nouveau_bios *,
+		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len, u8 *sub);
+
+struct nvbios_disp {
+	u16 data;
+};
+
+u16 nvbios_disp_entry(struct nouveau_bios *, u8 idx,
+		      u8 *ver, u8 *hdr__, u8 *sub);
+u16 nvbios_disp_parse(struct nouveau_bios *, u8 idx,
+		      u8 *ver, u8 *hdr__, u8 *sub,
+		      struct nvbios_disp *);
+
+struct nvbios_outp {
+	u16 type;
+	u16 mask;
+	u16 script[3];
+};
+
+u16 nvbios_outp_entry(struct nouveau_bios *, u8 idx,
+		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len);
+u16 nvbios_outp_parse(struct nouveau_bios *, u8 idx,
+		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		      struct nvbios_outp *);
+u16 nvbios_outp_match(struct nouveau_bios *, u16 type, u16 mask,
+		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		      struct nvbios_outp *);
+
+
+struct nvbios_ocfg {
+	u16 match;
+	u16 clkcmp[2];
+};
+
+u16 nvbios_ocfg_entry(struct nouveau_bios *, u16 outp, u8 idx,
+		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len);
+u16 nvbios_ocfg_parse(struct nouveau_bios *, u16 outp, u8 idx,
+		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		      struct nvbios_ocfg *);
+u16 nvbios_ocfg_match(struct nouveau_bios *, u16 outp, u16 type,
+		      u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		      struct nvbios_ocfg *);
+u16 nvbios_oclk_match(struct nouveau_bios *, u16 cmp, u32 khz);
+
+#endif
diff --git a/nvkm/include/subdev/bios/dp.h b/nvkm/include/subdev/bios/dp.h
new file mode 100644
index 000000000..6e54218b5
--- /dev/null
+++ b/nvkm/include/subdev/bios/dp.h
@@ -0,0 +1,34 @@
+#ifndef __NVBIOS_DP_H__
+#define __NVBIOS_DP_H__
+
+struct nvbios_dpout {
+	u16 type;
+	u16 mask;
+	u8  flags;
+	u32 script[5];
+	u32 lnkcmp;
+};
+
+u16 nvbios_dpout_parse(struct nouveau_bios *, u8 idx,
+		       u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		       struct nvbios_dpout *);
+u16 nvbios_dpout_match(struct nouveau_bios *, u16 type, u16 mask,
+		       u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		       struct nvbios_dpout *);
+
+struct nvbios_dpcfg {
+	u8 drv;
+	u8 pre;
+	u8 unk;
+};
+
+u16
+nvbios_dpcfg_parse(struct nouveau_bios *, u16 outp, u8 idx,
+		   u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		   struct nvbios_dpcfg *);
+u16
+nvbios_dpcfg_match(struct nouveau_bios *, u16 outp, u8 un, u8 vs, u8 pe,
+		   u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		   struct nvbios_dpcfg *);
+
+#endif
diff --git a/nvkm/include/subdev/bios/extdev.h b/nvkm/include/subdev/bios/extdev.h
new file mode 100644
index 000000000..949fee3af
--- /dev/null
+++ b/nvkm/include/subdev/bios/extdev.h
@@ -0,0 +1,30 @@
+#ifndef __NVBIOS_EXTDEV_H__
+#define __NVBIOS_EXTDEV_H__
+
+struct nouveau_bios;
+
+enum nvbios_extdev_type {
+	NVBIOS_EXTDEV_LM89		= 0x02,
+	NVBIOS_EXTDEV_VT1103M		= 0x40,
+	NVBIOS_EXTDEV_PX3540		= 0x41,
+	NVBIOS_EXTDEV_VT1105M		= 0x42, /* or close enough... */
+	NVBIOS_EXTDEV_ADT7473		= 0x70, /* can also be a LM64 */
+	NVBIOS_EXTDEV_HDCP_EEPROM	= 0x90,
+	NVBIOS_EXTDEV_NONE		= 0xff,
+};
+
+struct nvbios_extdev_func {
+	u8 type;
+	u8 addr;
+	u8 bus;
+};
+
+int
+nvbios_extdev_parse(struct nouveau_bios *, int, struct nvbios_extdev_func *);
+
+int
+nvbios_extdev_find(struct nouveau_bios *, enum nvbios_extdev_type,
+		   struct nvbios_extdev_func *);
+
+
+#endif
diff --git a/nvkm/include/subdev/bios/gpio.h b/nvkm/include/subdev/bios/gpio.h
new file mode 100644
index 000000000..e6563b5cb
--- /dev/null
+++ b/nvkm/include/subdev/bios/gpio.h
@@ -0,0 +1,35 @@
+#ifndef __NVBIOS_GPIO_H__
+#define __NVBIOS_GPIO_H__
+
+struct nouveau_bios;
+
+enum dcb_gpio_func_name {
+	DCB_GPIO_PANEL_POWER = 0x01,
+	DCB_GPIO_TVDAC0 = 0x0c,
+	DCB_GPIO_TVDAC1 = 0x2d,
+	DCB_GPIO_PWM_FAN = 0x09,
+	DCB_GPIO_FAN_SENSE = 0x3d,
+	DCB_GPIO_UNUSED = 0xff
+};
+
+struct dcb_gpio_func {
+	u8 func;
+	u8 line;
+	u8 log[2];
+
+	/* so far, "param" seems to only have an influence on PWM-related
+	 * GPIOs such as FAN_CONTROL and PANEL_BACKLIGHT_LEVEL.
+	 * if param equals 1, hardware PWM is available
+	 * if param equals 0, the host should toggle the GPIO itself
+	 */
+	u8 param;
+};
+
+u16 dcb_gpio_table(struct nouveau_bios *, u8 *ver, u8 *hdr, u8 *cnt, u8 *len);
+u16 dcb_gpio_entry(struct nouveau_bios *, int idx, int ent, u8 *ver, u8 *len);
+u16 dcb_gpio_parse(struct nouveau_bios *, int idx, int ent, u8 *ver, u8 *len,
+		   struct dcb_gpio_func *);
+u16 dcb_gpio_match(struct nouveau_bios *, int idx, u8 func, u8 line,
+		   u8 *ver, u8 *len, struct dcb_gpio_func *);
+
+#endif
diff --git a/nvkm/include/subdev/bios/i2c.h b/nvkm/include/subdev/bios/i2c.h
new file mode 100644
index 000000000..5079bedfd
--- /dev/null
+++ b/nvkm/include/subdev/bios/i2c.h
@@ -0,0 +1,25 @@
+#ifndef __NVBIOS_I2C_H__
+#define __NVBIOS_I2C_H__
+
+struct nouveau_bios;
+
+enum dcb_i2c_type {
+	DCB_I2C_NV04_BIT = 0,
+	DCB_I2C_NV4E_BIT = 4,
+	DCB_I2C_NVIO_BIT = 5,
+	DCB_I2C_NVIO_AUX = 6,
+	DCB_I2C_UNUSED = 0xff
+};
+
+struct dcb_i2c_entry {
+	enum dcb_i2c_type type;
+	u8 drive;
+	u8 sense;
+	u32 data;
+};
+
+u16 dcb_i2c_table(struct nouveau_bios *, u8 *ver, u8 *hdr, u8 *cnt, u8 *len);
+u16 dcb_i2c_entry(struct nouveau_bios *, u8 index, u8 *ver, u8 *len);
+int dcb_i2c_parse(struct nouveau_bios *, u8 index, struct dcb_i2c_entry *);
+
+#endif
diff --git a/nvkm/include/subdev/bios/init.h b/nvkm/include/subdev/bios/init.h
new file mode 100644
index 000000000..ca2f6bf37
--- /dev/null
+++ b/nvkm/include/subdev/bios/init.h
@@ -0,0 +1,22 @@
+#ifndef __NVBIOS_INIT_H__
+#define __NVBIOS_INIT_H__
+
+struct nvbios_init {
+	struct nouveau_subdev *subdev;
+	struct nouveau_bios *bios;
+	u16 offset;
+	struct dcb_output *outp;
+	int crtc;
+
+	/* internal state used during parsing */
+	u8 execute;
+	u32 nested;
+	u16 repeat;
+	u16 repend;
+	u32 ramcfg;
+};
+
+int nvbios_exec(struct nvbios_init *);
+int nvbios_init(struct nouveau_subdev *, bool execute);
+
+#endif
diff --git a/nvkm/include/subdev/bios/mxm.h b/nvkm/include/subdev/bios/mxm.h
new file mode 100644
index 000000000..5572e6041
--- /dev/null
+++ b/nvkm/include/subdev/bios/mxm.h
@@ -0,0 +1,9 @@
+#ifndef __NVBIOS_MXM_H__
+#define __NVBIOS_MXM_H__
+
+u16 mxm_table(struct nouveau_bios *, u8 *ver, u8 *hdr);
+
+u8  mxm_sor_map(struct nouveau_bios *, u8 conn);
+u8  mxm_ddc_map(struct nouveau_bios *, u8 port);
+
+#endif
diff --git a/nvkm/include/subdev/bios/perf.h b/nvkm/include/subdev/bios/perf.h
new file mode 100644
index 000000000..0b285e99b
--- /dev/null
+++ b/nvkm/include/subdev/bios/perf.h
@@ -0,0 +1,14 @@
+#ifndef __NVBIOS_PERF_H__
+#define __NVBIOS_PERF_H__
+
+struct nouveau_bios;
+
+struct nvbios_perf_fan {
+	u32 pwm_divisor;
+};
+
+int
+nvbios_perf_fan_parse(struct nouveau_bios *, struct nvbios_perf_fan *);
+
+
+#endif
diff --git a/nvkm/include/subdev/bios/pll.h b/nvkm/include/subdev/bios/pll.h
new file mode 100644
index 000000000..c34509759
--- /dev/null
+++ b/nvkm/include/subdev/bios/pll.h
@@ -0,0 +1,77 @@
+#ifndef __NVBIOS_PLL_H__
+#define __NVBIOS_PLL_H__
+
+/*XXX: kill me */
+struct nouveau_pll_vals {
+	union {
+		struct {
+#ifdef __BIG_ENDIAN
+			uint8_t N1, M1, N2, M2;
+#else
+			uint8_t M1, N1, M2, N2;
+#endif
+		};
+		struct {
+			uint16_t NM1, NM2;
+		} __attribute__((packed));
+	};
+	int log2P;
+
+	int refclk;
+};
+
+struct nouveau_bios;
+
+/* these match types in pll limits table version 0x40,
+ * nouveau uses them on all chipsets internally where a
+ * specific pll needs to be referenced, but the exact
+ * register isn't known.
+ */
+enum nvbios_pll_type {
+	PLL_CORE   = 0x01,
+	PLL_SHADER = 0x02,
+	PLL_UNK03  = 0x03,
+	PLL_MEMORY = 0x04,
+	PLL_VDEC   = 0x05,
+	PLL_UNK40  = 0x40,
+	PLL_UNK41  = 0x41,
+	PLL_UNK42  = 0x42,
+	PLL_VPLL0  = 0x80,
+	PLL_VPLL1  = 0x81,
+	PLL_MAX    = 0xff
+};
+
+struct nvbios_pll {
+	enum nvbios_pll_type type;
+	u32 reg;
+	u32 refclk;
+
+	u8 min_p;
+	u8 max_p;
+	u8 bias_p;
+
+	/*
+	 * for most pre nv50 cards setting a log2P of 7 (the common max_log2p
+	 * value) is no different to 6 (at least for vplls) so allowing the MNP
+	 * calc to use 7 causes the generated clock to be out by a factor of 2.
+	 * however, max_log2p cannot be fixed-up during parsing as the
+	 * unmodified max_log2p value is still needed for setting mplls, hence
+	 * an additional max_usable_log2p member
+	 */
+	u8 max_p_usable;
+
+	struct {
+		u32 min_freq;
+		u32 max_freq;
+		u32 min_inputfreq;
+		u32 max_inputfreq;
+		u8  min_m;
+		u8  max_m;
+		u8  min_n;
+		u8  max_n;
+	} vco1, vco2;
+};
+
+int nvbios_pll_parse(struct nouveau_bios *, u32 type, struct nvbios_pll *);
+
+#endif
diff --git a/nvkm/include/subdev/bios/therm.h b/nvkm/include/subdev/bios/therm.h
new file mode 100644
index 000000000..a2c4296fc
--- /dev/null
+++ b/nvkm/include/subdev/bios/therm.h
@@ -0,0 +1,46 @@
+#ifndef __NVBIOS_THERM_H__
+#define __NVBIOS_THERM_H__
+
+struct nouveau_bios;
+
+struct nvbios_therm_threshold {
+	u8 temp;
+	u8 hysteresis;
+};
+
+struct nvbios_therm_sensor {
+	/* diode */
+	s16 slope_mult;
+	s16 slope_div;
+	s16 offset_num;
+	s16 offset_den;
+	s8 offset_constant;
+
+	/* thresholds */
+	struct nvbios_therm_threshold thrs_fan_boost;
+	struct nvbios_therm_threshold thrs_down_clock;
+	struct nvbios_therm_threshold thrs_critical;
+	struct nvbios_therm_threshold thrs_shutdown;
+};
+
+struct nvbios_therm_fan {
+	u16 pwm_freq;
+
+	u8 min_duty;
+	u8 max_duty;
+};
+
+enum nvbios_therm_domain {
+	NVBIOS_THERM_DOMAIN_CORE,
+	NVBIOS_THERM_DOMAIN_AMBIENT,
+};
+
+int
+nvbios_therm_sensor_parse(struct nouveau_bios *, enum nvbios_therm_domain,
+			  struct nvbios_therm_sensor *);
+
+int
+nvbios_therm_fan_parse(struct nouveau_bios *, struct nvbios_therm_fan *);
+
+
+#endif
diff --git a/nvkm/include/subdev/clock.h b/nvkm/include/subdev/clock.h
new file mode 100644
index 000000000..41b7a6a76
--- /dev/null
+++ b/nvkm/include/subdev/clock.h
@@ -0,0 +1,60 @@
+#ifndef __NOUVEAU_CLOCK_H__
+#define __NOUVEAU_CLOCK_H__
+
+#include <core/device.h>
+#include <core/subdev.h>
+
+struct nouveau_pll_vals;
+struct nvbios_pll;
+
+struct nouveau_clock {
+	struct nouveau_subdev base;
+
+	int (*pll_set)(struct nouveau_clock *, u32 type, u32 freq);
+
+	/*XXX: die, these are here *only* to support the completely
+	 *     bat-shit insane what-was-nouveau_hw.c code
+	 */
+	int (*pll_calc)(struct nouveau_clock *, struct nvbios_pll *,
+			int clk, struct nouveau_pll_vals *pv);
+	int (*pll_prog)(struct nouveau_clock *, u32 reg1,
+			struct nouveau_pll_vals *pv);
+};
+
+static inline struct nouveau_clock *
+nouveau_clock(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_CLOCK];
+}
+
+#define nouveau_clock_create(p,e,o,d)                                          \
+	nouveau_subdev_create((p), (e), (o), 0, "CLOCK", "clock", d)
+#define nouveau_clock_destroy(p)                                               \
+	nouveau_subdev_destroy(&(p)->base)
+#define nouveau_clock_init(p)                                                  \
+	nouveau_subdev_init(&(p)->base)
+#define nouveau_clock_fini(p,s)                                                \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+int  nouveau_clock_create_(struct nouveau_object *, struct nouveau_object *,
+			   struct nouveau_oclass *, void *, u32, int, void **);
+
+#define _nouveau_clock_dtor _nouveau_subdev_dtor
+#define _nouveau_clock_init _nouveau_subdev_init
+#define _nouveau_clock_fini _nouveau_subdev_fini
+
+extern struct nouveau_oclass nv04_clock_oclass;
+extern struct nouveau_oclass nv40_clock_oclass;
+extern struct nouveau_oclass nv50_clock_oclass;
+extern struct nouveau_oclass nva3_clock_oclass;
+extern struct nouveau_oclass nvc0_clock_oclass;
+
+int nv04_clock_pll_set(struct nouveau_clock *, u32 type, u32 freq);
+int nv04_clock_pll_calc(struct nouveau_clock *, struct nvbios_pll *,
+			int clk, struct nouveau_pll_vals *);
+int nv04_clock_pll_prog(struct nouveau_clock *, u32 reg1,
+			struct nouveau_pll_vals *);
+int nva3_clock_pll_calc(struct nouveau_clock *, struct nvbios_pll *,
+			int clk, struct nouveau_pll_vals *);
+
+#endif
diff --git a/nvkm/include/subdev/device.h b/nvkm/include/subdev/device.h
new file mode 100644
index 000000000..c9e4c4afa
--- /dev/null
+++ b/nvkm/include/subdev/device.h
@@ -0,0 +1,24 @@
+#ifndef __NOUVEAU_SUBDEV_DEVICE_H__
+#define __NOUVEAU_SUBDEV_DEVICE_H__
+
+#include <core/device.h>
+
+#define nouveau_device_create(p,n,s,c,d,u)                                     \
+	nouveau_device_create_((p), (n), (s), (c), (d), sizeof(**u), (void **)u)
+
+int  nouveau_device_create_(struct pci_dev *, u64 name, const char *sname,
+			    const char *cfg, const char *dbg, int, void **);
+
+int nv04_identify(struct nouveau_device *);
+int nv10_identify(struct nouveau_device *);
+int nv20_identify(struct nouveau_device *);
+int nv30_identify(struct nouveau_device *);
+int nv40_identify(struct nouveau_device *);
+int nv50_identify(struct nouveau_device *);
+int nvc0_identify(struct nouveau_device *);
+int nve0_identify(struct nouveau_device *);
+
+extern struct nouveau_oclass nouveau_device_sclass[];
+struct nouveau_device *nouveau_device_find(u64 name);
+
+#endif
diff --git a/nvkm/include/subdev/devinit.h b/nvkm/include/subdev/devinit.h
new file mode 100644
index 000000000..29e4cc1f6
--- /dev/null
+++ b/nvkm/include/subdev/devinit.h
@@ -0,0 +1,40 @@
+#ifndef __NOUVEAU_DEVINIT_H__
+#define __NOUVEAU_DEVINIT_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+struct nouveau_devinit {
+	struct nouveau_subdev base;
+	bool post;
+	void (*meminit)(struct nouveau_devinit *);
+};
+
+static inline struct nouveau_devinit *
+nouveau_devinit(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_DEVINIT];
+}
+
+#define nouveau_devinit_create(p,e,o,d)                                        \
+	nouveau_devinit_create_((p), (e), (o), sizeof(**d), (void **)d)
+#define nouveau_devinit_destroy(p)                                             \
+	nouveau_subdev_destroy(&(p)->base)
+
+int nouveau_devinit_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, int, void **);
+int nouveau_devinit_init(struct nouveau_devinit *);
+int nouveau_devinit_fini(struct nouveau_devinit *, bool suspend);
+
+extern struct nouveau_oclass nv04_devinit_oclass;
+extern struct nouveau_oclass nv05_devinit_oclass;
+extern struct nouveau_oclass nv10_devinit_oclass;
+extern struct nouveau_oclass nv1a_devinit_oclass;
+extern struct nouveau_oclass nv20_devinit_oclass;
+extern struct nouveau_oclass nv50_devinit_oclass;
+
+void nv04_devinit_dtor(struct nouveau_object *);
+int  nv04_devinit_init(struct nouveau_object *);
+int  nv04_devinit_fini(struct nouveau_object *, bool);
+
+#endif
diff --git a/nvkm/include/subdev/fb.h b/nvkm/include/subdev/fb.h
new file mode 100644
index 000000000..da470e685
--- /dev/null
+++ b/nvkm/include/subdev/fb.h
@@ -0,0 +1,171 @@
+#ifndef __NOUVEAU_FB_H__
+#define __NOUVEAU_FB_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+#include <core/mm.h>
+
+#include <subdev/vm.h>
+
+/* memory type/access flags, do not match hardware values */
+#define NV_MEM_ACCESS_RO  1
+#define NV_MEM_ACCESS_WO  2
+#define NV_MEM_ACCESS_RW (NV_MEM_ACCESS_RO | NV_MEM_ACCESS_WO)
+#define NV_MEM_ACCESS_SYS 4
+#define NV_MEM_ACCESS_VM  8
+#define NV_MEM_ACCESS_NOSNOOP 16
+
+#define NV_MEM_TARGET_VRAM        0
+#define NV_MEM_TARGET_PCI         1
+#define NV_MEM_TARGET_PCI_NOSNOOP 2
+#define NV_MEM_TARGET_VM          3
+#define NV_MEM_TARGET_GART        4
+
+#define NV_MEM_TYPE_VM 0x7f
+#define NV_MEM_COMP_VM 0x03
+
+struct nouveau_mem {
+	struct drm_device *dev;
+
+	struct nouveau_vma bar_vma;
+	struct nouveau_vma vma[2];
+	u8  page_shift;
+
+	struct nouveau_mm_node *tag;
+	struct list_head regions;
+	dma_addr_t *pages;
+	u32 memtype;
+	u64 offset;
+	u64 size;
+	struct sg_table *sg;
+};
+
+struct nouveau_fb_tile {
+	struct nouveau_mm_node *tag;
+	u32 addr;
+	u32 limit;
+	u32 pitch;
+	u32 zcomp;
+};
+
+struct nouveau_fb {
+	struct nouveau_subdev base;
+
+	bool (*memtype_valid)(struct nouveau_fb *, u32 memtype);
+
+	struct {
+		enum {
+			NV_MEM_TYPE_UNKNOWN = 0,
+			NV_MEM_TYPE_STOLEN,
+			NV_MEM_TYPE_SGRAM,
+			NV_MEM_TYPE_SDRAM,
+			NV_MEM_TYPE_DDR1,
+			NV_MEM_TYPE_DDR2,
+			NV_MEM_TYPE_DDR3,
+			NV_MEM_TYPE_GDDR2,
+			NV_MEM_TYPE_GDDR3,
+			NV_MEM_TYPE_GDDR4,
+			NV_MEM_TYPE_GDDR5
+		} type;
+		u64 stolen;
+		u64 size;
+
+		int ranks;
+		int parts;
+
+		int  (*init)(struct nouveau_fb *);
+		int  (*get)(struct nouveau_fb *, u64 size, u32 align,
+			    u32 size_nc, u32 type, struct nouveau_mem **);
+		void (*put)(struct nouveau_fb *, struct nouveau_mem **);
+	} ram;
+
+	struct nouveau_mm vram;
+	struct nouveau_mm tags;
+
+	struct {
+		struct nouveau_fb_tile region[16];
+		int regions;
+		void (*init)(struct nouveau_fb *, int i, u32 addr, u32 size,
+			     u32 pitch, u32 flags, struct nouveau_fb_tile *);
+		void (*comp)(struct nouveau_fb *, int i, u32 size, u32 flags,
+			     struct nouveau_fb_tile *);
+		void (*fini)(struct nouveau_fb *, int i,
+			     struct nouveau_fb_tile *);
+		void (*prog)(struct nouveau_fb *, int i,
+			     struct nouveau_fb_tile *);
+	} tile;
+};
+
+static inline struct nouveau_fb *
+nouveau_fb(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_FB];
+}
+
+#define nouveau_fb_create(p,e,c,d)                                             \
+	nouveau_subdev_create((p), (e), (c), 0, "PFB", "fb", (d))
+int  nouveau_fb_preinit(struct nouveau_fb *);
+void nouveau_fb_destroy(struct nouveau_fb *);
+int  nouveau_fb_init(struct nouveau_fb *);
+#define nouveau_fb_fini(p,s)                                                   \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+void _nouveau_fb_dtor(struct nouveau_object *);
+int  _nouveau_fb_init(struct nouveau_object *);
+#define _nouveau_fb_fini _nouveau_subdev_fini
+
+extern struct nouveau_oclass nv04_fb_oclass;
+extern struct nouveau_oclass nv10_fb_oclass;
+extern struct nouveau_oclass nv1a_fb_oclass;
+extern struct nouveau_oclass nv20_fb_oclass;
+extern struct nouveau_oclass nv25_fb_oclass;
+extern struct nouveau_oclass nv30_fb_oclass;
+extern struct nouveau_oclass nv35_fb_oclass;
+extern struct nouveau_oclass nv36_fb_oclass;
+extern struct nouveau_oclass nv40_fb_oclass;
+extern struct nouveau_oclass nv41_fb_oclass;
+extern struct nouveau_oclass nv44_fb_oclass;
+extern struct nouveau_oclass nv46_fb_oclass;
+extern struct nouveau_oclass nv47_fb_oclass;
+extern struct nouveau_oclass nv49_fb_oclass;
+extern struct nouveau_oclass nv4e_fb_oclass;
+extern struct nouveau_oclass nv50_fb_oclass;
+extern struct nouveau_oclass nvc0_fb_oclass;
+
+struct nouveau_bios;
+int  nouveau_fb_bios_memtype(struct nouveau_bios *);
+
+bool nv04_fb_memtype_valid(struct nouveau_fb *, u32 memtype);
+
+void nv10_fb_tile_init(struct nouveau_fb *, int i, u32 addr, u32 size,
+		       u32 pitch, u32 flags, struct nouveau_fb_tile *);
+void nv10_fb_tile_fini(struct nouveau_fb *, int i, struct nouveau_fb_tile *);
+void nv10_fb_tile_prog(struct nouveau_fb *, int, struct nouveau_fb_tile *);
+
+int  nv20_fb_vram_init(struct nouveau_fb *);
+void nv20_fb_tile_init(struct nouveau_fb *, int i, u32 addr, u32 size,
+		       u32 pitch, u32 flags, struct nouveau_fb_tile *);
+void nv20_fb_tile_fini(struct nouveau_fb *, int i, struct nouveau_fb_tile *);
+void nv20_fb_tile_prog(struct nouveau_fb *, int, struct nouveau_fb_tile *);
+
+int  nv30_fb_init(struct nouveau_object *);
+void nv30_fb_tile_init(struct nouveau_fb *, int i, u32 addr, u32 size,
+		       u32 pitch, u32 flags, struct nouveau_fb_tile *);
+
+void nv40_fb_tile_comp(struct nouveau_fb *, int i, u32 size, u32 flags,
+		       struct nouveau_fb_tile *);
+
+int  nv41_fb_vram_init(struct nouveau_fb *);
+int  nv41_fb_init(struct nouveau_object *);
+void nv41_fb_tile_prog(struct nouveau_fb *, int, struct nouveau_fb_tile *);
+
+int  nv44_fb_vram_init(struct nouveau_fb *);
+int  nv44_fb_init(struct nouveau_object *);
+void nv44_fb_tile_prog(struct nouveau_fb *, int, struct nouveau_fb_tile *);
+
+void nv46_fb_tile_init(struct nouveau_fb *, int i, u32 addr, u32 size,
+		       u32 pitch, u32 flags, struct nouveau_fb_tile *);
+
+void nv50_fb_vram_del(struct nouveau_fb *, struct nouveau_mem **);
+
+#endif
diff --git a/nvkm/include/subdev/gpio.h b/nvkm/include/subdev/gpio.h
new file mode 100644
index 000000000..b75e8f18e
--- /dev/null
+++ b/nvkm/include/subdev/gpio.h
@@ -0,0 +1,64 @@
+#ifndef __NOUVEAU_GPIO_H__
+#define __NOUVEAU_GPIO_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/gpio.h>
+
+struct nouveau_gpio {
+	struct nouveau_subdev base;
+
+	/* hardware interfaces */
+	void (*reset)(struct nouveau_gpio *, u8 func);
+	int  (*drive)(struct nouveau_gpio *, int line, int dir, int out);
+	int  (*sense)(struct nouveau_gpio *, int line);
+	void (*irq_enable)(struct nouveau_gpio *, int line, bool);
+
+	/* software interfaces */
+	int  (*find)(struct nouveau_gpio *, int idx, u8 tag, u8 line,
+		     struct dcb_gpio_func *);
+	int  (*set)(struct nouveau_gpio *, int idx, u8 tag, u8 line, int state);
+	int  (*get)(struct nouveau_gpio *, int idx, u8 tag, u8 line);
+	int  (*irq)(struct nouveau_gpio *, int idx, u8 tag, u8 line, bool on);
+
+	/* interrupt handling */
+	struct list_head isr;
+	spinlock_t lock;
+
+	void (*isr_run)(struct nouveau_gpio *, int idx, u32 mask);
+	int  (*isr_add)(struct nouveau_gpio *, int idx, u8 tag, u8 line,
+			void (*)(void *, int state), void *data);
+	void (*isr_del)(struct nouveau_gpio *, int idx, u8 tag, u8 line,
+			void (*)(void *, int state), void *data);
+};
+
+static inline struct nouveau_gpio *
+nouveau_gpio(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_GPIO];
+}
+
+#define nouveau_gpio_create(p,e,o,d)                                           \
+	nouveau_gpio_create_((p), (e), (o), sizeof(**d), (void **)d)
+#define nouveau_gpio_destroy(p)                                                \
+	nouveau_subdev_destroy(&(p)->base)
+#define nouveau_gpio_fini(p,s)                                                 \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+int nouveau_gpio_create_(struct nouveau_object *, struct nouveau_object *,
+			 struct nouveau_oclass *, int, void **);
+int nouveau_gpio_init(struct nouveau_gpio *);
+
+extern struct nouveau_oclass nv10_gpio_oclass;
+extern struct nouveau_oclass nv50_gpio_oclass;
+extern struct nouveau_oclass nvd0_gpio_oclass;
+
+void nv50_gpio_dtor(struct nouveau_object *);
+int  nv50_gpio_init(struct nouveau_object *);
+int  nv50_gpio_fini(struct nouveau_object *, bool);
+void nv50_gpio_intr(struct nouveau_subdev *);
+void nv50_gpio_irq_enable(struct nouveau_gpio *, int line, bool);
+
+#endif
diff --git a/nvkm/include/subdev/i2c.h b/nvkm/include/subdev/i2c.h
new file mode 100644
index 000000000..b93ab01e3
--- /dev/null
+++ b/nvkm/include/subdev/i2c.h
@@ -0,0 +1,60 @@
+#ifndef __NOUVEAU_I2C_H__
+#define __NOUVEAU_I2C_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/i2c.h>
+
+#define NV_I2C_PORT(n)    (0x00 + (n))
+#define NV_I2C_DEFAULT(n) (0x80 + (n))
+
+struct nouveau_i2c_port {
+	struct i2c_adapter adapter;
+	struct nouveau_i2c *i2c;
+	struct i2c_algo_bit_data bit;
+	struct list_head head;
+	u8  index;
+	u8  type;
+	u32 dcb;
+	u32 drive;
+	u32 sense;
+	u32 state;
+};
+
+struct nouveau_i2c {
+	struct nouveau_subdev base;
+
+	struct nouveau_i2c_port *(*find)(struct nouveau_i2c *, u8 index);
+	int (*identify)(struct nouveau_i2c *, int index,
+			const char *what, struct i2c_board_info *,
+			bool (*match)(struct nouveau_i2c_port *,
+				      struct i2c_board_info *));
+	struct list_head ports;
+};
+
+static inline struct nouveau_i2c *
+nouveau_i2c(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_I2C];
+}
+
+extern struct nouveau_oclass nouveau_i2c_oclass;
+
+void nouveau_i2c_drive_scl(void *, int);
+void nouveau_i2c_drive_sda(void *, int);
+int  nouveau_i2c_sense_scl(void *);
+int  nouveau_i2c_sense_sda(void *);
+
+int  nv_rdi2cr(struct nouveau_i2c_port *, u8 addr, u8 reg);
+int  nv_wri2cr(struct nouveau_i2c_port *, u8 addr, u8 reg, u8 val);
+bool nv_probe_i2c(struct nouveau_i2c_port *, u8 addr);
+
+int nv_rdaux(struct nouveau_i2c_port *, u32 addr, u8 *data, u8 size);
+int nv_wraux(struct nouveau_i2c_port *, u32 addr, u8 *data, u8 size);
+
+extern const struct i2c_algorithm nouveau_i2c_bit_algo;
+extern const struct i2c_algorithm nouveau_i2c_aux_algo;
+
+#endif
diff --git a/nvkm/include/subdev/ibus.h b/nvkm/include/subdev/ibus.h
new file mode 100644
index 000000000..88814f159
--- /dev/null
+++ b/nvkm/include/subdev/ibus.h
@@ -0,0 +1,34 @@
+#ifndef __NOUVEAU_IBUS_H__
+#define __NOUVEAU_IBUS_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+struct nouveau_ibus {
+	struct nouveau_subdev base;
+};
+
+static inline struct nouveau_ibus *
+nouveau_ibus(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_IBUS];
+}
+
+#define nouveau_ibus_create(p,e,o,d)                                           \
+	nouveau_subdev_create_((p), (e), (o), 0, "PIBUS", "ibus",              \
+			       sizeof(**d), (void **)d)
+#define nouveau_ibus_destroy(p)                                                \
+	nouveau_subdev_destroy(&(p)->base)
+#define nouveau_ibus_init(p)                                                   \
+	nouveau_subdev_init(&(p)->base)
+#define nouveau_ibus_fini(p,s)                                                 \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+#define _nouveau_ibus_dtor _nouveau_subdev_dtor
+#define _nouveau_ibus_init _nouveau_subdev_init
+#define _nouveau_ibus_fini _nouveau_subdev_fini
+
+extern struct nouveau_oclass nvc0_ibus_oclass;
+extern struct nouveau_oclass nve0_ibus_oclass;
+
+#endif
diff --git a/nvkm/include/subdev/instmem.h b/nvkm/include/subdev/instmem.h
new file mode 100644
index 000000000..ec7a54e91
--- /dev/null
+++ b/nvkm/include/subdev/instmem.h
@@ -0,0 +1,73 @@
+#ifndef __NOUVEAU_INSTMEM_H__
+#define __NOUVEAU_INSTMEM_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+#include <core/mm.h>
+
+struct nouveau_instobj {
+	struct nouveau_object base;
+	struct list_head head;
+	u32 *suspend;
+	u64 addr;
+	u32 size;
+};
+
+static inline struct nouveau_instobj *
+nv_memobj(void *obj)
+{
+#if CONFIG_NOUVEAU_DEBUG >= NV_DBG_PARANOIA
+	if (unlikely(!nv_iclass(obj, NV_MEMOBJ_CLASS)))
+		nv_assert("BAD CAST -> NvMemObj, %08x", nv_hclass(obj));
+#endif
+	return obj;
+}
+
+#define nouveau_instobj_create(p,e,o,d)                                        \
+	nouveau_instobj_create_((p), (e), (o), sizeof(**d), (void **)d)
+#define nouveau_instobj_init(p)                                                \
+	nouveau_object_init(&(p)->base)
+#define nouveau_instobj_fini(p,s)                                              \
+	nouveau_object_fini(&(p)->base, (s))
+
+int  nouveau_instobj_create_(struct nouveau_object *, struct nouveau_object *,
+			     struct nouveau_oclass *, int, void **);
+void nouveau_instobj_destroy(struct nouveau_instobj *);
+
+void _nouveau_instobj_dtor(struct nouveau_object *);
+#define _nouveau_instobj_init nouveau_object_init
+#define _nouveau_instobj_fini nouveau_object_fini
+
+struct nouveau_instmem {
+	struct nouveau_subdev base;
+	struct list_head list;
+
+	u32 reserved;
+	int (*alloc)(struct nouveau_instmem *, struct nouveau_object *,
+		     u32 size, u32 align, struct nouveau_object **);
+};
+
+static inline struct nouveau_instmem *
+nouveau_instmem(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_INSTMEM];
+}
+
+#define nouveau_instmem_create(p,e,o,d)                                        \
+	nouveau_instmem_create_((p), (e), (o), sizeof(**d), (void **)d)
+#define nouveau_instmem_destroy(p)                                             \
+	nouveau_subdev_destroy(&(p)->base)
+int nouveau_instmem_create_(struct nouveau_object *, struct nouveau_object *,
+			    struct nouveau_oclass *, int, void **);
+int nouveau_instmem_init(struct nouveau_instmem *);
+int nouveau_instmem_fini(struct nouveau_instmem *, bool);
+
+#define _nouveau_instmem_dtor _nouveau_subdev_dtor
+int _nouveau_instmem_init(struct nouveau_object *);
+int _nouveau_instmem_fini(struct nouveau_object *, bool);
+
+extern struct nouveau_oclass nv04_instmem_oclass;
+extern struct nouveau_oclass nv40_instmem_oclass;
+extern struct nouveau_oclass nv50_instmem_oclass;
+
+#endif
diff --git a/nvkm/include/subdev/ltcg.h b/nvkm/include/subdev/ltcg.h
new file mode 100644
index 000000000..f351f63bc
--- /dev/null
+++ b/nvkm/include/subdev/ltcg.h
@@ -0,0 +1,33 @@
+#ifndef __NOUVEAU_LTCG_H__
+#define __NOUVEAU_LTCG_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+struct nouveau_ltcg {
+	struct nouveau_subdev base;
+};
+
+static inline struct nouveau_ltcg *
+nouveau_ltcg(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_LTCG];
+}
+
+#define nouveau_ltcg_create(p,e,o,d)                                           \
+	nouveau_subdev_create_((p), (e), (o), 0, "PLTCG", "level2",            \
+			       sizeof(**d), (void **)d)
+#define nouveau_ltcg_destroy(p)                                                \
+	nouveau_subdev_destroy(&(p)->base)
+#define nouveau_ltcg_init(p)                                                   \
+	nouveau_subdev_init(&(p)->base)
+#define nouveau_ltcg_fini(p,s)                                                 \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+#define _nouveau_ltcg_dtor _nouveau_subdev_dtor
+#define _nouveau_ltcg_init _nouveau_subdev_init
+#define _nouveau_ltcg_fini _nouveau_subdev_fini
+
+extern struct nouveau_oclass nvc0_ltcg_oclass;
+
+#endif
diff --git a/nvkm/include/subdev/mc.h b/nvkm/include/subdev/mc.h
new file mode 100644
index 000000000..fded97cea
--- /dev/null
+++ b/nvkm/include/subdev/mc.h
@@ -0,0 +1,49 @@
+#ifndef __NOUVEAU_MC_H__
+#define __NOUVEAU_MC_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+struct nouveau_mc_intr {
+	u32 stat;
+	u32 unit;
+};
+
+struct nouveau_mc {
+	struct nouveau_subdev base;
+	const struct nouveau_mc_intr *intr_map;
+};
+
+static inline struct nouveau_mc *
+nouveau_mc(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_MC];
+}
+
+#define nouveau_mc_create(p,e,o,d)                                             \
+	nouveau_subdev_create_((p), (e), (o), 0, "PMC", "master",              \
+			       sizeof(**d), (void **)d)
+#define nouveau_mc_destroy(p)                                                  \
+	nouveau_subdev_destroy(&(p)->base)
+#define nouveau_mc_init(p)                                                     \
+	nouveau_subdev_init(&(p)->base)
+#define nouveau_mc_fini(p,s)                                                   \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+#define _nouveau_mc_dtor _nouveau_subdev_dtor
+#define _nouveau_mc_init _nouveau_subdev_init
+#define _nouveau_mc_fini _nouveau_subdev_fini
+
+extern struct nouveau_oclass nv04_mc_oclass;
+extern struct nouveau_oclass nv44_mc_oclass;
+extern struct nouveau_oclass nv50_mc_oclass;
+extern struct nouveau_oclass nv98_mc_oclass;
+extern struct nouveau_oclass nvc0_mc_oclass;
+
+void nouveau_mc_intr(struct nouveau_subdev *);
+
+extern const struct nouveau_mc_intr nv04_mc_intr[];
+int nv04_mc_init(struct nouveau_object *);
+int nv50_mc_init(struct nouveau_object *);
+
+#endif
diff --git a/nvkm/include/subdev/mxm.h b/nvkm/include/subdev/mxm.h
new file mode 100644
index 000000000..b93b152cb
--- /dev/null
+++ b/nvkm/include/subdev/mxm.h
@@ -0,0 +1,37 @@
+#ifndef __NOUVEAU_MXM_H__
+#define __NOUVEAU_MXM_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+#define MXM_SANITISE_DCB 0x00000001
+
+struct nouveau_mxm {
+	struct nouveau_subdev base;
+	u32 action;
+	u8 *mxms;
+};
+
+static inline struct nouveau_mxm *
+nouveau_mxm(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_MXM];
+}
+
+#define nouveau_mxm_create(p,e,o,d)                                            \
+	nouveau_mxm_create_((p), (e), (o), sizeof(**d), (void **)d)
+#define nouveau_mxm_init(p)                                                    \
+	nouveau_subdev_init(&(p)->base)
+#define nouveau_mxm_fini(p,s)                                                  \
+	nouveau_subdev_fini(&(p)->base, (s))
+int  nouveau_mxm_create_(struct nouveau_object *, struct nouveau_object *,
+			 struct nouveau_oclass *, int, void **);
+void nouveau_mxm_destroy(struct nouveau_mxm *);
+
+#define _nouveau_mxm_dtor _nouveau_subdev_dtor
+#define _nouveau_mxm_init _nouveau_subdev_init
+#define _nouveau_mxm_fini _nouveau_subdev_fini
+
+extern struct nouveau_oclass nv50_mxm_oclass;
+
+#endif
diff --git a/nvkm/include/subdev/therm.h b/nvkm/include/subdev/therm.h
new file mode 100644
index 000000000..faee569fd
--- /dev/null
+++ b/nvkm/include/subdev/therm.h
@@ -0,0 +1,58 @@
+#ifndef __NOUVEAU_THERM_H__
+#define __NOUVEAU_THERM_H__
+
+#include <core/device.h>
+#include <core/subdev.h>
+
+enum nouveau_therm_fan_mode {
+	FAN_CONTROL_NONE = 0,
+	FAN_CONTROL_MANUAL = 1,
+	FAN_CONTROL_NR,
+};
+
+enum nouveau_therm_attr_type {
+	NOUVEAU_THERM_ATTR_FAN_MIN_DUTY = 0,
+	NOUVEAU_THERM_ATTR_FAN_MAX_DUTY = 1,
+	NOUVEAU_THERM_ATTR_FAN_MODE = 2,
+
+	NOUVEAU_THERM_ATTR_THRS_FAN_BOOST = 10,
+	NOUVEAU_THERM_ATTR_THRS_FAN_BOOST_HYST = 11,
+	NOUVEAU_THERM_ATTR_THRS_DOWN_CLK = 12,
+	NOUVEAU_THERM_ATTR_THRS_DOWN_CLK_HYST = 13,
+	NOUVEAU_THERM_ATTR_THRS_CRITICAL = 14,
+	NOUVEAU_THERM_ATTR_THRS_CRITICAL_HYST = 15,
+	NOUVEAU_THERM_ATTR_THRS_SHUTDOWN = 16,
+	NOUVEAU_THERM_ATTR_THRS_SHUTDOWN_HYST = 17,
+};
+
+struct nouveau_therm {
+	struct nouveau_subdev base;
+
+	int (*fan_get)(struct nouveau_therm *);
+	int (*fan_set)(struct nouveau_therm *, int);
+	int (*fan_sense)(struct nouveau_therm *);
+
+	int (*temp_get)(struct nouveau_therm *);
+
+	int (*attr_get)(struct nouveau_therm *, enum nouveau_therm_attr_type);
+	int (*attr_set)(struct nouveau_therm *,
+			enum nouveau_therm_attr_type, int);
+};
+
+static inline struct nouveau_therm *
+nouveau_therm(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_THERM];
+}
+
+#define nouveau_therm_create(p,e,o,d)                                          \
+	nouveau_subdev_create((p), (e), (o), 0, "THERM", "therm", d)
+#define nouveau_therm_destroy(p)                                               \
+	nouveau_subdev_destroy(&(p)->base)
+
+#define _nouveau_therm_dtor _nouveau_subdev_dtor
+
+extern struct nouveau_oclass nv40_therm_oclass;
+extern struct nouveau_oclass nv50_therm_oclass;
+
+#endif
diff --git a/nvkm/include/subdev/timer.h b/nvkm/include/subdev/timer.h
new file mode 100644
index 000000000..c24ec8ab3
--- /dev/null
+++ b/nvkm/include/subdev/timer.h
@@ -0,0 +1,53 @@
+#ifndef __NOUVEAU_TIMER_H__
+#define __NOUVEAU_TIMER_H__
+
+#include <core/subdev.h>
+#include <core/device.h>
+
+struct nouveau_alarm {
+	struct list_head head;
+	u64 timestamp;
+	void (*func)(struct nouveau_alarm *);
+};
+
+bool nouveau_timer_wait_eq(void *, u64 nsec, u32 addr, u32 mask, u32 data);
+bool nouveau_timer_wait_ne(void *, u64 nsec, u32 addr, u32 mask, u32 data);
+bool nouveau_timer_wait_cb(void *, u64 nsec, bool (*func)(void *), void *data);
+void nouveau_timer_alarm(void *, u32 nsec, struct nouveau_alarm *);
+
+#define NV_WAIT_DEFAULT 2000000000ULL
+#define nv_wait(o,a,m,v)                                                       \
+	nouveau_timer_wait_eq((o), NV_WAIT_DEFAULT, (a), (m), (v))
+#define nv_wait_ne(o,a,m,v)                                                    \
+	nouveau_timer_wait_ne((o), NV_WAIT_DEFAULT, (a), (m), (v))
+#define nv_wait_cb(o,c,d)                                                      \
+	nouveau_timer_wait_cb((o), NV_WAIT_DEFAULT, (c), (d))
+
+struct nouveau_timer {
+	struct nouveau_subdev base;
+	u64  (*read)(struct nouveau_timer *);
+	void (*alarm)(struct nouveau_timer *, u64 time, struct nouveau_alarm *);
+};
+
+static inline struct nouveau_timer *
+nouveau_timer(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_TIMER];
+}
+
+#define nouveau_timer_create(p,e,o,d)                                          \
+	nouveau_subdev_create_((p), (e), (o), 0, "PTIMER", "timer",            \
+			       sizeof(**d), (void **)d)
+#define nouveau_timer_destroy(p)                                               \
+	nouveau_subdev_destroy(&(p)->base)
+#define nouveau_timer_init(p)                                                  \
+	nouveau_subdev_init(&(p)->base)
+#define nouveau_timer_fini(p,s)                                                \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+int nouveau_timer_create_(struct nouveau_object *, struct nouveau_engine *,
+			  struct nouveau_oclass *, int size, void **);
+
+extern struct nouveau_oclass nv04_timer_oclass;
+
+#endif
diff --git a/nvkm/include/subdev/vga.h b/nvkm/include/subdev/vga.h
new file mode 100644
index 000000000..fee09ad81
--- /dev/null
+++ b/nvkm/include/subdev/vga.h
@@ -0,0 +1,30 @@
+#ifndef __NOUVEAU_VGA_H__
+#define __NOUVEAU_VGA_H__
+
+#include <core/os.h>
+
+/* access to various legacy io ports */
+u8   nv_rdport(void *obj, int head, u16 port);
+void nv_wrport(void *obj, int head, u16 port, u8 value);
+
+/* VGA Sequencer */
+u8   nv_rdvgas(void *obj, int head, u8 index);
+void nv_wrvgas(void *obj, int head, u8 index, u8 value);
+
+/* VGA Graphics */
+u8   nv_rdvgag(void *obj, int head, u8 index);
+void nv_wrvgag(void *obj, int head, u8 index, u8 value);
+
+/* VGA CRTC */
+u8   nv_rdvgac(void *obj, int head, u8 index);
+void nv_wrvgac(void *obj, int head, u8 index, u8 value);
+
+/* VGA indexed port access dispatcher */
+u8   nv_rdvgai(void *obj, int head, u16 port, u8 index);
+void nv_wrvgai(void *obj, int head, u16 port, u8 index, u8 value);
+
+bool nv_lockvgac(void *obj, bool lock);
+u8   nv_rdvgaowner(void *obj);
+void nv_wrvgaowner(void *obj, u8);
+
+#endif
diff --git a/nvkm/include/subdev/vm.h b/nvkm/include/subdev/vm.h
new file mode 100644
index 000000000..9d595efe6
--- /dev/null
+++ b/nvkm/include/subdev/vm.h
@@ -0,0 +1,142 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#ifndef __NOUVEAU_VM_H__
+#define __NOUVEAU_VM_H__
+
+#include <core/object.h>
+#include <core/subdev.h>
+#include <core/device.h>
+#include <core/mm.h>
+
+struct nouveau_vm_pgt {
+	struct nouveau_gpuobj *obj[2];
+	u32 refcount[2];
+};
+
+struct nouveau_vm_pgd {
+	struct list_head head;
+	struct nouveau_gpuobj *obj;
+};
+
+struct nouveau_gpuobj;
+struct nouveau_mem;
+
+struct nouveau_vma {
+	struct list_head head;
+	int refcount;
+	struct nouveau_vm *vm;
+	struct nouveau_mm_node *node;
+	u64 offset;
+	u32 access;
+};
+
+struct nouveau_vm {
+	struct nouveau_vmmgr *vmm;
+	struct nouveau_mm mm;
+	int refcount;
+
+	struct list_head pgd_list;
+	atomic_t engref[64]; //NVDEV_SUBDEV_NR];
+
+	struct nouveau_vm_pgt *pgt;
+	u32 fpde;
+	u32 lpde;
+};
+
+struct nouveau_vmmgr {
+	struct nouveau_subdev base;
+
+	u64 limit;
+	u8  dma_bits;
+	u32 pgt_bits;
+	u8  spg_shift;
+	u8  lpg_shift;
+
+	int  (*create)(struct nouveau_vmmgr *, u64 offset, u64 length,
+		       u64 mm_offset, struct nouveau_vm **);
+
+	void (*map_pgt)(struct nouveau_gpuobj *pgd, u32 pde,
+			struct nouveau_gpuobj *pgt[2]);
+	void (*map)(struct nouveau_vma *, struct nouveau_gpuobj *,
+		    struct nouveau_mem *, u32 pte, u32 cnt,
+		    u64 phys, u64 delta);
+	void (*map_sg)(struct nouveau_vma *, struct nouveau_gpuobj *,
+		       struct nouveau_mem *, u32 pte, u32 cnt, dma_addr_t *);
+	void (*unmap)(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt);
+	void (*flush)(struct nouveau_vm *);
+};
+
+static inline struct nouveau_vmmgr *
+nouveau_vmmgr(void *obj)
+{
+	return (void *)nv_device(obj)->subdev[NVDEV_SUBDEV_VM];
+}
+
+#define nouveau_vmmgr_create(p,e,o,i,f,d)                                      \
+	nouveau_subdev_create((p), (e), (o), 0, (i), (f), (d))
+#define nouveau_vmmgr_destroy(p)                                               \
+	nouveau_subdev_destroy(&(p)->base)
+#define nouveau_vmmgr_init(p)                                                  \
+	nouveau_subdev_init(&(p)->base)
+#define nouveau_vmmgr_fini(p,s)                                                \
+	nouveau_subdev_fini(&(p)->base, (s))
+
+#define _nouveau_vmmgr_dtor _nouveau_subdev_dtor
+#define _nouveau_vmmgr_init _nouveau_subdev_init
+#define _nouveau_vmmgr_fini _nouveau_subdev_fini
+
+extern struct nouveau_oclass nv04_vmmgr_oclass;
+extern struct nouveau_oclass nv41_vmmgr_oclass;
+extern struct nouveau_oclass nv44_vmmgr_oclass;
+extern struct nouveau_oclass nv50_vmmgr_oclass;
+extern struct nouveau_oclass nvc0_vmmgr_oclass;
+
+int  nv04_vm_create(struct nouveau_vmmgr *, u64, u64, u64,
+		    struct nouveau_vm **);
+void nv04_vmmgr_dtor(struct nouveau_object *);
+
+void nv50_vm_flush_engine(struct nouveau_subdev *, int engine);
+void nvc0_vm_flush_engine(struct nouveau_subdev *, u64 addr, int type);
+
+/* nouveau_vm.c */
+int  nouveau_vm_create(struct nouveau_vmmgr *, u64 offset, u64 length,
+		       u64 mm_offset, u32 block, struct nouveau_vm **);
+int  nouveau_vm_new(struct nouveau_device *, u64 offset, u64 length,
+		    u64 mm_offset, struct nouveau_vm **);
+int  nouveau_vm_ref(struct nouveau_vm *, struct nouveau_vm **,
+		    struct nouveau_gpuobj *pgd);
+int  nouveau_vm_get(struct nouveau_vm *, u64 size, u32 page_shift,
+		    u32 access, struct nouveau_vma *);
+void nouveau_vm_put(struct nouveau_vma *);
+void nouveau_vm_map(struct nouveau_vma *, struct nouveau_mem *);
+void nouveau_vm_map_at(struct nouveau_vma *, u64 offset, struct nouveau_mem *);
+void nouveau_vm_unmap(struct nouveau_vma *);
+void nouveau_vm_unmap_at(struct nouveau_vma *, u64 offset, u64 length);
+void nouveau_vm_map_sg(struct nouveau_vma *, u64 offset, u64 length,
+		       struct nouveau_mem *);
+void nouveau_vm_map_sg_table(struct nouveau_vma *vma, u64 delta, u64 length,
+		     struct nouveau_mem *mem);
+
+#endif
diff --git a/nvkm/subdev/Makefile.am b/nvkm/subdev/Makefile.am
new file mode 100644
index 000000000..9010528de
--- /dev/null
+++ b/nvkm/subdev/Makefile.am
@@ -0,0 +1,22 @@
+SUBDIRS = bar bios clock device devinit fb gpio i2c ibus instmem ltcg \
+	  mc mxm therm timer vm
+
+noinst_LTLIBRARIES = libsubdev.la
+
+libsubdev_la_SOURCES  =
+libsubdev_la_LIBADD   = bar/libbar.la \
+			bios/libbios.la \
+			clock/libclock.la \
+			device/libdevice.la \
+			devinit/libdevinit.la \
+			fb/libfb.la \
+			gpio/libgpio.la \
+			i2c/libi2c.la \
+			ibus/libibus.la \
+			instmem/libinstmem.la \
+			ltcg/libltcg.la \
+			mc/libmc.la \
+			mxm/libmxm.la \
+			therm/libtherm.la \
+			timer/libtimer.la \
+			vm/libvm.la
diff --git a/nvkm/subdev/Makefile.subdev b/nvkm/subdev/Makefile.subdev
new file mode 100644
index 000000000..513c6c87b
--- /dev/null
+++ b/nvkm/subdev/Makefile.subdev
@@ -0,0 +1,3 @@
+AM_CPPFLAGS = -I${top_srcdir}/nvkm \
+	      -I${top_srcdir}/nvkm/include \
+	      -I${top_srcdir}/lib
diff --git a/nvkm/subdev/bar/Makefile.am b/nvkm/subdev/bar/Makefile.am
new file mode 100644
index 000000000..c5108817c
--- /dev/null
+++ b/nvkm/subdev/bar/Makefile.am
@@ -0,0 +1,7 @@
+noinst_LTLIBRARIES = libbar.la
+
+libbar_la_SOURCES = base.c \
+		    nv50.c \
+		    nvc0.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/bar/base.c b/nvkm/subdev/bar/base.c
new file mode 100644
index 000000000..d70ba342a
--- /dev/null
+++ b/nvkm/subdev/bar/base.c
@@ -0,0 +1,135 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <subdev/bar.h>
+
+struct nouveau_barobj {
+	struct nouveau_object base;
+	struct nouveau_vma vma;
+	void __iomem *iomem;
+};
+
+static int
+nouveau_barobj_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *mem, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nouveau_bar *bar = (void *)engine;
+	struct nouveau_barobj *barobj;
+	int ret;
+
+	ret = nouveau_object_create(parent, engine, oclass, 0, &barobj);
+	*pobject = nv_object(barobj);
+	if (ret)
+		return ret;
+
+	ret = bar->kmap(bar, mem, NV_MEM_ACCESS_RW, &barobj->vma);
+	if (ret)
+		return ret;
+
+	barobj->iomem = bar->iomem + (u32)barobj->vma.offset;
+	return 0;
+}
+
+static void
+nouveau_barobj_dtor(struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = (void *)object->engine;
+	struct nouveau_barobj *barobj = (void *)object;
+	if (barobj->vma.node)
+		bar->unmap(bar, &barobj->vma);
+	nouveau_object_destroy(&barobj->base);
+}
+
+static u32
+nouveau_barobj_rd32(struct nouveau_object *object, u64 addr)
+{
+	struct nouveau_barobj *barobj = (void *)object;
+	return ioread32_native(barobj->iomem + addr);
+}
+
+static void
+nouveau_barobj_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	struct nouveau_barobj *barobj = (void *)object;
+	iowrite32_native(data, barobj->iomem + addr);
+}
+
+static struct nouveau_oclass
+nouveau_barobj_oclass = {
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nouveau_barobj_ctor,
+		.dtor = nouveau_barobj_dtor,
+		.init = nouveau_object_init,
+		.fini = nouveau_object_fini,
+		.rd32 = nouveau_barobj_rd32,
+		.wr32 = nouveau_barobj_wr32,
+	},
+};
+
+int
+nouveau_bar_alloc(struct nouveau_bar *bar, struct nouveau_object *parent,
+		  struct nouveau_mem *mem, struct nouveau_object **pobject)
+{
+	struct nouveau_object *engine = nv_object(bar);
+	return nouveau_object_ctor(parent, engine, &nouveau_barobj_oclass,
+				   mem, 0, pobject);
+}
+
+int
+nouveau_bar_create_(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, int length, void **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nouveau_bar *bar;
+	int ret;
+
+	ret = nouveau_subdev_create_(parent, engine, oclass, 0, "BARCTL",
+				     "bar", length, pobject);
+	bar = *pobject;
+	if (ret)
+		return ret;
+
+	bar->iomem = ioremap(pci_resource_start(device->pdev, 3),
+			     pci_resource_len(device->pdev, 3));
+	return 0;
+}
+
+void
+nouveau_bar_destroy(struct nouveau_bar *bar)
+{
+	if (bar->iomem)
+		iounmap(bar->iomem);
+	nouveau_subdev_destroy(&bar->base);
+}
+
+void
+_nouveau_bar_dtor(struct nouveau_object *object)
+{
+	struct nouveau_bar *bar = (void *)object;
+	nouveau_bar_destroy(bar);
+}
diff --git a/nvkm/subdev/bar/nv50.c b/nvkm/subdev/bar/nv50.c
new file mode 100644
index 000000000..c3acf5b70
--- /dev/null
+++ b/nvkm/subdev/bar/nv50.c
@@ -0,0 +1,263 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/gpuobj.h>
+
+#include <subdev/timer.h>
+#include <subdev/bar.h>
+#include <subdev/fb.h>
+#include <subdev/vm.h>
+
+struct nv50_bar_priv {
+	struct nouveau_bar base;
+	spinlock_t lock;
+	struct nouveau_gpuobj *mem;
+	struct nouveau_gpuobj *pad;
+	struct nouveau_gpuobj *pgd;
+	struct nouveau_vm *bar1_vm;
+	struct nouveau_gpuobj *bar1;
+	struct nouveau_vm *bar3_vm;
+	struct nouveau_gpuobj *bar3;
+};
+
+static int
+nv50_bar_kmap(struct nouveau_bar *bar, struct nouveau_mem *mem,
+	      u32 flags, struct nouveau_vma *vma)
+{
+	struct nv50_bar_priv *priv = (void *)bar;
+	int ret;
+
+	ret = nouveau_vm_get(priv->bar3_vm, mem->size << 12, 12, flags, vma);
+	if (ret)
+		return ret;
+
+	nouveau_vm_map(vma, mem);
+	nv50_vm_flush_engine(nv_subdev(bar), 6);
+	return 0;
+}
+
+static int
+nv50_bar_umap(struct nouveau_bar *bar, struct nouveau_mem *mem,
+	      u32 flags, struct nouveau_vma *vma)
+{
+	struct nv50_bar_priv *priv = (void *)bar;
+	int ret;
+
+	ret = nouveau_vm_get(priv->bar1_vm, mem->size << 12, 12, flags, vma);
+	if (ret)
+		return ret;
+
+	nouveau_vm_map(vma, mem);
+	nv50_vm_flush_engine(nv_subdev(bar), 6);
+	return 0;
+}
+
+static void
+nv50_bar_unmap(struct nouveau_bar *bar, struct nouveau_vma *vma)
+{
+	nouveau_vm_unmap(vma);
+	nv50_vm_flush_engine(nv_subdev(bar), 6);
+	nouveau_vm_put(vma);
+}
+
+static void
+nv50_bar_flush(struct nouveau_bar *bar)
+{
+	struct nv50_bar_priv *priv = (void *)bar;
+	unsigned long flags;
+	spin_lock_irqsave(&priv->lock, flags);
+	nv_wr32(priv, 0x00330c, 0x00000001);
+	if (!nv_wait(priv, 0x00330c, 0x00000002, 0x00000000))
+		nv_warn(priv, "flush timeout\n");
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+void
+nv84_bar_flush(struct nouveau_bar *bar)
+{
+	struct nv50_bar_priv *priv = (void *)bar;
+	unsigned long flags;
+	spin_lock_irqsave(&priv->lock, flags);
+	nv_wr32(bar, 0x070000, 0x00000001);
+	if (!nv_wait(priv, 0x070000, 0x00000002, 0x00000000))
+		nv_warn(priv, "flush timeout\n");
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+static int
+nv50_bar_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nouveau_object *heap;
+	struct nouveau_vm *vm;
+	struct nv50_bar_priv *priv;
+	u64 start, limit;
+	int ret;
+
+	ret = nouveau_bar_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x20000, 0, NVOBJ_FLAG_HEAP,
+				&priv->mem);
+	heap = nv_object(priv->mem);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, heap, (device->chipset == 0x50) ?
+				 0x1400 : 0x0200, 0, 0, &priv->pad);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, heap, 0x4000, 0, 0, &priv->pgd);
+	if (ret)
+		return ret;
+
+	/* BAR3 */
+	start = 0x0100000000ULL;
+	limit = start + pci_resource_len(device->pdev, 3);
+
+	ret = nouveau_vm_new(device, start, limit, start, &vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, heap, ((limit-- - start) >> 12) * 8,
+				 0x1000, NVOBJ_FLAG_ZERO_ALLOC,
+				 &vm->pgt[0].obj[0]);
+	vm->pgt[0].refcount[0] = 1;
+	if (ret)
+		return ret;
+
+	ret = nouveau_vm_ref(vm, &priv->bar3_vm, priv->pgd);
+	nouveau_vm_ref(NULL, &vm, NULL);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, heap, 24, 16, 0, &priv->bar3);
+	if (ret)
+		return ret;
+
+	nv_wo32(priv->bar3, 0x00, 0x7fc00000);
+	nv_wo32(priv->bar3, 0x04, lower_32_bits(limit));
+	nv_wo32(priv->bar3, 0x08, lower_32_bits(start));
+	nv_wo32(priv->bar3, 0x0c, upper_32_bits(limit) << 24 |
+				  upper_32_bits(start));
+	nv_wo32(priv->bar3, 0x10, 0x00000000);
+	nv_wo32(priv->bar3, 0x14, 0x00000000);
+
+	/* BAR1 */
+	start = 0x0000000000ULL;
+	limit = start + pci_resource_len(device->pdev, 1);
+
+	ret = nouveau_vm_new(device, start, limit--, start, &vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_vm_ref(vm, &priv->bar1_vm, priv->pgd);
+	nouveau_vm_ref(NULL, &vm, NULL);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, heap, 24, 16, 0, &priv->bar1);
+	if (ret)
+		return ret;
+
+	nv_wo32(priv->bar1, 0x00, 0x7fc00000);
+	nv_wo32(priv->bar1, 0x04, lower_32_bits(limit));
+	nv_wo32(priv->bar1, 0x08, lower_32_bits(start));
+	nv_wo32(priv->bar1, 0x0c, upper_32_bits(limit) << 24 |
+				  upper_32_bits(start));
+	nv_wo32(priv->bar1, 0x10, 0x00000000);
+	nv_wo32(priv->bar1, 0x14, 0x00000000);
+
+	priv->base.alloc = nouveau_bar_alloc;
+	priv->base.kmap = nv50_bar_kmap;
+	priv->base.umap = nv50_bar_umap;
+	priv->base.unmap = nv50_bar_unmap;
+	if (device->chipset == 0x50)
+		priv->base.flush = nv50_bar_flush;
+	else
+		priv->base.flush = nv84_bar_flush;
+	spin_lock_init(&priv->lock);
+	return 0;
+}
+
+static void
+nv50_bar_dtor(struct nouveau_object *object)
+{
+	struct nv50_bar_priv *priv = (void *)object;
+	nouveau_gpuobj_ref(NULL, &priv->bar1);
+	nouveau_vm_ref(NULL, &priv->bar1_vm, priv->pgd);
+	nouveau_gpuobj_ref(NULL, &priv->bar3);
+	if (priv->bar3_vm) {
+		nouveau_gpuobj_ref(NULL, &priv->bar3_vm->pgt[0].obj[0]);
+		nouveau_vm_ref(NULL, &priv->bar3_vm, priv->pgd);
+	}
+	nouveau_gpuobj_ref(NULL, &priv->pgd);
+	nouveau_gpuobj_ref(NULL, &priv->pad);
+	nouveau_gpuobj_ref(NULL, &priv->mem);
+	nouveau_bar_destroy(&priv->base);
+}
+
+static int
+nv50_bar_init(struct nouveau_object *object)
+{
+	struct nv50_bar_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_bar_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_mask(priv, 0x000200, 0x00000100, 0x00000000);
+	nv_mask(priv, 0x000200, 0x00000100, 0x00000100);
+	nv50_vm_flush_engine(nv_subdev(priv), 6);
+
+	nv_wr32(priv, 0x001704, 0x00000000 | priv->mem->addr >> 12);
+	nv_wr32(priv, 0x001704, 0x40000000 | priv->mem->addr >> 12);
+	nv_wr32(priv, 0x001708, 0x80000000 | priv->bar1->node->offset >> 4);
+	nv_wr32(priv, 0x00170c, 0x80000000 | priv->bar3->node->offset >> 4);
+	return 0;
+}
+
+static int
+nv50_bar_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_bar_priv *priv = (void *)object;
+	return nouveau_bar_fini(&priv->base, suspend);
+}
+
+struct nouveau_oclass
+nv50_bar_oclass = {
+	.handle = NV_SUBDEV(BAR, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_bar_ctor,
+		.dtor = nv50_bar_dtor,
+		.init = nv50_bar_init,
+		.fini = nv50_bar_fini,
+	},
+};
diff --git a/nvkm/subdev/bar/nvc0.c b/nvkm/subdev/bar/nvc0.c
new file mode 100644
index 000000000..77a6fb725
--- /dev/null
+++ b/nvkm/subdev/bar/nvc0.c
@@ -0,0 +1,215 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/gpuobj.h>
+
+#include <subdev/timer.h>
+#include <subdev/bar.h>
+#include <subdev/fb.h>
+#include <subdev/vm.h>
+
+struct nvc0_bar_priv {
+	struct nouveau_bar base;
+	spinlock_t lock;
+	struct {
+		struct nouveau_gpuobj *mem;
+		struct nouveau_gpuobj *pgd;
+		struct nouveau_vm *vm;
+	} bar[2];
+};
+
+static int
+nvc0_bar_kmap(struct nouveau_bar *bar, struct nouveau_mem *mem,
+	      u32 flags, struct nouveau_vma *vma)
+{
+	struct nvc0_bar_priv *priv = (void *)bar;
+	int ret;
+
+	ret = nouveau_vm_get(priv->bar[0].vm, mem->size << 12, 12, flags, vma);
+	if (ret)
+		return ret;
+
+	nouveau_vm_map(vma, mem);
+	nvc0_vm_flush_engine(nv_subdev(bar), priv->bar[0].pgd->addr, 5);
+	return 0;
+}
+
+static int
+nvc0_bar_umap(struct nouveau_bar *bar, struct nouveau_mem *mem,
+	      u32 flags, struct nouveau_vma *vma)
+{
+	struct nvc0_bar_priv *priv = (void *)bar;
+	int ret;
+
+	ret = nouveau_vm_get(priv->bar[1].vm, mem->size << 12,
+			     mem->page_shift, flags, vma);
+	if (ret)
+		return ret;
+
+	nouveau_vm_map(vma, mem);
+	nvc0_vm_flush_engine(nv_subdev(bar), priv->bar[1].pgd->addr, 5);
+	return 0;
+}
+
+static void
+nvc0_bar_unmap(struct nouveau_bar *bar, struct nouveau_vma *vma)
+{
+	struct nvc0_bar_priv *priv = (void *)bar;
+	int i = !(vma->vm == priv->bar[0].vm);
+
+	nouveau_vm_unmap(vma);
+	nvc0_vm_flush_engine(nv_subdev(bar), priv->bar[i].pgd->addr, 5);
+	nouveau_vm_put(vma);
+}
+
+static int
+nvc0_bar_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct pci_dev *pdev = device->pdev;
+	struct nvc0_bar_priv *priv;
+	struct nouveau_gpuobj *mem;
+	struct nouveau_vm *vm;
+	int ret;
+
+	ret = nouveau_bar_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	/* BAR3 */
+	ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 0, 0, &priv->bar[0].mem);
+	mem = priv->bar[0].mem;
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x8000, 0, 0, &priv->bar[0].pgd);
+	if (ret)
+		return ret;
+
+	ret = nouveau_vm_new(device, 0, pci_resource_len(pdev, 3), 0, &vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL,
+				 (pci_resource_len(pdev, 3) >> 12) * 8,
+				 0x1000, NVOBJ_FLAG_ZERO_ALLOC,
+				 &vm->pgt[0].obj[0]);
+	vm->pgt[0].refcount[0] = 1;
+	if (ret)
+		return ret;
+
+	ret = nouveau_vm_ref(vm, &priv->bar[0].vm, priv->bar[0].pgd);
+	nouveau_vm_ref(NULL, &vm, NULL);
+	if (ret)
+		return ret;
+
+	nv_wo32(mem, 0x0200, lower_32_bits(priv->bar[0].pgd->addr));
+	nv_wo32(mem, 0x0204, upper_32_bits(priv->bar[0].pgd->addr));
+	nv_wo32(mem, 0x0208, lower_32_bits(pci_resource_len(pdev, 3) - 1));
+	nv_wo32(mem, 0x020c, upper_32_bits(pci_resource_len(pdev, 3) - 1));
+
+	/* BAR1 */
+	ret = nouveau_gpuobj_new(parent, NULL, 0x1000, 0, 0, &priv->bar[1].mem);
+	mem = priv->bar[1].mem;
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL, 0x8000, 0, 0, &priv->bar[1].pgd);
+	if (ret)
+		return ret;
+
+	ret = nouveau_vm_new(device, 0, pci_resource_len(pdev, 1), 0, &vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_vm_ref(vm, &priv->bar[1].vm, priv->bar[1].pgd);
+	nouveau_vm_ref(NULL, &vm, NULL);
+	if (ret)
+		return ret;
+
+	nv_wo32(mem, 0x0200, lower_32_bits(priv->bar[1].pgd->addr));
+	nv_wo32(mem, 0x0204, upper_32_bits(priv->bar[1].pgd->addr));
+	nv_wo32(mem, 0x0208, lower_32_bits(pci_resource_len(pdev, 1) - 1));
+	nv_wo32(mem, 0x020c, upper_32_bits(pci_resource_len(pdev, 1) - 1));
+
+	priv->base.alloc = nouveau_bar_alloc;
+	priv->base.kmap = nvc0_bar_kmap;
+	priv->base.umap = nvc0_bar_umap;
+	priv->base.unmap = nvc0_bar_unmap;
+	priv->base.flush = nv84_bar_flush;
+	spin_lock_init(&priv->lock);
+	return 0;
+}
+
+static void
+nvc0_bar_dtor(struct nouveau_object *object)
+{
+	struct nvc0_bar_priv *priv = (void *)object;
+
+	nouveau_vm_ref(NULL, &priv->bar[1].vm, priv->bar[1].pgd);
+	nouveau_gpuobj_ref(NULL, &priv->bar[1].pgd);
+	nouveau_gpuobj_ref(NULL, &priv->bar[1].mem);
+
+	if (priv->bar[0].vm) {
+		nouveau_gpuobj_ref(NULL, &priv->bar[0].vm->pgt[0].obj[0]);
+		nouveau_vm_ref(NULL, &priv->bar[0].vm, priv->bar[0].pgd);
+	}
+	nouveau_gpuobj_ref(NULL, &priv->bar[0].pgd);
+	nouveau_gpuobj_ref(NULL, &priv->bar[0].mem);
+
+	nouveau_bar_destroy(&priv->base);
+}
+
+static int
+nvc0_bar_init(struct nouveau_object *object)
+{
+	struct nvc0_bar_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_bar_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_mask(priv, 0x000200, 0x00000100, 0x00000000);
+	nv_mask(priv, 0x000200, 0x00000100, 0x00000100);
+	nv_mask(priv, 0x100c80, 0x00000001, 0x00000000);
+
+	nv_wr32(priv, 0x001704, 0x80000000 | priv->bar[1].mem->addr >> 12);
+	nv_wr32(priv, 0x001714, 0xc0000000 | priv->bar[0].mem->addr >> 12);
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_bar_oclass = {
+	.handle = NV_SUBDEV(BAR, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_bar_ctor,
+		.dtor = nvc0_bar_dtor,
+		.init = nvc0_bar_init,
+		.fini = _nouveau_bar_fini,
+	},
+};
diff --git a/nvkm/subdev/bios/Makefile.am b/nvkm/subdev/bios/Makefile.am
new file mode 100644
index 000000000..68a52a7db
--- /dev/null
+++ b/nvkm/subdev/bios/Makefile.am
@@ -0,0 +1,18 @@
+noinst_LTLIBRARIES = libbios.la
+
+libbios_la_SOURCES = base.c \
+		     bit.c \
+		     conn.c \
+		     dcb.c \
+		     disp.c \
+		     dp.c \
+		     extdev.c \
+		     gpio.c \
+		     i2c.c \
+		     init.c \
+		     mxm.c \
+		     perf.c \
+		     pll.c \
+		     therm.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/bios/base.c b/nvkm/subdev/bios/base.c
new file mode 100644
index 000000000..f621f69fa
--- /dev/null
+++ b/nvkm/subdev/bios/base.c
@@ -0,0 +1,503 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/device.h>
+#include <core/subdev.h>
+#include <core/option.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/bmp.h>
+#include <subdev/bios/bit.h>
+
+u8
+nvbios_checksum(const u8 *data, int size)
+{
+	u8 sum = 0;
+	while (size--)
+		sum += *data++;
+	return sum;
+}
+
+u16
+nvbios_findstr(const u8 *data, int size, const char *str, int len)
+{
+	int i, j;
+
+	for (i = 0; i <= (size - len); i++) {
+		for (j = 0; j < len; j++)
+			if ((char)data[i + j] != str[j])
+				break;
+		if (j == len)
+			return i;
+	}
+
+	return 0;
+}
+
+#if defined(__powerpc__)
+static void
+nouveau_bios_shadow_of(struct nouveau_bios *bios)
+{
+	struct pci_dev *pdev = nv_device(bios)->pdev;
+	struct device_node *dn;
+	const u32 *data;
+	int size;
+
+	dn = pci_device_to_OF_node(pdev);
+	if (!dn) {
+		nv_info(bios, "Unable to get the OF node\n");
+		return;
+	}
+
+	data = of_get_property(dn, "NVDA,BMP", &size);
+	if (data && size) {
+		bios->size = size;
+		bios->data = kmalloc(bios->size, GFP_KERNEL);
+		if (bios->data)
+			memcpy(bios->data, data, size);
+	}
+}
+#endif
+
+static void
+nouveau_bios_shadow_pramin(struct nouveau_bios *bios)
+{
+	struct nouveau_device *device = nv_device(bios);
+	u32 bar0 = 0;
+	int i;
+
+	if (device->card_type >= NV_50) {
+		u64 addr = (u64)(nv_rd32(bios, 0x619f04) & 0xffffff00) << 8;
+		if (!addr) {
+			addr  = (u64)nv_rd32(bios, 0x001700) << 16;
+			addr += 0xf0000;
+		}
+
+		bar0 = nv_mask(bios, 0x001700, 0xffffffff, addr >> 16);
+	}
+
+	/* bail if no rom signature */
+	if (nv_rd08(bios, 0x700000) != 0x55 ||
+	    nv_rd08(bios, 0x700001) != 0xaa)
+		goto out;
+
+	bios->size = nv_rd08(bios, 0x700002) * 512;
+	if (!bios->size)
+		goto out;
+
+	bios->data = kmalloc(bios->size, GFP_KERNEL);
+	if (bios->data) {
+		for (i = 0; i < bios->size; i++)
+			nv_wo08(bios, i, nv_rd08(bios, 0x700000 + i));
+	}
+
+out:
+	if (device->card_type >= NV_50)
+		nv_wr32(bios, 0x001700, bar0);
+}
+
+static void
+nouveau_bios_shadow_prom(struct nouveau_bios *bios)
+{
+	struct nouveau_device *device = nv_device(bios);
+	u32 pcireg, access;
+	u16 pcir;
+	int i;
+
+	/* enable access to rom */
+	if (device->card_type >= NV_50)
+		pcireg = 0x088050;
+	else
+		pcireg = 0x001850;
+	access = nv_mask(bios, pcireg, 0x00000001, 0x00000000);
+
+	/* bail if no rom signature, with a workaround for a PROM reading
+	 * issue on some chipsets.  the first read after a period of
+	 * inactivity returns the wrong result, so retry the first header
+	 * byte a few times before giving up as a workaround
+	 */
+	i = 16;
+	do {
+		if (nv_rd08(bios, 0x300000) == 0x55)
+			break;
+	} while (i--);
+
+	if (!i || nv_rd08(bios, 0x300001) != 0xaa)
+		goto out;
+
+	/* additional check (see note below) - read PCI record header */
+	pcir = nv_rd08(bios, 0x300018) |
+	       nv_rd08(bios, 0x300019) << 8;
+	if (nv_rd08(bios, 0x300000 + pcir) != 'P' ||
+	    nv_rd08(bios, 0x300001 + pcir) != 'C' ||
+	    nv_rd08(bios, 0x300002 + pcir) != 'I' ||
+	    nv_rd08(bios, 0x300003 + pcir) != 'R')
+		goto out;
+
+	/* read entire bios image to system memory */
+	bios->size = nv_rd08(bios, 0x300002) * 512;
+	if (!bios->size)
+		goto out;
+
+	bios->data = kmalloc(bios->size, GFP_KERNEL);
+	if (bios->data) {
+		for (i = 0; i < bios->size; i++)
+			nv_wo08(bios, i, nv_rd08(bios, 0x300000 + i));
+	}
+
+out:
+	/* disable access to rom */
+	nv_wr32(bios, pcireg, access);
+}
+
+#if defined(CONFIG_ACPI)
+int nouveau_acpi_get_bios_chunk(uint8_t *bios, int offset, int len);
+bool nouveau_acpi_rom_supported(struct pci_dev *pdev);
+#else
+static inline bool
+nouveau_acpi_rom_supported(struct pci_dev *pdev) {
+	return false;
+}
+
+static inline int
+nouveau_acpi_get_bios_chunk(uint8_t *bios, int offset, int len) {
+	return -EINVAL;
+}
+#endif
+
+static void
+nouveau_bios_shadow_acpi(struct nouveau_bios *bios)
+{
+	struct pci_dev *pdev = nv_device(bios)->pdev;
+	int ret, cnt, i;
+
+	if (!nouveau_acpi_rom_supported(pdev)) {
+		bios->data = NULL;
+		return;
+	}
+
+	bios->size = 0;
+	bios->data = kmalloc(4096, GFP_KERNEL);
+	if (bios->data) {
+		if (nouveau_acpi_get_bios_chunk(bios->data, 0, 4096) == 4096)
+			bios->size = bios->data[2] * 512;
+		kfree(bios->data);
+	}
+
+	if (!bios->size)
+		return;
+
+	bios->data = kmalloc(bios->size, GFP_KERNEL);
+	if (bios->data) {
+		/* disobey the acpi spec - much faster on at least w530 ... */
+		ret = nouveau_acpi_get_bios_chunk(bios->data, 0, bios->size);
+		if (ret != bios->size ||
+		    nvbios_checksum(bios->data, bios->size)) {
+			/* ... that didn't work, ok, i'll be good now */
+			for (i = 0; i < bios->size; i += cnt) {
+				cnt = min((bios->size - i), (u32)4096);
+				ret = nouveau_acpi_get_bios_chunk(bios->data, i, cnt);
+				if (ret != cnt)
+					break;
+			}
+		}
+	}
+}
+
+static void
+nouveau_bios_shadow_pci(struct nouveau_bios *bios)
+{
+	struct pci_dev *pdev = nv_device(bios)->pdev;
+	size_t size;
+
+	if (!pci_enable_rom(pdev)) {
+		void __iomem *rom = pci_map_rom(pdev, &size);
+		if (rom && size) {
+			bios->data = kmalloc(size, GFP_KERNEL);
+			if (bios->data) {
+				memcpy_fromio(bios->data, rom, size);
+				bios->size = size;
+			}
+		}
+		if (rom)
+			pci_unmap_rom(pdev, rom);
+
+		pci_disable_rom(pdev);
+	}
+}
+
+static int
+nouveau_bios_score(struct nouveau_bios *bios, const bool writeable)
+{
+	if (bios->size < 3 || !bios->data || bios->data[0] != 0x55 ||
+			bios->data[1] != 0xAA) {
+		nv_info(bios, "... signature not found\n");
+		return 0;
+	}
+
+	if (nvbios_checksum(bios->data,
+			min_t(u32, bios->data[2] * 512, bios->size))) {
+		nv_info(bios, "... checksum invalid\n");
+		/* if a ro image is somewhat bad, it's probably all rubbish */
+		return writeable ? 2 : 1;
+	}
+
+	nv_info(bios, "... appears to be valid\n");
+	return 3;
+}
+
+struct methods {
+	const char desc[16];
+	void (*shadow)(struct nouveau_bios *);
+	const bool rw;
+	int score;
+	u32 size;
+	u8 *data;
+};
+
+static int
+nouveau_bios_shadow(struct nouveau_bios *bios)
+{
+	struct methods shadow_methods[] = {
+#if defined(__powerpc__)
+		{ "OpenFirmware", nouveau_bios_shadow_of, true, 0, 0, NULL },
+#endif
+		{ "PRAMIN", nouveau_bios_shadow_pramin, true, 0, 0, NULL },
+		{ "PROM", nouveau_bios_shadow_prom, false, 0, 0, NULL },
+		{ "ACPI", nouveau_bios_shadow_acpi, true, 0, 0, NULL },
+		{ "PCIROM", nouveau_bios_shadow_pci, true, 0, 0, NULL },
+		{}
+	};
+	struct methods *mthd, *best;
+	const struct firmware *fw;
+	const char *optarg;
+	int optlen, ret;
+	char *source;
+
+	optarg = nouveau_stropt(nv_device(bios)->cfgopt, "NvBios", &optlen);
+	source = optarg ? kstrndup(optarg, optlen, GFP_KERNEL) : NULL;
+	if (source) {
+		/* try to match one of the built-in methods */
+		mthd = shadow_methods;
+		do {
+			if (strcasecmp(source, mthd->desc))
+				continue;
+			nv_info(bios, "source: %s\n", mthd->desc);
+
+			mthd->shadow(bios);
+			mthd->score = nouveau_bios_score(bios, mthd->rw);
+			if (mthd->score) {
+				kfree(source);
+				return 0;
+			}
+		} while ((++mthd)->shadow);
+
+		/* attempt to load firmware image */
+		ret = request_firmware(&fw, source, &nv_device(bios)->pdev->dev);
+		if (ret == 0) {
+			bios->size = fw->size;
+			bios->data = kmemdup(fw->data, fw->size, GFP_KERNEL);
+			release_firmware(fw);
+
+			nv_info(bios, "image: %s\n", source);
+			if (nouveau_bios_score(bios, 1)) {
+				kfree(source);
+				return 0;
+			}
+
+			kfree(bios->data);
+			bios->data = NULL;
+		}
+
+		nv_error(bios, "source \'%s\' invalid\n", source);
+		kfree(source);
+	}
+
+	mthd = shadow_methods;
+	do {
+		nv_info(bios, "checking %s for image...\n", mthd->desc);
+		mthd->shadow(bios);
+		mthd->score = nouveau_bios_score(bios, mthd->rw);
+		mthd->size = bios->size;
+		mthd->data = bios->data;
+		bios->data = NULL;
+	} while (mthd->score != 3 && (++mthd)->shadow);
+
+	mthd = shadow_methods;
+	best = mthd;
+	do {
+		if (mthd->score > best->score) {
+			kfree(best->data);
+			best = mthd;
+		}
+	} while ((++mthd)->shadow);
+
+	if (best->score) {
+		nv_info(bios, "using image from %s\n", best->desc);
+		bios->size = best->size;
+		bios->data = best->data;
+		return 0;
+	}
+
+	nv_error(bios, "unable to locate usable image\n");
+	return -EINVAL;
+}
+
+static u8
+nouveau_bios_rd08(struct nouveau_object *object, u64 addr)
+{
+	struct nouveau_bios *bios = (void *)object;
+	return bios->data[addr];
+}
+
+static u16
+nouveau_bios_rd16(struct nouveau_object *object, u64 addr)
+{
+	struct nouveau_bios *bios = (void *)object;
+	return get_unaligned_le16(&bios->data[addr]);
+}
+
+static u32
+nouveau_bios_rd32(struct nouveau_object *object, u64 addr)
+{
+	struct nouveau_bios *bios = (void *)object;
+	return get_unaligned_le32(&bios->data[addr]);
+}
+
+static void
+nouveau_bios_wr08(struct nouveau_object *object, u64 addr, u8 data)
+{
+	struct nouveau_bios *bios = (void *)object;
+	bios->data[addr] = data;
+}
+
+static void
+nouveau_bios_wr16(struct nouveau_object *object, u64 addr, u16 data)
+{
+	struct nouveau_bios *bios = (void *)object;
+	put_unaligned_le16(data, &bios->data[addr]);
+}
+
+static void
+nouveau_bios_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	struct nouveau_bios *bios = (void *)object;
+	put_unaligned_le32(data, &bios->data[addr]);
+}
+
+static int
+nouveau_bios_ctor(struct nouveau_object *parent,
+		  struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nouveau_bios *bios;
+	struct bit_entry bit_i;
+	int ret;
+
+	ret = nouveau_subdev_create(parent, engine, oclass, 0,
+				    "VBIOS", "bios", &bios);
+	*pobject = nv_object(bios);
+	if (ret)
+		return ret;
+
+	ret = nouveau_bios_shadow(bios);
+	if (ret)
+		return ret;
+
+	/* detect type of vbios we're dealing with */
+	bios->bmp_offset = nvbios_findstr(bios->data, bios->size,
+					  "\xff\x7f""NV\0", 5);
+	if (bios->bmp_offset) {
+		nv_info(bios, "BMP version %x.%x\n",
+			bmp_version(bios) >> 8,
+			bmp_version(bios) & 0xff);
+	}
+
+	bios->bit_offset = nvbios_findstr(bios->data, bios->size,
+					  "\xff\xb8""BIT", 5);
+	if (bios->bit_offset)
+		nv_info(bios, "BIT signature found\n");
+
+	/* determine the vbios version number */
+	if (!bit_entry(bios, 'i', &bit_i) && bit_i.length >= 4) {
+		bios->version.major = nv_ro08(bios, bit_i.offset + 3);
+		bios->version.chip  = nv_ro08(bios, bit_i.offset + 2);
+		bios->version.minor = nv_ro08(bios, bit_i.offset + 1);
+		bios->version.micro = nv_ro08(bios, bit_i.offset + 0);
+		bios->version.patch = nv_ro08(bios, bit_i.offset + 4);
+	} else
+	if (bmp_version(bios)) {
+		bios->version.major = nv_ro08(bios, bios->bmp_offset + 13);
+		bios->version.chip  = nv_ro08(bios, bios->bmp_offset + 12);
+		bios->version.minor = nv_ro08(bios, bios->bmp_offset + 11);
+		bios->version.micro = nv_ro08(bios, bios->bmp_offset + 10);
+	}
+
+	nv_info(bios, "version %02x.%02x.%02x.%02x.%02x\n",
+		bios->version.major, bios->version.chip,
+		bios->version.minor, bios->version.micro, bios->version.patch);
+
+	return 0;
+}
+
+static void
+nouveau_bios_dtor(struct nouveau_object *object)
+{
+	struct nouveau_bios *bios = (void *)object;
+	kfree(bios->data);
+	nouveau_subdev_destroy(&bios->base);
+}
+
+static int
+nouveau_bios_init(struct nouveau_object *object)
+{
+	struct nouveau_bios *bios = (void *)object;
+	return nouveau_subdev_init(&bios->base);
+}
+
+static int
+nouveau_bios_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nouveau_bios *bios = (void *)object;
+	return nouveau_subdev_fini(&bios->base, suspend);
+}
+
+struct nouveau_oclass
+nouveau_bios_oclass = {
+	.handle = NV_SUBDEV(VBIOS, 0x00),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nouveau_bios_ctor,
+		.dtor = nouveau_bios_dtor,
+		.init = nouveau_bios_init,
+		.fini = nouveau_bios_fini,
+		.rd08 = nouveau_bios_rd08,
+		.rd16 = nouveau_bios_rd16,
+		.rd32 = nouveau_bios_rd32,
+		.wr08 = nouveau_bios_wr08,
+		.wr16 = nouveau_bios_wr16,
+		.wr32 = nouveau_bios_wr32,
+	},
+};
diff --git a/nvkm/subdev/bios/bit.c b/nvkm/subdev/bios/bit.c
new file mode 100644
index 000000000..1d03a3f2b
--- /dev/null
+++ b/nvkm/subdev/bios/bit.c
@@ -0,0 +1,52 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "core/object.h"
+
+#include "subdev/bios.h"
+#include "subdev/bios/bit.h"
+
+int
+bit_entry(struct nouveau_bios *bios, u8 id, struct bit_entry *bit)
+{
+	if (likely(bios->bit_offset)) {
+		u8  entries = nv_ro08(bios, bios->bit_offset + 10);
+		u32 entry   = bios->bit_offset + 12;
+		while (entries--) {
+			if (nv_ro08(bios, entry + 0) == id) {
+				bit->id      = nv_ro08(bios, entry + 0);
+				bit->version = nv_ro08(bios, entry + 1);
+				bit->length  = nv_ro16(bios, entry + 2);
+				bit->offset  = nv_ro16(bios, entry + 4);
+				return 0;
+			}
+
+			entry += nv_ro08(bios, bios->bit_offset + 9);
+		}
+
+		return -ENOENT;
+	}
+
+	return -EINVAL;
+}
diff --git a/nvkm/subdev/bios/conn.c b/nvkm/subdev/bios/conn.c
new file mode 100644
index 000000000..5ac010efd
--- /dev/null
+++ b/nvkm/subdev/bios/conn.c
@@ -0,0 +1,56 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/device.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/conn.h>
+
+u16
+dcb_conntab(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	u16 dcb = dcb_table(bios, ver, hdr, cnt, len);
+	if (dcb && *ver >= 0x30 && *hdr >= 0x16) {
+		u16 data = nv_ro16(bios, dcb + 0x14);
+		if (data) {
+			*ver = nv_ro08(bios, data + 0);
+			*hdr = nv_ro08(bios, data + 1);
+			*cnt = nv_ro08(bios, data + 2);
+			*len = nv_ro08(bios, data + 3);
+			return data;
+		}
+	}
+	return 0x0000;
+}
+
+u16
+dcb_conn(struct nouveau_bios *bios, u8 idx, u8 *ver, u8 *len)
+{
+	u8  hdr, cnt;
+	u16 data = dcb_conntab(bios, ver, &hdr, &cnt, len);
+	if (data && idx < cnt)
+		return data + hdr + (idx * *len);
+	return 0x0000;
+}
diff --git a/nvkm/subdev/bios/dcb.c b/nvkm/subdev/bios/dcb.c
new file mode 100644
index 000000000..0fd87df99
--- /dev/null
+++ b/nvkm/subdev/bios/dcb.c
@@ -0,0 +1,198 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "core/device.h"
+
+#include "subdev/bios.h"
+#include "subdev/bios/dcb.h"
+
+u16
+dcb_table(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	struct nouveau_device *device = nv_device(bios);
+	u16 dcb = 0x0000;
+
+	if (device->card_type > NV_04)
+		dcb = nv_ro16(bios, 0x36);
+	if (!dcb) {
+		nv_warn(bios, "DCB table not found\n");
+		return dcb;
+	}
+
+	*ver = nv_ro08(bios, dcb);
+
+	if (*ver >= 0x41) {
+		nv_warn(bios, "DCB version 0x%02x unknown\n", *ver);
+		return 0x0000;
+	} else
+	if (*ver >= 0x30) {
+		if (nv_ro32(bios, dcb + 6) == 0x4edcbdcb) {
+			*hdr = nv_ro08(bios, dcb + 1);
+			*cnt = nv_ro08(bios, dcb + 2);
+			*len = nv_ro08(bios, dcb + 3);
+			return dcb;
+		}
+	} else
+	if (*ver >= 0x20) {
+		if (nv_ro32(bios, dcb + 4) == 0x4edcbdcb) {
+			u16 i2c = nv_ro16(bios, dcb + 2);
+			*hdr = 8;
+			*cnt = (i2c - dcb) / 8;
+			*len = 8;
+			return dcb;
+		}
+	} else
+	if (*ver >= 0x15) {
+		if (!nv_memcmp(bios, dcb - 7, "DEV_REC", 7)) {
+			u16 i2c = nv_ro16(bios, dcb + 2);
+			*hdr = 4;
+			*cnt = (i2c - dcb) / 10;
+			*len = 10;
+			return dcb;
+		}
+	} else {
+		/*
+		 * v1.4 (some NV15/16, NV11+) seems the same as v1.5, but
+		 * always has the same single (crt) entry, even when tv-out
+		 * present, so the conclusion is this version cannot really
+		 * be used.
+		 *
+		 * v1.2 tables (some NV6/10, and NV15+) normally have the
+		 * same 5 entries, which are not specific to the card and so
+		 * no use.
+		 *
+		 * v1.2 does have an I2C table that read_dcb_i2c_table can
+		 * handle, but cards exist (nv11 in #14821) with a bad i2c
+		 * table pointer, so use the indices parsed in
+		 * parse_bmp_structure.
+		 *
+		 * v1.1 (NV5+, maybe some NV4) is entirely unhelpful
+		 */
+		nv_warn(bios, "DCB contains no useful data\n");
+		return 0x0000;
+	}
+
+	nv_warn(bios, "DCB header validation failed\n");
+	return 0x0000;
+}
+
+u16
+dcb_outp(struct nouveau_bios *bios, u8 idx, u8 *ver, u8 *len)
+{
+	u8  hdr, cnt;
+	u16 dcb = dcb_table(bios, ver, &hdr, &cnt, len);
+	if (dcb && idx < cnt)
+		return dcb + hdr + (idx * *len);
+	return 0x0000;
+}
+
+u16
+dcb_outp_parse(struct nouveau_bios *bios, u8 idx, u8 *ver, u8 *len,
+	       struct dcb_output *outp)
+{
+	u16 dcb = dcb_outp(bios, idx, ver, len);
+	if (dcb) {
+		if (*ver >= 0x20) {
+			u32 conn = nv_ro32(bios, dcb + 0x00);
+			outp->or        = (conn & 0x0f000000) >> 24;
+			outp->location  = (conn & 0x00300000) >> 20;
+			outp->bus       = (conn & 0x000f0000) >> 16;
+			outp->connector = (conn & 0x0000f000) >> 12;
+			outp->heads     = (conn & 0x00000f00) >> 8;
+			outp->i2c_index = (conn & 0x000000f0) >> 4;
+			outp->type      = (conn & 0x0000000f);
+			outp->link      = 0;
+		} else {
+			dcb = 0x0000;
+		}
+
+		if (*ver >= 0x40) {
+			u32 conf = nv_ro32(bios, dcb + 0x04);
+			switch (outp->type) {
+			case DCB_OUTPUT_TMDS:
+			case DCB_OUTPUT_LVDS:
+			case DCB_OUTPUT_DP:
+				outp->link = (conf & 0x00000030) >> 4;
+				outp->sorconf.link = outp->link; /*XXX*/
+				break;
+			default:
+				break;
+			}
+		}
+	}
+	return dcb;
+}
+
+static inline u16
+dcb_outp_hasht(struct dcb_output *outp)
+{
+	return outp->type;
+}
+
+static inline u16
+dcb_outp_hashm(struct dcb_output *outp)
+{
+	return (outp->heads << 8) | (outp->link << 6) | outp->or;
+}
+
+u16
+dcb_outp_match(struct nouveau_bios *bios, u16 type, u16 mask,
+	       u8 *ver, u8 *len, struct dcb_output *outp)
+{
+	u16 dcb, idx = 0;
+	while ((dcb = dcb_outp_parse(bios, idx++, ver, len, outp))) {
+		if (dcb_outp_hasht(outp) == type) {
+			if ((dcb_outp_hashm(outp) & mask) == mask)
+				break;
+		}
+	}
+	return dcb;
+}
+
+int
+dcb_outp_foreach(struct nouveau_bios *bios, void *data,
+		 int (*exec)(struct nouveau_bios *, void *, int, u16))
+{
+	int ret, idx = -1;
+	u8  ver, len;
+	u16 outp;
+
+	while ((outp = dcb_outp(bios, ++idx, &ver, &len))) {
+		if (nv_ro32(bios, outp) == 0x00000000)
+			break; /* seen on an NV11 with DCB v1.5 */
+		if (nv_ro32(bios, outp) == 0xffffffff)
+			break; /* seen on an NV17 with DCB v2.0 */
+
+		if (nv_ro08(bios, outp) == DCB_OUTPUT_UNUSED)
+			continue;
+		if (nv_ro08(bios, outp) == DCB_OUTPUT_EOL)
+			break;
+
+		ret = exec(bios, data, idx, outp);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/bios/disp.c b/nvkm/subdev/bios/disp.c
new file mode 100644
index 000000000..7f16e52d9
--- /dev/null
+++ b/nvkm/subdev/bios/disp.c
@@ -0,0 +1,178 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/bios.h>
+#include <subdev/bios/bit.h>
+#include <subdev/bios/disp.h>
+
+u16
+nvbios_disp_table(struct nouveau_bios *bios,
+		  u8 *ver, u8 *hdr, u8 *cnt, u8 *len, u8 *sub)
+{
+	struct bit_entry U;
+
+	if (!bit_entry(bios, 'U', &U)) {
+		if (U.version == 1) {
+			u16 data = nv_ro16(bios, U.offset);
+			if (data) {
+				*ver = nv_ro08(bios, data + 0x00);
+				switch (*ver) {
+				case 0x20:
+				case 0x21:
+					*hdr = nv_ro08(bios, data + 0x01);
+					*len = nv_ro08(bios, data + 0x02);
+					*cnt = nv_ro08(bios, data + 0x03);
+					*sub = nv_ro08(bios, data + 0x04);
+					return data;
+				default:
+					break;
+				}
+			}
+		}
+	}
+
+	return 0x0000;
+}
+
+u16
+nvbios_disp_entry(struct nouveau_bios *bios, u8 idx,
+		  u8 *ver, u8 *len, u8 *sub)
+{
+	u8  hdr, cnt;
+	u16 data = nvbios_disp_table(bios, ver, &hdr, &cnt, len, sub);
+	if (data && idx < cnt)
+		return data + hdr + (idx * *len);
+	*ver = 0x00;
+	return 0x0000;
+}
+
+u16
+nvbios_disp_parse(struct nouveau_bios *bios, u8 idx,
+		  u8 *ver, u8 *len, u8 *sub,
+		  struct nvbios_disp *info)
+{
+	u16 data = nvbios_disp_entry(bios, idx, ver, len, sub);
+	if (data && *len >= 2) {
+		info->data = nv_ro16(bios, data + 0);
+		return data;
+	}
+	return 0x0000;
+}
+
+u16
+nvbios_outp_entry(struct nouveau_bios *bios, u8 idx,
+		  u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	struct nvbios_disp info;
+	u16 data = nvbios_disp_parse(bios, idx, ver, len, hdr, &info);
+	if (data) {
+		*cnt = nv_ro08(bios, info.data + 0x05);
+		*len = 0x06;
+		data = info.data;
+	}
+	return data;
+}
+
+u16
+nvbios_outp_parse(struct nouveau_bios *bios, u8 idx,
+		  u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		  struct nvbios_outp *info)
+{
+	u16 data = nvbios_outp_entry(bios, idx, ver, hdr, cnt, len);
+	if (data && *hdr >= 0x0a) {
+		info->type      = nv_ro16(bios, data + 0x00);
+		info->mask      = nv_ro32(bios, data + 0x02);
+		if (*ver <= 0x20) /* match any link */
+			info->mask |= 0x00c0;
+		info->script[0] = nv_ro16(bios, data + 0x06);
+		info->script[1] = nv_ro16(bios, data + 0x08);
+		info->script[2] = 0x0000;
+		if (*hdr >= 0x0c)
+			info->script[2] = nv_ro16(bios, data + 0x0a);
+		return data;
+	}
+	return 0x0000;
+}
+
+u16
+nvbios_outp_match(struct nouveau_bios *bios, u16 type, u16 mask,
+		  u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		  struct nvbios_outp *info)
+{
+	u16 data, idx = 0;
+	while ((data = nvbios_outp_parse(bios, idx++, ver, hdr, cnt, len, info)) || *ver) {
+		if (data && info->type == type) {
+			if ((info->mask & mask) == mask)
+				break;
+		}
+	}
+	return data;
+}
+
+u16
+nvbios_ocfg_entry(struct nouveau_bios *bios, u16 outp, u8 idx,
+		  u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	if (idx < *cnt)
+		return outp + *hdr + (idx * *len);
+	return 0x0000;
+}
+
+u16
+nvbios_ocfg_parse(struct nouveau_bios *bios, u16 outp, u8 idx,
+		  u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		  struct nvbios_ocfg *info)
+{
+	u16 data = nvbios_ocfg_entry(bios, outp, idx, ver, hdr, cnt, len);
+	if (data) {
+		info->match     = nv_ro16(bios, data + 0x00);
+		info->clkcmp[0] = nv_ro16(bios, data + 0x02);
+		info->clkcmp[1] = nv_ro16(bios, data + 0x04);
+	}
+	return data;
+}
+
+u16
+nvbios_ocfg_match(struct nouveau_bios *bios, u16 outp, u16 type,
+		  u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		  struct nvbios_ocfg *info)
+{
+	u16 data, idx = 0;
+	while ((data = nvbios_ocfg_parse(bios, outp, idx++, ver, hdr, cnt, len, info))) {
+		if (info->match == type)
+			break;
+	}
+	return data;
+}
+
+u16
+nvbios_oclk_match(struct nouveau_bios *bios, u16 cmp, u32 khz)
+{
+	while (cmp) {
+		if (khz / 10 >= nv_ro16(bios, cmp + 0x00))
+			return  nv_ro16(bios, cmp + 0x02);
+		cmp += 0x04;
+	}
+	return 0x0000;
+}
diff --git a/nvkm/subdev/bios/dp.c b/nvkm/subdev/bios/dp.c
new file mode 100644
index 000000000..663853bcc
--- /dev/null
+++ b/nvkm/subdev/bios/dp.c
@@ -0,0 +1,204 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+
+#include "subdev/bios.h"
+#include "subdev/bios/bit.h"
+#include "subdev/bios/dp.h"
+
+static u16
+nvbios_dp_table(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	struct bit_entry d;
+
+	if (!bit_entry(bios, 'd', &d)) {
+		if (d.version == 1 && d.length >= 2) {
+			u16 data = nv_ro16(bios, d.offset);
+			if (data) {
+				*ver = nv_ro08(bios, data + 0x00);
+				switch (*ver) {
+				case 0x21:
+				case 0x30:
+				case 0x40:
+					*hdr = nv_ro08(bios, data + 0x01);
+					*len = nv_ro08(bios, data + 0x02);
+					*cnt = nv_ro08(bios, data + 0x03);
+					return data;
+				default:
+					break;
+				}
+			}
+		}
+	}
+
+	return 0x0000;
+}
+
+static u16
+nvbios_dpout_entry(struct nouveau_bios *bios, u8 idx,
+		   u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	u16 data = nvbios_dp_table(bios, ver, hdr, cnt, len);
+	if (data && idx < *cnt) {
+		u16 outp = nv_ro16(bios, data + *hdr + idx * *len);
+		switch (*ver * !!outp) {
+		case 0x21:
+		case 0x30:
+			*hdr = nv_ro08(bios, data + 0x04);
+			*len = nv_ro08(bios, data + 0x05);
+			*cnt = nv_ro08(bios, outp + 0x04);
+			break;
+		case 0x40:
+			*hdr = nv_ro08(bios, data + 0x04);
+			*cnt = 0;
+			*len = 0;
+			break;
+		default:
+			break;
+		}
+		return outp;
+	}
+	*ver = 0x00;
+	return 0x0000;
+}
+
+u16
+nvbios_dpout_parse(struct nouveau_bios *bios, u8 idx,
+		   u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		   struct nvbios_dpout *info)
+{
+	u16 data = nvbios_dpout_entry(bios, idx, ver, hdr, cnt, len);
+	if (data && *ver) {
+		info->type = nv_ro16(bios, data + 0x00);
+		info->mask = nv_ro16(bios, data + 0x02);
+		switch (*ver) {
+		case 0x21:
+		case 0x30:
+			info->flags     = nv_ro08(bios, data + 0x05);
+			info->script[0] = nv_ro16(bios, data + 0x06);
+			info->script[1] = nv_ro16(bios, data + 0x08);
+			info->lnkcmp    = nv_ro16(bios, data + 0x0a);
+			info->script[2] = nv_ro16(bios, data + 0x0c);
+			info->script[3] = nv_ro16(bios, data + 0x0e);
+			info->script[4] = nv_ro16(bios, data + 0x10);
+			break;
+		case 0x40:
+			info->flags     = nv_ro08(bios, data + 0x04);
+			info->script[0] = nv_ro16(bios, data + 0x05);
+			info->script[1] = nv_ro16(bios, data + 0x07);
+			info->lnkcmp    = nv_ro16(bios, data + 0x09);
+			info->script[2] = nv_ro16(bios, data + 0x0b);
+			info->script[3] = nv_ro16(bios, data + 0x0d);
+			info->script[4] = nv_ro16(bios, data + 0x0f);
+			break;
+		default:
+			data = 0x0000;
+			break;
+		}
+	}
+	return data;
+}
+
+u16
+nvbios_dpout_match(struct nouveau_bios *bios, u16 type, u16 mask,
+		   u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		   struct nvbios_dpout *info)
+{
+	u16 data, idx = 0;
+	while ((data = nvbios_dpout_parse(bios, idx++, ver, hdr, cnt, len, info)) || *ver) {
+		if (data && info->type == type) {
+			if ((info->mask & mask) == mask)
+				break;
+		}
+	}
+	return data;
+}
+
+static u16
+nvbios_dpcfg_entry(struct nouveau_bios *bios, u16 outp, u8 idx,
+		   u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	if (*ver >= 0x40) {
+		outp = nvbios_dp_table(bios, ver, hdr, cnt, len);
+		*hdr = *hdr + (*len * * cnt);
+		*len = nv_ro08(bios, outp + 0x06);
+		*cnt = nv_ro08(bios, outp + 0x07);
+	}
+
+	if (idx < *cnt)
+		return outp + *hdr + (idx * *len);
+
+	return 0x0000;
+}
+
+u16
+nvbios_dpcfg_parse(struct nouveau_bios *bios, u16 outp, u8 idx,
+		   u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		   struct nvbios_dpcfg *info)
+{
+	u16 data = nvbios_dpcfg_entry(bios, outp, idx, ver, hdr, cnt, len);
+	if (data) {
+		switch (*ver) {
+		case 0x21:
+			info->drv = nv_ro08(bios, data + 0x02);
+			info->pre = nv_ro08(bios, data + 0x03);
+			info->unk = nv_ro08(bios, data + 0x04);
+			break;
+		case 0x30:
+		case 0x40:
+			info->drv = nv_ro08(bios, data + 0x01);
+			info->pre = nv_ro08(bios, data + 0x02);
+			info->unk = nv_ro08(bios, data + 0x03);
+			break;
+		default:
+			data = 0x0000;
+			break;
+		}
+	}
+	return data;
+}
+
+u16
+nvbios_dpcfg_match(struct nouveau_bios *bios, u16 outp, u8 un, u8 vs, u8 pe,
+		   u8 *ver, u8 *hdr, u8 *cnt, u8 *len,
+		   struct nvbios_dpcfg *info)
+{
+	u8 idx = 0xff;
+	u16 data;
+
+	if (*ver >= 0x30) {
+		const u8 vsoff[] = { 0, 4, 7, 9 };
+		idx = (un * 10) + vsoff[vs] + pe;
+	} else {
+		while ((data = nvbios_dpcfg_entry(bios, outp, idx,
+						  ver, hdr, cnt, len))) {
+			if (nv_ro08(bios, data + 0x00) == vs &&
+			    nv_ro08(bios, data + 0x01) == pe)
+				break;
+			idx++;
+		}
+	}
+
+	return nvbios_dpcfg_parse(bios, outp, pe, ver, hdr, cnt, len, info);
+}
diff --git a/nvkm/subdev/bios/extdev.c b/nvkm/subdev/bios/extdev.c
new file mode 100644
index 000000000..5afb568b2
--- /dev/null
+++ b/nvkm/subdev/bios/extdev.c
@@ -0,0 +1,100 @@
+/*
+ * Copyright 2012 Nouveau Community
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Martin Peres
+ */
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/extdev.h>
+
+static u16
+extdev_table(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *len, u8 *cnt)
+{
+	u8  dcb_ver, dcb_hdr, dcb_cnt, dcb_len;
+	u16 dcb, extdev = 0;
+
+	dcb = dcb_table(bios, &dcb_ver, &dcb_hdr, &dcb_cnt, &dcb_len);
+	if (!dcb || (dcb_ver != 0x30 && dcb_ver != 0x40))
+		return 0x0000;
+
+	extdev = nv_ro16(bios, dcb + 18);
+	if (!extdev)
+		return 0x0000;
+
+	*ver = nv_ro08(bios, extdev + 0);
+	*hdr = nv_ro08(bios, extdev + 1);
+	*cnt = nv_ro08(bios, extdev + 2);
+	*len = nv_ro08(bios, extdev + 3);
+
+	return extdev + *hdr;
+}
+
+u16
+nvbios_extdev_entry(struct nouveau_bios *bios, int idx, u8 *ver, u8 *len)
+{
+	u8 hdr, cnt;
+	u16 extdev = extdev_table(bios, ver, &hdr, len, &cnt);
+	if (extdev && idx < cnt)
+		return extdev + idx * *len;
+	return 0x0000;
+}
+
+static void
+extdev_parse_entry(struct nouveau_bios *bios, u16 offset,
+			  struct nvbios_extdev_func *entry)
+{
+	entry->type = nv_ro08(bios, offset + 0);
+	entry->addr = nv_ro08(bios, offset + 1);
+	entry->bus = (nv_ro08(bios, offset + 2) >> 4) & 1;
+}
+
+int
+nvbios_extdev_parse(struct nouveau_bios *bios, int idx,
+		    struct nvbios_extdev_func *func)
+{
+	u8 ver, len;
+	u16 entry;
+
+	if (!(entry = nvbios_extdev_entry(bios, idx, &ver, &len)))
+		return -EINVAL;
+
+	extdev_parse_entry(bios, entry, func);
+
+	return 0;
+}
+
+int
+nvbios_extdev_find(struct nouveau_bios *bios, enum nvbios_extdev_type type,
+		   struct nvbios_extdev_func *func)
+{
+	u8 ver, len, i;
+	u16 entry;
+
+	i = 0;
+	while (!(entry = nvbios_extdev_entry(bios, i++, &ver, &len))) {
+		extdev_parse_entry(bios, entry, func);
+		if (func->type == type)
+			return 0;
+	}
+
+	return -EINVAL;
+}
diff --git a/nvkm/subdev/bios/gpio.c b/nvkm/subdev/bios/gpio.c
new file mode 100644
index 000000000..c84e93fa6
--- /dev/null
+++ b/nvkm/subdev/bios/gpio.c
@@ -0,0 +1,143 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/gpio.h>
+
+u16
+dcb_gpio_table(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	u16 data = 0x0000;
+	u16 dcb = dcb_table(bios, ver, hdr, cnt, len);
+	if (dcb) {
+		if (*ver >= 0x30 && *hdr >= 0x0c)
+			data = nv_ro16(bios, dcb + 0x0a);
+		else
+		if (*ver >= 0x22 && nv_ro08(bios, dcb - 1) >= 0x13)
+			data = nv_ro16(bios, dcb - 0x0f);
+
+		if (data) {
+			*ver = nv_ro08(bios, data + 0x00);
+			if (*ver < 0x30) {
+				*hdr = 3;
+				*cnt = nv_ro08(bios, data + 0x02);
+				*len = nv_ro08(bios, data + 0x01);
+			} else
+			if (*ver <= 0x41) {
+				*hdr = nv_ro08(bios, data + 0x01);
+				*cnt = nv_ro08(bios, data + 0x02);
+				*len = nv_ro08(bios, data + 0x03);
+			} else {
+				data = 0x0000;
+			}
+		}
+	}
+	return data;
+}
+
+u16
+dcb_gpio_entry(struct nouveau_bios *bios, int idx, int ent, u8 *ver, u8 *len)
+{
+	u8  hdr, cnt;
+	u16 gpio = !idx ? dcb_gpio_table(bios, ver, &hdr, &cnt, len) : 0x0000;
+	if (gpio && ent < cnt)
+		return gpio + hdr + (ent * *len);
+	return 0x0000;
+}
+
+u16
+dcb_gpio_parse(struct nouveau_bios *bios, int idx, int ent, u8 *ver, u8 *len,
+	       struct dcb_gpio_func *gpio)
+{
+	u16 data = dcb_gpio_entry(bios, idx, ent, ver, len);
+	if (data) {
+		if (*ver < 0x40) {
+			u16 info = nv_ro16(bios, data);
+			*gpio = (struct dcb_gpio_func) {
+				.line = (info & 0x001f) >> 0,
+				.func = (info & 0x07e0) >> 5,
+				.log[0] = (info & 0x1800) >> 11,
+				.log[1] = (info & 0x6000) >> 13,
+				.param = !!(info & 0x8000),
+			};
+		} else
+		if (*ver < 0x41) {
+			u32 info = nv_ro32(bios, data);
+			*gpio = (struct dcb_gpio_func) {
+				.line = (info & 0x0000001f) >> 0,
+				.func = (info & 0x0000ff00) >> 8,
+				.log[0] = (info & 0x18000000) >> 27,
+				.log[1] = (info & 0x60000000) >> 29,
+				.param = !!(info & 0x80000000),
+			};
+		} else {
+			u32 info = nv_ro32(bios, data + 0);
+			u8 info1 = nv_ro32(bios, data + 4);
+			*gpio = (struct dcb_gpio_func) {
+				.line = (info & 0x0000003f) >> 0,
+				.func = (info & 0x0000ff00) >> 8,
+				.log[0] = (info1 & 0x30) >> 4,
+				.log[1] = (info1 & 0xc0) >> 6,
+				.param = !!(info & 0x80000000),
+			};
+		}
+	}
+
+	return data;
+}
+
+u16
+dcb_gpio_match(struct nouveau_bios *bios, int idx, u8 func, u8 line,
+	       u8 *ver, u8 *len, struct dcb_gpio_func *gpio)
+{
+	u8  hdr, cnt, i = 0;
+	u16 data;
+
+	while ((data = dcb_gpio_parse(bios, idx, i++, ver, len, gpio))) {
+		if ((line == 0xff || line == gpio->line) &&
+		    (func == 0xff || func == gpio->func))
+			return data;
+	}
+
+	/* DCB 2.2, fixed TVDAC GPIO data */
+	if ((data = dcb_table(bios, ver, &hdr, &cnt, len))) {
+		if (*ver >= 0x22 && *ver < 0x30 && func == DCB_GPIO_TVDAC0) {
+			u8 conf = nv_ro08(bios, data - 5);
+			u8 addr = nv_ro08(bios, data - 4);
+			if (conf & 0x01) {
+				*gpio = (struct dcb_gpio_func) {
+					.func = DCB_GPIO_TVDAC0,
+					.line = addr >> 4,
+					.log[0] = !!(conf & 0x02),
+					.log[1] =  !(conf & 0x02),
+				};
+				*ver = 0x00;
+				return data;
+			}
+		}
+	}
+
+	return 0x0000;
+}
diff --git a/nvkm/subdev/bios/i2c.c b/nvkm/subdev/bios/i2c.c
new file mode 100644
index 000000000..ad577db83
--- /dev/null
+++ b/nvkm/subdev/bios/i2c.c
@@ -0,0 +1,129 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+
+#include "subdev/bios.h"
+#include "subdev/bios/dcb.h"
+#include "subdev/bios/i2c.h"
+
+u16
+dcb_i2c_table(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	u16 i2c = 0x0000;
+	u16 dcb = dcb_table(bios, ver, hdr, cnt, len);
+	if (dcb) {
+		if (*ver >= 0x15)
+			i2c = nv_ro16(bios, dcb + 2);
+		if (*ver >= 0x30)
+			i2c = nv_ro16(bios, dcb + 4);
+	}
+
+	if (i2c && *ver >= 0x30) {
+		*ver = nv_ro08(bios, i2c + 0);
+		*hdr = nv_ro08(bios, i2c + 1);
+		*cnt = nv_ro08(bios, i2c + 2);
+		*len = nv_ro08(bios, i2c + 3);
+	} else {
+		*ver = *ver; /* use DCB version */
+		*hdr = 0;
+		*cnt = 16;
+		*len = 4;
+	}
+
+	return i2c;
+}
+
+u16
+dcb_i2c_entry(struct nouveau_bios *bios, u8 idx, u8 *ver, u8 *len)
+{
+	u8  hdr, cnt;
+	u16 i2c = dcb_i2c_table(bios, ver, &hdr, &cnt, len);
+	if (i2c && idx < cnt)
+		return i2c + hdr + (idx * *len);
+	return 0x0000;
+}
+
+int
+dcb_i2c_parse(struct nouveau_bios *bios, u8 idx, struct dcb_i2c_entry *info)
+{
+	u8  ver, len;
+	u16 ent = dcb_i2c_entry(bios, idx, &ver, &len);
+	if (ent) {
+		info->data = nv_ro32(bios, ent + 0);
+		info->type = nv_ro08(bios, ent + 3);
+		if (ver < 0x30) {
+			info->type &= 0x07;
+			if (info->type == 0x07)
+				info->type = 0xff;
+		}
+
+		switch (info->type) {
+		case DCB_I2C_NV04_BIT:
+			info->drive = nv_ro08(bios, ent + 0);
+			info->sense = nv_ro08(bios, ent + 1);
+			return 0;
+		case DCB_I2C_NV4E_BIT:
+			info->drive = nv_ro08(bios, ent + 1);
+			return 0;
+		case DCB_I2C_NVIO_BIT:
+		case DCB_I2C_NVIO_AUX:
+			info->drive = nv_ro08(bios, ent + 0);
+			return 0;
+		case DCB_I2C_UNUSED:
+			return 0;
+		default:
+			nv_warn(bios, "unknown i2c type %d\n", info->type);
+			info->type = DCB_I2C_UNUSED;
+			return 0;
+		}
+	}
+
+	if (bios->bmp_offset && idx < 2) {
+		/* BMP (from v4.0 has i2c info in the structure, it's in a
+		 * fixed location on earlier VBIOS
+		 */
+		if (nv_ro08(bios, bios->bmp_offset + 5) < 4)
+			ent = 0x0048;
+		else
+			ent = 0x0036 + bios->bmp_offset;
+
+		if (idx == 0) {
+			info->drive = nv_ro08(bios, ent + 4);
+			if (!info->drive) info->drive = 0x3f;
+			info->sense = nv_ro08(bios, ent + 5);
+			if (!info->sense) info->sense = 0x3e;
+		} else
+		if (idx == 1) {
+			info->drive = nv_ro08(bios, ent + 6);
+			if (!info->drive) info->drive = 0x37;
+			info->sense = nv_ro08(bios, ent + 7);
+			if (!info->sense) info->sense = 0x36;
+		}
+
+		info->type = DCB_I2C_NV04_BIT;
+		return 0;
+	}
+
+	return -ENOENT;
+}
diff --git a/nvkm/subdev/bios/init.c b/nvkm/subdev/bios/init.c
new file mode 100644
index 000000000..2917d5526
--- /dev/null
+++ b/nvkm/subdev/bios/init.c
@@ -0,0 +1,2182 @@
+#include <core/engine.h>
+#include <core/device.h>
+
+#include <subdev/bios.h>
+#include <subdev/bios/bmp.h>
+#include <subdev/bios/bit.h>
+#include <subdev/bios/conn.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/dp.h>
+#include <subdev/bios/gpio.h>
+#include <subdev/bios/init.h>
+#include <subdev/devinit.h>
+#include <subdev/clock.h>
+#include <subdev/i2c.h>
+#include <subdev/vga.h>
+#include <subdev/gpio.h>
+
+#define bioslog(lvl, fmt, args...) do {                                        \
+	nv_printk(init->bios, lvl, "0x%04x[%c]: "fmt, init->offset,            \
+		  init_exec(init) ? '0' + (init->nested - 1) : ' ', ##args);   \
+} while(0)
+#define cont(fmt, args...) do {                                                \
+	if (nv_subdev(init->bios)->debug >= NV_DBG_TRACE)                      \
+		printk(fmt, ##args);                                           \
+} while(0)
+#define trace(fmt, args...) bioslog(TRACE, fmt, ##args)
+#define warn(fmt, args...) bioslog(WARN, fmt, ##args)
+#define error(fmt, args...) bioslog(ERROR, fmt, ##args)
+
+/******************************************************************************
+ * init parser control flow helpers
+ *****************************************************************************/
+
+static inline bool
+init_exec(struct nvbios_init *init)
+{
+	return (init->execute == 1) || ((init->execute & 5) == 5);
+}
+
+static inline void
+init_exec_set(struct nvbios_init *init, bool exec)
+{
+	if (exec) init->execute &= 0xfd;
+	else      init->execute |= 0x02;
+}
+
+static inline void
+init_exec_inv(struct nvbios_init *init)
+{
+	init->execute ^= 0x02;
+}
+
+static inline void
+init_exec_force(struct nvbios_init *init, bool exec)
+{
+	if (exec) init->execute |= 0x04;
+	else      init->execute &= 0xfb;
+}
+
+/******************************************************************************
+ * init parser wrappers for normal register/i2c/whatever accessors
+ *****************************************************************************/
+
+static inline int
+init_or(struct nvbios_init *init)
+{
+	if (init->outp)
+		return ffs(init->outp->or) - 1;
+	error("script needs OR!!\n");
+	return 0;
+}
+
+static inline int
+init_link(struct nvbios_init *init)
+{
+	if (init->outp)
+		return !(init->outp->sorconf.link & 1);
+	error("script needs OR link\n");
+	return 0;
+}
+
+static inline int
+init_crtc(struct nvbios_init *init)
+{
+	if (init->crtc >= 0)
+		return init->crtc;
+	error("script needs crtc\n");
+	return 0;
+}
+
+static u8
+init_conn(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+
+	if (init->outp) {
+		u8  ver, len;
+		u16 conn = dcb_conn(bios, init->outp->connector, &ver, &len);
+		if (conn)
+			return nv_ro08(bios, conn);
+	}
+
+	error("script needs connector type\n");
+	return 0x00;
+}
+
+static inline u32
+init_nvreg(struct nvbios_init *init, u32 reg)
+{
+	/* C51 (at least) sometimes has the lower bits set which the VBIOS
+	 * interprets to mean that access needs to go through certain IO
+	 * ports instead.  The NVIDIA binary driver has been seen to access
+	 * these through the NV register address, so lets assume we can
+	 * do the same
+	 */
+	reg &= ~0x00000003;
+
+	/* GF8+ display scripts need register addresses mangled a bit to
+	 * select a specific CRTC/OR
+	 */
+	if (nv_device(init->bios)->card_type >= NV_50) {
+		if (reg & 0x80000000) {
+			reg += init_crtc(init) * 0x800;
+			reg &= ~0x80000000;
+		}
+
+		if (reg & 0x40000000) {
+			reg += init_or(init) * 0x800;
+			reg &= ~0x40000000;
+			if (reg & 0x20000000) {
+				reg += init_link(init) * 0x80;
+				reg &= ~0x20000000;
+			}
+		}
+	}
+
+	if (reg & ~0x00fffffc)
+		warn("unknown bits in register 0x%08x\n", reg);
+	return reg;
+}
+
+static u32
+init_rd32(struct nvbios_init *init, u32 reg)
+{
+	reg = init_nvreg(init, reg);
+	if (init_exec(init))
+		return nv_rd32(init->subdev, reg);
+	return 0x00000000;
+}
+
+static void
+init_wr32(struct nvbios_init *init, u32 reg, u32 val)
+{
+	reg = init_nvreg(init, reg);
+	if (init_exec(init))
+		nv_wr32(init->subdev, reg, val);
+}
+
+static u32
+init_mask(struct nvbios_init *init, u32 reg, u32 mask, u32 val)
+{
+	reg = init_nvreg(init, reg);
+	if (init_exec(init)) {
+		u32 tmp = nv_rd32(init->subdev, reg);
+		nv_wr32(init->subdev, reg, (tmp & ~mask) | val);
+		return tmp;
+	}
+	return 0x00000000;
+}
+
+static u8
+init_rdport(struct nvbios_init *init, u16 port)
+{
+	if (init_exec(init))
+		return nv_rdport(init->subdev, init->crtc, port);
+	return 0x00;
+}
+
+static void
+init_wrport(struct nvbios_init *init, u16 port, u8 value)
+{
+	if (init_exec(init))
+		nv_wrport(init->subdev, init->crtc, port, value);
+}
+
+static u8
+init_rdvgai(struct nvbios_init *init, u16 port, u8 index)
+{
+	struct nouveau_subdev *subdev = init->subdev;
+	if (init_exec(init)) {
+		int head = init->crtc < 0 ? 0 : init->crtc;
+		return nv_rdvgai(subdev, head, port, index);
+	}
+	return 0x00;
+}
+
+static void
+init_wrvgai(struct nvbios_init *init, u16 port, u8 index, u8 value)
+{
+	/* force head 0 for updates to cr44, it only exists on first head */
+	if (nv_device(init->subdev)->card_type < NV_50) {
+		if (port == 0x03d4 && index == 0x44)
+			init->crtc = 0;
+	}
+
+	if (init_exec(init)) {
+		int head = init->crtc < 0 ? 0 : init->crtc;
+		nv_wrvgai(init->subdev, head, port, index, value);
+	}
+
+	/* select head 1 if cr44 write selected it */
+	if (nv_device(init->subdev)->card_type < NV_50) {
+		if (port == 0x03d4 && index == 0x44 && value == 3)
+			init->crtc = 1;
+	}
+}
+
+static struct nouveau_i2c_port *
+init_i2c(struct nvbios_init *init, int index)
+{
+	struct nouveau_i2c *i2c = nouveau_i2c(init->bios);
+
+	if (index == 0xff) {
+		index = NV_I2C_DEFAULT(0);
+		if (init->outp && init->outp->i2c_upper_default)
+			index = NV_I2C_DEFAULT(1);
+	} else
+	if (index < 0) {
+		if (!init->outp) {
+			error("script needs output for i2c\n");
+			return NULL;
+		}
+
+		index = init->outp->i2c_index;
+	}
+
+	return i2c->find(i2c, index);
+}
+
+static int
+init_rdi2cr(struct nvbios_init *init, u8 index, u8 addr, u8 reg)
+{
+	struct nouveau_i2c_port *port = init_i2c(init, index);
+	if (port && init_exec(init))
+		return nv_rdi2cr(port, addr, reg);
+	return -ENODEV;
+}
+
+static int
+init_wri2cr(struct nvbios_init *init, u8 index, u8 addr, u8 reg, u8 val)
+{
+	struct nouveau_i2c_port *port = init_i2c(init, index);
+	if (port && init_exec(init))
+		return nv_wri2cr(port, addr, reg, val);
+	return -ENODEV;
+}
+
+static int
+init_rdauxr(struct nvbios_init *init, u32 addr)
+{
+	struct nouveau_i2c_port *port = init_i2c(init, -1);
+	u8 data;
+
+	if (port && init_exec(init)) {
+		int ret = nv_rdaux(port, addr, &data, 1);
+		if (ret)
+			return ret;
+		return data;
+	}
+
+	return -ENODEV;
+}
+
+static int
+init_wrauxr(struct nvbios_init *init, u32 addr, u8 data)
+{
+	struct nouveau_i2c_port *port = init_i2c(init, -1);
+	if (port && init_exec(init))
+		return nv_wraux(port, addr, &data, 1);
+	return -ENODEV;
+}
+
+static void
+init_prog_pll(struct nvbios_init *init, u32 id, u32 freq)
+{
+	struct nouveau_clock *clk = nouveau_clock(init->bios);
+	if (clk && clk->pll_set && init_exec(init)) {
+		int ret = clk->pll_set(clk, id, freq);
+		if (ret)
+			warn("failed to prog pll 0x%08x to %dkHz\n", id, freq);
+	}
+}
+
+/******************************************************************************
+ * parsing of bios structures that are required to execute init tables
+ *****************************************************************************/
+
+static u16
+init_table(struct nouveau_bios *bios, u16 *len)
+{
+	struct bit_entry bit_I;
+
+	if (!bit_entry(bios, 'I', &bit_I)) {
+		*len = bit_I.length;
+		return bit_I.offset;
+	}
+
+	if (bmp_version(bios) >= 0x0510) {
+		*len = 14;
+		return bios->bmp_offset + 75;
+	}
+
+	return 0x0000;
+}
+
+static u16
+init_table_(struct nvbios_init *init, u16 offset, const char *name)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 len, data = init_table(bios, &len);
+	if (data) {
+		if (len >= offset + 2) {
+			data = nv_ro16(bios, data + offset);
+			if (data)
+				return data;
+
+			warn("%s pointer invalid\n", name);
+			return 0x0000;
+		}
+
+		warn("init data too short for %s pointer", name);
+		return 0x0000;
+	}
+
+	warn("init data not found\n");
+	return 0x0000;
+}
+
+#define init_script_table(b) init_table_((b), 0x00, "script table")
+#define init_macro_index_table(b) init_table_((b), 0x02, "macro index table")
+#define init_macro_table(b) init_table_((b), 0x04, "macro table")
+#define init_condition_table(b) init_table_((b), 0x06, "condition table")
+#define init_io_condition_table(b) init_table_((b), 0x08, "io condition table")
+#define init_io_flag_condition_table(b) init_table_((b), 0x0a, "io flag conditon table")
+#define init_function_table(b) init_table_((b), 0x0c, "function table")
+#define init_xlat_table(b) init_table_((b), 0x10, "xlat table");
+
+static u16
+init_script(struct nouveau_bios *bios, int index)
+{
+	struct nvbios_init init = { .bios = bios };
+	u16 data;
+
+	if (bmp_version(bios) && bmp_version(bios) < 0x0510) {
+		if (index > 1)
+			return 0x0000;
+
+		data = bios->bmp_offset + (bios->version.major < 2 ? 14 : 18);
+		return nv_ro16(bios, data + (index * 2));
+	}
+
+	data = init_script_table(&init);
+	if (data)
+		return nv_ro16(bios, data + (index * 2));
+
+	return 0x0000;
+}
+
+static u16
+init_unknown_script(struct nouveau_bios *bios)
+{
+	u16 len, data = init_table(bios, &len);
+	if (data && len >= 16)
+		return nv_ro16(bios, data + 14);
+	return 0x0000;
+}
+
+static u16
+init_ram_restrict_table(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	struct bit_entry bit_M;
+	u16 data = 0x0000;
+
+	if (!bit_entry(bios, 'M', &bit_M)) {
+		if (bit_M.version == 1 && bit_M.length >= 5)
+			data = nv_ro16(bios, bit_M.offset + 3);
+		if (bit_M.version == 2 && bit_M.length >= 3)
+			data = nv_ro16(bios, bit_M.offset + 1);
+	}
+
+	if (data == 0x0000)
+		warn("ram restrict table not found\n");
+	return data;
+}
+
+static u8
+init_ram_restrict_group_count(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	struct bit_entry bit_M;
+
+	if (!bit_entry(bios, 'M', &bit_M)) {
+		if (bit_M.version == 1 && bit_M.length >= 5)
+			return nv_ro08(bios, bit_M.offset + 2);
+		if (bit_M.version == 2 && bit_M.length >= 3)
+			return nv_ro08(bios, bit_M.offset + 0);
+	}
+
+	return 0x00;
+}
+
+static u8
+init_ram_restrict_strap(struct nvbios_init *init)
+{
+	/* This appears to be the behaviour of the VBIOS parser, and *is*
+	 * important to cache the NV_PEXTDEV_BOOT0 on later chipsets to
+	 * avoid fucking up the memory controller (somehow) by reading it
+	 * on every INIT_RAM_RESTRICT_ZM_GROUP opcode.
+	 *
+	 * Preserving the non-caching behaviour on earlier chipsets just
+	 * in case *not* re-reading the strap causes similar breakage.
+	 */
+	if (!init->ramcfg || init->bios->version.major < 0x70)
+		init->ramcfg = init_rd32(init, 0x101000);
+	return (init->ramcfg & 0x00000003c) >> 2;
+}
+
+static u8
+init_ram_restrict(struct nvbios_init *init)
+{
+	u8  strap = init_ram_restrict_strap(init);
+	u16 table = init_ram_restrict_table(init);
+	if (table)
+		return nv_ro08(init->bios, table + strap);
+	return 0x00;
+}
+
+static u8
+init_xlat_(struct nvbios_init *init, u8 index, u8 offset)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 table = init_xlat_table(init);
+	if (table) {
+		u16 data = nv_ro16(bios, table + (index * 2));
+		if (data)
+			return nv_ro08(bios, data + offset);
+		warn("xlat table pointer %d invalid\n", index);
+	}
+	return 0x00;
+}
+
+/******************************************************************************
+ * utility functions used by various init opcode handlers
+ *****************************************************************************/
+
+static bool
+init_condition_met(struct nvbios_init *init, u8 cond)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 table = init_condition_table(init);
+	if (table) {
+		u32 reg = nv_ro32(bios, table + (cond * 12) + 0);
+		u32 msk = nv_ro32(bios, table + (cond * 12) + 4);
+		u32 val = nv_ro32(bios, table + (cond * 12) + 8);
+		trace("\t[0x%02x] (R[0x%06x] & 0x%08x) == 0x%08x\n",
+		      cond, reg, msk, val);
+		return (init_rd32(init, reg) & msk) == val;
+	}
+	return false;
+}
+
+static bool
+init_io_condition_met(struct nvbios_init *init, u8 cond)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 table = init_io_condition_table(init);
+	if (table) {
+		u16 port = nv_ro16(bios, table + (cond * 5) + 0);
+		u8 index = nv_ro08(bios, table + (cond * 5) + 2);
+		u8  mask = nv_ro08(bios, table + (cond * 5) + 3);
+		u8 value = nv_ro08(bios, table + (cond * 5) + 4);
+		trace("\t[0x%02x] (0x%04x[0x%02x] & 0x%02x) == 0x%02x\n",
+		      cond, port, index, mask, value);
+		return (init_rdvgai(init, port, index) & mask) == value;
+	}
+	return false;
+}
+
+static bool
+init_io_flag_condition_met(struct nvbios_init *init, u8 cond)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 table = init_io_flag_condition_table(init);
+	if (table) {
+		u16 port = nv_ro16(bios, table + (cond * 9) + 0);
+		u8 index = nv_ro08(bios, table + (cond * 9) + 2);
+		u8  mask = nv_ro08(bios, table + (cond * 9) + 3);
+		u8 shift = nv_ro08(bios, table + (cond * 9) + 4);
+		u16 data = nv_ro16(bios, table + (cond * 9) + 5);
+		u8 dmask = nv_ro08(bios, table + (cond * 9) + 7);
+		u8 value = nv_ro08(bios, table + (cond * 9) + 8);
+		u8 ioval = (init_rdvgai(init, port, index) & mask) >> shift;
+		return (nv_ro08(bios, data + ioval) & dmask) == value;
+	}
+	return false;
+}
+
+static inline u32
+init_shift(u32 data, u8 shift)
+{
+	if (shift < 0x80)
+		return data >> shift;
+	return data << (0x100 - shift);
+}
+
+static u32
+init_tmds_reg(struct nvbios_init *init, u8 tmds)
+{
+	/* For mlv < 0x80, it is an index into a table of TMDS base addresses.
+	 * For mlv == 0x80 use the "or" value of the dcb_entry indexed by
+	 * CR58 for CR57 = 0 to index a table of offsets to the basic
+	 * 0x6808b0 address.
+	 * For mlv == 0x81 use the "or" value of the dcb_entry indexed by
+	 * CR58 for CR57 = 0 to index a table of offsets to the basic
+	 * 0x6808b0 address, and then flip the offset by 8.
+	 */
+
+	const int pramdac_offset[13] = {
+		0, 0, 0x8, 0, 0x2000, 0, 0, 0, 0x2008, 0, 0, 0, 0x2000 };
+	const u32 pramdac_table[4] = {
+		0x6808b0, 0x6808b8, 0x6828b0, 0x6828b8 };
+
+	if (tmds >= 0x80) {
+		if (init->outp) {
+			u32 dacoffset = pramdac_offset[init->outp->or];
+			if (tmds == 0x81)
+				dacoffset ^= 8;
+			return 0x6808b0 + dacoffset;
+		}
+
+		error("tmds opcodes need dcb\n");
+	} else {
+		if (tmds < ARRAY_SIZE(pramdac_table))
+			return pramdac_table[tmds];
+
+		error("tmds selector 0x%02x unknown\n", tmds);
+	}
+
+	return 0;
+}
+
+/******************************************************************************
+ * init opcode handlers
+ *****************************************************************************/
+
+/**
+ * init_reserved - stub for various unknown/unused single-byte opcodes
+ *
+ */
+static void
+init_reserved(struct nvbios_init *init)
+{
+	u8 opcode = nv_ro08(init->bios, init->offset);
+	trace("RESERVED\t0x%02x\n", opcode);
+	init->offset += 1;
+}
+
+/**
+ * INIT_DONE - opcode 0x71
+ *
+ */
+static void
+init_done(struct nvbios_init *init)
+{
+	trace("DONE\n");
+	init->offset = 0x0000;
+}
+
+/**
+ * INIT_IO_RESTRICT_PROG - opcode 0x32
+ *
+ */
+static void
+init_io_restrict_prog(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 port = nv_ro16(bios, init->offset + 1);
+	u8 index = nv_ro08(bios, init->offset + 3);
+	u8  mask = nv_ro08(bios, init->offset + 4);
+	u8 shift = nv_ro08(bios, init->offset + 5);
+	u8 count = nv_ro08(bios, init->offset + 6);
+	u32  reg = nv_ro32(bios, init->offset + 7);
+	u8 conf, i;
+
+	trace("IO_RESTRICT_PROG\tR[0x%06x] = "
+	      "((0x%04x[0x%02x] & 0x%02x) >> %d) [{\n",
+	      reg, port, index, mask, shift);
+	init->offset += 11;
+
+	conf = (init_rdvgai(init, port, index) & mask) >> shift;
+	for (i = 0; i < count; i++) {
+		u32 data = nv_ro32(bios, init->offset);
+
+		if (i == conf) {
+			trace("\t0x%08x *\n", data);
+			init_wr32(init, reg, data);
+		} else {
+			trace("\t0x%08x\n", data);
+		}
+
+		init->offset += 4;
+	}
+	trace("}]\n");
+}
+
+/**
+ * INIT_REPEAT - opcode 0x33
+ *
+ */
+static void
+init_repeat(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 count = nv_ro08(bios, init->offset + 1);
+	u16 repeat = init->repeat;
+
+	trace("REPEAT\t0x%02x\n", count);
+	init->offset += 2;
+
+	init->repeat = init->offset;
+	init->repend = init->offset;
+	while (count--) {
+		init->offset = init->repeat;
+		nvbios_exec(init);
+		if (count)
+			trace("REPEAT\t0x%02x\n", count);
+	}
+	init->offset = init->repend;
+	init->repeat = repeat;
+}
+
+/**
+ * INIT_IO_RESTRICT_PLL - opcode 0x34
+ *
+ */
+static void
+init_io_restrict_pll(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 port = nv_ro16(bios, init->offset + 1);
+	u8 index = nv_ro08(bios, init->offset + 3);
+	u8  mask = nv_ro08(bios, init->offset + 4);
+	u8 shift = nv_ro08(bios, init->offset + 5);
+	s8  iofc = nv_ro08(bios, init->offset + 6);
+	u8 count = nv_ro08(bios, init->offset + 7);
+	u32  reg = nv_ro32(bios, init->offset + 8);
+	u8 conf, i;
+
+	trace("IO_RESTRICT_PLL\tR[0x%06x] =PLL= "
+	      "((0x%04x[0x%02x] & 0x%02x) >> 0x%02x) IOFCOND 0x%02x [{\n",
+	      reg, port, index, mask, shift, iofc);
+	init->offset += 12;
+
+	conf = (init_rdvgai(init, port, index) & mask) >> shift;
+	for (i = 0; i < count; i++) {
+		u32 freq = nv_ro16(bios, init->offset) * 10;
+
+		if (i == conf) {
+			trace("\t%dkHz *\n", freq);
+			if (iofc > 0 && init_io_flag_condition_met(init, iofc))
+				freq *= 2;
+			init_prog_pll(init, reg, freq);
+		} else {
+			trace("\t%dkHz\n", freq);
+		}
+
+		init->offset += 2;
+	}
+	trace("}]\n");
+}
+
+/**
+ * INIT_END_REPEAT - opcode 0x36
+ *
+ */
+static void
+init_end_repeat(struct nvbios_init *init)
+{
+	trace("END_REPEAT\n");
+	init->offset += 1;
+
+	if (init->repeat) {
+		init->repend = init->offset;
+		init->offset = 0;
+	}
+}
+
+/**
+ * INIT_COPY - opcode 0x37
+ *
+ */
+static void
+init_copy(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32  reg = nv_ro32(bios, init->offset + 1);
+	u8 shift = nv_ro08(bios, init->offset + 5);
+	u8 smask = nv_ro08(bios, init->offset + 6);
+	u16 port = nv_ro16(bios, init->offset + 7);
+	u8 index = nv_ro08(bios, init->offset + 9);
+	u8  mask = nv_ro08(bios, init->offset + 10);
+	u8  data;
+
+	trace("COPY\t0x%04x[0x%02x] &= 0x%02x |= "
+	      "((R[0x%06x] %s 0x%02x) & 0x%02x)\n",
+	      port, index, mask, reg, (shift & 0x80) ? "<<" : ">>",
+	      (shift & 0x80) ? (0x100 - shift) : shift, smask);
+	init->offset += 11;
+
+	data  = init_rdvgai(init, port, index) & mask;
+	data |= init_shift(init_rd32(init, reg), shift) & smask;
+	init_wrvgai(init, port, index, data);
+}
+
+/**
+ * INIT_NOT - opcode 0x38
+ *
+ */
+static void
+init_not(struct nvbios_init *init)
+{
+	trace("NOT\n");
+	init->offset += 1;
+	init_exec_inv(init);
+}
+
+/**
+ * INIT_IO_FLAG_CONDITION - opcode 0x39
+ *
+ */
+static void
+init_io_flag_condition(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 cond = nv_ro08(bios, init->offset + 1);
+
+	trace("IO_FLAG_CONDITION\t0x%02x\n", cond);
+	init->offset += 2;
+
+	if (!init_io_flag_condition_met(init, cond))
+		init_exec_set(init, false);
+}
+
+/**
+ * INIT_DP_CONDITION - opcode 0x3a
+ *
+ */
+static void
+init_dp_condition(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	struct nvbios_dpout info;
+	u8  cond = nv_ro08(bios, init->offset + 1);
+	u8  unkn = nv_ro08(bios, init->offset + 2);
+	u8  ver, hdr, cnt, len;
+	u16 data;
+
+	trace("DP_CONDITION\t0x%02x 0x%02x\n", cond, unkn);
+	init->offset += 3;
+
+	switch (cond) {
+	case 0:
+		if (init_conn(init) != DCB_CONNECTOR_eDP)
+			init_exec_set(init, false);
+		break;
+	case 1:
+	case 2:
+		if ( init->outp &&
+		    (data = nvbios_dpout_match(bios, DCB_OUTPUT_DP,
+					       (init->outp->or << 0) |
+					       (init->outp->sorconf.link << 6),
+					       &ver, &hdr, &cnt, &len, &info)))
+		{
+			if (!(info.flags & cond))
+				init_exec_set(init, false);
+			break;
+		}
+
+		warn("script needs dp output table data\n");
+		break;
+	case 5:
+		if (!(init_rdauxr(init, 0x0d) & 1))
+			init_exec_set(init, false);
+		break;
+	default:
+		warn("unknown dp condition 0x%02x\n", cond);
+		break;
+	}
+}
+
+/**
+ * INIT_IO_MASK_OR - opcode 0x3b
+ *
+ */
+static void
+init_io_mask_or(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 index = nv_ro08(bios, init->offset + 1);
+	u8    or = init_or(init);
+	u8  data;
+
+	trace("IO_MASK_OR\t0x03d4[0x%02x] &= ~(1 << 0x%02x)", index, or);
+	init->offset += 2;
+
+	data = init_rdvgai(init, 0x03d4, index);
+	init_wrvgai(init, 0x03d4, index, data &= ~(1 << or));
+}
+
+/**
+ * INIT_IO_OR - opcode 0x3c
+ *
+ */
+static void
+init_io_or(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 index = nv_ro08(bios, init->offset + 1);
+	u8    or = init_or(init);
+	u8  data;
+
+	trace("IO_OR\t0x03d4[0x%02x] |= (1 << 0x%02x)", index, or);
+	init->offset += 2;
+
+	data = init_rdvgai(init, 0x03d4, index);
+	init_wrvgai(init, 0x03d4, index, data | (1 << or));
+}
+
+/**
+ * INIT_INDEX_ADDRESS_LATCHED - opcode 0x49
+ *
+ */
+static void
+init_idx_addr_latched(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 creg = nv_ro32(bios, init->offset + 1);
+	u32 dreg = nv_ro32(bios, init->offset + 5);
+	u32 mask = nv_ro32(bios, init->offset + 9);
+	u32 data = nv_ro32(bios, init->offset + 13);
+	u8 count = nv_ro08(bios, init->offset + 17);
+
+	trace("INDEX_ADDRESS_LATCHED\t"
+	      "R[0x%06x] : R[0x%06x]\n\tCTRL &= 0x%08x |= 0x%08x\n",
+	      creg, dreg, mask, data);
+	init->offset += 18;
+
+	while (count--) {
+		u8 iaddr = nv_ro08(bios, init->offset + 0);
+		u8 idata = nv_ro08(bios, init->offset + 1);
+
+		trace("\t[0x%02x] = 0x%02x\n", iaddr, idata);
+		init->offset += 2;
+
+		init_wr32(init, dreg, idata);
+		init_mask(init, creg, ~mask, data | idata);
+	}
+}
+
+/**
+ * INIT_IO_RESTRICT_PLL2 - opcode 0x4a
+ *
+ */
+static void
+init_io_restrict_pll2(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 port = nv_ro16(bios, init->offset + 1);
+	u8 index = nv_ro08(bios, init->offset + 3);
+	u8  mask = nv_ro08(bios, init->offset + 4);
+	u8 shift = nv_ro08(bios, init->offset + 5);
+	u8 count = nv_ro08(bios, init->offset + 6);
+	u32  reg = nv_ro32(bios, init->offset + 7);
+	u8  conf, i;
+
+	trace("IO_RESTRICT_PLL2\t"
+	      "R[0x%06x] =PLL= ((0x%04x[0x%02x] & 0x%02x) >> 0x%02x) [{\n",
+	      reg, port, index, mask, shift);
+	init->offset += 11;
+
+	conf = (init_rdvgai(init, port, index) & mask) >> shift;
+	for (i = 0; i < count; i++) {
+		u32 freq = nv_ro32(bios, init->offset);
+		if (i == conf) {
+			trace("\t%dkHz *\n", freq);
+			init_prog_pll(init, reg, freq);
+		} else {
+			trace("\t%dkHz\n", freq);
+		}
+		init->offset += 4;
+	}
+	trace("}]\n");
+}
+
+/**
+ * INIT_PLL2 - opcode 0x4b
+ *
+ */
+static void
+init_pll2(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32  reg = nv_ro32(bios, init->offset + 1);
+	u32 freq = nv_ro32(bios, init->offset + 5);
+
+	trace("PLL2\tR[0x%06x] =PLL= %dkHz\n", reg, freq);
+	init->offset += 9;
+
+	init_prog_pll(init, reg, freq);
+}
+
+/**
+ * INIT_I2C_BYTE - opcode 0x4c
+ *
+ */
+static void
+init_i2c_byte(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 index = nv_ro08(bios, init->offset + 1);
+	u8  addr = nv_ro08(bios, init->offset + 2) >> 1;
+	u8 count = nv_ro08(bios, init->offset + 3);
+
+	trace("I2C_BYTE\tI2C[0x%02x][0x%02x]\n", index, addr);
+	init->offset += 4;
+
+	while (count--) {
+		u8  reg = nv_ro08(bios, init->offset + 0);
+		u8 mask = nv_ro08(bios, init->offset + 1);
+		u8 data = nv_ro08(bios, init->offset + 2);
+		int val;
+
+		trace("\t[0x%02x] &= 0x%02x |= 0x%02x\n", reg, mask, data);
+		init->offset += 3;
+
+		val = init_rdi2cr(init, index, addr, reg);
+		if (val < 0)
+			continue;
+		init_wri2cr(init, index, addr, reg, (val & mask) | data);
+	}
+}
+
+/**
+ * INIT_ZM_I2C_BYTE - opcode 0x4d
+ *
+ */
+static void
+init_zm_i2c_byte(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 index = nv_ro08(bios, init->offset + 1);
+	u8  addr = nv_ro08(bios, init->offset + 2) >> 1;
+	u8 count = nv_ro08(bios, init->offset + 3);
+
+	trace("ZM_I2C_BYTE\tI2C[0x%02x][0x%02x]\n", index, addr);
+	init->offset += 4;
+
+	while (count--) {
+		u8  reg = nv_ro08(bios, init->offset + 0);
+		u8 data = nv_ro08(bios, init->offset + 1);
+
+		trace("\t[0x%02x] = 0x%02x\n", reg, data);
+		init->offset += 2;
+
+		init_wri2cr(init, index, addr, reg, data);
+	}
+
+}
+
+/**
+ * INIT_ZM_I2C - opcode 0x4e
+ *
+ */
+static void
+init_zm_i2c(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 index = nv_ro08(bios, init->offset + 1);
+	u8  addr = nv_ro08(bios, init->offset + 2) >> 1;
+	u8 count = nv_ro08(bios, init->offset + 3);
+	u8 data[256], i;
+
+	trace("ZM_I2C\tI2C[0x%02x][0x%02x]\n", index, addr);
+	init->offset += 4;
+
+	for (i = 0; i < count; i++) {
+		data[i] = nv_ro08(bios, init->offset);
+		trace("\t0x%02x\n", data[i]);
+		init->offset++;
+	}
+
+	if (init_exec(init)) {
+		struct nouveau_i2c_port *port = init_i2c(init, index);
+		struct i2c_msg msg = {
+			.addr = addr, .flags = 0, .len = count, .buf = data,
+		};
+		int ret;
+
+		if (port && (ret = i2c_transfer(&port->adapter, &msg, 1)) != 1)
+			warn("i2c wr failed, %d\n", ret);
+	}
+}
+
+/**
+ * INIT_TMDS - opcode 0x4f
+ *
+ */
+static void
+init_tmds(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 tmds = nv_ro08(bios, init->offset + 1);
+	u8 addr = nv_ro08(bios, init->offset + 2);
+	u8 mask = nv_ro08(bios, init->offset + 3);
+	u8 data = nv_ro08(bios, init->offset + 4);
+	u32 reg = init_tmds_reg(init, tmds);
+
+	trace("TMDS\tT[0x%02x][0x%02x] &= 0x%02x |= 0x%02x\n",
+	      tmds, addr, mask, data);
+	init->offset += 5;
+
+	if (reg == 0)
+		return;
+
+	init_wr32(init, reg + 0, addr | 0x00010000);
+	init_wr32(init, reg + 4, data | (init_rd32(init, reg + 4) & mask));
+	init_wr32(init, reg + 0, addr);
+}
+
+/**
+ * INIT_ZM_TMDS_GROUP - opcode 0x50
+ *
+ */
+static void
+init_zm_tmds_group(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8  tmds = nv_ro08(bios, init->offset + 1);
+	u8 count = nv_ro08(bios, init->offset + 2);
+	u32  reg = init_tmds_reg(init, tmds);
+
+	trace("TMDS_ZM_GROUP\tT[0x%02x]\n", tmds);
+	init->offset += 3;
+
+	while (count--) {
+		u8 addr = nv_ro08(bios, init->offset + 0);
+		u8 data = nv_ro08(bios, init->offset + 1);
+
+		trace("\t[0x%02x] = 0x%02x\n", addr, data);
+		init->offset += 2;
+
+		init_wr32(init, reg + 4, data);
+		init_wr32(init, reg + 0, addr);
+	}
+}
+
+/**
+ * INIT_CR_INDEX_ADDRESS_LATCHED - opcode 0x51
+ *
+ */
+static void
+init_cr_idx_adr_latch(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 addr0 = nv_ro08(bios, init->offset + 1);
+	u8 addr1 = nv_ro08(bios, init->offset + 2);
+	u8  base = nv_ro08(bios, init->offset + 3);
+	u8 count = nv_ro08(bios, init->offset + 4);
+	u8 save0;
+
+	trace("CR_INDEX_ADDR C[%02x] C[%02x]\n", addr0, addr1);
+	init->offset += 5;
+
+	save0 = init_rdvgai(init, 0x03d4, addr0);
+	while (count--) {
+		u8 data = nv_ro08(bios, init->offset);
+
+		trace("\t\t[0x%02x] = 0x%02x\n", base, data);
+		init->offset += 1;
+
+		init_wrvgai(init, 0x03d4, addr0, base++);
+		init_wrvgai(init, 0x03d4, addr1, data);
+	}
+	init_wrvgai(init, 0x03d4, addr0, save0);
+}
+
+/**
+ * INIT_CR - opcode 0x52
+ *
+ */
+static void
+init_cr(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 addr = nv_ro08(bios, init->offset + 1);
+	u8 mask = nv_ro08(bios, init->offset + 2);
+	u8 data = nv_ro08(bios, init->offset + 3);
+	u8 val;
+
+	trace("CR\t\tC[0x%02x] &= 0x%02x |= 0x%02x\n", addr, mask, data);
+	init->offset += 4;
+
+	val = init_rdvgai(init, 0x03d4, addr) & mask;
+	init_wrvgai(init, 0x03d4, addr, val | data);
+}
+
+/**
+ * INIT_ZM_CR - opcode 0x53
+ *
+ */
+static void
+init_zm_cr(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 addr = nv_ro08(bios, init->offset + 1);
+	u8 data = nv_ro08(bios, init->offset + 2);
+
+	trace("ZM_CR\tC[0x%02x] = 0x%02x\n", addr,  data);
+	init->offset += 3;
+
+	init_wrvgai(init, 0x03d4, addr, data);
+}
+
+/**
+ * INIT_ZM_CR_GROUP - opcode 0x54
+ *
+ */
+static void
+init_zm_cr_group(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 count = nv_ro08(bios, init->offset + 1);
+
+	trace("ZM_CR_GROUP\n");
+	init->offset += 2;
+
+	while (count--) {
+		u8 addr = nv_ro08(bios, init->offset + 0);
+		u8 data = nv_ro08(bios, init->offset + 1);
+
+		trace("\t\tC[0x%02x] = 0x%02x\n", addr, data);
+		init->offset += 2;
+
+		init_wrvgai(init, 0x03d4, addr, data);
+	}
+}
+
+/**
+ * INIT_CONDITION_TIME - opcode 0x56
+ *
+ */
+static void
+init_condition_time(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8  cond = nv_ro08(bios, init->offset + 1);
+	u8 retry = nv_ro08(bios, init->offset + 2);
+	u8  wait = min((u16)retry * 50, 100);
+
+	trace("CONDITION_TIME\t0x%02x 0x%02x\n", cond, retry);
+	init->offset += 3;
+
+	if (!init_exec(init))
+		return;
+
+	while (wait--) {
+		if (init_condition_met(init, cond))
+			return;
+		mdelay(20);
+	}
+
+	init_exec_set(init, false);
+}
+
+/**
+ * INIT_LTIME - opcode 0x57
+ *
+ */
+static void
+init_ltime(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 msec = nv_ro16(bios, init->offset + 1);
+
+	trace("LTIME\t0x%04x\n", msec);
+	init->offset += 3;
+
+	if (init_exec(init))
+		mdelay(msec);
+}
+
+/**
+ * INIT_ZM_REG_SEQUENCE - opcode 0x58
+ *
+ */
+static void
+init_zm_reg_sequence(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 base = nv_ro32(bios, init->offset + 1);
+	u8 count = nv_ro08(bios, init->offset + 5);
+
+	trace("ZM_REG_SEQUENCE\t0x%02x\n", count);
+	init->offset += 6;
+
+	while (count--) {
+		u32 data = nv_ro32(bios, init->offset);
+
+		trace("\t\tR[0x%06x] = 0x%08x\n", base, data);
+		init->offset += 4;
+
+		init_wr32(init, base, data);
+		base += 4;
+	}
+}
+
+/**
+ * INIT_SUB_DIRECT - opcode 0x5b
+ *
+ */
+static void
+init_sub_direct(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 addr = nv_ro16(bios, init->offset + 1);
+	u16 save;
+
+	trace("SUB_DIRECT\t0x%04x\n", addr);
+
+	if (init_exec(init)) {
+		save = init->offset;
+		init->offset = addr;
+		if (nvbios_exec(init)) {
+			error("error parsing sub-table\n");
+			return;
+		}
+		init->offset = save;
+	}
+
+	init->offset += 3;
+}
+
+/**
+ * INIT_JUMP - opcode 0x5c
+ *
+ */
+static void
+init_jump(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 offset = nv_ro16(bios, init->offset + 1);
+
+	trace("JUMP\t0x%04x\n", offset);
+	init->offset = offset;
+}
+
+/**
+ * INIT_I2C_IF - opcode 0x5e
+ *
+ */
+static void
+init_i2c_if(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 index = nv_ro08(bios, init->offset + 1);
+	u8  addr = nv_ro08(bios, init->offset + 2);
+	u8   reg = nv_ro08(bios, init->offset + 3);
+	u8  mask = nv_ro08(bios, init->offset + 4);
+	u8  data = nv_ro08(bios, init->offset + 5);
+	u8 value;
+
+	trace("I2C_IF\tI2C[0x%02x][0x%02x][0x%02x] & 0x%02x == 0x%02x\n",
+	      index, addr, reg, mask, data);
+	init->offset += 6;
+	init_exec_force(init, true);
+
+	value = init_rdi2cr(init, index, addr, reg);
+	if ((value & mask) != data)
+		init_exec_set(init, false);
+
+	init_exec_force(init, false);
+}
+
+/**
+ * INIT_COPY_NV_REG - opcode 0x5f
+ *
+ */
+static void
+init_copy_nv_reg(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32  sreg = nv_ro32(bios, init->offset + 1);
+	u8  shift = nv_ro08(bios, init->offset + 5);
+	u32 smask = nv_ro32(bios, init->offset + 6);
+	u32  sxor = nv_ro32(bios, init->offset + 10);
+	u32  dreg = nv_ro32(bios, init->offset + 14);
+	u32 dmask = nv_ro32(bios, init->offset + 18);
+	u32 data;
+
+	trace("COPY_NV_REG\tR[0x%06x] &= 0x%08x |= "
+	      "((R[0x%06x] %s 0x%02x) & 0x%08x ^ 0x%08x)\n",
+	      dreg, dmask, sreg, (shift & 0x80) ? "<<" : ">>",
+	      (shift & 0x80) ? (0x100 - shift) : shift, smask, sxor);
+	init->offset += 22;
+
+	data = init_shift(init_rd32(init, sreg), shift);
+	init_mask(init, dreg, ~dmask, (data & smask) ^ sxor);
+}
+
+/**
+ * INIT_ZM_INDEX_IO - opcode 0x62
+ *
+ */
+static void
+init_zm_index_io(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 port = nv_ro16(bios, init->offset + 1);
+	u8 index = nv_ro08(bios, init->offset + 3);
+	u8  data = nv_ro08(bios, init->offset + 4);
+
+	trace("ZM_INDEX_IO\tI[0x%04x][0x%02x] = 0x%02x\n", port, index, data);
+	init->offset += 5;
+
+	init_wrvgai(init, port, index, data);
+}
+
+/**
+ * INIT_COMPUTE_MEM - opcode 0x63
+ *
+ */
+static void
+init_compute_mem(struct nvbios_init *init)
+{
+	struct nouveau_devinit *devinit = nouveau_devinit(init->bios);
+
+	trace("COMPUTE_MEM\n");
+	init->offset += 1;
+
+	init_exec_force(init, true);
+	if (init_exec(init) && devinit->meminit)
+		devinit->meminit(devinit);
+	init_exec_force(init, false);
+}
+
+/**
+ * INIT_RESET - opcode 0x65
+ *
+ */
+static void
+init_reset(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32   reg = nv_ro32(bios, init->offset + 1);
+	u32 data1 = nv_ro32(bios, init->offset + 5);
+	u32 data2 = nv_ro32(bios, init->offset + 9);
+	u32 savepci19;
+
+	trace("RESET\tR[0x%08x] = 0x%08x, 0x%08x", reg, data1, data2);
+	init->offset += 13;
+	init_exec_force(init, true);
+
+	savepci19 = init_mask(init, 0x00184c, 0x00000f00, 0x00000000);
+	init_wr32(init, reg, data1);
+	udelay(10);
+	init_wr32(init, reg, data2);
+	init_wr32(init, 0x00184c, savepci19);
+	init_mask(init, 0x001850, 0x00000001, 0x00000000);
+
+	init_exec_force(init, false);
+}
+
+/**
+ * INIT_CONFIGURE_MEM - opcode 0x66
+ *
+ */
+static u16
+init_configure_mem_clk(struct nvbios_init *init)
+{
+	u16 mdata = bmp_mem_init_table(init->bios);
+	if (mdata)
+		mdata += (init_rdvgai(init, 0x03d4, 0x3c) >> 4) * 66;
+	return mdata;
+}
+
+static void
+init_configure_mem(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 mdata, sdata;
+	u32 addr, data;
+
+	trace("CONFIGURE_MEM\n");
+	init->offset += 1;
+
+	if (bios->version.major > 2) {
+		init_done(init);
+		return;
+	}
+	init_exec_force(init, true);
+
+	mdata = init_configure_mem_clk(init);
+	sdata = bmp_sdr_seq_table(bios);
+	if (nv_ro08(bios, mdata) & 0x01)
+		sdata = bmp_ddr_seq_table(bios);
+	mdata += 6; /* skip to data */
+
+	data = init_rdvgai(init, 0x03c4, 0x01);
+	init_wrvgai(init, 0x03c4, 0x01, data | 0x20);
+
+	while ((addr = nv_ro32(bios, sdata)) != 0xffffffff) {
+		switch (addr) {
+		case 0x10021c: /* CKE_NORMAL */
+		case 0x1002d0: /* CMD_REFRESH */
+		case 0x1002d4: /* CMD_PRECHARGE */
+			data = 0x00000001;
+			break;
+		default:
+			data = nv_ro32(bios, mdata);
+			mdata += 4;
+			if (data == 0xffffffff)
+				continue;
+			break;
+		}
+
+		init_wr32(init, addr, data);
+	}
+
+	init_exec_force(init, false);
+}
+
+/**
+ * INIT_CONFIGURE_CLK - opcode 0x67
+ *
+ */
+static void
+init_configure_clk(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 mdata, clock;
+
+	trace("CONFIGURE_CLK\n");
+	init->offset += 1;
+
+	if (bios->version.major > 2) {
+		init_done(init);
+		return;
+	}
+	init_exec_force(init, true);
+
+	mdata = init_configure_mem_clk(init);
+
+	/* NVPLL */
+	clock = nv_ro16(bios, mdata + 4) * 10;
+	init_prog_pll(init, 0x680500, clock);
+
+	/* MPLL */
+	clock = nv_ro16(bios, mdata + 2) * 10;
+	if (nv_ro08(bios, mdata) & 0x01)
+		clock *= 2;
+	init_prog_pll(init, 0x680504, clock);
+
+	init_exec_force(init, false);
+}
+
+/**
+ * INIT_CONFIGURE_PREINIT - opcode 0x68
+ *
+ */
+static void
+init_configure_preinit(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 strap;
+
+	trace("CONFIGURE_PREINIT\n");
+	init->offset += 1;
+
+	if (bios->version.major > 2) {
+		init_done(init);
+		return;
+	}
+	init_exec_force(init, true);
+
+	strap = init_rd32(init, 0x101000);
+	strap = ((strap << 2) & 0xf0) | ((strap & 0x40) >> 6);
+	init_wrvgai(init, 0x03d4, 0x3c, strap);
+
+	init_exec_force(init, false);
+}
+
+/**
+ * INIT_IO - opcode 0x69
+ *
+ */
+static void
+init_io(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 port = nv_ro16(bios, init->offset + 1);
+	u8  mask = nv_ro16(bios, init->offset + 3);
+	u8  data = nv_ro16(bios, init->offset + 4);
+	u8 value;
+
+	trace("IO\t\tI[0x%04x] &= 0x%02x |= 0x%02x\n", port, mask, data);
+	init->offset += 5;
+
+	/* ummm.. yes.. should really figure out wtf this is and why it's
+	 * needed some day..  it's almost certainly wrong, but, it also
+	 * somehow makes things work...
+	 */
+	if (nv_device(init->bios)->card_type >= NV_50 &&
+	    port == 0x03c3 && data == 0x01) {
+		init_mask(init, 0x614100, 0xf0800000, 0x00800000);
+		init_mask(init, 0x00e18c, 0x00020000, 0x00020000);
+		init_mask(init, 0x614900, 0xf0800000, 0x00800000);
+		init_mask(init, 0x000200, 0x40000000, 0x00000000);
+		mdelay(10);
+		init_mask(init, 0x00e18c, 0x00020000, 0x00000000);
+		init_mask(init, 0x000200, 0x40000000, 0x40000000);
+		init_wr32(init, 0x614100, 0x00800018);
+		init_wr32(init, 0x614900, 0x00800018);
+		mdelay(10);
+		init_wr32(init, 0x614100, 0x10000018);
+		init_wr32(init, 0x614900, 0x10000018);
+		return;
+	}
+
+	value = init_rdport(init, port) & mask;
+	init_wrport(init, port, data | value);
+}
+
+/**
+ * INIT_SUB - opcode 0x6b
+ *
+ */
+static void
+init_sub(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 index = nv_ro08(bios, init->offset + 1);
+	u16 addr, save;
+
+	trace("SUB\t0x%02x\n", index);
+
+	addr = init_script(bios, index);
+	if (addr && init_exec(init)) {
+		save = init->offset;
+		init->offset = addr;
+		if (nvbios_exec(init)) {
+			error("error parsing sub-table\n");
+			return;
+		}
+		init->offset = save;
+	}
+
+	init->offset += 2;
+}
+
+/**
+ * INIT_RAM_CONDITION - opcode 0x6d
+ *
+ */
+static void
+init_ram_condition(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8  mask = nv_ro08(bios, init->offset + 1);
+	u8 value = nv_ro08(bios, init->offset + 2);
+
+	trace("RAM_CONDITION\t"
+	      "(R[0x100000] & 0x%02x) == 0x%02x\n", mask, value);
+	init->offset += 3;
+
+	if ((init_rd32(init, 0x100000) & mask) != value)
+		init_exec_set(init, false);
+}
+
+/**
+ * INIT_NV_REG - opcode 0x6e
+ *
+ */
+static void
+init_nv_reg(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32  reg = nv_ro32(bios, init->offset + 1);
+	u32 mask = nv_ro32(bios, init->offset + 5);
+	u32 data = nv_ro32(bios, init->offset + 9);
+
+	trace("NV_REG\tR[0x%06x] &= 0x%08x |= 0x%08x\n", reg, mask, data);
+	init->offset += 13;
+
+	init_mask(init, reg, ~mask, data);
+}
+
+/**
+ * INIT_MACRO - opcode 0x6f
+ *
+ */
+static void
+init_macro(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8  macro = nv_ro08(bios, init->offset + 1);
+	u16 table;
+
+	trace("MACRO\t0x%02x\n", macro);
+
+	table = init_macro_table(init);
+	if (table) {
+		u32 addr = nv_ro32(bios, table + (macro * 8) + 0);
+		u32 data = nv_ro32(bios, table + (macro * 8) + 4);
+		trace("\t\tR[0x%06x] = 0x%08x\n", addr, data);
+		init_wr32(init, addr, data);
+	}
+
+	init->offset += 2;
+}
+
+/**
+ * INIT_RESUME - opcode 0x72
+ *
+ */
+static void
+init_resume(struct nvbios_init *init)
+{
+	trace("RESUME\n");
+	init->offset += 1;
+	init_exec_set(init, true);
+}
+
+/**
+ * INIT_TIME - opcode 0x74
+ *
+ */
+static void
+init_time(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 usec = nv_ro16(bios, init->offset + 1);
+
+	trace("TIME\t0x%04x\n", usec);
+	init->offset += 3;
+
+	if (init_exec(init)) {
+		if (usec < 1000)
+			udelay(usec);
+		else
+			mdelay((usec + 900) / 1000);
+	}
+}
+
+/**
+ * INIT_CONDITION - opcode 0x75
+ *
+ */
+static void
+init_condition(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 cond = nv_ro08(bios, init->offset + 1);
+
+	trace("CONDITION\t0x%02x\n", cond);
+	init->offset += 2;
+
+	if (!init_condition_met(init, cond))
+		init_exec_set(init, false);
+}
+
+/**
+ * INIT_IO_CONDITION - opcode 0x76
+ *
+ */
+static void
+init_io_condition(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 cond = nv_ro08(bios, init->offset + 1);
+
+	trace("IO_CONDITION\t0x%02x\n", cond);
+	init->offset += 2;
+
+	if (!init_io_condition_met(init, cond))
+		init_exec_set(init, false);
+}
+
+/**
+ * INIT_INDEX_IO - opcode 0x78
+ *
+ */
+static void
+init_index_io(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u16 port = nv_ro16(bios, init->offset + 1);
+	u8 index = nv_ro16(bios, init->offset + 3);
+	u8  mask = nv_ro08(bios, init->offset + 4);
+	u8  data = nv_ro08(bios, init->offset + 5);
+	u8 value;
+
+	trace("INDEX_IO\tI[0x%04x][0x%02x] &= 0x%02x |= 0x%02x\n",
+	      port, index, mask, data);
+	init->offset += 6;
+
+	value = init_rdvgai(init, port, index) & mask;
+	init_wrvgai(init, port, index, data | value);
+}
+
+/**
+ * INIT_PLL - opcode 0x79
+ *
+ */
+static void
+init_pll(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32  reg = nv_ro32(bios, init->offset + 1);
+	u32 freq = nv_ro16(bios, init->offset + 5) * 10;
+
+	trace("PLL\tR[0x%06x] =PLL= %dkHz\n", reg, freq);
+	init->offset += 7;
+
+	init_prog_pll(init, reg, freq);
+}
+
+/**
+ * INIT_ZM_REG - opcode 0x7a
+ *
+ */
+static void
+init_zm_reg(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 addr = nv_ro32(bios, init->offset + 1);
+	u32 data = nv_ro32(bios, init->offset + 5);
+
+	trace("ZM_REG\tR[0x%06x] = 0x%08x\n", addr, data);
+	init->offset += 9;
+
+	if (addr == 0x000200)
+		data |= 0x00000001;
+
+	init_wr32(init, addr, data);
+}
+
+/**
+ * INIT_RAM_RESTRICT_PLL - opcde 0x87
+ *
+ */
+static void
+init_ram_restrict_pll(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8  type = nv_ro08(bios, init->offset + 1);
+	u8 count = init_ram_restrict_group_count(init);
+	u8 strap = init_ram_restrict(init);
+	u8 cconf;
+
+	trace("RAM_RESTRICT_PLL\t0x%02x\n", type);
+	init->offset += 2;
+
+	for (cconf = 0; cconf < count; cconf++) {
+		u32 freq = nv_ro32(bios, init->offset);
+
+		if (cconf == strap) {
+			trace("%dkHz *\n", freq);
+			init_prog_pll(init, type, freq);
+		} else {
+			trace("%dkHz\n", freq);
+		}
+
+		init->offset += 4;
+	}
+}
+
+/**
+ * INIT_GPIO - opcode 0x8e
+ *
+ */
+static void
+init_gpio(struct nvbios_init *init)
+{
+	struct nouveau_gpio *gpio = nouveau_gpio(init->bios);
+
+	trace("GPIO\n");
+	init->offset += 1;
+
+	if (init_exec(init) && gpio && gpio->reset)
+		gpio->reset(gpio, DCB_GPIO_UNUSED);
+}
+
+/**
+ * INIT_RAM_RESTRICT_ZM_GROUP - opcode 0x8f
+ *
+ */
+static void
+init_ram_restrict_zm_reg_group(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 addr = nv_ro32(bios, init->offset + 1);
+	u8  incr = nv_ro08(bios, init->offset + 5);
+	u8   num = nv_ro08(bios, init->offset + 6);
+	u8 count = init_ram_restrict_group_count(init);
+	u8 index = init_ram_restrict(init);
+	u8 i, j;
+
+	trace("RAM_RESTRICT_ZM_REG_GROUP\t"
+	      "R[%08x] 0x%02x 0x%02x\n", addr, incr, num);
+	init->offset += 7;
+
+	for (i = 0; i < num; i++) {
+		trace("\tR[0x%06x] = {\n", addr);
+		for (j = 0; j < count; j++) {
+			u32 data = nv_ro32(bios, init->offset);
+
+			if (j == index) {
+				trace("\t\t0x%08x *\n", data);
+				init_wr32(init, addr, data);
+			} else {
+				trace("\t\t0x%08x\n", data);
+			}
+
+			init->offset += 4;
+		}
+		trace("\t}\n");
+		addr += incr;
+	}
+}
+
+/**
+ * INIT_COPY_ZM_REG - opcode 0x90
+ *
+ */
+static void
+init_copy_zm_reg(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 sreg = nv_ro32(bios, init->offset + 1);
+	u32 dreg = nv_ro32(bios, init->offset + 5);
+
+	trace("COPY_ZM_REG\tR[0x%06x] = R[0x%06x]\n", sreg, dreg);
+	init->offset += 9;
+
+	init_wr32(init, dreg, init_rd32(init, sreg));
+}
+
+/**
+ * INIT_ZM_REG_GROUP - opcode 0x91
+ *
+ */
+static void
+init_zm_reg_group(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 addr = nv_ro32(bios, init->offset + 1);
+	u8 count = nv_ro08(bios, init->offset + 5);
+
+	trace("ZM_REG_GROUP\tR[0x%06x] =\n");
+	init->offset += 6;
+
+	while (count--) {
+		u32 data = nv_ro32(bios, init->offset);
+		trace("\t0x%08x\n", data);
+		init_wr32(init, addr, data);
+		init->offset += 4;
+	}
+}
+
+/**
+ * INIT_XLAT - opcode 0x96
+ *
+ */
+static void
+init_xlat(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 saddr = nv_ro32(bios, init->offset + 1);
+	u8 sshift = nv_ro08(bios, init->offset + 5);
+	u8  smask = nv_ro08(bios, init->offset + 6);
+	u8  index = nv_ro08(bios, init->offset + 7);
+	u32 daddr = nv_ro32(bios, init->offset + 8);
+	u32 dmask = nv_ro32(bios, init->offset + 12);
+	u8  shift = nv_ro08(bios, init->offset + 16);
+	u32 data;
+
+	trace("INIT_XLAT\tR[0x%06x] &= 0x%08x |= "
+	      "(X%02x((R[0x%06x] %s 0x%02x) & 0x%02x) << 0x%02x)\n",
+	      daddr, dmask, index, saddr, (sshift & 0x80) ? "<<" : ">>",
+	      (sshift & 0x80) ? (0x100 - sshift) : sshift, smask, shift);
+	init->offset += 17;
+
+	data = init_shift(init_rd32(init, saddr), sshift) & smask;
+	data = init_xlat_(init, index, data) << shift;
+	init_mask(init, daddr, ~dmask, data);
+}
+
+/**
+ * INIT_ZM_MASK_ADD - opcode 0x97
+ *
+ */
+static void
+init_zm_mask_add(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 addr = nv_ro32(bios, init->offset + 1);
+	u32 mask = nv_ro32(bios, init->offset + 5);
+	u32  add = nv_ro32(bios, init->offset + 9);
+	u32 data;
+
+	trace("ZM_MASK_ADD\tR[0x%06x] &= 0x%08x += 0x%08x\n", addr, mask, add);
+	init->offset += 13;
+
+	data  =  init_rd32(init, addr) & mask;
+	data |= ((data + add) & ~mask);
+	init_wr32(init, addr, data);
+}
+
+/**
+ * INIT_AUXCH - opcode 0x98
+ *
+ */
+static void
+init_auxch(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 addr = nv_ro32(bios, init->offset + 1);
+	u8 count = nv_ro08(bios, init->offset + 5);
+
+	trace("AUXCH\tAUX[0x%08x] 0x%02x\n", addr, count);
+	init->offset += 6;
+
+	while (count--) {
+		u8 mask = nv_ro08(bios, init->offset + 0);
+		u8 data = nv_ro08(bios, init->offset + 1);
+		trace("\tAUX[0x%08x] &= 0x%02x |= 0x%02x\n", addr, mask, data);
+		mask = init_rdauxr(init, addr) & mask;
+		init_wrauxr(init, addr, mask | data);
+		init->offset += 2;
+	}
+}
+
+/**
+ * INIT_AUXCH - opcode 0x99
+ *
+ */
+static void
+init_zm_auxch(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u32 addr = nv_ro32(bios, init->offset + 1);
+	u8 count = nv_ro08(bios, init->offset + 5);
+
+	trace("ZM_AUXCH\tAUX[0x%08x] 0x%02x\n", addr, count);
+	init->offset += 6;
+
+	while (count--) {
+		u8 data = nv_ro08(bios, init->offset + 0);
+		trace("\tAUX[0x%08x] = 0x%02x\n", addr, data);
+		init_wrauxr(init, addr, data);
+		init->offset += 1;
+	}
+}
+
+/**
+ * INIT_I2C_LONG_IF - opcode 0x9a
+ *
+ */
+static void
+init_i2c_long_if(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	u8 index = nv_ro08(bios, init->offset + 1);
+	u8  addr = nv_ro08(bios, init->offset + 2) >> 1;
+	u8 reglo = nv_ro08(bios, init->offset + 3);
+	u8 reghi = nv_ro08(bios, init->offset + 4);
+	u8  mask = nv_ro08(bios, init->offset + 5);
+	u8  data = nv_ro08(bios, init->offset + 6);
+	struct nouveau_i2c_port *port;
+
+	trace("I2C_LONG_IF\t"
+	      "I2C[0x%02x][0x%02x][0x%02x%02x] & 0x%02x == 0x%02x\n",
+	      index, addr, reglo, reghi, mask, data);
+	init->offset += 7;
+
+	port = init_i2c(init, index);
+	if (port) {
+		u8 i[2] = { reghi, reglo };
+		u8 o[1] = {};
+		struct i2c_msg msg[] = {
+			{ .addr = addr, .flags = 0, .len = 2, .buf = i },
+			{ .addr = addr, .flags = I2C_M_RD, .len = 1, .buf = o }
+		};
+		int ret;
+
+		ret = i2c_transfer(&port->adapter, msg, 2);
+		if (ret == 2 && ((o[0] & mask) == data))
+			return;
+	}
+
+	init_exec_set(init, false);
+}
+
+/**
+ * INIT_GPIO_NE - opcode 0xa9
+ *
+ */
+static void
+init_gpio_ne(struct nvbios_init *init)
+{
+	struct nouveau_bios *bios = init->bios;
+	struct nouveau_gpio *gpio = nouveau_gpio(bios);
+	struct dcb_gpio_func func;
+	u8 count = nv_ro08(bios, init->offset + 1);
+	u8 idx = 0, ver, len;
+	u16 data, i;
+
+	trace("GPIO_NE\t");
+	init->offset += 2;
+
+	for (i = init->offset; i < init->offset + count; i++)
+		cont("0x%02x ", nv_ro08(bios, i));
+	cont("\n");
+
+	while ((data = dcb_gpio_parse(bios, 0, idx++, &ver, &len, &func))) {
+		if (func.func != DCB_GPIO_UNUSED) {
+			for (i = init->offset; i < init->offset + count; i++) {
+				if (func.func == nv_ro08(bios, i))
+					break;
+			}
+
+			trace("\tFUNC[0x%02x]", func.func);
+			if (i == (init->offset + count)) {
+				cont(" *");
+				if (init_exec(init) && gpio && gpio->reset)
+					gpio->reset(gpio, func.func);
+			}
+			cont("\n");
+		}
+	}
+
+	init->offset += count;
+}
+
+static struct nvbios_init_opcode {
+	void (*exec)(struct nvbios_init *);
+} init_opcode[] = {
+	[0x32] = { init_io_restrict_prog },
+	[0x33] = { init_repeat },
+	[0x34] = { init_io_restrict_pll },
+	[0x36] = { init_end_repeat },
+	[0x37] = { init_copy },
+	[0x38] = { init_not },
+	[0x39] = { init_io_flag_condition },
+	[0x3a] = { init_dp_condition },
+	[0x3b] = { init_io_mask_or },
+	[0x3c] = { init_io_or },
+	[0x49] = { init_idx_addr_latched },
+	[0x4a] = { init_io_restrict_pll2 },
+	[0x4b] = { init_pll2 },
+	[0x4c] = { init_i2c_byte },
+	[0x4d] = { init_zm_i2c_byte },
+	[0x4e] = { init_zm_i2c },
+	[0x4f] = { init_tmds },
+	[0x50] = { init_zm_tmds_group },
+	[0x51] = { init_cr_idx_adr_latch },
+	[0x52] = { init_cr },
+	[0x53] = { init_zm_cr },
+	[0x54] = { init_zm_cr_group },
+	[0x56] = { init_condition_time },
+	[0x57] = { init_ltime },
+	[0x58] = { init_zm_reg_sequence },
+	[0x5b] = { init_sub_direct },
+	[0x5c] = { init_jump },
+	[0x5e] = { init_i2c_if },
+	[0x5f] = { init_copy_nv_reg },
+	[0x62] = { init_zm_index_io },
+	[0x63] = { init_compute_mem },
+	[0x65] = { init_reset },
+	[0x66] = { init_configure_mem },
+	[0x67] = { init_configure_clk },
+	[0x68] = { init_configure_preinit },
+	[0x69] = { init_io },
+	[0x6b] = { init_sub },
+	[0x6d] = { init_ram_condition },
+	[0x6e] = { init_nv_reg },
+	[0x6f] = { init_macro },
+	[0x71] = { init_done },
+	[0x72] = { init_resume },
+	[0x74] = { init_time },
+	[0x75] = { init_condition },
+	[0x76] = { init_io_condition },
+	[0x78] = { init_index_io },
+	[0x79] = { init_pll },
+	[0x7a] = { init_zm_reg },
+	[0x87] = { init_ram_restrict_pll },
+	[0x8c] = { init_reserved },
+	[0x8d] = { init_reserved },
+	[0x8e] = { init_gpio },
+	[0x8f] = { init_ram_restrict_zm_reg_group },
+	[0x90] = { init_copy_zm_reg },
+	[0x91] = { init_zm_reg_group },
+	[0x92] = { init_reserved },
+	[0x96] = { init_xlat },
+	[0x97] = { init_zm_mask_add },
+	[0x98] = { init_auxch },
+	[0x99] = { init_zm_auxch },
+	[0x9a] = { init_i2c_long_if },
+	[0xa9] = { init_gpio_ne },
+};
+
+#define init_opcode_nr (sizeof(init_opcode) / sizeof(init_opcode[0]))
+
+int
+nvbios_exec(struct nvbios_init *init)
+{
+	init->nested++;
+	while (init->offset) {
+		u8 opcode = nv_ro08(init->bios, init->offset);
+		if (opcode >= init_opcode_nr || !init_opcode[opcode].exec) {
+			error("unknown opcode 0x%02x\n", opcode);
+			return -EINVAL;
+		}
+
+		init_opcode[opcode].exec(init);
+	}
+	init->nested--;
+	return 0;
+}
+
+int
+nvbios_init(struct nouveau_subdev *subdev, bool execute)
+{
+	struct nouveau_bios *bios = nouveau_bios(subdev);
+	int ret = 0;
+	int i = -1;
+	u16 data;
+
+	if (execute)
+		nv_info(bios, "running init tables\n");
+	while (!ret && (data = (init_script(bios, ++i)))) {
+		struct nvbios_init init = {
+			.subdev = subdev,
+			.bios = bios,
+			.offset = data,
+			.outp = NULL,
+			.crtc = -1,
+			.execute = execute ? 1 : 0,
+		};
+
+		ret = nvbios_exec(&init);
+	}
+
+	/* the vbios parser will run this right after the normal init
+	 * tables, whereas the binary driver appears to run it later.
+	 */
+	if (!ret && (data = init_unknown_script(bios))) {
+		struct nvbios_init init = {
+			.subdev = subdev,
+			.bios = bios,
+			.offset = data,
+			.outp = NULL,
+			.crtc = -1,
+			.execute = execute ? 1 : 0,
+		};
+
+		ret = nvbios_exec(&init);
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/bios/mxm.c b/nvkm/subdev/bios/mxm.c
new file mode 100644
index 000000000..2610b11a9
--- /dev/null
+++ b/nvkm/subdev/bios/mxm.c
@@ -0,0 +1,135 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/bios.h>
+#include <subdev/bios/bit.h>
+#include <subdev/bios/mxm.h>
+
+u16
+mxm_table(struct nouveau_bios *bios, u8 *ver, u8 *hdr)
+{
+	struct bit_entry x;
+
+	if (bit_entry(bios, 'x', &x)) {
+		nv_debug(bios, "BIT 'x' table not present\n");
+		return 0x0000;
+	}
+
+	*ver = x.version;
+	*hdr = x.length;
+	if (*ver != 1 || *hdr < 3) {
+		nv_warn(bios, "BIT 'x' table %d/%d unknown\n", *ver, *hdr);
+		return 0x0000;
+	}
+
+	return x.offset;
+}
+
+/* These map MXM v2.x digital connection values to the appropriate SOR/link,
+ * hopefully they're correct for all boards within the same chipset...
+ *
+ * MXM v3.x VBIOS are nicer and provide pointers to these tables.
+ */
+static u8 nv84_sor_map[16] = {
+	0x00, 0x12, 0x22, 0x11, 0x32, 0x31, 0x11, 0x31,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+
+static u8 nv92_sor_map[16] = {
+	0x00, 0x12, 0x22, 0x11, 0x32, 0x31, 0x11, 0x31,
+	0x11, 0x31, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+
+static u8 nv94_sor_map[16] = {
+	0x00, 0x14, 0x24, 0x11, 0x34, 0x31, 0x11, 0x31,
+	0x11, 0x31, 0x12, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+
+static u8 nv98_sor_map[16] = {
+	0x00, 0x14, 0x12, 0x11, 0x00, 0x31, 0x11, 0x31,
+	0x11, 0x31, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00
+};
+
+u8
+mxm_sor_map(struct nouveau_bios *bios, u8 conn)
+{
+	u8  ver, hdr;
+	u16 mxm = mxm_table(bios, &ver, &hdr);
+	if (mxm && hdr >= 6) {
+		u16 map = nv_ro16(bios, mxm + 4);
+		if (map) {
+			ver = nv_ro08(bios, map);
+			if (ver == 0x10) {
+				if (conn < nv_ro08(bios, map + 3)) {
+					map += nv_ro08(bios, map + 1);
+					map += conn;
+					return nv_ro08(bios, map);
+				}
+
+				return 0x00;
+			}
+
+			nv_warn(bios, "unknown sor map v%02x\n", ver);
+		}
+	}
+
+	if (bios->version.chip == 0x84 || bios->version.chip == 0x86)
+		return nv84_sor_map[conn];
+	if (bios->version.chip == 0x92)
+		return nv92_sor_map[conn];
+	if (bios->version.chip == 0x94 || bios->version.chip == 0x96)
+		return nv94_sor_map[conn];
+	if (bios->version.chip == 0x98)
+		return nv98_sor_map[conn];
+
+	nv_warn(bios, "missing sor map\n");
+	return 0x00;
+}
+
+u8
+mxm_ddc_map(struct nouveau_bios *bios, u8 port)
+{
+	u8  ver, hdr;
+	u16 mxm = mxm_table(bios, &ver, &hdr);
+	if (mxm && hdr >= 8) {
+		u16 map = nv_ro16(bios, mxm + 6);
+		if (map) {
+			ver = nv_ro08(bios, map);
+			if (ver == 0x10) {
+				if (port < nv_ro08(bios, map + 3)) {
+					map += nv_ro08(bios, map + 1);
+					map += port;
+					return nv_ro08(bios, map);
+				}
+
+				return 0x00;
+			}
+
+			nv_warn(bios, "unknown ddc map v%02x\n", ver);
+		}
+	}
+
+	/* v2.x: directly write port as dcb i2cidx */
+	return (port << 4) | port;
+}
diff --git a/nvkm/subdev/bios/perf.c b/nvkm/subdev/bios/perf.c
new file mode 100644
index 000000000..bcbb056c2
--- /dev/null
+++ b/nvkm/subdev/bios/perf.c
@@ -0,0 +1,75 @@
+/*
+ * Copyright 2012 Nouveau Community
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Martin Peres
+ */
+
+#include <subdev/bios.h>
+#include <subdev/bios/bit.h>
+#include <subdev/bios/perf.h>
+
+static u16
+perf_table(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	struct bit_entry bit_P;
+	u16 perf = 0x0000;
+
+	if (!bit_entry(bios, 'P', &bit_P)) {
+		if (bit_P.version <= 2) {
+			perf = nv_ro16(bios, bit_P.offset + 0);
+			if (perf) {
+				*ver = nv_ro08(bios, perf + 0);
+				*hdr = nv_ro08(bios, perf + 1);
+			}
+		} else
+			nv_error(bios, "unknown offset for perf in BIT P %d\n",
+				bit_P.version);
+	}
+
+	if (bios->bmp_offset) {
+		if (nv_ro08(bios, bios->bmp_offset + 6) >= 0x25) {
+			perf = nv_ro16(bios, bios->bmp_offset + 0x94);
+			if (perf) {
+				*hdr = nv_ro08(bios, perf + 0);
+				*ver = nv_ro08(bios, perf + 1);
+			}
+		}
+	}
+
+	return perf;
+}
+
+int
+nvbios_perf_fan_parse(struct nouveau_bios *bios,
+		      struct nvbios_perf_fan *fan)
+{
+	u8 ver = 0, hdr = 0, cnt = 0, len = 0;
+	u16 perf = perf_table(bios, &ver, &hdr, &cnt, &len);
+	if (!perf)
+		return -ENODEV;
+
+	if (ver >= 0x20 && ver < 0x40 && hdr > 6)
+		fan->pwm_divisor = nv_ro16(bios, perf + 6);
+	else
+		fan->pwm_divisor = 0;
+
+	return 0;
+}
diff --git a/nvkm/subdev/bios/pll.c b/nvkm/subdev/bios/pll.c
new file mode 100644
index 000000000..f83550120
--- /dev/null
+++ b/nvkm/subdev/bios/pll.c
@@ -0,0 +1,415 @@
+/*
+ * Copyright 2005-2006 Erik Waling
+ * Copyright 2006 Stephane Marchesin
+ * Copyright 2007-2009 Stuart Bennett
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
+ * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <subdev/vga.h>
+#include <subdev/bios.h>
+#include <subdev/bios/bit.h>
+#include <subdev/bios/bmp.h>
+#include <subdev/bios/pll.h>
+
+struct pll_mapping {
+	u8  type;
+	u32 reg;
+};
+
+static struct pll_mapping
+nv04_pll_mapping[] = {
+	{ PLL_CORE  , 0x680500 },
+	{ PLL_MEMORY, 0x680504 },
+	{ PLL_VPLL0 , 0x680508 },
+	{ PLL_VPLL1 , 0x680520 },
+	{}
+};
+
+static struct pll_mapping
+nv40_pll_mapping[] = {
+	{ PLL_CORE  , 0x004000 },
+	{ PLL_MEMORY, 0x004020 },
+	{ PLL_VPLL0 , 0x680508 },
+	{ PLL_VPLL1 , 0x680520 },
+	{}
+};
+
+static struct pll_mapping
+nv50_pll_mapping[] = {
+	{ PLL_CORE  , 0x004028 },
+	{ PLL_SHADER, 0x004020 },
+	{ PLL_UNK03 , 0x004000 },
+	{ PLL_MEMORY, 0x004008 },
+	{ PLL_UNK40 , 0x00e810 },
+	{ PLL_UNK41 , 0x00e818 },
+	{ PLL_UNK42 , 0x00e824 },
+	{ PLL_VPLL0 , 0x614100 },
+	{ PLL_VPLL1 , 0x614900 },
+	{}
+};
+
+static struct pll_mapping
+nv84_pll_mapping[] = {
+	{ PLL_CORE  , 0x004028 },
+	{ PLL_SHADER, 0x004020 },
+	{ PLL_MEMORY, 0x004008 },
+	{ PLL_VDEC  , 0x004030 },
+	{ PLL_UNK41 , 0x00e818 },
+	{ PLL_VPLL0 , 0x614100 },
+	{ PLL_VPLL1 , 0x614900 },
+	{}
+};
+
+static u16
+pll_limits_table(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *cnt, u8 *len)
+{
+	struct bit_entry bit_C;
+
+	if (!bit_entry(bios, 'C', &bit_C) && bit_C.length >= 10) {
+		u16 data = nv_ro16(bios, bit_C.offset + 8);
+		if (data) {
+			*ver = nv_ro08(bios, data + 0);
+			*hdr = nv_ro08(bios, data + 1);
+			*len = nv_ro08(bios, data + 2);
+			*cnt = nv_ro08(bios, data + 3);
+			return data;
+		}
+	}
+
+	if (bmp_version(bios) >= 0x0524) {
+		u16 data = nv_ro16(bios, bios->bmp_offset + 142);
+		if (data) {
+			*ver = nv_ro08(bios, data + 0);
+			*hdr = 1;
+			*cnt = 1;
+			*len = 0x18;
+			return data;
+		}
+	}
+
+	*ver = 0x00;
+	return 0x0000;
+}
+
+static struct pll_mapping *
+pll_map(struct nouveau_bios *bios)
+{
+	switch (nv_device(bios)->card_type) {
+	case NV_04:
+	case NV_10:
+	case NV_20:
+	case NV_30:
+		return nv04_pll_mapping;
+		break;
+	case NV_40:
+		return nv40_pll_mapping;
+	case NV_50:
+		if (nv_device(bios)->chipset == 0x50)
+			return nv50_pll_mapping;
+		else
+		if (nv_device(bios)->chipset <  0xa3 ||
+		    nv_device(bios)->chipset == 0xaa ||
+		    nv_device(bios)->chipset == 0xac)
+			return nv84_pll_mapping;
+	default:
+		return NULL;
+	}
+}
+
+static u16
+pll_map_reg(struct nouveau_bios *bios, u32 reg, u32 *type, u8 *ver, u8 *len)
+{
+	struct pll_mapping *map;
+	u8  hdr, cnt;
+	u16 data;
+
+	data = pll_limits_table(bios, ver, &hdr, &cnt, len);
+	if (data && *ver >= 0x30) {
+		data += hdr;
+		while (cnt--) {
+			if (nv_ro32(bios, data + 3) == reg) {
+				*type = nv_ro08(bios, data + 0);
+				return data;
+			}
+			data += *len;
+		}
+		return 0x0000;
+	}
+
+	map = pll_map(bios);
+	while (map->reg) {
+		if (map->reg == reg && *ver >= 0x20) {
+			u16 addr = (data += hdr);
+			*type = map->type;
+			while (cnt--) {
+				if (nv_ro32(bios, data) == map->reg)
+					return data;
+				data += *len;
+			}
+			return addr;
+		} else
+		if (map->reg == reg) {
+			*type = map->type;
+			return data + 1;
+		}
+		map++;
+	}
+
+	return 0x0000;
+}
+
+static u16
+pll_map_type(struct nouveau_bios *bios, u8 type, u32 *reg, u8 *ver, u8 *len)
+{
+	struct pll_mapping *map;
+	u8  hdr, cnt;
+	u16 data;
+
+	data = pll_limits_table(bios, ver, &hdr, &cnt, len);
+	if (data && *ver >= 0x30) {
+		data += hdr;
+		while (cnt--) {
+			if (nv_ro08(bios, data + 0) == type) {
+				*reg = nv_ro32(bios, data + 3);
+				return data;
+			}
+			data += *len;
+		}
+		return 0x0000;
+	}
+
+	map = pll_map(bios);
+	while (map->reg) {
+		if (map->type == type && *ver >= 0x20) {
+			u16 addr = (data += hdr);
+			*reg = map->reg;
+			while (cnt--) {
+				if (nv_ro32(bios, data) == map->reg)
+					return data;
+				data += *len;
+			}
+			return addr;
+		} else
+		if (map->type == type) {
+			*reg = map->reg;
+			return data + 1;
+		}
+		map++;
+	}
+
+	return 0x0000;
+}
+
+int
+nvbios_pll_parse(struct nouveau_bios *bios, u32 type, struct nvbios_pll *info)
+{
+	u8  ver, len;
+	u32 reg = type;
+	u16 data;
+
+	if (type > PLL_MAX) {
+		reg  = type;
+		data = pll_map_reg(bios, reg, &type, &ver, &len);
+	} else {
+		data = pll_map_type(bios, type, &reg, &ver, &len);
+	}
+
+	if (ver && !data)
+		return -ENOENT;
+
+	memset(info, 0, sizeof(*info));
+	info->type = type;
+	info->reg = reg;
+
+	switch (ver) {
+	case 0x00:
+		break;
+	case 0x10:
+	case 0x11:
+		info->vco1.min_freq = nv_ro32(bios, data + 0);
+		info->vco1.max_freq = nv_ro32(bios, data + 4);
+		info->vco2.min_freq = nv_ro32(bios, data + 8);
+		info->vco2.max_freq = nv_ro32(bios, data + 12);
+		info->vco1.min_inputfreq = nv_ro32(bios, data + 16);
+		info->vco2.min_inputfreq = nv_ro32(bios, data + 20);
+		info->vco1.max_inputfreq = INT_MAX;
+		info->vco2.max_inputfreq = INT_MAX;
+
+		info->max_p = 0x7;
+		info->max_p_usable = 0x6;
+
+		/* these values taken from nv30/31/36 */
+		switch (bios->version.chip) {
+		case 0x36:
+			info->vco1.min_n = 0x5;
+			break;
+		default:
+			info->vco1.min_n = 0x1;
+			break;
+		}
+		info->vco1.max_n = 0xff;
+		info->vco1.min_m = 0x1;
+		info->vco1.max_m = 0xd;
+
+		/*
+		 * On nv30, 31, 36 (i.e. all cards with two stage PLLs with this
+		 * table version (apart from nv35)), N2 is compared to
+		 * maxN2 (0x46) and 10 * maxM2 (0x4), so set maxN2 to 0x28 and
+		 * save a comparison
+		 */
+		info->vco2.min_n = 0x4;
+		switch (bios->version.chip) {
+		case 0x30:
+		case 0x35:
+			info->vco2.max_n = 0x1f;
+			break;
+		default:
+			info->vco2.max_n = 0x28;
+			break;
+		}
+		info->vco2.min_m = 0x1;
+		info->vco2.max_m = 0x4;
+		break;
+	case 0x20:
+	case 0x21:
+		info->vco1.min_freq = nv_ro16(bios, data + 4) * 1000;
+		info->vco1.max_freq = nv_ro16(bios, data + 6) * 1000;
+		info->vco2.min_freq = nv_ro16(bios, data + 8) * 1000;
+		info->vco2.max_freq = nv_ro16(bios, data + 10) * 1000;
+		info->vco1.min_inputfreq = nv_ro16(bios, data + 12) * 1000;
+		info->vco2.min_inputfreq = nv_ro16(bios, data + 14) * 1000;
+		info->vco1.max_inputfreq = nv_ro16(bios, data + 16) * 1000;
+		info->vco2.max_inputfreq = nv_ro16(bios, data + 18) * 1000;
+		info->vco1.min_n = nv_ro08(bios, data + 20);
+		info->vco1.max_n = nv_ro08(bios, data + 21);
+		info->vco1.min_m = nv_ro08(bios, data + 22);
+		info->vco1.max_m = nv_ro08(bios, data + 23);
+		info->vco2.min_n = nv_ro08(bios, data + 24);
+		info->vco2.max_n = nv_ro08(bios, data + 25);
+		info->vco2.min_m = nv_ro08(bios, data + 26);
+		info->vco2.max_m = nv_ro08(bios, data + 27);
+
+		info->max_p = nv_ro08(bios, data + 29);
+		info->max_p_usable = info->max_p;
+		if (bios->version.chip < 0x60)
+			info->max_p_usable = 0x6;
+		info->bias_p = nv_ro08(bios, data + 30);
+
+		if (len > 0x22)
+			info->refclk = nv_ro32(bios, data + 31);
+		break;
+	case 0x30:
+		data = nv_ro16(bios, data + 1);
+
+		info->vco1.min_freq = nv_ro16(bios, data + 0) * 1000;
+		info->vco1.max_freq = nv_ro16(bios, data + 2) * 1000;
+		info->vco2.min_freq = nv_ro16(bios, data + 4) * 1000;
+		info->vco2.max_freq = nv_ro16(bios, data + 6) * 1000;
+		info->vco1.min_inputfreq = nv_ro16(bios, data + 8) * 1000;
+		info->vco2.min_inputfreq = nv_ro16(bios, data + 10) * 1000;
+		info->vco1.max_inputfreq = nv_ro16(bios, data + 12) * 1000;
+		info->vco2.max_inputfreq = nv_ro16(bios, data + 14) * 1000;
+		info->vco1.min_n = nv_ro08(bios, data + 16);
+		info->vco1.max_n = nv_ro08(bios, data + 17);
+		info->vco1.min_m = nv_ro08(bios, data + 18);
+		info->vco1.max_m = nv_ro08(bios, data + 19);
+		info->vco2.min_n = nv_ro08(bios, data + 20);
+		info->vco2.max_n = nv_ro08(bios, data + 21);
+		info->vco2.min_m = nv_ro08(bios, data + 22);
+		info->vco2.max_m = nv_ro08(bios, data + 23);
+		info->max_p_usable = info->max_p = nv_ro08(bios, data + 25);
+		info->bias_p = nv_ro08(bios, data + 27);
+		info->refclk = nv_ro32(bios, data + 28);
+		break;
+	case 0x40:
+		info->refclk = nv_ro16(bios, data + 9) * 1000;
+		data = nv_ro16(bios, data + 1);
+
+		info->vco1.min_freq = nv_ro16(bios, data + 0) * 1000;
+		info->vco1.max_freq = nv_ro16(bios, data + 2) * 1000;
+		info->vco1.min_inputfreq = nv_ro16(bios, data + 4) * 1000;
+		info->vco1.max_inputfreq = nv_ro16(bios, data + 6) * 1000;
+		info->vco1.min_m = nv_ro08(bios, data + 8);
+		info->vco1.max_m = nv_ro08(bios, data + 9);
+		info->vco1.min_n = nv_ro08(bios, data + 10);
+		info->vco1.max_n = nv_ro08(bios, data + 11);
+		info->min_p = nv_ro08(bios, data + 12);
+		info->max_p = nv_ro08(bios, data + 13);
+		break;
+	default:
+		nv_error(bios, "unknown pll limits version 0x%02x\n", ver);
+		return -EINVAL;
+	}
+
+	if (!info->refclk) {
+		info->refclk = nv_device(bios)->crystal;
+		if (bios->version.chip == 0x51) {
+			u32 sel_clk = nv_rd32(bios, 0x680524);
+			if ((info->reg == 0x680508 && sel_clk & 0x20) ||
+			    (info->reg == 0x680520 && sel_clk & 0x80)) {
+				if (nv_rdvgac(bios, 0, 0x27) < 0xa3)
+					info->refclk = 200000;
+				else
+					info->refclk = 25000;
+			}
+		}
+	}
+
+	/*
+	 * By now any valid limit table ought to have set a max frequency for
+	 * vco1, so if it's zero it's either a pre limit table bios, or one
+	 * with an empty limit table (seen on nv18)
+	 */
+	if (!info->vco1.max_freq) {
+		info->vco1.max_freq = nv_ro32(bios, bios->bmp_offset + 67);
+		info->vco1.min_freq = nv_ro32(bios, bios->bmp_offset + 71);
+		if (bmp_version(bios) < 0x0506) {
+			info->vco1.max_freq = 256000;
+			info->vco1.min_freq = 128000;
+		}
+
+		info->vco1.min_inputfreq = 0;
+		info->vco1.max_inputfreq = INT_MAX;
+		info->vco1.min_n = 0x1;
+		info->vco1.max_n = 0xff;
+		info->vco1.min_m = 0x1;
+
+		if (nv_device(bios)->crystal == 13500) {
+			/* nv05 does this, nv11 doesn't, nv10 unknown */
+			if (bios->version.chip < 0x11)
+				info->vco1.min_m = 0x7;
+			info->vco1.max_m = 0xd;
+		} else {
+			if (bios->version.chip < 0x11)
+				info->vco1.min_m = 0x8;
+			info->vco1.max_m = 0xe;
+		}
+
+		if (bios->version.chip <  0x17 ||
+		    bios->version.chip == 0x1a ||
+		    bios->version.chip == 0x20)
+			info->max_p = 4;
+		else
+			info->max_p = 5;
+		info->max_p_usable = info->max_p;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/bios/therm.c b/nvkm/subdev/bios/therm.c
new file mode 100644
index 000000000..862a08a2a
--- /dev/null
+++ b/nvkm/subdev/bios/therm.c
@@ -0,0 +1,177 @@
+/*
+ * Copyright 2012 Nouveau Community
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Martin Peres
+ */
+
+#include <subdev/bios.h>
+#include <subdev/bios/bit.h>
+#include <subdev/bios/therm.h>
+
+static u16
+therm_table(struct nouveau_bios *bios, u8 *ver, u8 *hdr, u8 *len, u8 *cnt)
+{
+	struct bit_entry bit_P;
+	u16 therm = 0;
+
+	if (!bit_entry(bios, 'P', &bit_P)) {
+		if (bit_P.version == 1)
+			therm = nv_ro16(bios, bit_P.offset + 12);
+		else if (bit_P.version == 2)
+			therm = nv_ro16(bios, bit_P.offset + 16);
+		else
+			nv_error(bios,
+				"unknown offset for thermal in BIT P %d\n",
+				bit_P.version);
+	}
+
+	/* exit now if we haven't found the thermal table */
+	if (!therm)
+		return 0x0000;
+
+	*ver = nv_ro08(bios, therm + 0);
+	*hdr = nv_ro08(bios, therm + 1);
+	*len = nv_ro08(bios, therm + 2);
+	*cnt = nv_ro08(bios, therm + 3);
+
+	return therm + nv_ro08(bios, therm + 1);
+}
+
+u16
+nvbios_therm_entry(struct nouveau_bios *bios, int idx, u8 *ver, u8 *len)
+{
+	u8 hdr, cnt;
+	u16 therm = therm_table(bios, ver, &hdr, len, &cnt);
+	if (therm && idx < cnt)
+		return therm + idx * *len;
+	return 0x0000;
+}
+
+int
+nvbios_therm_sensor_parse(struct nouveau_bios *bios,
+			  enum nvbios_therm_domain domain,
+			  struct nvbios_therm_sensor *sensor)
+{
+	s8 thrs_section, sensor_section, offset;
+	u8 ver, len, i;
+	u16 entry;
+
+	/* we only support the core domain for now */
+	if (domain != NVBIOS_THERM_DOMAIN_CORE)
+		return -EINVAL;
+
+	/* Read the entries from the table */
+	thrs_section = 0;
+	sensor_section = -1;
+	i = 0;
+	while ((entry = nvbios_therm_entry(bios, i++, &ver, &len))) {
+		s16 value = nv_ro16(bios, entry + 1);
+
+		switch (nv_ro08(bios, entry + 0)) {
+		case 0x0:
+			thrs_section = value;
+			if (value > 0)
+				return 0; /* we do not try to support ambient */
+			break;
+		case 0x01:
+			sensor_section++;
+			if (sensor_section == 0) {
+				offset = ((s8) nv_ro08(bios, entry + 2)) / 2;
+				sensor->offset_constant = offset;
+			}
+			break;
+
+		case 0x04:
+			if (thrs_section == 0) {
+				sensor->thrs_critical.temp = (value & 0xff0) >> 4;
+				sensor->thrs_critical.hysteresis = value & 0xf;
+			}
+			break;
+
+		case 0x07:
+			if (thrs_section == 0) {
+				sensor->thrs_down_clock.temp = (value & 0xff0) >> 4;
+				sensor->thrs_down_clock.hysteresis = value & 0xf;
+			}
+			break;
+
+		case 0x08:
+			if (thrs_section == 0) {
+				sensor->thrs_fan_boost.temp = (value & 0xff0) >> 4;
+				sensor->thrs_fan_boost.hysteresis = value & 0xf;
+			}
+			break;
+
+		case 0x10:
+			if (sensor_section == 0)
+				sensor->offset_num = value;
+			break;
+
+		case 0x11:
+			if (sensor_section == 0)
+				sensor->offset_den = value;
+			break;
+
+		case 0x12:
+			if (sensor_section == 0)
+				sensor->slope_mult = value;
+			break;
+
+		case 0x13:
+			if (sensor_section == 0)
+				sensor->slope_div = value;
+			break;
+		case 0x32:
+			if (thrs_section == 0) {
+				sensor->thrs_shutdown.temp = (value & 0xff0) >> 4;
+				sensor->thrs_shutdown.hysteresis = value & 0xf;
+			}
+			break;
+		}
+	}
+
+	return 0;
+}
+
+int
+nvbios_therm_fan_parse(struct nouveau_bios *bios,
+			  struct nvbios_therm_fan *fan)
+{
+	u8 ver, len, i;
+	u16 entry;
+
+	i = 0;
+	while ((entry = nvbios_therm_entry(bios, i++, &ver, &len))) {
+		s16 value = nv_ro16(bios, entry + 1);
+
+		switch (nv_ro08(bios, entry + 0)) {
+		case 0x22:
+			fan->min_duty = value & 0xff;
+			fan->max_duty = (value & 0xff00) >> 8;
+			break;
+		case 0x26:
+			fan->pwm_freq = value;
+			break;
+		}
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/clock/Makefile.am b/nvkm/subdev/clock/Makefile.am
new file mode 100644
index 000000000..13927719e
--- /dev/null
+++ b/nvkm/subdev/clock/Makefile.am
@@ -0,0 +1,11 @@
+noinst_LTLIBRARIES = libclock.la
+
+libclock_la_SOURCES = nv04.c \
+		      nv40.c \
+		      nv50.c \
+		      nva3.c \
+		      nvc0.c \
+		      pllnv04.c \
+		      pllnva3.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/clock/nv04.c b/nvkm/subdev/clock/nv04.c
new file mode 100644
index 000000000..b7fd11511
--- /dev/null
+++ b/nvkm/subdev/clock/nv04.c
@@ -0,0 +1,359 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/clock.h>
+#include <subdev/bios.h>
+#include <subdev/bios/pll.h>
+
+#include "pll.h"
+
+struct nv04_clock_priv {
+	struct nouveau_clock base;
+};
+
+static int
+powerctrl_1_shift(int chip_version, int reg)
+{
+	int shift = -4;
+
+	if (chip_version < 0x17 || chip_version == 0x1a || chip_version == 0x20)
+		return shift;
+
+	switch (reg) {
+	case 0x680520:
+		shift += 4;
+	case 0x680508:
+		shift += 4;
+	case 0x680504:
+		shift += 4;
+	case 0x680500:
+		shift += 4;
+	}
+
+	/*
+	 * the shift for vpll regs is only used for nv3x chips with a single
+	 * stage pll
+	 */
+	if (shift > 4 && (chip_version < 0x32 || chip_version == 0x35 ||
+			  chip_version == 0x36 || chip_version >= 0x40))
+		shift = -4;
+
+	return shift;
+}
+
+static void
+setPLL_single(struct nv04_clock_priv *priv, u32 reg,
+	      struct nouveau_pll_vals *pv)
+{
+	int chip_version = nouveau_bios(priv)->version.chip;
+	uint32_t oldpll = nv_rd32(priv, reg);
+	int oldN = (oldpll >> 8) & 0xff, oldM = oldpll & 0xff;
+	uint32_t pll = (oldpll & 0xfff80000) | pv->log2P << 16 | pv->NM1;
+	uint32_t saved_powerctrl_1 = 0;
+	int shift_powerctrl_1 = powerctrl_1_shift(chip_version, reg);
+
+	if (oldpll == pll)
+		return;	/* already set */
+
+	if (shift_powerctrl_1 >= 0) {
+		saved_powerctrl_1 = nv_rd32(priv, 0x001584);
+		nv_wr32(priv, 0x001584,
+			(saved_powerctrl_1 & ~(0xf << shift_powerctrl_1)) |
+			1 << shift_powerctrl_1);
+	}
+
+	if (oldM && pv->M1 && (oldN / oldM < pv->N1 / pv->M1))
+		/* upclock -- write new post divider first */
+		nv_wr32(priv, reg, pv->log2P << 16 | (oldpll & 0xffff));
+	else
+		/* downclock -- write new NM first */
+		nv_wr32(priv, reg, (oldpll & 0xffff0000) | pv->NM1);
+
+	if (chip_version < 0x17 && chip_version != 0x11)
+		/* wait a bit on older chips */
+		msleep(64);
+	nv_rd32(priv, reg);
+
+	/* then write the other half as well */
+	nv_wr32(priv, reg, pll);
+
+	if (shift_powerctrl_1 >= 0)
+		nv_wr32(priv, 0x001584, saved_powerctrl_1);
+}
+
+static uint32_t
+new_ramdac580(uint32_t reg1, bool ss, uint32_t ramdac580)
+{
+	bool head_a = (reg1 == 0x680508);
+
+	if (ss)	/* single stage pll mode */
+		ramdac580 |= head_a ? 0x00000100 : 0x10000000;
+	else
+		ramdac580 &= head_a ? 0xfffffeff : 0xefffffff;
+
+	return ramdac580;
+}
+
+static void
+setPLL_double_highregs(struct nv04_clock_priv *priv, u32 reg1,
+		       struct nouveau_pll_vals *pv)
+{
+	int chip_version = nouveau_bios(priv)->version.chip;
+	bool nv3035 = chip_version == 0x30 || chip_version == 0x35;
+	uint32_t reg2 = reg1 + ((reg1 == 0x680520) ? 0x5c : 0x70);
+	uint32_t oldpll1 = nv_rd32(priv, reg1);
+	uint32_t oldpll2 = !nv3035 ? nv_rd32(priv, reg2) : 0;
+	uint32_t pll1 = (oldpll1 & 0xfff80000) | pv->log2P << 16 | pv->NM1;
+	uint32_t pll2 = (oldpll2 & 0x7fff0000) | 1 << 31 | pv->NM2;
+	uint32_t oldramdac580 = 0, ramdac580 = 0;
+	bool single_stage = !pv->NM2 || pv->N2 == pv->M2;	/* nv41+ only */
+	uint32_t saved_powerctrl_1 = 0, savedc040 = 0;
+	int shift_powerctrl_1 = powerctrl_1_shift(chip_version, reg1);
+
+	/* model specific additions to generic pll1 and pll2 set up above */
+	if (nv3035) {
+		pll1 = (pll1 & 0xfcc7ffff) | (pv->N2 & 0x18) << 21 |
+		       (pv->N2 & 0x7) << 19 | 8 << 4 | (pv->M2 & 7) << 4;
+		pll2 = 0;
+	}
+	if (chip_version > 0x40 && reg1 >= 0x680508) { /* !nv40 */
+		oldramdac580 = nv_rd32(priv, 0x680580);
+		ramdac580 = new_ramdac580(reg1, single_stage, oldramdac580);
+		if (oldramdac580 != ramdac580)
+			oldpll1 = ~0;	/* force mismatch */
+		if (single_stage)
+			/* magic value used by nvidia in single stage mode */
+			pll2 |= 0x011f;
+	}
+	if (chip_version > 0x70)
+		/* magic bits set by the blob (but not the bios) on g71-73 */
+		pll1 = (pll1 & 0x7fffffff) | (single_stage ? 0x4 : 0xc) << 28;
+
+	if (oldpll1 == pll1 && oldpll2 == pll2)
+		return;	/* already set */
+
+	if (shift_powerctrl_1 >= 0) {
+		saved_powerctrl_1 = nv_rd32(priv, 0x001584);
+		nv_wr32(priv, 0x001584,
+			(saved_powerctrl_1 & ~(0xf << shift_powerctrl_1)) |
+			1 << shift_powerctrl_1);
+	}
+
+	if (chip_version >= 0x40) {
+		int shift_c040 = 14;
+
+		switch (reg1) {
+		case 0x680504:
+			shift_c040 += 2;
+		case 0x680500:
+			shift_c040 += 2;
+		case 0x680520:
+			shift_c040 += 2;
+		case 0x680508:
+			shift_c040 += 2;
+		}
+
+		savedc040 = nv_rd32(priv, 0xc040);
+		if (shift_c040 != 14)
+			nv_wr32(priv, 0xc040, savedc040 & ~(3 << shift_c040));
+	}
+
+	if (oldramdac580 != ramdac580)
+		nv_wr32(priv, 0x680580, ramdac580);
+
+	if (!nv3035)
+		nv_wr32(priv, reg2, pll2);
+	nv_wr32(priv, reg1, pll1);
+
+	if (shift_powerctrl_1 >= 0)
+		nv_wr32(priv, 0x001584, saved_powerctrl_1);
+	if (chip_version >= 0x40)
+		nv_wr32(priv, 0xc040, savedc040);
+}
+
+static void
+setPLL_double_lowregs(struct nv04_clock_priv *priv, u32 NMNMreg,
+		      struct nouveau_pll_vals *pv)
+{
+	/* When setting PLLs, there is a merry game of disabling and enabling
+	 * various bits of hardware during the process. This function is a
+	 * synthesis of six nv4x traces, nearly each card doing a subtly
+	 * different thing. With luck all the necessary bits for each card are
+	 * combined herein. Without luck it deviates from each card's formula
+	 * so as to not work on any :)
+	 */
+
+	uint32_t Preg = NMNMreg - 4;
+	bool mpll = Preg == 0x4020;
+	uint32_t oldPval = nv_rd32(priv, Preg);
+	uint32_t NMNM = pv->NM2 << 16 | pv->NM1;
+	uint32_t Pval = (oldPval & (mpll ? ~(0x77 << 16) : ~(7 << 16))) |
+			0xc << 28 | pv->log2P << 16;
+	uint32_t saved4600 = 0;
+	/* some cards have different maskc040s */
+	uint32_t maskc040 = ~(3 << 14), savedc040;
+	bool single_stage = !pv->NM2 || pv->N2 == pv->M2;
+
+	if (nv_rd32(priv, NMNMreg) == NMNM && (oldPval & 0xc0070000) == Pval)
+		return;
+
+	if (Preg == 0x4000)
+		maskc040 = ~0x333;
+	if (Preg == 0x4058)
+		maskc040 = ~(0xc << 24);
+
+	if (mpll) {
+		struct nvbios_pll info;
+		uint8_t Pval2;
+
+		if (nvbios_pll_parse(nouveau_bios(priv), Preg, &info))
+			return;
+
+		Pval2 = pv->log2P + info.bias_p;
+		if (Pval2 > info.max_p)
+			Pval2 = info.max_p;
+		Pval |= 1 << 28 | Pval2 << 20;
+
+		saved4600 = nv_rd32(priv, 0x4600);
+		nv_wr32(priv, 0x4600, saved4600 | 8 << 28);
+	}
+	if (single_stage)
+		Pval |= mpll ? 1 << 12 : 1 << 8;
+
+	nv_wr32(priv, Preg, oldPval | 1 << 28);
+	nv_wr32(priv, Preg, Pval & ~(4 << 28));
+	if (mpll) {
+		Pval |= 8 << 20;
+		nv_wr32(priv, 0x4020, Pval & ~(0xc << 28));
+		nv_wr32(priv, 0x4038, Pval & ~(0xc << 28));
+	}
+
+	savedc040 = nv_rd32(priv, 0xc040);
+	nv_wr32(priv, 0xc040, savedc040 & maskc040);
+
+	nv_wr32(priv, NMNMreg, NMNM);
+	if (NMNMreg == 0x4024)
+		nv_wr32(priv, 0x403c, NMNM);
+
+	nv_wr32(priv, Preg, Pval);
+	if (mpll) {
+		Pval &= ~(8 << 20);
+		nv_wr32(priv, 0x4020, Pval);
+		nv_wr32(priv, 0x4038, Pval);
+		nv_wr32(priv, 0x4600, saved4600);
+	}
+
+	nv_wr32(priv, 0xc040, savedc040);
+
+	if (mpll) {
+		nv_wr32(priv, 0x4020, Pval & ~(1 << 28));
+		nv_wr32(priv, 0x4038, Pval & ~(1 << 28));
+	}
+}
+
+int
+nv04_clock_pll_set(struct nouveau_clock *clk, u32 type, u32 freq)
+{
+	struct nv04_clock_priv *priv = (void *)clk;
+	struct nouveau_pll_vals pv;
+	struct nvbios_pll info;
+	int ret;
+
+	ret = nvbios_pll_parse(nouveau_bios(priv), type > 0x405c ?
+			       type : type - 4, &info);
+	if (ret)
+		return ret;
+
+	ret = clk->pll_calc(clk, &info, freq, &pv);
+	if (!ret)
+		return ret;
+
+	return clk->pll_prog(clk, type, &pv);
+}
+
+int
+nv04_clock_pll_calc(struct nouveau_clock *clock, struct nvbios_pll *info,
+		    int clk, struct nouveau_pll_vals *pv)
+{
+	int N1, M1, N2, M2, P;
+	int ret = nv04_pll_calc(clock, info, clk, &N1, &M1, &N2, &M2, &P);
+	if (ret) {
+		pv->refclk = info->refclk;
+		pv->N1 = N1;
+		pv->M1 = M1;
+		pv->N2 = N2;
+		pv->M2 = M2;
+		pv->log2P = P;
+	}
+	return ret;
+}
+
+int
+nv04_clock_pll_prog(struct nouveau_clock *clk, u32 reg1,
+		    struct nouveau_pll_vals *pv)
+{
+	struct nv04_clock_priv *priv = (void *)clk;
+	int cv = nouveau_bios(clk)->version.chip;
+
+	if (cv == 0x30 || cv == 0x31 || cv == 0x35 || cv == 0x36 ||
+	    cv >= 0x40) {
+		if (reg1 > 0x405c)
+			setPLL_double_highregs(priv, reg1, pv);
+		else
+			setPLL_double_lowregs(priv, reg1, pv);
+	} else
+		setPLL_single(priv, reg1, pv);
+
+	return 0;
+}
+
+static int
+nv04_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv04_clock_priv *priv;
+	int ret;
+
+	ret = nouveau_clock_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.pll_set = nv04_clock_pll_set;
+	priv->base.pll_calc = nv04_clock_pll_calc;
+	priv->base.pll_prog = nv04_clock_pll_prog;
+	return 0;
+}
+
+struct nouveau_oclass
+nv04_clock_oclass = {
+	.handle = NV_SUBDEV(CLOCK, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_clock_ctor,
+		.dtor = _nouveau_clock_dtor,
+		.init = _nouveau_clock_init,
+		.fini = _nouveau_clock_fini,
+	},
+};
diff --git a/nvkm/subdev/clock/nv40.c b/nvkm/subdev/clock/nv40.c
new file mode 100644
index 000000000..a4b2b7ebf
--- /dev/null
+++ b/nvkm/subdev/clock/nv40.c
@@ -0,0 +1,59 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/clock.h>
+
+struct nv40_clock_priv {
+	struct nouveau_clock base;
+};
+
+static int
+nv40_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv40_clock_priv *priv;
+	int ret;
+
+	ret = nouveau_clock_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.pll_set = nv04_clock_pll_set;
+	priv->base.pll_calc = nv04_clock_pll_calc;
+	priv->base.pll_prog = nv04_clock_pll_prog;
+	return 0;
+}
+
+struct nouveau_oclass
+nv40_clock_oclass = {
+	.handle = NV_SUBDEV(CLOCK, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv40_clock_ctor,
+		.dtor = _nouveau_clock_dtor,
+		.init = _nouveau_clock_init,
+		.fini = _nouveau_clock_fini,
+	},
+};
diff --git a/nvkm/subdev/clock/nv50.c b/nvkm/subdev/clock/nv50.c
new file mode 100644
index 000000000..f4147f67e
--- /dev/null
+++ b/nvkm/subdev/clock/nv50.c
@@ -0,0 +1,106 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/clock.h>
+#include <subdev/bios.h>
+#include <subdev/bios/pll.h>
+
+#include "pll.h"
+
+struct nv50_clock_priv {
+	struct nouveau_clock base;
+};
+
+static int
+nv50_clock_pll_set(struct nouveau_clock *clk, u32 type, u32 freq)
+{
+	struct nv50_clock_priv *priv = (void *)clk;
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct nvbios_pll info;
+	int N1, M1, N2, M2, P;
+	int ret;
+
+	ret = nvbios_pll_parse(bios, type, &info);
+	if (ret) {
+		nv_error(clk, "failed to retrieve pll data, %d\n", ret);
+		return ret;
+	}
+
+	ret = nv04_pll_calc(clk, &info, freq, &N1, &M1, &N2, &M2, &P);
+	if (!ret) {
+		nv_error(clk, "failed pll calculation\n");
+		return ret;
+	}
+
+	switch (info.type) {
+	case PLL_VPLL0:
+	case PLL_VPLL1:
+		nv_wr32(priv, info.reg + 0, 0x10000611);
+		nv_mask(priv, info.reg + 4, 0x00ff00ff, (M1 << 16) | N1);
+		nv_mask(priv, info.reg + 8, 0x7fff00ff, (P  << 28) |
+							(M2 << 16) | N2);
+		break;
+	case PLL_MEMORY:
+		nv_mask(priv, info.reg + 0, 0x01ff0000, (P << 22) |
+						        (info.bias_p << 19) |
+							(P << 16));
+		nv_wr32(priv, info.reg + 4, (N1 << 8) | M1);
+		break;
+	default:
+		nv_mask(priv, info.reg + 0, 0x00070000, (P << 16));
+		nv_wr32(priv, info.reg + 4, (N1 << 8) | M1);
+		break;
+	}
+
+	return 0;
+}
+
+static int
+nv50_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv50_clock_priv *priv;
+	int ret;
+
+	ret = nouveau_clock_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.pll_set = nv50_clock_pll_set;
+	priv->base.pll_calc = nv04_clock_pll_calc;
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_clock_oclass = {
+	.handle = NV_SUBDEV(CLOCK, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_clock_ctor,
+		.dtor = _nouveau_clock_dtor,
+		.init = _nouveau_clock_init,
+		.fini = _nouveau_clock_fini,
+	},
+};
diff --git a/nvkm/subdev/clock/nva3.c b/nvkm/subdev/clock/nva3.c
new file mode 100644
index 000000000..9068c98b9
--- /dev/null
+++ b/nvkm/subdev/clock/nva3.c
@@ -0,0 +1,114 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/clock.h>
+#include <subdev/bios.h>
+#include <subdev/bios/pll.h>
+
+#include "pll.h"
+
+struct nva3_clock_priv {
+	struct nouveau_clock base;
+};
+
+static int
+nva3_clock_pll_set(struct nouveau_clock *clk, u32 type, u32 freq)
+{
+	struct nva3_clock_priv *priv = (void *)clk;
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct nvbios_pll info;
+	int N, fN, M, P;
+	int ret;
+
+	ret = nvbios_pll_parse(bios, type, &info);
+	if (ret)
+		return ret;
+
+	ret = nva3_pll_calc(clk, &info, freq, &N, &fN, &M, &P);
+	if (ret < 0)
+		return ret;
+
+	switch (info.type) {
+	case PLL_VPLL0:
+	case PLL_VPLL1:
+		nv_wr32(priv, info.reg + 0, 0x50000610);
+		nv_mask(priv, info.reg + 4, 0x003fffff,
+					    (P << 16) | (M << 8) | N);
+		nv_wr32(priv, info.reg + 8, fN);
+		break;
+	default:
+		nv_warn(priv, "0x%08x/%dKhz unimplemented\n", type, freq);
+		ret = -EINVAL;
+		break;
+	}
+
+	return ret;
+}
+
+int
+nva3_clock_pll_calc(struct nouveau_clock *clock, struct nvbios_pll *info,
+		    int clk, struct nouveau_pll_vals *pv)
+{
+	int ret, N, M, P;
+
+	ret = nva3_pll_calc(clock, info, clk, &N, NULL, &M, &P);
+
+	if (ret > 0) {
+		pv->refclk = info->refclk;
+		pv->N1 = N;
+		pv->M1 = M;
+		pv->log2P = P;
+	}
+	return ret;
+}
+
+
+static int
+nva3_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nva3_clock_priv *priv;
+	int ret;
+
+	ret = nouveau_clock_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.pll_set = nva3_clock_pll_set;
+	priv->base.pll_calc = nva3_clock_pll_calc;
+	return 0;
+}
+
+struct nouveau_oclass
+nva3_clock_oclass = {
+	.handle = NV_SUBDEV(CLOCK, 0xa3),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nva3_clock_ctor,
+		.dtor = _nouveau_clock_dtor,
+		.init = _nouveau_clock_init,
+		.fini = _nouveau_clock_fini,
+	},
+};
diff --git a/nvkm/subdev/clock/nvc0.c b/nvkm/subdev/clock/nvc0.c
new file mode 100644
index 000000000..f6962c9b6
--- /dev/null
+++ b/nvkm/subdev/clock/nvc0.c
@@ -0,0 +1,95 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/clock.h>
+#include <subdev/bios.h>
+#include <subdev/bios/pll.h>
+
+#include "pll.h"
+
+struct nvc0_clock_priv {
+	struct nouveau_clock base;
+};
+
+static int
+nvc0_clock_pll_set(struct nouveau_clock *clk, u32 type, u32 freq)
+{
+	struct nvc0_clock_priv *priv = (void *)clk;
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct nvbios_pll info;
+	int N, fN, M, P;
+	int ret;
+
+	ret = nvbios_pll_parse(bios, type, &info);
+	if (ret)
+		return ret;
+
+	ret = nva3_pll_calc(clk, &info, freq, &N, &fN, &M, &P);
+	if (ret < 0)
+		return ret;
+
+	switch (info.type) {
+	case PLL_VPLL0:
+	case PLL_VPLL1:
+		nv_mask(priv, info.reg + 0x0c, 0x00000000, 0x00000100);
+		nv_wr32(priv, info.reg + 0x04, (P << 16) | (N << 8) | M);
+		nv_wr32(priv, info.reg + 0x10, fN << 16);
+		break;
+	default:
+		nv_warn(priv, "0x%08x/%dKhz unimplemented\n", type, freq);
+		ret = -EINVAL;
+		break;
+	}
+
+	return ret;
+}
+
+static int
+nvc0_clock_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nvc0_clock_priv *priv;
+	int ret;
+
+	ret = nouveau_clock_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.pll_set = nvc0_clock_pll_set;
+	priv->base.pll_calc = nva3_clock_pll_calc;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_clock_oclass = {
+	.handle = NV_SUBDEV(CLOCK, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_clock_ctor,
+		.dtor = _nouveau_clock_dtor,
+		.init = _nouveau_clock_init,
+		.fini = _nouveau_clock_fini,
+	},
+};
diff --git a/nvkm/subdev/clock/pll.h b/nvkm/subdev/clock/pll.h
new file mode 100644
index 000000000..ef2c0078f
--- /dev/null
+++ b/nvkm/subdev/clock/pll.h
@@ -0,0 +1,9 @@
+#ifndef __NOUVEAU_PLL_H__
+#define __NOUVEAU_PLL_H__
+
+int nv04_pll_calc(struct nouveau_clock *, struct nvbios_pll *, u32 freq,
+		  int *N1, int *M1, int *N2, int *M2, int *P);
+int nva3_pll_calc(struct nouveau_clock *, struct nvbios_pll *, u32 freq,
+		  int *N, int *fN, int *M, int *P);
+
+#endif
diff --git a/nvkm/subdev/clock/pllnv04.c b/nvkm/subdev/clock/pllnv04.c
new file mode 100644
index 000000000..a2ab6d051
--- /dev/null
+++ b/nvkm/subdev/clock/pllnv04.c
@@ -0,0 +1,242 @@
+/*
+ * Copyright 1993-2003 NVIDIA, Corporation
+ * Copyright 2007-2009 Stuart Bennett
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
+ * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <subdev/clock.h>
+#include <subdev/bios.h>
+#include <subdev/bios/pll.h>
+
+#include "pll.h"
+
+static int
+getMNP_single(struct nouveau_clock *clock, struct nvbios_pll *info, int clk,
+	      int *pN, int *pM, int *pP)
+{
+	/* Find M, N and P for a single stage PLL
+	 *
+	 * Note that some bioses (NV3x) have lookup tables of precomputed MNP
+	 * values, but we're too lazy to use those atm
+	 *
+	 * "clk" parameter in kHz
+	 * returns calculated clock
+	 */
+	int cv = nouveau_bios(clock)->version.chip;
+	int minvco = info->vco1.min_freq, maxvco = info->vco1.max_freq;
+	int minM = info->vco1.min_m, maxM = info->vco1.max_m;
+	int minN = info->vco1.min_n, maxN = info->vco1.max_n;
+	int minU = info->vco1.min_inputfreq;
+	int maxU = info->vco1.max_inputfreq;
+	int minP = info->min_p;
+	int maxP = info->max_p_usable;
+	int crystal = info->refclk;
+	int M, N, thisP, P;
+	int clkP, calcclk;
+	int delta, bestdelta = INT_MAX;
+	int bestclk = 0;
+
+	/* this division verified for nv20, nv18, nv28 (Haiku), and nv34 */
+	/* possibly correlated with introduction of 27MHz crystal */
+	if (cv < 0x17 || cv == 0x1a || cv == 0x20) {
+		if (clk > 250000)
+			maxM = 6;
+		if (clk > 340000)
+			maxM = 2;
+	} else if (cv < 0x40) {
+		if (clk > 150000)
+			maxM = 6;
+		if (clk > 200000)
+			maxM = 4;
+		if (clk > 340000)
+			maxM = 2;
+	}
+
+	P = 1 << maxP;
+	if ((clk * P) < minvco) {
+		minvco = clk * maxP;
+		maxvco = minvco * 2;
+	}
+
+	if (clk + clk/200 > maxvco)	/* +0.5% */
+		maxvco = clk + clk/200;
+
+	/* NV34 goes maxlog2P->0, NV20 goes 0->maxlog2P */
+	for (thisP = minP; thisP <= maxP; thisP++) {
+		P = 1 << thisP;
+		clkP = clk * P;
+
+		if (clkP < minvco)
+			continue;
+		if (clkP > maxvco)
+			return bestclk;
+
+		for (M = minM; M <= maxM; M++) {
+			if (crystal/M < minU)
+				return bestclk;
+			if (crystal/M > maxU)
+				continue;
+
+			/* add crystal/2 to round better */
+			N = (clkP * M + crystal/2) / crystal;
+
+			if (N < minN)
+				continue;
+			if (N > maxN)
+				break;
+
+			/* more rounding additions */
+			calcclk = ((N * crystal + P/2) / P + M/2) / M;
+			delta = abs(calcclk - clk);
+			/* we do an exhaustive search rather than terminating
+			 * on an optimality condition...
+			 */
+			if (delta < bestdelta) {
+				bestdelta = delta;
+				bestclk = calcclk;
+				*pN = N;
+				*pM = M;
+				*pP = thisP;
+				if (delta == 0)	/* except this one */
+					return bestclk;
+			}
+		}
+	}
+
+	return bestclk;
+}
+
+static int
+getMNP_double(struct nouveau_clock *clock, struct nvbios_pll *info, int clk,
+	      int *pN1, int *pM1, int *pN2, int *pM2, int *pP)
+{
+	/* Find M, N and P for a two stage PLL
+	 *
+	 * Note that some bioses (NV30+) have lookup tables of precomputed MNP
+	 * values, but we're too lazy to use those atm
+	 *
+	 * "clk" parameter in kHz
+	 * returns calculated clock
+	 */
+	int chip_version = nouveau_bios(clock)->version.chip;
+	int minvco1 = info->vco1.min_freq, maxvco1 = info->vco1.max_freq;
+	int minvco2 = info->vco2.min_freq, maxvco2 = info->vco2.max_freq;
+	int minU1 = info->vco1.min_inputfreq, minU2 = info->vco2.min_inputfreq;
+	int maxU1 = info->vco1.max_inputfreq, maxU2 = info->vco2.max_inputfreq;
+	int minM1 = info->vco1.min_m, maxM1 = info->vco1.max_m;
+	int minN1 = info->vco1.min_n, maxN1 = info->vco1.max_n;
+	int minM2 = info->vco2.min_m, maxM2 = info->vco2.max_m;
+	int minN2 = info->vco2.min_n, maxN2 = info->vco2.max_n;
+	int maxlog2P = info->max_p_usable;
+	int crystal = info->refclk;
+	bool fixedgain2 = (minM2 == maxM2 && minN2 == maxN2);
+	int M1, N1, M2, N2, log2P;
+	int clkP, calcclk1, calcclk2, calcclkout;
+	int delta, bestdelta = INT_MAX;
+	int bestclk = 0;
+
+	int vco2 = (maxvco2 - maxvco2/200) / 2;
+	for (log2P = 0; clk && log2P < maxlog2P && clk <= (vco2 >> log2P); log2P++)
+		;
+	clkP = clk << log2P;
+
+	if (maxvco2 < clk + clk/200)	/* +0.5% */
+		maxvco2 = clk + clk/200;
+
+	for (M1 = minM1; M1 <= maxM1; M1++) {
+		if (crystal/M1 < minU1)
+			return bestclk;
+		if (crystal/M1 > maxU1)
+			continue;
+
+		for (N1 = minN1; N1 <= maxN1; N1++) {
+			calcclk1 = crystal * N1 / M1;
+			if (calcclk1 < minvco1)
+				continue;
+			if (calcclk1 > maxvco1)
+				break;
+
+			for (M2 = minM2; M2 <= maxM2; M2++) {
+				if (calcclk1/M2 < minU2)
+					break;
+				if (calcclk1/M2 > maxU2)
+					continue;
+
+				/* add calcclk1/2 to round better */
+				N2 = (clkP * M2 + calcclk1/2) / calcclk1;
+				if (N2 < minN2)
+					continue;
+				if (N2 > maxN2)
+					break;
+
+				if (!fixedgain2) {
+					if (chip_version < 0x60)
+						if (N2/M2 < 4 || N2/M2 > 10)
+							continue;
+
+					calcclk2 = calcclk1 * N2 / M2;
+					if (calcclk2 < minvco2)
+						break;
+					if (calcclk2 > maxvco2)
+						continue;
+				} else
+					calcclk2 = calcclk1;
+
+				calcclkout = calcclk2 >> log2P;
+				delta = abs(calcclkout - clk);
+				/* we do an exhaustive search rather than terminating
+				 * on an optimality condition...
+				 */
+				if (delta < bestdelta) {
+					bestdelta = delta;
+					bestclk = calcclkout;
+					*pN1 = N1;
+					*pM1 = M1;
+					*pN2 = N2;
+					*pM2 = M2;
+					*pP = log2P;
+					if (delta == 0)	/* except this one */
+						return bestclk;
+				}
+			}
+		}
+	}
+
+	return bestclk;
+}
+
+int
+nv04_pll_calc(struct nouveau_clock *clk, struct nvbios_pll *info, u32 freq,
+	      int *N1, int *M1, int *N2, int *M2, int *P)
+{
+	int ret;
+
+	if (!info->vco2.max_freq) {
+		ret = getMNP_single(clk, info, freq, N1, M1, P);
+		*N2 = 1;
+		*M2 = 1;
+	} else {
+		ret = getMNP_double(clk, info, freq, N1, M1, N2, M2, P);
+	}
+
+	if (!ret)
+		nv_error(clk, "unable to compute acceptable pll values\n");
+	return ret;
+}
diff --git a/nvkm/subdev/clock/pllnva3.c b/nvkm/subdev/clock/pllnva3.c
new file mode 100644
index 000000000..eed5c16cf
--- /dev/null
+++ b/nvkm/subdev/clock/pllnva3.c
@@ -0,0 +1,80 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/clock.h>
+#include <subdev/bios.h>
+#include <subdev/bios/pll.h>
+
+#include "pll.h"
+
+int
+nva3_pll_calc(struct nouveau_clock *clock, struct nvbios_pll *info,
+	      u32 freq, int *pN, int *pfN, int *pM, int *P)
+{
+	u32 best_err = ~0, err;
+	int M, lM, hM, N, fN;
+
+	*P = info->vco1.max_freq / freq;
+	if (*P > info->max_p)
+		*P = info->max_p;
+	if (*P < info->min_p)
+		*P = info->min_p;
+
+	lM = (info->refclk + info->vco1.max_inputfreq) / info->vco1.max_inputfreq;
+	lM = max(lM, (int)info->vco1.min_m);
+	hM = (info->refclk + info->vco1.min_inputfreq) / info->vco1.min_inputfreq;
+	hM = min(hM, (int)info->vco1.max_m);
+
+	for (M = lM; M <= hM; M++) {
+		u32 tmp = freq * *P * M;
+		N  = tmp / info->refclk;
+		fN = tmp % info->refclk;
+		if (!pfN && fN >= info->refclk / 2)
+			N++;
+
+		if (N < info->vco1.min_n)
+			continue;
+		if (N > info->vco1.max_n)
+			break;
+
+		err = abs(freq - (info->refclk * N / M / *P));
+		if (err < best_err) {
+			best_err = err;
+			*pN = N;
+			*pM = M;
+		}
+
+		if (pfN) {
+			*pfN = (((fN << 13) / info->refclk) - 4096) & 0xffff;
+			return freq;
+		}
+	}
+
+	if (unlikely(best_err == ~0)) {
+		nv_error(clock, "unable to find matching pll values\n");
+		return -EINVAL;
+	}
+
+	return info->refclk * *pN / *pM / *P;
+}
diff --git a/nvkm/subdev/device/Makefile.am b/nvkm/subdev/device/Makefile.am
new file mode 100644
index 000000000..5a687f22c
--- /dev/null
+++ b/nvkm/subdev/device/Makefile.am
@@ -0,0 +1,13 @@
+noinst_LTLIBRARIES = libdevice.la
+
+libdevice_la_SOURCES = base.c \
+		       nv04.c \
+		       nv10.c \
+		       nv20.c \
+		       nv30.c \
+		       nv40.c \
+		       nv50.c \
+		       nvc0.c \
+		       nve0.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/device/base.c b/nvkm/subdev/device/base.c
new file mode 100644
index 000000000..f8a7ed416
--- /dev/null
+++ b/nvkm/subdev/device/base.c
@@ -0,0 +1,476 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/device.h>
+#include <core/client.h>
+#include <core/option.h>
+
+#include <core/class.h>
+
+#include <subdev/device.h>
+
+static DEFINE_MUTEX(nv_devices_mutex);
+static LIST_HEAD(nv_devices);
+
+struct nouveau_device *
+nouveau_device_find(u64 name)
+{
+	struct nouveau_device *device, *match = NULL;
+	mutex_lock(&nv_devices_mutex);
+	list_for_each_entry(device, &nv_devices, head) {
+		if (device->handle == name) {
+			match = device;
+			break;
+		}
+	}
+	mutex_unlock(&nv_devices_mutex);
+	return match;
+}
+
+/******************************************************************************
+ * nouveau_devobj (0x0080): class implementation
+ *****************************************************************************/
+struct nouveau_devobj {
+	struct nouveau_parent base;
+	struct nouveau_object *subdev[NVDEV_SUBDEV_NR];
+	bool created;
+};
+
+static const u64 disable_map[] = {
+	[NVDEV_SUBDEV_VBIOS]	= NV_DEVICE_DISABLE_VBIOS,
+	[NVDEV_SUBDEV_DEVINIT]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_GPIO]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_I2C]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_CLOCK]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_MXM]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_MC]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_TIMER]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_FB]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_LTCG]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_IBUS]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_INSTMEM]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_VM]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_BAR]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_VOLT]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_SUBDEV_THERM]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_ENGINE_DMAOBJ]	= NV_DEVICE_DISABLE_CORE,
+	[NVDEV_ENGINE_FIFO]	= NV_DEVICE_DISABLE_FIFO,
+	[NVDEV_ENGINE_SW]	= NV_DEVICE_DISABLE_FIFO,
+	[NVDEV_ENGINE_GR]	= NV_DEVICE_DISABLE_GRAPH,
+	[NVDEV_ENGINE_MPEG]	= NV_DEVICE_DISABLE_MPEG,
+	[NVDEV_ENGINE_ME]	= NV_DEVICE_DISABLE_ME,
+	[NVDEV_ENGINE_VP]	= NV_DEVICE_DISABLE_VP,
+	[NVDEV_ENGINE_CRYPT]	= NV_DEVICE_DISABLE_CRYPT,
+	[NVDEV_ENGINE_BSP]	= NV_DEVICE_DISABLE_BSP,
+	[NVDEV_ENGINE_PPP]	= NV_DEVICE_DISABLE_PPP,
+	[NVDEV_ENGINE_COPY0]	= NV_DEVICE_DISABLE_COPY0,
+	[NVDEV_ENGINE_COPY1]	= NV_DEVICE_DISABLE_COPY1,
+	[NVDEV_ENGINE_UNK1C1]	= NV_DEVICE_DISABLE_UNK1C1,
+	[NVDEV_ENGINE_VENC]	= NV_DEVICE_DISABLE_VENC,
+	[NVDEV_ENGINE_DISP]	= NV_DEVICE_DISABLE_DISP,
+	[NVDEV_SUBDEV_NR]	= 0,
+};
+
+static int
+nouveau_devobj_ctor(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, void *data, u32 size,
+		    struct nouveau_object **pobject)
+{
+	struct nouveau_client *client = nv_client(parent);
+	struct nouveau_device *device;
+	struct nouveau_devobj *devobj;
+	struct nv_device_class *args = data;
+	u64 disable, boot0, strap;
+	u64 mmio_base, mmio_size;
+	void __iomem *map;
+	int ret, i, c;
+
+	if (size < sizeof(struct nv_device_class))
+		return -EINVAL;
+
+	/* find the device subdev that matches what the client requested */
+	device = nv_device(client->device);
+	if (args->device != ~0) {
+		device = nouveau_device_find(args->device);
+		if (!device)
+			return -ENODEV;
+	}
+
+	ret = nouveau_parent_create(parent, nv_object(device), oclass, 0, NULL,
+				    (1ULL << NVDEV_ENGINE_DMAOBJ) |
+				    (1ULL << NVDEV_ENGINE_FIFO) |
+				    (1ULL << NVDEV_ENGINE_DISP), &devobj);
+	*pobject = nv_object(devobj);
+	if (ret)
+		return ret;
+
+	mmio_base = pci_resource_start(device->pdev, 0);
+	mmio_size = pci_resource_len(device->pdev, 0);
+
+	/* translate api disable mask into internal mapping */
+	disable = args->debug0;
+	for (i = 0; i < NVDEV_SUBDEV_NR; i++) {
+		if (args->disable & disable_map[i])
+			disable |= (1ULL << i);
+	}
+
+	/* identify the chipset, and determine classes of subdev/engines */
+	if (!(args->disable & NV_DEVICE_DISABLE_IDENTIFY) &&
+	    !device->card_type) {
+		map = ioremap(mmio_base, 0x102000);
+		if (map == NULL)
+			return -ENOMEM;
+
+		/* switch mmio to cpu's native endianness */
+#ifndef __BIG_ENDIAN
+		if (ioread32_native(map + 0x000004) != 0x00000000)
+#else
+		if (ioread32_native(map + 0x000004) == 0x00000000)
+#endif
+			iowrite32_native(0x01000001, map + 0x000004);
+
+		/* read boot0 and strapping information */
+		boot0 = ioread32_native(map + 0x000000);
+		strap = ioread32_native(map + 0x101000);
+		iounmap(map);
+
+		/* determine chipset and derive architecture from it */
+		if ((boot0 & 0x0f000000) > 0) {
+			device->chipset = (boot0 & 0xff00000) >> 20;
+			switch (device->chipset & 0xf0) {
+			case 0x10: device->card_type = NV_10; break;
+			case 0x20: device->card_type = NV_20; break;
+			case 0x30: device->card_type = NV_30; break;
+			case 0x40:
+			case 0x60: device->card_type = NV_40; break;
+			case 0x50:
+			case 0x80:
+			case 0x90:
+			case 0xa0: device->card_type = NV_50; break;
+			case 0xc0: device->card_type = NV_C0; break;
+			case 0xd0: device->card_type = NV_D0; break;
+			case 0xe0: device->card_type = NV_E0; break;
+			default:
+				break;
+			}
+		} else
+		if ((boot0 & 0xff00fff0) == 0x20004000) {
+			if (boot0 & 0x00f00000)
+				device->chipset = 0x05;
+			else
+				device->chipset = 0x04;
+			device->card_type = NV_04;
+		}
+
+		switch (device->card_type) {
+		case NV_04: ret = nv04_identify(device); break;
+		case NV_10: ret = nv10_identify(device); break;
+		case NV_20: ret = nv20_identify(device); break;
+		case NV_30: ret = nv30_identify(device); break;
+		case NV_40: ret = nv40_identify(device); break;
+		case NV_50: ret = nv50_identify(device); break;
+		case NV_C0:
+		case NV_D0: ret = nvc0_identify(device); break;
+		case NV_E0: ret = nve0_identify(device); break;
+		default:
+			ret = -EINVAL;
+			break;
+		}
+
+		if (ret) {
+			nv_error(device, "unknown chipset, 0x%08x\n", boot0);
+			return ret;
+		}
+
+		nv_info(device, "BOOT0  : 0x%08x\n", boot0);
+		nv_info(device, "Chipset: %s (NV%02X)\n",
+			device->cname, device->chipset);
+		nv_info(device, "Family : NV%02X\n", device->card_type);
+
+		/* determine frequency of timing crystal */
+		if ( device->chipset < 0x17 ||
+		    (device->chipset >= 0x20 && device->chipset < 0x25))
+			strap &= 0x00000040;
+		else
+			strap &= 0x00400040;
+
+		switch (strap) {
+		case 0x00000000: device->crystal = 13500; break;
+		case 0x00000040: device->crystal = 14318; break;
+		case 0x00400000: device->crystal = 27000; break;
+		case 0x00400040: device->crystal = 25000; break;
+		}
+
+		nv_debug(device, "crystal freq: %dKHz\n", device->crystal);
+	}
+
+	if (!(args->disable & NV_DEVICE_DISABLE_MMIO) &&
+	    !nv_subdev(device)->mmio) {
+		nv_subdev(device)->mmio  = ioremap(mmio_base, mmio_size);
+		if (!nv_subdev(device)->mmio) {
+			nv_error(device, "unable to map device registers\n");
+			return -ENOMEM;
+		}
+	}
+
+	/* ensure requested subsystems are available for use */
+	for (i = 0, c = 0; i < NVDEV_SUBDEV_NR; i++) {
+		if (!(oclass = device->oclass[i]) || (disable & (1ULL << i)))
+			continue;
+
+		if (!device->subdev[i]) {
+			ret = nouveau_object_ctor(nv_object(device), NULL,
+						  oclass, NULL, i,
+						  &devobj->subdev[i]);
+			if (ret == -ENODEV)
+				continue;
+			if (ret)
+				return ret;
+
+			if (nv_iclass(devobj->subdev[i], NV_ENGINE_CLASS))
+				nouveau_subdev_reset(devobj->subdev[i]);
+		} else {
+			nouveau_object_ref(device->subdev[i],
+					  &devobj->subdev[i]);
+		}
+
+		/* note: can't init *any* subdevs until devinit has been run
+		 * due to not knowing exactly what the vbios init tables will
+		 * mess with.  devinit also can't be run until all of its
+		 * dependencies have been created.
+		 *
+		 * this code delays init of any subdev until all of devinit's
+		 * dependencies have been created, and then initialises each
+		 * subdev in turn as they're created.
+		 */
+		while (i >= NVDEV_SUBDEV_DEVINIT_LAST && c <= i) {
+			struct nouveau_object *subdev = devobj->subdev[c++];
+			if (subdev && !nv_iclass(subdev, NV_ENGINE_CLASS)) {
+				ret = nouveau_object_inc(subdev);
+				if (ret)
+					return ret;
+			}
+		}
+	}
+
+	return 0;
+}
+
+static void
+nouveau_devobj_dtor(struct nouveau_object *object)
+{
+	struct nouveau_devobj *devobj = (void *)object;
+	int i;
+
+	for (i = NVDEV_SUBDEV_NR - 1; i >= 0; i--)
+		nouveau_object_ref(NULL, &devobj->subdev[i]);
+
+	nouveau_parent_destroy(&devobj->base);
+}
+
+static int
+nouveau_devobj_init(struct nouveau_object *object)
+{
+	struct nouveau_devobj *devobj = (void *)object;
+	struct nouveau_object *subdev;
+	int ret, i;
+
+	ret = nouveau_parent_init(&devobj->base);
+	if (ret)
+		return ret;
+
+	for (i = 0; devobj->created && i < NVDEV_SUBDEV_NR; i++) {
+		if ((subdev = devobj->subdev[i])) {
+			if (!nv_iclass(subdev, NV_ENGINE_CLASS)) {
+				ret = nouveau_object_inc(subdev);
+				if (ret)
+					goto fail;
+			}
+		}
+	}
+
+	devobj->created = true;
+	return 0;
+
+fail:
+	for (--i; i >= 0; i--) {
+		if ((subdev = devobj->subdev[i])) {
+			if (!nv_iclass(subdev, NV_ENGINE_CLASS))
+				nouveau_object_dec(subdev, false);
+		}
+	}
+
+	return ret;
+}
+
+static int
+nouveau_devobj_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nouveau_devobj *devobj = (void *)object;
+	struct nouveau_object *subdev;
+	int ret, i;
+
+	for (i = NVDEV_SUBDEV_NR - 1; i >= 0; i--) {
+		if ((subdev = devobj->subdev[i])) {
+			if (!nv_iclass(subdev, NV_ENGINE_CLASS)) {
+				ret = nouveau_object_dec(subdev, suspend);
+				if (ret && suspend)
+					goto fail;
+			}
+		}
+	}
+
+	ret = nouveau_parent_fini(&devobj->base, suspend);
+fail:
+	for (; ret && suspend && i < NVDEV_SUBDEV_NR; i++) {
+		if ((subdev = devobj->subdev[i])) {
+			if (!nv_iclass(subdev, NV_ENGINE_CLASS)) {
+				ret = nouveau_object_inc(subdev);
+				if (ret) {
+					/* XXX */
+				}
+			}
+		}
+	}
+
+	return ret;
+}
+
+static u8
+nouveau_devobj_rd08(struct nouveau_object *object, u64 addr)
+{
+	return nv_rd08(object->engine, addr);
+}
+
+static u16
+nouveau_devobj_rd16(struct nouveau_object *object, u64 addr)
+{
+	return nv_rd16(object->engine, addr);
+}
+
+static u32
+nouveau_devobj_rd32(struct nouveau_object *object, u64 addr)
+{
+	return nv_rd32(object->engine, addr);
+}
+
+static void
+nouveau_devobj_wr08(struct nouveau_object *object, u64 addr, u8 data)
+{
+	nv_wr08(object->engine, addr, data);
+}
+
+static void
+nouveau_devobj_wr16(struct nouveau_object *object, u64 addr, u16 data)
+{
+	nv_wr16(object->engine, addr, data);
+}
+
+static void
+nouveau_devobj_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	nv_wr32(object->engine, addr, data);
+}
+
+static struct nouveau_ofuncs
+nouveau_devobj_ofuncs = {
+	.ctor = nouveau_devobj_ctor,
+	.dtor = nouveau_devobj_dtor,
+	.init = nouveau_devobj_init,
+	.fini = nouveau_devobj_fini,
+	.rd08 = nouveau_devobj_rd08,
+	.rd16 = nouveau_devobj_rd16,
+	.rd32 = nouveau_devobj_rd32,
+	.wr08 = nouveau_devobj_wr08,
+	.wr16 = nouveau_devobj_wr16,
+	.wr32 = nouveau_devobj_wr32,
+};
+
+/******************************************************************************
+ * nouveau_device: engine functions
+ *****************************************************************************/
+struct nouveau_oclass
+nouveau_device_sclass[] = {
+	{ 0x0080, &nouveau_devobj_ofuncs },
+	{}
+};
+
+static void
+nouveau_device_dtor(struct nouveau_object *object)
+{
+	struct nouveau_device *device = (void *)object;
+
+	mutex_lock(&nv_devices_mutex);
+	list_del(&device->head);
+	mutex_unlock(&nv_devices_mutex);
+
+	if (device->base.mmio)
+		iounmap(device->base.mmio);
+
+	nouveau_subdev_destroy(&device->base);
+}
+
+static struct nouveau_oclass
+nouveau_device_oclass = {
+	.handle = NV_SUBDEV(DEVICE, 0x00),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.dtor = nouveau_device_dtor,
+	},
+};
+
+int
+nouveau_device_create_(struct pci_dev *pdev, u64 name, const char *sname,
+		       const char *cfg, const char *dbg,
+		       int length, void **pobject)
+{
+	struct nouveau_device *device;
+	int ret = -EEXIST;
+
+	mutex_lock(&nv_devices_mutex);
+	list_for_each_entry(device, &nv_devices, head) {
+		if (device->handle == name)
+			goto done;
+	}
+
+	ret = nouveau_subdev_create_(NULL, NULL, &nouveau_device_oclass, 0,
+				     "DEVICE", "device", length, pobject);
+	device = *pobject;
+	if (ret)
+		goto done;
+
+	atomic_set(&nv_object(device)->usecount, 2);
+	device->pdev = pdev;
+	device->handle = name;
+	device->cfgopt = cfg;
+	device->dbgopt = dbg;
+	device->name = sname;
+
+	nv_subdev(device)->debug = nouveau_dbgopt(device->dbgopt, "DEVICE");
+	list_add(&device->head, &nv_devices);
+done:
+	mutex_unlock(&nv_devices_mutex);
+	return ret;
+}
diff --git a/nvkm/subdev/device/nv04.c b/nvkm/subdev/device/nv04.c
new file mode 100644
index 000000000..8626d0d6c
--- /dev/null
+++ b/nvkm/subdev/device/nv04.c
@@ -0,0 +1,86 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/device.h>
+#include <subdev/bios.h>
+#include <subdev/i2c.h>
+#include <subdev/clock.h>
+#include <subdev/devinit.h>
+#include <subdev/mc.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/instmem.h>
+#include <subdev/vm.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+#include <engine/software.h>
+#include <engine/graph.h>
+#include <engine/disp.h>
+
+int
+nv04_identify(struct nouveau_device *device)
+{
+	switch (device->chipset) {
+	case 0x04:
+		device->cname = "NV04";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv04_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv04_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv04_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv04_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv04_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x05:
+		device->cname = "NV05";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv05_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv04_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv04_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv04_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv04_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	default:
+		nv_fatal(device, "unknown RIVA chipset\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/device/nv10.c b/nvkm/subdev/device/nv10.c
new file mode 100644
index 000000000..9c40b0fb2
--- /dev/null
+++ b/nvkm/subdev/device/nv10.c
@@ -0,0 +1,195 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/device.h>
+#include <subdev/bios.h>
+#include <subdev/gpio.h>
+#include <subdev/i2c.h>
+#include <subdev/clock.h>
+#include <subdev/devinit.h>
+#include <subdev/mc.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/instmem.h>
+#include <subdev/vm.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+#include <engine/software.h>
+#include <engine/graph.h>
+#include <engine/disp.h>
+
+int
+nv10_identify(struct nouveau_device *device)
+{
+	switch (device->chipset) {
+	case 0x10:
+		device->cname = "NV10";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv10_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv10_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv10_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x15:
+		device->cname = "NV15";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv10_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv10_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv10_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv10_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x16:
+		device->cname = "NV16";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv10_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv10_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv10_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv10_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x1a:
+		device->cname = "nForce";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv1a_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv10_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv10_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x11:
+		device->cname = "NV11";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv10_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv10_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv10_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv10_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x17:
+		device->cname = "NV17";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv10_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv10_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv10_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x1f:
+		device->cname = "nForce2";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv1a_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv10_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x18:
+		device->cname = "NV18";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv10_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv10_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv10_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	default:
+		nv_fatal(device, "unknown Celsius chipset\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/device/nv20.c b/nvkm/subdev/device/nv20.c
new file mode 100644
index 000000000..74f88f48e
--- /dev/null
+++ b/nvkm/subdev/device/nv20.c
@@ -0,0 +1,126 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/device.h>
+#include <subdev/bios.h>
+#include <subdev/gpio.h>
+#include <subdev/i2c.h>
+#include <subdev/clock.h>
+#include <subdev/therm.h>
+#include <subdev/devinit.h>
+#include <subdev/mc.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/instmem.h>
+#include <subdev/vm.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+#include <engine/software.h>
+#include <engine/graph.h>
+#include <engine/disp.h>
+
+int
+nv20_identify(struct nouveau_device *device)
+{
+	switch (device->chipset) {
+	case 0x20:
+		device->cname = "NV20";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv20_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv20_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv20_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x25:
+		device->cname = "NV25";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv20_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv25_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv25_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x28:
+		device->cname = "NV28";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv20_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv25_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv25_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x2a:
+		device->cname = "NV2A";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv20_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv25_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv2a_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	default:
+		nv_fatal(device, "unknown Kelvin chipset\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/device/nv30.c b/nvkm/subdev/device/nv30.c
new file mode 100644
index 000000000..0ac1b2c4f
--- /dev/null
+++ b/nvkm/subdev/device/nv30.c
@@ -0,0 +1,147 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/device.h>
+#include <subdev/bios.h>
+#include <subdev/gpio.h>
+#include <subdev/i2c.h>
+#include <subdev/clock.h>
+#include <subdev/devinit.h>
+#include <subdev/mc.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/instmem.h>
+#include <subdev/vm.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+#include <engine/software.h>
+#include <engine/graph.h>
+#include <engine/mpeg.h>
+#include <engine/disp.h>
+
+int
+nv30_identify(struct nouveau_device *device)
+{
+	switch (device->chipset) {
+	case 0x30:
+		device->cname = "NV30";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv20_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv30_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv30_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x35:
+		device->cname = "NV35";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv20_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv35_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv35_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x31:
+		device->cname = "NV31";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv20_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv30_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv30_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv31_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x36:
+		device->cname = "NV36";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv20_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv36_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv35_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv31_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x34:
+		device->cname = "NV34";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv04_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv10_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv10_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv04_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv17_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv34_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv31_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	default:
+		nv_fatal(device, "unknown Rankine chipset\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/device/nv40.c b/nvkm/subdev/device/nv40.c
new file mode 100644
index 000000000..41d59689a
--- /dev/null
+++ b/nvkm/subdev/device/nv40.c
@@ -0,0 +1,375 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/device.h>
+#include <subdev/bios.h>
+#include <subdev/gpio.h>
+#include <subdev/i2c.h>
+#include <subdev/clock.h>
+#include <subdev/therm.h>
+#include <subdev/devinit.h>
+#include <subdev/mc.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/instmem.h>
+#include <subdev/vm.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+#include <engine/software.h>
+#include <engine/graph.h>
+#include <engine/mpeg.h>
+#include <engine/disp.h>
+
+int
+nv40_identify(struct nouveau_device *device)
+{
+	switch (device->chipset) {
+	case 0x40:
+		device->cname = "NV40";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv40_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x41:
+		device->cname = "NV41";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv41_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv41_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x42:
+		device->cname = "NV42";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv41_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv41_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x43:
+		device->cname = "NV43";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv41_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv41_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x45:
+		device->cname = "NV45";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv40_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv04_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x47:
+		device->cname = "G70";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv47_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv41_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x49:
+		device->cname = "G71";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv49_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv41_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x4b:
+		device->cname = "G73";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv04_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv49_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv41_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x44:
+		device->cname = "NV44";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv44_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv44_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv44_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x46:
+		device->cname = "G72";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv44_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv46_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv44_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x4a:
+		device->cname = "NV44A";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv44_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv44_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv44_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x4c:
+		device->cname = "C61";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv44_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv46_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv44_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x4e:
+		device->cname = "C51";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv44_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv4e_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv44_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x63:
+		device->cname = "C73";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv44_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv46_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv44_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x67:
+		device->cname = "C67";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv44_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv46_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv44_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	case 0x68:
+		device->cname = "C68";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv10_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv40_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv40_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv1a_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv44_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv46_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv40_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv44_vmmgr_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv04_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv40_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv10_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv40_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv40_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv04_disp_oclass;
+		break;
+	default:
+		nv_fatal(device, "unknown Curie chipset\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/device/nv50.c b/nvkm/subdev/device/nv50.c
new file mode 100644
index 000000000..6ccfd8585
--- /dev/null
+++ b/nvkm/subdev/device/nv50.c
@@ -0,0 +1,410 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/device.h>
+#include <subdev/bios.h>
+#include <subdev/gpio.h>
+#include <subdev/i2c.h>
+#include <subdev/clock.h>
+#include <subdev/therm.h>
+#include <subdev/mxm.h>
+#include <subdev/devinit.h>
+#include <subdev/mc.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/instmem.h>
+#include <subdev/vm.h>
+#include <subdev/bar.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+#include <engine/software.h>
+#include <engine/graph.h>
+#include <engine/mpeg.h>
+#include <engine/vp.h>
+#include <engine/crypt.h>
+#include <engine/bsp.h>
+#include <engine/ppp.h>
+#include <engine/copy.h>
+#include <engine/disp.h>
+
+int
+nv50_identify(struct nouveau_device *device)
+{
+	switch (device->chipset) {
+	case 0x50:
+		device->cname = "G80";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv50_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv50_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv50_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv50_disp_oclass;
+		break;
+	case 0x84:
+		device->cname = "G84";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv50_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv84_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_CRYPT  ] = &nv84_crypt_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv84_disp_oclass;
+		break;
+	case 0x86:
+		device->cname = "G86";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv50_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv84_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_CRYPT  ] = &nv84_crypt_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv84_disp_oclass;
+		break;
+	case 0x92:
+		device->cname = "G92";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv50_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv84_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_CRYPT  ] = &nv84_crypt_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv84_disp_oclass;
+		break;
+	case 0x94:
+		device->cname = "G94";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv50_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv84_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_CRYPT  ] = &nv84_crypt_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv94_disp_oclass;
+		break;
+	case 0x96:
+		device->cname = "G96";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv50_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv84_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_CRYPT  ] = &nv84_crypt_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv94_disp_oclass;
+		break;
+	case 0x98:
+		device->cname = "G98";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv98_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_CRYPT  ] = &nv98_crypt_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nv98_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv94_disp_oclass;
+		break;
+	case 0xa0:
+		device->cname = "G200";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv98_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv84_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_CRYPT  ] = &nv84_crypt_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva0_disp_oclass;
+		break;
+	case 0xaa:
+		device->cname = "MCP77/MCP78";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv98_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_CRYPT  ] = &nv98_crypt_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nv98_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv94_disp_oclass;
+		break;
+	case 0xac:
+		device->cname = "MCP79/MCP7A";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nv50_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv98_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_CRYPT  ] = &nv98_crypt_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nv98_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nv94_disp_oclass;
+		break;
+	case 0xa3:
+		device->cname = "GT215";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nva3_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv98_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_MPEG   ] = &nv84_mpeg_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nv98_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nva3_copy_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xa5:
+		device->cname = "GT216";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nva3_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv98_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nv98_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nva3_copy_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xa8:
+		device->cname = "GT218";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nva3_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv98_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nv98_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nva3_copy_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xaf:
+		device->cname = "MCP89";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nva3_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nv98_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nv50_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nv50_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nv50_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nv50_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nv84_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nv50_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nv50_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nv84_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nv84_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nv98_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nva3_copy_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	default:
+		nv_fatal(device, "unknown Tesla chipset\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/device/nvc0.c b/nvkm/subdev/device/nvc0.c
new file mode 100644
index 000000000..f0461685a
--- /dev/null
+++ b/nvkm/subdev/device/nvc0.c
@@ -0,0 +1,285 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/device.h>
+#include <subdev/bios.h>
+#include <subdev/gpio.h>
+#include <subdev/i2c.h>
+#include <subdev/clock.h>
+#include <subdev/therm.h>
+#include <subdev/mxm.h>
+#include <subdev/devinit.h>
+#include <subdev/mc.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/ltcg.h>
+#include <subdev/ibus.h>
+#include <subdev/instmem.h>
+#include <subdev/vm.h>
+#include <subdev/bar.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+#include <engine/software.h>
+#include <engine/graph.h>
+#include <engine/vp.h>
+#include <engine/bsp.h>
+#include <engine/ppp.h>
+#include <engine/copy.h>
+#include <engine/disp.h>
+
+int
+nvc0_identify(struct nouveau_device *device)
+{
+	switch (device->chipset) {
+	case 0xc0:
+		device->cname = "GF100";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nvc0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvc0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nvc0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nvc0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nvc0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nvc0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nvc0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nvc0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xc4:
+		device->cname = "GF104";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nvc0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvc0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nvc0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nvc0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nvc0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nvc0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nvc0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nvc0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xc3:
+		device->cname = "GF106";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nvc0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvc0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nvc0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nvc0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nvc0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nvc0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nvc0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nvc0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xce:
+		device->cname = "GF114";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nvc0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvc0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nvc0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nvc0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nvc0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nvc0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nvc0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nvc0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xcf:
+		device->cname = "GF116";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nvc0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvc0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nvc0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nvc0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nvc0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nvc0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nvc0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nvc0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xc1:
+		device->cname = "GF108";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nvc0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvc0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nvc0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nvc0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nvc0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nvc0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nvc0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nvc0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xc8:
+		device->cname = "GF110";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nv50_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nvc0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvc0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nvc0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nvc0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nvc0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nvc0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nvc0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nvc0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nva3_disp_oclass;
+		break;
+	case 0xd9:
+		device->cname = "GF119";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nvd0_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nvc0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvd0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nvc0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nvc0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nvc0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nvc0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nvc0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nvd0_disp_oclass;
+		break;
+	default:
+		nv_fatal(device, "unknown Fermi chipset\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/device/nve0.c b/nvkm/subdev/device/nve0.c
new file mode 100644
index 000000000..03a652876
--- /dev/null
+++ b/nvkm/subdev/device/nve0.c
@@ -0,0 +1,146 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/device.h>
+#include <subdev/bios.h>
+#include <subdev/gpio.h>
+#include <subdev/i2c.h>
+#include <subdev/clock.h>
+#include <subdev/therm.h>
+#include <subdev/mxm.h>
+#include <subdev/devinit.h>
+#include <subdev/mc.h>
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/ltcg.h>
+#include <subdev/ibus.h>
+#include <subdev/instmem.h>
+#include <subdev/vm.h>
+#include <subdev/bar.h>
+
+#include <engine/dmaobj.h>
+#include <engine/fifo.h>
+#include <engine/software.h>
+#include <engine/graph.h>
+#include <engine/disp.h>
+#include <engine/copy.h>
+#include <engine/bsp.h>
+#include <engine/vp.h>
+#include <engine/ppp.h>
+
+int
+nve0_identify(struct nouveau_device *device)
+{
+	switch (device->chipset) {
+	case 0xe4:
+		device->cname = "GK104";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nvd0_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nve0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvd0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nve0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nve0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nve0_disp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nve0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nve0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nve0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nve0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		break;
+	case 0xe7:
+		device->cname = "GK107";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nvd0_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nve0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvd0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nve0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nve0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nve0_disp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nve0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nve0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nve0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nve0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		break;
+	case 0xe6:
+		device->cname = "GK106";
+		device->oclass[NVDEV_SUBDEV_VBIOS  ] = &nouveau_bios_oclass;
+		device->oclass[NVDEV_SUBDEV_GPIO   ] = &nvd0_gpio_oclass;
+		device->oclass[NVDEV_SUBDEV_I2C    ] = &nouveau_i2c_oclass;
+		device->oclass[NVDEV_SUBDEV_CLOCK  ] = &nvc0_clock_oclass;
+		device->oclass[NVDEV_SUBDEV_THERM  ] = &nv50_therm_oclass;
+		device->oclass[NVDEV_SUBDEV_MXM    ] = &nv50_mxm_oclass;
+		device->oclass[NVDEV_SUBDEV_DEVINIT] = &nv50_devinit_oclass;
+		device->oclass[NVDEV_SUBDEV_MC     ] = &nvc0_mc_oclass;
+		device->oclass[NVDEV_SUBDEV_TIMER  ] = &nv04_timer_oclass;
+		device->oclass[NVDEV_SUBDEV_FB     ] = &nvc0_fb_oclass;
+		device->oclass[NVDEV_SUBDEV_LTCG   ] = &nvc0_ltcg_oclass;
+		device->oclass[NVDEV_SUBDEV_IBUS   ] = &nve0_ibus_oclass;
+		device->oclass[NVDEV_SUBDEV_INSTMEM] = &nv50_instmem_oclass;
+		device->oclass[NVDEV_SUBDEV_VM     ] = &nvc0_vmmgr_oclass;
+		device->oclass[NVDEV_SUBDEV_BAR    ] = &nvc0_bar_oclass;
+		device->oclass[NVDEV_ENGINE_DMAOBJ ] = &nvd0_dmaeng_oclass;
+		device->oclass[NVDEV_ENGINE_FIFO   ] = &nve0_fifo_oclass;
+		device->oclass[NVDEV_ENGINE_SW     ] = &nvc0_software_oclass;
+		device->oclass[NVDEV_ENGINE_GR     ] = &nve0_graph_oclass;
+		device->oclass[NVDEV_ENGINE_DISP   ] = &nve0_disp_oclass;
+		device->oclass[NVDEV_ENGINE_COPY0  ] = &nve0_copy0_oclass;
+		device->oclass[NVDEV_ENGINE_COPY1  ] = &nve0_copy1_oclass;
+		device->oclass[NVDEV_ENGINE_BSP    ] = &nve0_bsp_oclass;
+		device->oclass[NVDEV_ENGINE_VP     ] = &nve0_vp_oclass;
+		device->oclass[NVDEV_ENGINE_PPP    ] = &nvc0_ppp_oclass;
+		break;
+	default:
+		nv_fatal(device, "unknown Kepler chipset\n");
+		return -EINVAL;
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/devinit/Makefile.am b/nvkm/subdev/devinit/Makefile.am
new file mode 100644
index 000000000..f34cb05fa
--- /dev/null
+++ b/nvkm/subdev/devinit/Makefile.am
@@ -0,0 +1,11 @@
+noinst_LTLIBRARIES = libdevinit.la
+
+libdevinit_la_SOURCES = base.c \
+			nv04.c \
+			nv05.c \
+			nv10.c \
+			nv1a.c \
+			nv20.c \
+			nv50.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/devinit/base.c b/nvkm/subdev/devinit/base.c
new file mode 100644
index 000000000..5a07a39c1
--- /dev/null
+++ b/nvkm/subdev/devinit/base.c
@@ -0,0 +1,69 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/option.h>
+
+#include <subdev/devinit.h>
+#include <subdev/bios.h>
+#include <subdev/bios/init.h>
+
+int
+nouveau_devinit_init(struct nouveau_devinit *devinit)
+{
+	int ret = nouveau_subdev_init(&devinit->base);
+	if (ret)
+		return ret;
+
+	return nvbios_init(&devinit->base, devinit->post);
+}
+
+int
+nouveau_devinit_fini(struct nouveau_devinit *devinit, bool suspend)
+{
+	/* force full reinit on resume */
+	if (suspend)
+		devinit->post = true;
+
+	return nouveau_subdev_fini(&devinit->base, suspend);
+}
+
+int
+nouveau_devinit_create_(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass,
+			int size, void **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nouveau_devinit *devinit;
+	int ret;
+
+	ret = nouveau_subdev_create_(parent, engine, oclass, 0, "DEVINIT",
+				     "init", size, pobject);
+	devinit = *pobject;
+	if (ret)
+		return ret;
+
+	devinit->post = nouveau_boolopt(device->cfgopt, "NvForcePost", false);
+	return 0;
+}
diff --git a/nvkm/subdev/devinit/fbmem.h b/nvkm/subdev/devinit/fbmem.h
new file mode 100644
index 000000000..6b56a0f4c
--- /dev/null
+++ b/nvkm/subdev/devinit/fbmem.h
@@ -0,0 +1,98 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#define NV04_PFB_BOOT_0						0x00100000
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT			0x00000003
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT_32MB			0x00000000
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT_4MB			0x00000001
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT_8MB			0x00000002
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT_16MB			0x00000003
+#	define NV04_PFB_BOOT_0_RAM_WIDTH_128			0x00000004
+#	define NV04_PFB_BOOT_0_RAM_TYPE				0x00000028
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SGRAM_8MBIT		0x00000000
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SGRAM_16MBIT		0x00000008
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SGRAM_16MBIT_4BANK	0x00000010
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SDRAM_16MBIT		0x00000018
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SDRAM_64MBIT		0x00000020
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SDRAM_64MBITX16		0x00000028
+#	define NV04_PFB_BOOT_0_UMA_ENABLE			0x00000100
+#	define NV04_PFB_BOOT_0_UMA_SIZE				0x0000f000
+#define NV04_PFB_DEBUG_0					0x00100080
+#	define NV04_PFB_DEBUG_0_PAGE_MODE			0x00000001
+#	define NV04_PFB_DEBUG_0_REFRESH_OFF			0x00000010
+#	define NV04_PFB_DEBUG_0_REFRESH_COUNTX64		0x00003f00
+#	define NV04_PFB_DEBUG_0_REFRESH_SLOW_CLK		0x00004000
+#	define NV04_PFB_DEBUG_0_SAFE_MODE			0x00008000
+#	define NV04_PFB_DEBUG_0_ALOM_ENABLE			0x00010000
+#	define NV04_PFB_DEBUG_0_CASOE				0x00100000
+#	define NV04_PFB_DEBUG_0_CKE_INVERT			0x10000000
+#	define NV04_PFB_DEBUG_0_REFINC				0x20000000
+#	define NV04_PFB_DEBUG_0_SAVE_POWER_OFF			0x40000000
+#define NV04_PFB_CFG0						0x00100200
+#	define NV04_PFB_CFG0_SCRAMBLE				0x20000000
+#define NV04_PFB_CFG1						0x00100204
+#define NV04_PFB_SCRAMBLE(i)                         (0x00100400 + 4 * (i))
+
+#define NV10_PFB_REFCTRL					0x00100210
+#	define NV10_PFB_REFCTRL_VALID_1				(1 << 31)
+
+static inline struct io_mapping *
+fbmem_init(struct pci_dev *pdev)
+{
+	return io_mapping_create_wc(pci_resource_start(pdev, 1),
+				    pci_resource_len(pdev, 1));
+}
+
+static inline void
+fbmem_fini(struct io_mapping *fb)
+{
+	io_mapping_free(fb);
+}
+
+static inline u32
+fbmem_peek(struct io_mapping *fb, u32 off)
+{
+	u8 __iomem *p = io_mapping_map_atomic_wc(fb, off & PAGE_MASK);
+	u32 val = ioread32(p + (off & ~PAGE_MASK));
+	io_mapping_unmap_atomic(p);
+	return val;
+}
+
+static inline void
+fbmem_poke(struct io_mapping *fb, u32 off, u32 val)
+{
+	u8 __iomem *p = io_mapping_map_atomic_wc(fb, off & PAGE_MASK);
+	iowrite32(val, p + (off & ~PAGE_MASK));
+	wmb();
+	io_mapping_unmap_atomic(p);
+}
+
+static inline bool
+fbmem_readback(struct io_mapping *fb, u32 off, u32 val)
+{
+	fbmem_poke(fb, off, val);
+	return val == fbmem_peek(fb, off);
+}
diff --git a/nvkm/subdev/devinit/nv04.c b/nvkm/subdev/devinit/nv04.c
new file mode 100644
index 000000000..7a72d9394
--- /dev/null
+++ b/nvkm/subdev/devinit/nv04.c
@@ -0,0 +1,189 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/devinit.h>
+#include <subdev/vga.h>
+
+#include "fbmem.h"
+
+struct nv04_devinit_priv {
+	struct nouveau_devinit base;
+	int owner;
+};
+
+static void
+nv04_devinit_meminit(struct nouveau_devinit *devinit)
+{
+	struct nv04_devinit_priv *priv = (void *)devinit;
+	u32 patt = 0xdeadbeef;
+	struct io_mapping *fb;
+	int i;
+
+	/* Map the framebuffer aperture */
+	fb = fbmem_init(nv_device(priv)->pdev);
+	if (!fb) {
+		nv_error(priv, "failed to map fb\n");
+		return;
+	}
+
+	/* Sequencer and refresh off */
+	nv_wrvgas(priv, 0, 1, nv_rdvgas(priv, 0, 1) | 0x20);
+	nv_mask(priv, NV04_PFB_DEBUG_0, 0, NV04_PFB_DEBUG_0_REFRESH_OFF);
+
+	nv_mask(priv, NV04_PFB_BOOT_0, ~0,
+		      NV04_PFB_BOOT_0_RAM_AMOUNT_16MB |
+		      NV04_PFB_BOOT_0_RAM_WIDTH_128 |
+		      NV04_PFB_BOOT_0_RAM_TYPE_SGRAM_16MBIT);
+
+	for (i = 0; i < 4; i++)
+		fbmem_poke(fb, 4 * i, patt);
+
+	fbmem_poke(fb, 0x400000, patt + 1);
+
+	if (fbmem_peek(fb, 0) == patt + 1) {
+		nv_mask(priv, NV04_PFB_BOOT_0,
+			      NV04_PFB_BOOT_0_RAM_TYPE,
+			      NV04_PFB_BOOT_0_RAM_TYPE_SDRAM_16MBIT);
+		nv_mask(priv, NV04_PFB_DEBUG_0,
+			      NV04_PFB_DEBUG_0_REFRESH_OFF, 0);
+
+		for (i = 0; i < 4; i++)
+			fbmem_poke(fb, 4 * i, patt);
+
+		if ((fbmem_peek(fb, 0xc) & 0xffff) != (patt & 0xffff))
+			nv_mask(priv, NV04_PFB_BOOT_0,
+				      NV04_PFB_BOOT_0_RAM_WIDTH_128 |
+				      NV04_PFB_BOOT_0_RAM_AMOUNT,
+				      NV04_PFB_BOOT_0_RAM_AMOUNT_8MB);
+	} else
+	if ((fbmem_peek(fb, 0xc) & 0xffff0000) != (patt & 0xffff0000)) {
+		nv_mask(priv, NV04_PFB_BOOT_0,
+			      NV04_PFB_BOOT_0_RAM_WIDTH_128 |
+			      NV04_PFB_BOOT_0_RAM_AMOUNT,
+			      NV04_PFB_BOOT_0_RAM_AMOUNT_4MB);
+	} else
+	if (fbmem_peek(fb, 0) != patt) {
+		if (fbmem_readback(fb, 0x800000, patt))
+			nv_mask(priv, NV04_PFB_BOOT_0,
+				      NV04_PFB_BOOT_0_RAM_AMOUNT,
+				      NV04_PFB_BOOT_0_RAM_AMOUNT_8MB);
+		else
+			nv_mask(priv, NV04_PFB_BOOT_0,
+				      NV04_PFB_BOOT_0_RAM_AMOUNT,
+				      NV04_PFB_BOOT_0_RAM_AMOUNT_4MB);
+
+		nv_mask(priv, NV04_PFB_BOOT_0, NV04_PFB_BOOT_0_RAM_TYPE,
+			      NV04_PFB_BOOT_0_RAM_TYPE_SGRAM_8MBIT);
+	} else
+	if (!fbmem_readback(fb, 0x800000, patt)) {
+		nv_mask(priv, NV04_PFB_BOOT_0, NV04_PFB_BOOT_0_RAM_AMOUNT,
+			      NV04_PFB_BOOT_0_RAM_AMOUNT_8MB);
+
+	}
+
+	/* Refresh on, sequencer on */
+	nv_mask(priv, NV04_PFB_DEBUG_0, NV04_PFB_DEBUG_0_REFRESH_OFF, 0);
+	nv_wrvgas(priv, 0, 1, nv_rdvgas(priv, 0, 1) & ~0x20);
+	fbmem_fini(fb);
+}
+
+static int
+nv04_devinit_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv04_devinit_priv *priv;
+	int ret;
+
+	ret = nouveau_devinit_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.meminit = nv04_devinit_meminit;
+	priv->owner = -1;
+	return 0;
+}
+
+void
+nv04_devinit_dtor(struct nouveau_object *object)
+{
+	struct nv04_devinit_priv *priv = (void *)object;
+
+	/* restore vga owner saved at first init, and lock crtc regs  */
+	nv_wrvgaowner(priv, priv->owner);
+	nv_lockvgac(priv, true);
+
+	nouveau_devinit_destroy(&priv->base);
+}
+
+int
+nv04_devinit_init(struct nouveau_object *object)
+{
+	struct nv04_devinit_priv *priv = (void *)object;
+
+	if (!priv->base.post) {
+		u32 htotal = nv_rdvgac(priv, 0, 0x06);
+		htotal |= (nv_rdvgac(priv, 0, 0x07) & 0x01) << 8;
+		htotal |= (nv_rdvgac(priv, 0, 0x07) & 0x20) << 4;
+		htotal |= (nv_rdvgac(priv, 0, 0x25) & 0x01) << 10;
+		htotal |= (nv_rdvgac(priv, 0, 0x41) & 0x01) << 11;
+		if (!htotal) {
+			nv_info(priv, "adaptor not initialised\n");
+			priv->base.post = true;
+		}
+	}
+
+	return nouveau_devinit_init(&priv->base);
+}
+
+int
+nv04_devinit_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv04_devinit_priv *priv = (void *)object;
+
+	/* make i2c busses accessible */
+	nv_mask(priv, 0x000200, 0x00000001, 0x00000001);
+
+	/* unlock extended vga crtc regs, and unslave crtcs */
+	nv_lockvgac(priv, false);
+	if (priv->owner < 0)
+		priv->owner = nv_rdvgaowner(priv);
+	nv_wrvgaowner(priv, 0);
+
+	return nouveau_devinit_fini(&priv->base, suspend);
+}
+
+struct nouveau_oclass
+nv04_devinit_oclass = {
+	.handle = NV_SUBDEV(DEVINIT, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_devinit_ctor,
+		.dtor = nv04_devinit_dtor,
+		.init = nv04_devinit_init,
+		.fini = nv04_devinit_fini,
+	},
+};
diff --git a/nvkm/subdev/devinit/nv05.c b/nvkm/subdev/devinit/nv05.c
new file mode 100644
index 000000000..191447d0d
--- /dev/null
+++ b/nvkm/subdev/devinit/nv05.c
@@ -0,0 +1,159 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/devinit.h>
+#include <subdev/bios.h>
+#include <subdev/bios/bmp.h>
+#include <subdev/vga.h>
+
+#include "fbmem.h"
+
+struct nv05_devinit_priv {
+	struct nouveau_devinit base;
+	u8 owner;
+};
+
+static void
+nv05_devinit_meminit(struct nouveau_devinit *devinit)
+{
+	static const u8 default_config_tab[][2] = {
+		{ 0x24, 0x00 },
+		{ 0x28, 0x00 },
+		{ 0x24, 0x01 },
+		{ 0x1f, 0x00 },
+		{ 0x0f, 0x00 },
+		{ 0x17, 0x00 },
+		{ 0x06, 0x00 },
+		{ 0x00, 0x00 }
+	};
+	struct nv05_devinit_priv *priv = (void *)devinit;
+	struct nouveau_bios *bios = nouveau_bios(priv);
+	struct io_mapping *fb;
+	u32 patt = 0xdeadbeef;
+	u16 data;
+	u8 strap, ramcfg[2];
+	int i, v;
+
+	/* Map the framebuffer aperture */
+	fb = fbmem_init(nv_device(priv)->pdev);
+	if (!fb) {
+		nv_error(priv, "failed to map fb\n");
+		return;
+	}
+
+	strap = (nv_rd32(priv, 0x101000) & 0x0000003c) >> 2;
+	if ((data = bmp_mem_init_table(bios))) {
+		ramcfg[0] = nv_ro08(bios, data + 2 * strap + 0);
+		ramcfg[1] = nv_ro08(bios, data + 2 * strap + 1);
+	} else {
+		ramcfg[0] = default_config_tab[strap][0];
+		ramcfg[1] = default_config_tab[strap][1];
+	}
+
+	/* Sequencer off */
+	nv_wrvgas(priv, 0, 1, nv_rdvgas(priv, 0, 1) | 0x20);
+
+	if (nv_rd32(priv, NV04_PFB_BOOT_0) & NV04_PFB_BOOT_0_UMA_ENABLE)
+		goto out;
+
+	nv_mask(priv, NV04_PFB_DEBUG_0, NV04_PFB_DEBUG_0_REFRESH_OFF, 0);
+
+	/* If present load the hardcoded scrambling table */
+	if (data) {
+		for (i = 0, data += 0x10; i < 8; i++, data += 4) {
+			u32 scramble = nv_ro32(bios, data);
+			nv_wr32(priv, NV04_PFB_SCRAMBLE(i), scramble);
+		}
+	}
+
+	/* Set memory type/width/length defaults depending on the straps */
+	nv_mask(priv, NV04_PFB_BOOT_0, 0x3f, ramcfg[0]);
+
+	if (ramcfg[1] & 0x80)
+		nv_mask(priv, NV04_PFB_CFG0, 0, NV04_PFB_CFG0_SCRAMBLE);
+
+	nv_mask(priv, NV04_PFB_CFG1, 0x700001, (ramcfg[1] & 1) << 20);
+	nv_mask(priv, NV04_PFB_CFG1, 0, 1);
+
+	/* Probe memory bus width */
+	for (i = 0; i < 4; i++)
+		fbmem_poke(fb, 4 * i, patt);
+
+	if (fbmem_peek(fb, 0xc) != patt)
+		nv_mask(priv, NV04_PFB_BOOT_0,
+			  NV04_PFB_BOOT_0_RAM_WIDTH_128, 0);
+
+	/* Probe memory length */
+	v = nv_rd32(priv, NV04_PFB_BOOT_0) & NV04_PFB_BOOT_0_RAM_AMOUNT;
+
+	if (v == NV04_PFB_BOOT_0_RAM_AMOUNT_32MB &&
+	    (!fbmem_readback(fb, 0x1000000, ++patt) ||
+	     !fbmem_readback(fb, 0, ++patt)))
+		nv_mask(priv, NV04_PFB_BOOT_0, NV04_PFB_BOOT_0_RAM_AMOUNT,
+			  NV04_PFB_BOOT_0_RAM_AMOUNT_16MB);
+
+	if (v == NV04_PFB_BOOT_0_RAM_AMOUNT_16MB &&
+	    !fbmem_readback(fb, 0x800000, ++patt))
+		nv_mask(priv, NV04_PFB_BOOT_0, NV04_PFB_BOOT_0_RAM_AMOUNT,
+			  NV04_PFB_BOOT_0_RAM_AMOUNT_8MB);
+
+	if (!fbmem_readback(fb, 0x400000, ++patt))
+		nv_mask(priv, NV04_PFB_BOOT_0, NV04_PFB_BOOT_0_RAM_AMOUNT,
+			  NV04_PFB_BOOT_0_RAM_AMOUNT_4MB);
+
+out:
+	/* Sequencer on */
+	nv_wrvgas(priv, 0, 1, nv_rdvgas(priv, 0, 1) & ~0x20);
+	fbmem_fini(fb);
+}
+
+static int
+nv05_devinit_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv05_devinit_priv *priv;
+	int ret;
+
+	ret = nouveau_devinit_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.meminit = nv05_devinit_meminit;
+	return 0;
+}
+
+struct nouveau_oclass
+nv05_devinit_oclass = {
+	.handle = NV_SUBDEV(DEVINIT, 0x05),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv05_devinit_ctor,
+		.dtor = nv04_devinit_dtor,
+		.init = nv04_devinit_init,
+		.fini = nv04_devinit_fini,
+	},
+};
diff --git a/nvkm/subdev/devinit/nv10.c b/nvkm/subdev/devinit/nv10.c
new file mode 100644
index 000000000..eb76ffab6
--- /dev/null
+++ b/nvkm/subdev/devinit/nv10.c
@@ -0,0 +1,124 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/devinit.h>
+#include <subdev/vga.h>
+
+#include "fbmem.h"
+
+struct nv10_devinit_priv {
+	struct nouveau_devinit base;
+	u8 owner;
+};
+
+static void
+nv10_devinit_meminit(struct nouveau_devinit *devinit)
+{
+	struct nv10_devinit_priv *priv = (void *)devinit;
+	const int mem_width[] = { 0x10, 0x00, 0x20 };
+	const int mem_width_count = nv_device(priv)->chipset >= 0x17 ? 3 : 2;
+	uint32_t patt = 0xdeadbeef;
+	struct io_mapping *fb;
+	int i, j, k;
+
+	/* Map the framebuffer aperture */
+	fb = fbmem_init(nv_device(priv)->pdev);
+	if (!fb) {
+		nv_error(priv, "failed to map fb\n");
+		return;
+	}
+
+	nv_wr32(priv, NV10_PFB_REFCTRL, NV10_PFB_REFCTRL_VALID_1);
+
+	/* Probe memory bus width */
+	for (i = 0; i < mem_width_count; i++) {
+		nv_mask(priv, NV04_PFB_CFG0, 0x30, mem_width[i]);
+
+		for (j = 0; j < 4; j++) {
+			for (k = 0; k < 4; k++)
+				fbmem_poke(fb, 0x1c, 0);
+
+			fbmem_poke(fb, 0x1c, patt);
+			fbmem_poke(fb, 0x3c, 0);
+
+			if (fbmem_peek(fb, 0x1c) == patt)
+				goto mem_width_found;
+		}
+	}
+
+mem_width_found:
+	patt <<= 1;
+
+	/* Probe amount of installed memory */
+	for (i = 0; i < 4; i++) {
+		int off = nv_rd32(priv, 0x10020c) - 0x100000;
+
+		fbmem_poke(fb, off, patt);
+		fbmem_poke(fb, 0, 0);
+
+		fbmem_peek(fb, 0);
+		fbmem_peek(fb, 0);
+		fbmem_peek(fb, 0);
+		fbmem_peek(fb, 0);
+
+		if (fbmem_peek(fb, off) == patt)
+			goto amount_found;
+	}
+
+	/* IC missing - disable the upper half memory space. */
+	nv_mask(priv, NV04_PFB_CFG0, 0x1000, 0);
+
+amount_found:
+	fbmem_fini(fb);
+}
+
+static int
+nv10_devinit_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv10_devinit_priv *priv;
+	int ret;
+
+	ret = nouveau_devinit_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.meminit = nv10_devinit_meminit;
+	return 0;
+}
+
+struct nouveau_oclass
+nv10_devinit_oclass = {
+	.handle = NV_SUBDEV(DEVINIT, 0x10),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv10_devinit_ctor,
+		.dtor = nv04_devinit_dtor,
+		.init = nv04_devinit_init,
+		.fini = nv04_devinit_fini,
+	},
+};
diff --git a/nvkm/subdev/devinit/nv1a.c b/nvkm/subdev/devinit/nv1a.c
new file mode 100644
index 000000000..5b2ba630d
--- /dev/null
+++ b/nvkm/subdev/devinit/nv1a.c
@@ -0,0 +1,58 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/devinit.h>
+#include <subdev/vga.h>
+
+struct nv1a_devinit_priv {
+	struct nouveau_devinit base;
+	u8 owner;
+};
+
+static int
+nv1a_devinit_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv1a_devinit_priv *priv;
+	int ret;
+
+	ret = nouveau_devinit_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+struct nouveau_oclass
+nv1a_devinit_oclass = {
+	.handle = NV_SUBDEV(DEVINIT, 0x1a),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv1a_devinit_ctor,
+		.dtor = nv04_devinit_dtor,
+		.init = nv04_devinit_init,
+		.fini = nv04_devinit_fini,
+	},
+};
diff --git a/nvkm/subdev/devinit/nv20.c b/nvkm/subdev/devinit/nv20.c
new file mode 100644
index 000000000..eb32e9900
--- /dev/null
+++ b/nvkm/subdev/devinit/nv20.c
@@ -0,0 +1,96 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/devinit.h>
+#include <subdev/vga.h>
+
+#include "fbmem.h"
+
+struct nv20_devinit_priv {
+	struct nouveau_devinit base;
+	u8 owner;
+};
+
+static void
+nv20_devinit_meminit(struct nouveau_devinit *devinit)
+{
+	struct nv20_devinit_priv *priv = (void *)devinit;
+	struct nouveau_device *device = nv_device(priv);
+	uint32_t mask = (device->chipset >= 0x25 ? 0x300 : 0x900);
+	uint32_t amount, off;
+	struct io_mapping *fb;
+
+	/* Map the framebuffer aperture */
+	fb = fbmem_init(nv_device(priv)->pdev);
+	if (!fb) {
+		nv_error(priv, "failed to map fb\n");
+		return;
+	}
+
+	nv_wr32(priv, NV10_PFB_REFCTRL, NV10_PFB_REFCTRL_VALID_1);
+
+	/* Allow full addressing */
+	nv_mask(priv, NV04_PFB_CFG0, 0, mask);
+
+	amount = nv_rd32(priv, 0x10020c);
+	for (off = amount; off > 0x2000000; off -= 0x2000000)
+		fbmem_poke(fb, off - 4, off);
+
+	amount = nv_rd32(priv, 0x10020c);
+	if (amount != fbmem_peek(fb, amount - 4))
+		/* IC missing - disable the upper half memory space. */
+		nv_mask(priv, NV04_PFB_CFG0, mask, 0);
+
+	fbmem_fini(fb);
+}
+
+static int
+nv20_devinit_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv20_devinit_priv *priv;
+	int ret;
+
+	ret = nouveau_devinit_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.meminit = nv20_devinit_meminit;
+	return 0;
+}
+
+struct nouveau_oclass
+nv20_devinit_oclass = {
+	.handle = NV_SUBDEV(DEVINIT, 0x20),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv20_devinit_ctor,
+		.dtor = nv04_devinit_dtor,
+		.init = nv04_devinit_init,
+		.fini = nv04_devinit_fini,
+	},
+};
diff --git a/nvkm/subdev/devinit/nv50.c b/nvkm/subdev/devinit/nv50.c
new file mode 100644
index 000000000..ae7249b09
--- /dev/null
+++ b/nvkm/subdev/devinit/nv50.c
@@ -0,0 +1,119 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/bios.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/disp.h>
+#include <subdev/bios/init.h>
+#include <subdev/devinit.h>
+#include <subdev/vga.h>
+
+struct nv50_devinit_priv {
+	struct nouveau_devinit base;
+};
+
+static int
+nv50_devinit_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv50_devinit_priv *priv;
+	int ret;
+
+	ret = nouveau_devinit_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static void
+nv50_devinit_dtor(struct nouveau_object *object)
+{
+	struct nv50_devinit_priv *priv = (void *)object;
+	nouveau_devinit_destroy(&priv->base);
+}
+
+static int
+nv50_devinit_init(struct nouveau_object *object)
+{
+	struct nouveau_bios *bios = nouveau_bios(object);
+	struct nv50_devinit_priv *priv = (void *)object;
+	struct nvbios_outp info;
+	struct dcb_output outp;
+	u8  ver = 0xff, hdr, cnt, len;
+	int ret, i = 0;
+
+	if (!priv->base.post) {
+		if (!nv_rdvgac(priv, 0, 0x00) &&
+		    !nv_rdvgac(priv, 0, 0x1a)) {
+			nv_info(priv, "adaptor not initialised\n");
+			priv->base.post = true;
+		}
+	}
+
+	ret = nouveau_devinit_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* if we ran the init tables, execute first script pointer for each
+	 * display table output entry that has a matching dcb entry.
+	 */
+	while (priv->base.post && ver) {
+		u16 data = nvbios_outp_parse(bios, i++, &ver, &hdr, &cnt, &len, &info);
+		if (data && dcb_outp_match(bios, info.type, info.mask, &ver, &len, &outp)) {
+			struct nvbios_init init = {
+				.subdev = nv_subdev(priv),
+				.bios = bios,
+				.offset = info.script[0],
+				.outp = &outp,
+				.crtc = -1,
+				.execute = 1,
+			};
+
+			nvbios_exec(&init);
+		}
+	};
+
+	return 0;
+}
+
+static int
+nv50_devinit_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_devinit_priv *priv = (void *)object;
+	return nouveau_devinit_fini(&priv->base, suspend);
+}
+
+struct nouveau_oclass
+nv50_devinit_oclass = {
+	.handle = NV_SUBDEV(DEVINIT, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_devinit_ctor,
+		.dtor = nv50_devinit_dtor,
+		.init = nv50_devinit_init,
+		.fini = nv50_devinit_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/Makefile.am b/nvkm/subdev/fb/Makefile.am
new file mode 100644
index 000000000..6355a55cd
--- /dev/null
+++ b/nvkm/subdev/fb/Makefile.am
@@ -0,0 +1,12 @@
+noinst_LTLIBRARIES = libfb.la
+
+libfb_la_SOURCES = base.c \
+		   nv04.c \
+		   nv10.c nv1a.c \
+		   nv20.c nv25.c \
+		   nv30.c nv35.c nv36.c \
+		   nv40.c nv41.c nv44.c nv46.c nv47.c nv49.c nv4e.c \
+		   nv50.c \
+		   nvc0.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/fb/base.c b/nvkm/subdev/fb/base.c
new file mode 100644
index 000000000..d6d16007e
--- /dev/null
+++ b/nvkm/subdev/fb/base.c
@@ -0,0 +1,140 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "subdev/fb.h"
+#include "subdev/bios.h"
+#include "subdev/bios/bit.h"
+
+int
+nouveau_fb_bios_memtype(struct nouveau_bios *bios)
+{
+	struct bit_entry M;
+	u8 ramcfg;
+
+	ramcfg = (nv_rd32(bios, 0x101000) & 0x0000003c) >> 2;
+	if (!bit_entry(bios, 'M', &M) && M.version == 2 && M.length >= 5) {
+		u16 table   = nv_ro16(bios, M.offset + 3);
+		u8  version = nv_ro08(bios, table + 0);
+		u8  header  = nv_ro08(bios, table + 1);
+		u8  record  = nv_ro08(bios, table + 2);
+		u8  entries = nv_ro08(bios, table + 3);
+		if (table && version == 0x10 && ramcfg < entries) {
+			u16 entry = table + header + (ramcfg * record);
+			switch (nv_ro08(bios, entry) & 0x0f) {
+			case 0: return NV_MEM_TYPE_DDR2;
+			case 1: return NV_MEM_TYPE_DDR3;
+			case 2: return NV_MEM_TYPE_GDDR3;
+			case 3: return NV_MEM_TYPE_GDDR5;
+			default:
+				break;
+			}
+
+		}
+	}
+
+	return NV_MEM_TYPE_UNKNOWN;
+}
+
+int
+nouveau_fb_preinit(struct nouveau_fb *pfb)
+{
+	static const char *name[] = {
+		[NV_MEM_TYPE_UNKNOWN] = "unknown",
+		[NV_MEM_TYPE_STOLEN ] = "stolen system memory",
+		[NV_MEM_TYPE_SGRAM  ] = "SGRAM",
+		[NV_MEM_TYPE_SDRAM  ] = "SDRAM",
+		[NV_MEM_TYPE_DDR1   ] = "DDR1",
+		[NV_MEM_TYPE_DDR2   ] = "DDR2",
+		[NV_MEM_TYPE_DDR3   ] = "DDR3",
+		[NV_MEM_TYPE_GDDR2  ] = "GDDR2",
+		[NV_MEM_TYPE_GDDR3  ] = "GDDR3",
+		[NV_MEM_TYPE_GDDR4  ] = "GDDR4",
+		[NV_MEM_TYPE_GDDR5  ] = "GDDR5",
+	};
+	int ret, tags;
+
+	tags = pfb->ram.init(pfb);
+	if (tags < 0 || !pfb->ram.size) {
+		nv_fatal(pfb, "error detecting memory configuration!!\n");
+		return (tags < 0) ? tags : -ERANGE;
+	}
+
+	if (!nouveau_mm_initialised(&pfb->vram)) {
+		ret = nouveau_mm_init(&pfb->vram, 0, pfb->ram.size >> 12, 1);
+		if (ret)
+			return ret;
+	}
+
+	if (!nouveau_mm_initialised(&pfb->tags) && tags) {
+		ret = nouveau_mm_init(&pfb->tags, 0, ++tags, 1);
+		if (ret)
+			return ret;
+	}
+
+	nv_info(pfb, "RAM type: %s\n", name[pfb->ram.type]);
+	nv_info(pfb, "RAM size: %d MiB\n", (int)(pfb->ram.size >> 20));
+	nv_info(pfb, "   ZCOMP: %d tags\n", tags);
+	return 0;
+}
+
+void
+nouveau_fb_destroy(struct nouveau_fb *pfb)
+{
+	int i;
+
+	for (i = 0; i < pfb->tile.regions; i++)
+		pfb->tile.fini(pfb, i, &pfb->tile.region[i]);
+	nouveau_mm_fini(&pfb->tags);
+	nouveau_mm_fini(&pfb->vram);
+
+	nouveau_subdev_destroy(&pfb->base);
+}
+
+void
+_nouveau_fb_dtor(struct nouveau_object *object)
+{
+	struct nouveau_fb *pfb = (void *)object;
+	nouveau_fb_destroy(pfb);
+}
+int
+nouveau_fb_init(struct nouveau_fb *pfb)
+{
+	int ret, i;
+
+	ret = nouveau_subdev_init(&pfb->base);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < pfb->tile.regions; i++)
+		pfb->tile.prog(pfb, i, &pfb->tile.region[i]);
+
+	return 0;
+}
+
+int
+_nouveau_fb_init(struct nouveau_object *object)
+{
+	struct nouveau_fb *pfb = (void *)object;
+	return nouveau_fb_init(pfb);
+}
diff --git a/nvkm/subdev/fb/nv04.c b/nvkm/subdev/fb/nv04.c
new file mode 100644
index 000000000..6e369f853
--- /dev/null
+++ b/nvkm/subdev/fb/nv04.c
@@ -0,0 +1,134 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/fb.h>
+
+#define NV04_PFB_BOOT_0						0x00100000
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT			0x00000003
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT_32MB			0x00000000
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT_4MB			0x00000001
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT_8MB			0x00000002
+#	define NV04_PFB_BOOT_0_RAM_AMOUNT_16MB			0x00000003
+#	define NV04_PFB_BOOT_0_RAM_WIDTH_128			0x00000004
+#	define NV04_PFB_BOOT_0_RAM_TYPE				0x00000028
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SGRAM_8MBIT		0x00000000
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SGRAM_16MBIT		0x00000008
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SGRAM_16MBIT_4BANK	0x00000010
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SDRAM_16MBIT		0x00000018
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SDRAM_64MBIT		0x00000020
+#	define NV04_PFB_BOOT_0_RAM_TYPE_SDRAM_64MBITX16		0x00000028
+#	define NV04_PFB_BOOT_0_UMA_ENABLE			0x00000100
+#	define NV04_PFB_BOOT_0_UMA_SIZE				0x0000f000
+#define NV04_PFB_CFG0						0x00100200
+
+struct nv04_fb_priv {
+	struct nouveau_fb base;
+};
+
+bool
+nv04_fb_memtype_valid(struct nouveau_fb *pfb, u32 tile_flags)
+{
+	if (!(tile_flags & 0xff00))
+		return true;
+
+	return false;
+}
+
+static int
+nv04_fb_vram_init(struct nouveau_fb *pfb)
+{
+	u32 boot0 = nv_rd32(pfb, NV04_PFB_BOOT_0);
+	if (boot0 & 0x00000100) {
+		pfb->ram.size  = ((boot0 >> 12) & 0xf) * 2 + 2;
+		pfb->ram.size *= 1024 * 1024;
+	} else {
+		switch (boot0 & NV04_PFB_BOOT_0_RAM_AMOUNT) {
+		case NV04_PFB_BOOT_0_RAM_AMOUNT_32MB:
+			pfb->ram.size = 32 * 1024 * 1024;
+			break;
+		case NV04_PFB_BOOT_0_RAM_AMOUNT_16MB:
+			pfb->ram.size = 16 * 1024 * 1024;
+			break;
+		case NV04_PFB_BOOT_0_RAM_AMOUNT_8MB:
+			pfb->ram.size = 8 * 1024 * 1024;
+			break;
+		case NV04_PFB_BOOT_0_RAM_AMOUNT_4MB:
+			pfb->ram.size = 4 * 1024 * 1024;
+			break;
+		}
+	}
+
+	if ((boot0 & 0x00000038) <= 0x10)
+		pfb->ram.type = NV_MEM_TYPE_SGRAM;
+	else
+		pfb->ram.type = NV_MEM_TYPE_SDRAM;
+	return 0;
+}
+
+static int
+nv04_fb_init(struct nouveau_object *object)
+{
+	struct nv04_fb_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_fb_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* This is what the DDX did for NV_ARCH_04, but a mmio-trace shows
+	 * nvidia reading PFB_CFG_0, then writing back its original value.
+	 * (which was 0x701114 in this case)
+	 */
+	nv_wr32(priv, NV04_PFB_CFG0, 0x1114);
+	return 0;
+}
+
+static int
+nv04_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv04_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv04_fb_vram_init;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+struct nouveau_oclass
+nv04_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv04_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv10.c b/nvkm/subdev/fb/nv10.c
new file mode 100644
index 000000000..edbbe26e8
--- /dev/null
+++ b/nvkm/subdev/fb/nv10.c
@@ -0,0 +1,104 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv10_fb_priv {
+	struct nouveau_fb base;
+};
+
+static int
+nv10_fb_vram_init(struct nouveau_fb *pfb)
+{
+	u32 cfg0 = nv_rd32(pfb, 0x100200);
+	if (cfg0 & 0x00000001)
+		pfb->ram.type = NV_MEM_TYPE_DDR1;
+	else
+		pfb->ram.type = NV_MEM_TYPE_SDRAM;
+
+	pfb->ram.size = nv_rd32(pfb, 0x10020c) & 0xff000000;
+	return 0;
+}
+
+void
+nv10_fb_tile_init(struct nouveau_fb *pfb, int i, u32 addr, u32 size, u32 pitch,
+		  u32 flags, struct nouveau_fb_tile *tile)
+{
+	tile->addr  = 0x80000000 | addr;
+	tile->limit = max(1u, addr + size) - 1;
+	tile->pitch = pitch;
+}
+
+void
+nv10_fb_tile_fini(struct nouveau_fb *pfb, int i, struct nouveau_fb_tile *tile)
+{
+	tile->addr  = 0;
+	tile->limit = 0;
+	tile->pitch = 0;
+	tile->zcomp = 0;
+}
+
+void
+nv10_fb_tile_prog(struct nouveau_fb *pfb, int i, struct nouveau_fb_tile *tile)
+{
+	nv_wr32(pfb, 0x100244 + (i * 0x10), tile->limit);
+	nv_wr32(pfb, 0x100248 + (i * 0x10), tile->pitch);
+	nv_wr32(pfb, 0x100240 + (i * 0x10), tile->addr);
+	nv_rd32(pfb, 0x100240 + (i * 0x10));
+}
+
+static int
+nv10_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv10_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv10_fb_vram_init;
+	priv->base.tile.regions = 8;
+	priv->base.tile.init = nv10_fb_tile_init;
+	priv->base.tile.fini = nv10_fb_tile_fini;
+	priv->base.tile.prog = nv10_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+struct nouveau_oclass
+nv10_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x10),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv10_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = _nouveau_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv1a.c b/nvkm/subdev/fb/nv1a.c
new file mode 100644
index 000000000..48366841d
--- /dev/null
+++ b/nvkm/subdev/fb/nv1a.c
@@ -0,0 +1,89 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv1a_fb_priv {
+	struct nouveau_fb base;
+};
+
+static int
+nv1a_fb_vram_init(struct nouveau_fb *pfb)
+{
+	struct pci_dev *bridge;
+	u32 mem, mib;
+
+	bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0, 1));
+	if (!bridge) {
+		nv_fatal(pfb, "no bridge device\n");
+		return -ENODEV;
+	}
+
+	if (nv_device(pfb)->chipset == 0x1a) {
+		pci_read_config_dword(bridge, 0x7c, &mem);
+		mib = ((mem >> 6) & 31) + 1;
+	} else {
+		pci_read_config_dword(bridge, 0x84, &mem);
+		mib = ((mem >> 4) & 127) + 1;
+	}
+
+	pfb->ram.type = NV_MEM_TYPE_STOLEN;
+	pfb->ram.size = mib * 1024 * 1024;
+	return 0;
+}
+
+static int
+nv1a_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv1a_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv1a_fb_vram_init;
+	priv->base.tile.regions = 8;
+	priv->base.tile.init = nv10_fb_tile_init;
+	priv->base.tile.fini = nv10_fb_tile_fini;
+	priv->base.tile.prog = nv10_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+struct nouveau_oclass
+nv1a_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x1a),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv1a_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = _nouveau_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv20.c b/nvkm/subdev/fb/nv20.c
new file mode 100644
index 000000000..5d14612a2
--- /dev/null
+++ b/nvkm/subdev/fb/nv20.c
@@ -0,0 +1,132 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv20_fb_priv {
+	struct nouveau_fb base;
+};
+
+int
+nv20_fb_vram_init(struct nouveau_fb *pfb)
+{
+	u32 pbus1218 = nv_rd32(pfb, 0x001218);
+
+	switch (pbus1218 & 0x00000300) {
+	case 0x00000000: pfb->ram.type = NV_MEM_TYPE_SDRAM; break;
+	case 0x00000100: pfb->ram.type = NV_MEM_TYPE_DDR1; break;
+	case 0x00000200: pfb->ram.type = NV_MEM_TYPE_GDDR3; break;
+	case 0x00000300: pfb->ram.type = NV_MEM_TYPE_GDDR2; break;
+	}
+	pfb->ram.size  = (nv_rd32(pfb, 0x10020c) & 0xff000000);
+	pfb->ram.parts = (nv_rd32(pfb, 0x100200) & 0x00000003) + 1;
+
+	return nv_rd32(pfb, 0x100320);
+}
+
+void
+nv20_fb_tile_init(struct nouveau_fb *pfb, int i, u32 addr, u32 size, u32 pitch,
+		  u32 flags, struct nouveau_fb_tile *tile)
+{
+	tile->addr  = 0x00000001 | addr;
+	tile->limit = max(1u, addr + size) - 1;
+	tile->pitch = pitch;
+	if (flags & 4) {
+		pfb->tile.comp(pfb, i, size, flags, tile);
+		tile->addr |= 2;
+	}
+}
+
+static void
+nv20_fb_tile_comp(struct nouveau_fb *pfb, int i, u32 size, u32 flags,
+		  struct nouveau_fb_tile *tile)
+{
+	u32 tiles = DIV_ROUND_UP(size, 0x40);
+	u32 tags  = round_up(tiles / pfb->ram.parts, 0x40);
+	if (!nouveau_mm_head(&pfb->tags, 1, tags, tags, 1, &tile->tag)) {
+		if (!(flags & 2)) tile->zcomp = 0x00000000; /* Z16 */
+		else              tile->zcomp = 0x04000000; /* Z24S8 */
+		tile->zcomp |= tile->tag->offset;
+		tile->zcomp |= 0x80000000; /* enable */
+#ifdef __BIG_ENDIAN
+		tile->zcomp |= 0x08000000;
+#endif
+	}
+}
+
+void
+nv20_fb_tile_fini(struct nouveau_fb *pfb, int i, struct nouveau_fb_tile *tile)
+{
+	tile->addr  = 0;
+	tile->limit = 0;
+	tile->pitch = 0;
+	tile->zcomp = 0;
+	nouveau_mm_free(&pfb->tags, &tile->tag);
+}
+
+void
+nv20_fb_tile_prog(struct nouveau_fb *pfb, int i, struct nouveau_fb_tile *tile)
+{
+	nv_wr32(pfb, 0x100244 + (i * 0x10), tile->limit);
+	nv_wr32(pfb, 0x100248 + (i * 0x10), tile->pitch);
+	nv_wr32(pfb, 0x100240 + (i * 0x10), tile->addr);
+	nv_rd32(pfb, 0x100240 + (i * 0x10));
+	nv_wr32(pfb, 0x100300 + (i * 0x04), tile->zcomp);
+}
+
+static int
+nv20_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv20_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv20_fb_vram_init;
+	priv->base.tile.regions = 8;
+	priv->base.tile.init = nv20_fb_tile_init;
+	priv->base.tile.comp = nv20_fb_tile_comp;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv20_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+struct nouveau_oclass
+nv20_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x20),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv20_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = _nouveau_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv25.c b/nvkm/subdev/fb/nv25.c
new file mode 100644
index 000000000..0042ace6b
--- /dev/null
+++ b/nvkm/subdev/fb/nv25.c
@@ -0,0 +1,81 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv25_fb_priv {
+	struct nouveau_fb base;
+};
+
+static void
+nv25_fb_tile_comp(struct nouveau_fb *pfb, int i, u32 size, u32 flags,
+		  struct nouveau_fb_tile *tile)
+{
+	u32 tiles = DIV_ROUND_UP(size, 0x40);
+	u32 tags  = round_up(tiles / pfb->ram.parts, 0x40);
+	if (!nouveau_mm_head(&pfb->tags, 1, tags, tags, 1, &tile->tag)) {
+		if (!(flags & 2)) tile->zcomp = 0x00100000; /* Z16 */
+		else              tile->zcomp = 0x00200000; /* Z24S8 */
+		tile->zcomp |= tile->tag->offset;
+#ifdef __BIG_ENDIAN
+		tile->zcomp |= 0x01000000;
+#endif
+	}
+}
+
+static int
+nv25_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv25_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv20_fb_vram_init;
+	priv->base.tile.regions = 8;
+	priv->base.tile.init = nv20_fb_tile_init;
+	priv->base.tile.comp = nv25_fb_tile_comp;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv20_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+struct nouveau_oclass
+nv25_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x25),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv25_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = _nouveau_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv30.c b/nvkm/subdev/fb/nv30.c
new file mode 100644
index 000000000..a7ba0d048
--- /dev/null
+++ b/nvkm/subdev/fb/nv30.c
@@ -0,0 +1,159 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv30_fb_priv {
+	struct nouveau_fb base;
+};
+
+void
+nv30_fb_tile_init(struct nouveau_fb *pfb, int i, u32 addr, u32 size, u32 pitch,
+		  u32 flags, struct nouveau_fb_tile *tile)
+{
+	/* for performance, select alternate bank offset for zeta */
+	if (!(flags & 4)) {
+		tile->addr = (0 << 4);
+	} else {
+		if (pfb->tile.comp) /* z compression */
+			pfb->tile.comp(pfb, i, size, flags, tile);
+		tile->addr = (1 << 4);
+	}
+
+	tile->addr |= 0x00000001; /* enable */
+	tile->addr |= addr;
+	tile->limit = max(1u, addr + size) - 1;
+	tile->pitch = pitch;
+}
+
+static void
+nv30_fb_tile_comp(struct nouveau_fb *pfb, int i, u32 size, u32 flags,
+		  struct nouveau_fb_tile *tile)
+{
+	u32 tiles = DIV_ROUND_UP(size, 0x40);
+	u32 tags  = round_up(tiles / pfb->ram.parts, 0x40);
+	if (!nouveau_mm_head(&pfb->tags, 1, tags, tags, 1, &tile->tag)) {
+		if (flags & 2) tile->zcomp |= 0x01000000; /* Z16 */
+		else           tile->zcomp |= 0x02000000; /* Z24S8 */
+		tile->zcomp |= ((tile->tag->offset           ) >> 6);
+		tile->zcomp |= ((tile->tag->offset + tags - 1) >> 6) << 12;
+#ifdef __BIG_ENDIAN
+		tile->zcomp |= 0x10000000;
+#endif
+	}
+}
+
+static int
+calc_bias(struct nv30_fb_priv *priv, int k, int i, int j)
+{
+	struct nouveau_device *device = nv_device(priv);
+	int b = (device->chipset > 0x30 ?
+		 nv_rd32(priv, 0x122c + 0x10 * k + 0x4 * j) >> (4 * (i ^ 1)) :
+		 0) & 0xf;
+
+	return 2 * (b & 0x8 ? b - 0x10 : b);
+}
+
+static int
+calc_ref(struct nv30_fb_priv *priv, int l, int k, int i)
+{
+	int j, x = 0;
+
+	for (j = 0; j < 4; j++) {
+		int m = (l >> (8 * i) & 0xff) + calc_bias(priv, k, i, j);
+
+		x |= (0x80 | clamp(m, 0, 0x1f)) << (8 * j);
+	}
+
+	return x;
+}
+
+int
+nv30_fb_init(struct nouveau_object *object)
+{
+	struct nouveau_device *device = nv_device(object);
+	struct nv30_fb_priv *priv = (void *)object;
+	int ret, i, j;
+
+	ret = nouveau_fb_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* Init the memory timing regs at 0x10037c/0x1003ac */
+	if (device->chipset == 0x30 ||
+	    device->chipset == 0x31 ||
+	    device->chipset == 0x35) {
+		/* Related to ROP count */
+		int n = (device->chipset == 0x31 ? 2 : 4);
+		int l = nv_rd32(priv, 0x1003d0);
+
+		for (i = 0; i < n; i++) {
+			for (j = 0; j < 3; j++)
+				nv_wr32(priv, 0x10037c + 0xc * i + 0x4 * j,
+					calc_ref(priv, l, 0, j));
+
+			for (j = 0; j < 2; j++)
+				nv_wr32(priv, 0x1003ac + 0x8 * i + 0x4 * j,
+					calc_ref(priv, l, 1, j));
+		}
+	}
+
+	return 0;
+}
+
+static int
+nv30_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv30_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv20_fb_vram_init;
+	priv->base.tile.regions = 8;
+	priv->base.tile.init = nv30_fb_tile_init;
+	priv->base.tile.comp = nv30_fb_tile_comp;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv20_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+struct nouveau_oclass
+nv30_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x30),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv30_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv30_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv35.c b/nvkm/subdev/fb/nv35.c
new file mode 100644
index 000000000..092f6f4f3
--- /dev/null
+++ b/nvkm/subdev/fb/nv35.c
@@ -0,0 +1,82 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv35_fb_priv {
+	struct nouveau_fb base;
+};
+
+static void
+nv35_fb_tile_comp(struct nouveau_fb *pfb, int i, u32 size, u32 flags,
+		  struct nouveau_fb_tile *tile)
+{
+	u32 tiles = DIV_ROUND_UP(size, 0x40);
+	u32 tags  = round_up(tiles / pfb->ram.parts, 0x40);
+	if (!nouveau_mm_head(&pfb->tags, 1, tags, tags, 1, &tile->tag)) {
+		if (flags & 2) tile->zcomp |= 0x04000000; /* Z16 */
+		else           tile->zcomp |= 0x08000000; /* Z24S8 */
+		tile->zcomp |= ((tile->tag->offset           ) >> 6);
+		tile->zcomp |= ((tile->tag->offset + tags - 1) >> 6) << 13;
+#ifdef __BIG_ENDIAN
+		tile->zcomp |= 0x40000000;
+#endif
+	}
+}
+
+static int
+nv35_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv35_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv20_fb_vram_init;
+	priv->base.tile.regions = 8;
+	priv->base.tile.init = nv30_fb_tile_init;
+	priv->base.tile.comp = nv35_fb_tile_comp;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv20_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+struct nouveau_oclass
+nv35_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x35),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv35_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv30_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv36.c b/nvkm/subdev/fb/nv36.c
new file mode 100644
index 000000000..797ab3b82
--- /dev/null
+++ b/nvkm/subdev/fb/nv36.c
@@ -0,0 +1,82 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv36_fb_priv {
+	struct nouveau_fb base;
+};
+
+static void
+nv36_fb_tile_comp(struct nouveau_fb *pfb, int i, u32 size, u32 flags,
+		  struct nouveau_fb_tile *tile)
+{
+	u32 tiles = DIV_ROUND_UP(size, 0x40);
+	u32 tags  = round_up(tiles / pfb->ram.parts, 0x40);
+	if (!nouveau_mm_head(&pfb->tags, 1, tags, tags, 1, &tile->tag)) {
+		if (flags & 2) tile->zcomp |= 0x10000000; /* Z16 */
+		else           tile->zcomp |= 0x20000000; /* Z24S8 */
+		tile->zcomp |= ((tile->tag->offset           ) >> 6);
+		tile->zcomp |= ((tile->tag->offset + tags - 1) >> 6) << 14;
+#ifdef __BIG_ENDIAN
+		tile->zcomp |= 0x80000000;
+#endif
+	}
+}
+
+static int
+nv36_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv36_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv20_fb_vram_init;
+	priv->base.tile.regions = 8;
+	priv->base.tile.init = nv30_fb_tile_init;
+	priv->base.tile.comp = nv36_fb_tile_comp;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv20_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+struct nouveau_oclass
+nv36_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x36),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv36_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv30_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv40.c b/nvkm/subdev/fb/nv40.c
new file mode 100644
index 000000000..65e131b90
--- /dev/null
+++ b/nvkm/subdev/fb/nv40.c
@@ -0,0 +1,113 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv40_fb_priv {
+	struct nouveau_fb base;
+};
+
+static int
+nv40_fb_vram_init(struct nouveau_fb *pfb)
+{
+	u32 pbus1218 = nv_rd32(pfb, 0x001218);
+	switch (pbus1218 & 0x00000300) {
+	case 0x00000000: pfb->ram.type = NV_MEM_TYPE_SDRAM; break;
+	case 0x00000100: pfb->ram.type = NV_MEM_TYPE_DDR1; break;
+	case 0x00000200: pfb->ram.type = NV_MEM_TYPE_GDDR3; break;
+	case 0x00000300: pfb->ram.type = NV_MEM_TYPE_DDR2; break;
+	}
+
+	pfb->ram.size  =  nv_rd32(pfb, 0x10020c) & 0xff000000;
+	pfb->ram.parts = (nv_rd32(pfb, 0x100200) & 0x00000003) + 1;
+	return nv_rd32(pfb, 0x100320);
+}
+
+void
+nv40_fb_tile_comp(struct nouveau_fb *pfb, int i, u32 size, u32 flags,
+		  struct nouveau_fb_tile *tile)
+{
+	u32 tiles = DIV_ROUND_UP(size, 0x80);
+	u32 tags  = round_up(tiles / pfb->ram.parts, 0x100);
+	if ( (flags & 2) &&
+	    !nouveau_mm_head(&pfb->tags, 1, tags, tags, 1, &tile->tag)) {
+		tile->zcomp  = 0x28000000; /* Z24S8_SPLIT_GRAD */
+		tile->zcomp |= ((tile->tag->offset           ) >> 8);
+		tile->zcomp |= ((tile->tag->offset + tags - 1) >> 8) << 13;
+#ifdef __BIG_ENDIAN
+		tile->zcomp |= 0x40000000;
+#endif
+	}
+}
+
+static int
+nv40_fb_init(struct nouveau_object *object)
+{
+	struct nv40_fb_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_fb_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_mask(priv, 0x10033c, 0x00008000, 0x00000000);
+	return 0;
+}
+
+static int
+nv40_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv40_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv40_fb_vram_init;
+	priv->base.tile.regions = 8;
+	priv->base.tile.init = nv30_fb_tile_init;
+	priv->base.tile.comp = nv40_fb_tile_comp;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv20_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+
+struct nouveau_oclass
+nv40_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv40_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv40_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv41.c b/nvkm/subdev/fb/nv41.c
new file mode 100644
index 000000000..e9e5a08c4
--- /dev/null
+++ b/nvkm/subdev/fb/nv41.c
@@ -0,0 +1,106 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv41_fb_priv {
+	struct nouveau_fb base;
+};
+
+int
+nv41_fb_vram_init(struct nouveau_fb *pfb)
+{
+	u32 pfb474 = nv_rd32(pfb, 0x100474);
+	if (pfb474 & 0x00000004)
+		pfb->ram.type = NV_MEM_TYPE_GDDR3;
+	if (pfb474 & 0x00000002)
+		pfb->ram.type = NV_MEM_TYPE_DDR2;
+	if (pfb474 & 0x00000001)
+		pfb->ram.type = NV_MEM_TYPE_DDR1;
+
+	pfb->ram.size =   nv_rd32(pfb, 0x10020c) & 0xff000000;
+	pfb->ram.parts = (nv_rd32(pfb, 0x100200) & 0x00000003) + 1;
+	return nv_rd32(pfb, 0x100320);
+}
+
+void
+nv41_fb_tile_prog(struct nouveau_fb *pfb, int i, struct nouveau_fb_tile *tile)
+{
+	nv_wr32(pfb, 0x100604 + (i * 0x10), tile->limit);
+	nv_wr32(pfb, 0x100608 + (i * 0x10), tile->pitch);
+	nv_wr32(pfb, 0x100600 + (i * 0x10), tile->addr);
+	nv_rd32(pfb, 0x100600 + (i * 0x10));
+	nv_wr32(pfb, 0x100700 + (i * 0x04), tile->zcomp);
+}
+
+int
+nv41_fb_init(struct nouveau_object *object)
+{
+	struct nv41_fb_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_fb_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x100800, 0x00000001);
+	return 0;
+}
+
+static int
+nv41_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv41_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv41_fb_vram_init;
+	priv->base.tile.regions = 12;
+	priv->base.tile.init = nv30_fb_tile_init;
+	priv->base.tile.comp = nv40_fb_tile_comp;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv41_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+
+struct nouveau_oclass
+nv41_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x41),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv41_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv41_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv44.c b/nvkm/subdev/fb/nv44.c
new file mode 100644
index 000000000..ae89b5006
--- /dev/null
+++ b/nvkm/subdev/fb/nv44.c
@@ -0,0 +1,114 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv44_fb_priv {
+	struct nouveau_fb base;
+};
+
+int
+nv44_fb_vram_init(struct nouveau_fb *pfb)
+{
+	u32 pfb474 = nv_rd32(pfb, 0x100474);
+	if (pfb474 & 0x00000004)
+		pfb->ram.type = NV_MEM_TYPE_GDDR3;
+	if (pfb474 & 0x00000002)
+		pfb->ram.type = NV_MEM_TYPE_DDR2;
+	if (pfb474 & 0x00000001)
+		pfb->ram.type = NV_MEM_TYPE_DDR1;
+
+	pfb->ram.size = nv_rd32(pfb, 0x10020c) & 0xff000000;
+	return 0;
+}
+
+static void
+nv44_fb_tile_init(struct nouveau_fb *pfb, int i, u32 addr, u32 size, u32 pitch,
+		  u32 flags, struct nouveau_fb_tile *tile)
+{
+	tile->addr  = 0x00000001; /* mode = vram */
+	tile->addr |= addr;
+	tile->limit = max(1u, addr + size) - 1;
+	tile->pitch = pitch;
+}
+
+void
+nv44_fb_tile_prog(struct nouveau_fb *pfb, int i, struct nouveau_fb_tile *tile)
+{
+	nv_wr32(pfb, 0x100604 + (i * 0x10), tile->limit);
+	nv_wr32(pfb, 0x100608 + (i * 0x10), tile->pitch);
+	nv_wr32(pfb, 0x100600 + (i * 0x10), tile->addr);
+	nv_rd32(pfb, 0x100600 + (i * 0x10));
+}
+
+int
+nv44_fb_init(struct nouveau_object *object)
+{
+	struct nv44_fb_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_fb_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x100850, 0x80000000);
+	nv_wr32(priv, 0x100800, 0x00000001);
+	return 0;
+}
+
+static int
+nv44_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv44_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv44_fb_vram_init;
+	priv->base.tile.regions = 12;
+	priv->base.tile.init = nv44_fb_tile_init;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv44_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+
+struct nouveau_oclass
+nv44_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x44),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv44_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv44_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv46.c b/nvkm/subdev/fb/nv46.c
new file mode 100644
index 000000000..589b93ea2
--- /dev/null
+++ b/nvkm/subdev/fb/nv46.c
@@ -0,0 +1,79 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv46_fb_priv {
+	struct nouveau_fb base;
+};
+
+void
+nv46_fb_tile_init(struct nouveau_fb *pfb, int i, u32 addr, u32 size, u32 pitch,
+		  u32 flags, struct nouveau_fb_tile *tile)
+{
+	/* for performance, select alternate bank offset for zeta */
+	if (!(flags & 4)) tile->addr = (0 << 3);
+	else              tile->addr = (1 << 3);
+
+	tile->addr |= 0x00000001; /* mode = vram */
+	tile->addr |= addr;
+	tile->limit = max(1u, addr + size) - 1;
+	tile->pitch = pitch;
+}
+
+static int
+nv46_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv46_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv44_fb_vram_init;
+	priv->base.tile.regions = 15;
+	priv->base.tile.init = nv46_fb_tile_init;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv44_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+
+struct nouveau_oclass
+nv46_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x46),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv46_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv44_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv47.c b/nvkm/subdev/fb/nv47.c
new file mode 100644
index 000000000..818bba35b
--- /dev/null
+++ b/nvkm/subdev/fb/nv47.c
@@ -0,0 +1,66 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv47_fb_priv {
+	struct nouveau_fb base;
+};
+
+static int
+nv47_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv47_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv41_fb_vram_init;
+	priv->base.tile.regions = 15;
+	priv->base.tile.init = nv30_fb_tile_init;
+	priv->base.tile.comp = nv40_fb_tile_comp;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv41_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+
+struct nouveau_oclass
+nv47_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x47),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv47_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv41_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv49.c b/nvkm/subdev/fb/nv49.c
new file mode 100644
index 000000000..84a31af16
--- /dev/null
+++ b/nvkm/subdev/fb/nv49.c
@@ -0,0 +1,84 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv49_fb_priv {
+	struct nouveau_fb base;
+};
+
+static int
+nv49_fb_vram_init(struct nouveau_fb *pfb)
+{
+	u32 pfb914 = nv_rd32(pfb, 0x100914);
+
+	switch (pfb914 & 0x00000003) {
+	case 0x00000000: pfb->ram.type = NV_MEM_TYPE_DDR1; break;
+	case 0x00000001: pfb->ram.type = NV_MEM_TYPE_DDR2; break;
+	case 0x00000002: pfb->ram.type = NV_MEM_TYPE_GDDR3; break;
+	case 0x00000003: break;
+	}
+
+	pfb->ram.size =   nv_rd32(pfb, 0x10020c) & 0xff000000;
+	pfb->ram.parts = (nv_rd32(pfb, 0x100200) & 0x00000003) + 1;
+	return nv_rd32(pfb, 0x100320);
+}
+
+static int
+nv49_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv49_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv49_fb_vram_init;
+	priv->base.tile.regions = 15;
+	priv->base.tile.init = nv30_fb_tile_init;
+	priv->base.tile.comp = nv40_fb_tile_comp;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv41_fb_tile_prog;
+
+	return nouveau_fb_preinit(&priv->base);
+}
+
+
+struct nouveau_oclass
+nv49_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x49),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv49_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv41_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv4e.c b/nvkm/subdev/fb/nv4e.c
new file mode 100644
index 000000000..797fd5581
--- /dev/null
+++ b/nvkm/subdev/fb/nv4e.c
@@ -0,0 +1,72 @@
+/*
+ * Copyright (C) 2010 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/fb.h>
+
+struct nv4e_fb_priv {
+	struct nouveau_fb base;
+};
+
+static int
+nv4e_fb_vram_init(struct nouveau_fb *pfb)
+{
+	pfb->ram.size = nv_rd32(pfb, 0x10020c) & 0xff000000;
+	pfb->ram.type = NV_MEM_TYPE_STOLEN;
+	return 0;
+}
+
+static int
+nv4e_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv4e_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nv04_fb_memtype_valid;
+	priv->base.ram.init = nv4e_fb_vram_init;
+	priv->base.tile.regions = 12;
+	priv->base.tile.init = nv46_fb_tile_init;
+	priv->base.tile.fini = nv20_fb_tile_fini;
+	priv->base.tile.prog = nv44_fb_tile_prog;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+struct nouveau_oclass
+nv4e_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x4e),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv4e_fb_ctor,
+		.dtor = _nouveau_fb_dtor,
+		.init = nv44_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nv50.c b/nvkm/subdev/fb/nv50.c
new file mode 100644
index 000000000..487cb8c6c
--- /dev/null
+++ b/nvkm/subdev/fb/nv50.c
@@ -0,0 +1,497 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/object.h>
+#include <core/enum.h>
+
+#include <subdev/fb.h>
+#include <subdev/bios.h>
+
+struct nv50_fb_priv {
+	struct nouveau_fb base;
+	struct page *r100c08_page;
+	dma_addr_t r100c08;
+};
+
+static int types[0x80] = {
+	1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	1, 1, 1, 1, 0, 0, 0, 0, 2, 2, 2, 2, 0, 0, 0, 0,
+	1, 1, 1, 1, 1, 1, 1, 0, 2, 2, 2, 2, 2, 2, 2, 0,
+	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 0, 0,
+	0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 2, 2, 2, 2,
+	1, 0, 2, 0, 1, 0, 2, 0, 1, 1, 2, 2, 1, 1, 0, 0
+};
+
+static bool
+nv50_fb_memtype_valid(struct nouveau_fb *pfb, u32 memtype)
+{
+	return types[(memtype & 0xff00) >> 8] != 0;
+}
+
+static u32
+nv50_fb_vram_rblock(struct nouveau_fb *pfb)
+{
+	int i, parts, colbits, rowbitsa, rowbitsb, banks;
+	u64 rowsize, predicted;
+	u32 r0, r4, rt, ru, rblock_size;
+
+	r0 = nv_rd32(pfb, 0x100200);
+	r4 = nv_rd32(pfb, 0x100204);
+	rt = nv_rd32(pfb, 0x100250);
+	ru = nv_rd32(pfb, 0x001540);
+	nv_debug(pfb, "memcfg 0x%08x 0x%08x 0x%08x 0x%08x\n", r0, r4, rt, ru);
+
+	for (i = 0, parts = 0; i < 8; i++) {
+		if (ru & (0x00010000 << i))
+			parts++;
+	}
+
+	colbits  =  (r4 & 0x0000f000) >> 12;
+	rowbitsa = ((r4 & 0x000f0000) >> 16) + 8;
+	rowbitsb = ((r4 & 0x00f00000) >> 20) + 8;
+	banks    = 1 << (((r4 & 0x03000000) >> 24) + 2);
+
+	rowsize = parts * banks * (1 << colbits) * 8;
+	predicted = rowsize << rowbitsa;
+	if (r0 & 0x00000004)
+		predicted += rowsize << rowbitsb;
+
+	if (predicted != pfb->ram.size) {
+		nv_warn(pfb, "memory controller reports %d MiB VRAM\n",
+			(u32)(pfb->ram.size >> 20));
+	}
+
+	rblock_size = rowsize;
+	if (rt & 1)
+		rblock_size *= 3;
+
+	nv_debug(pfb, "rblock %d bytes\n", rblock_size);
+	return rblock_size;
+}
+
+static int
+nv50_fb_vram_init(struct nouveau_fb *pfb)
+{
+	struct nouveau_device *device = nv_device(pfb);
+	struct nouveau_bios *bios = nouveau_bios(device);
+	const u32 rsvd_head = ( 256 * 1024) >> 12; /* vga memory */
+	const u32 rsvd_tail = (1024 * 1024) >> 12; /* vbios etc */
+	u32 size;
+	int ret;
+
+	pfb->ram.size = nv_rd32(pfb, 0x10020c);
+	pfb->ram.size = (pfb->ram.size & 0xffffff00) |
+		       ((pfb->ram.size & 0x000000ff) << 32);
+
+	size = (pfb->ram.size >> 12) - rsvd_head - rsvd_tail;
+	switch (device->chipset) {
+	case 0xaa:
+	case 0xac:
+	case 0xaf: /* IGPs, no reordering, no real VRAM */
+		ret = nouveau_mm_init(&pfb->vram, rsvd_head, size, 1);
+		if (ret)
+			return ret;
+
+		pfb->ram.type   = NV_MEM_TYPE_STOLEN;
+		pfb->ram.stolen = (u64)nv_rd32(pfb, 0x100e10) << 12;
+		break;
+	default:
+		switch (nv_rd32(pfb, 0x100714) & 0x00000007) {
+		case 0: pfb->ram.type = NV_MEM_TYPE_DDR1; break;
+		case 1:
+			if (nouveau_fb_bios_memtype(bios) == NV_MEM_TYPE_DDR3)
+				pfb->ram.type = NV_MEM_TYPE_DDR3;
+			else
+				pfb->ram.type = NV_MEM_TYPE_DDR2;
+			break;
+		case 2: pfb->ram.type = NV_MEM_TYPE_GDDR3; break;
+		case 3: pfb->ram.type = NV_MEM_TYPE_GDDR4; break;
+		case 4: pfb->ram.type = NV_MEM_TYPE_GDDR5; break;
+		default:
+			break;
+		}
+
+		ret = nouveau_mm_init(&pfb->vram, rsvd_head, size,
+				      nv50_fb_vram_rblock(pfb) >> 12);
+		if (ret)
+			return ret;
+
+		pfb->ram.ranks = (nv_rd32(pfb, 0x100200) & 0x4) ? 2 : 1;
+		break;
+	}
+
+	return nv_rd32(pfb, 0x100320);
+}
+
+static int
+nv50_fb_vram_new(struct nouveau_fb *pfb, u64 size, u32 align, u32 ncmin,
+		 u32 memtype, struct nouveau_mem **pmem)
+{
+	struct nv50_fb_priv *priv = (void *)pfb;
+	struct nouveau_mm *heap = &priv->base.vram;
+	struct nouveau_mm *tags = &priv->base.tags;
+	struct nouveau_mm_node *r;
+	struct nouveau_mem *mem;
+	int comp = (memtype & 0x300) >> 8;
+	int type = (memtype & 0x07f);
+	int back = (memtype & 0x800);
+	int min, max, ret;
+
+	max = (size >> 12);
+	min = ncmin ? (ncmin >> 12) : max;
+	align >>= 12;
+
+	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
+	if (!mem)
+		return -ENOMEM;
+
+	mutex_lock(&pfb->base.mutex);
+	if (comp) {
+		if (align == 16) {
+			int n = (max >> 4) * comp;
+
+			ret = nouveau_mm_head(tags, 1, n, n, 1, &mem->tag);
+			if (ret)
+				mem->tag = NULL;
+		}
+
+		if (unlikely(!mem->tag))
+			comp = 0;
+	}
+
+	INIT_LIST_HEAD(&mem->regions);
+	mem->memtype = (comp << 7) | type;
+	mem->size = max;
+
+	type = types[type];
+	do {
+		if (back)
+			ret = nouveau_mm_tail(heap, type, max, min, align, &r);
+		else
+			ret = nouveau_mm_head(heap, type, max, min, align, &r);
+		if (ret) {
+			mutex_unlock(&pfb->base.mutex);
+			pfb->ram.put(pfb, &mem);
+			return ret;
+		}
+
+		list_add_tail(&r->rl_entry, &mem->regions);
+		max -= r->length;
+	} while (max);
+	mutex_unlock(&pfb->base.mutex);
+
+	r = list_first_entry(&mem->regions, struct nouveau_mm_node, rl_entry);
+	mem->offset = (u64)r->offset << 12;
+	*pmem = mem;
+	return 0;
+}
+
+void
+nv50_fb_vram_del(struct nouveau_fb *pfb, struct nouveau_mem **pmem)
+{
+	struct nv50_fb_priv *priv = (void *)pfb;
+	struct nouveau_mm_node *this;
+	struct nouveau_mem *mem;
+
+	mem = *pmem;
+	*pmem = NULL;
+	if (unlikely(mem == NULL))
+		return;
+
+	mutex_lock(&pfb->base.mutex);
+	while (!list_empty(&mem->regions)) {
+		this = list_first_entry(&mem->regions, typeof(*this), rl_entry);
+
+		list_del(&this->rl_entry);
+		nouveau_mm_free(&priv->base.vram, &this);
+	}
+
+	nouveau_mm_free(&priv->base.tags, &mem->tag);
+	mutex_unlock(&pfb->base.mutex);
+
+	kfree(mem);
+}
+
+static const struct nouveau_enum vm_dispatch_subclients[] = {
+	{ 0x00000000, "GRCTX", NULL },
+	{ 0x00000001, "NOTIFY", NULL },
+	{ 0x00000002, "QUERY", NULL },
+	{ 0x00000003, "COND", NULL },
+	{ 0x00000004, "M2M_IN", NULL },
+	{ 0x00000005, "M2M_OUT", NULL },
+	{ 0x00000006, "M2M_NOTIFY", NULL },
+	{}
+};
+
+static const struct nouveau_enum vm_ccache_subclients[] = {
+	{ 0x00000000, "CB", NULL },
+	{ 0x00000001, "TIC", NULL },
+	{ 0x00000002, "TSC", NULL },
+	{}
+};
+
+static const struct nouveau_enum vm_prop_subclients[] = {
+	{ 0x00000000, "RT0", NULL },
+	{ 0x00000001, "RT1", NULL },
+	{ 0x00000002, "RT2", NULL },
+	{ 0x00000003, "RT3", NULL },
+	{ 0x00000004, "RT4", NULL },
+	{ 0x00000005, "RT5", NULL },
+	{ 0x00000006, "RT6", NULL },
+	{ 0x00000007, "RT7", NULL },
+	{ 0x00000008, "ZETA", NULL },
+	{ 0x00000009, "LOCAL", NULL },
+	{ 0x0000000a, "GLOBAL", NULL },
+	{ 0x0000000b, "STACK", NULL },
+	{ 0x0000000c, "DST2D", NULL },
+	{}
+};
+
+static const struct nouveau_enum vm_pfifo_subclients[] = {
+	{ 0x00000000, "PUSHBUF", NULL },
+	{ 0x00000001, "SEMAPHORE", NULL },
+	{}
+};
+
+static const struct nouveau_enum vm_bar_subclients[] = {
+	{ 0x00000000, "FB", NULL },
+	{ 0x00000001, "IN", NULL },
+	{}
+};
+
+static const struct nouveau_enum vm_client[] = {
+	{ 0x00000000, "STRMOUT", NULL },
+	{ 0x00000003, "DISPATCH", vm_dispatch_subclients },
+	{ 0x00000004, "PFIFO_WRITE", NULL },
+	{ 0x00000005, "CCACHE", vm_ccache_subclients },
+	{ 0x00000006, "PPPP", NULL },
+	{ 0x00000007, "CLIPID", NULL },
+	{ 0x00000008, "PFIFO_READ", NULL },
+	{ 0x00000009, "VFETCH", NULL },
+	{ 0x0000000a, "TEXTURE", NULL },
+	{ 0x0000000b, "PROP", vm_prop_subclients },
+	{ 0x0000000c, "PVP", NULL },
+	{ 0x0000000d, "PBSP", NULL },
+	{ 0x0000000e, "PCRYPT", NULL },
+	{ 0x0000000f, "PCOUNTER", NULL },
+	{ 0x00000011, "PDAEMON", NULL },
+	{}
+};
+
+static const struct nouveau_enum vm_engine[] = {
+	{ 0x00000000, "PGRAPH", NULL },
+	{ 0x00000001, "PVP", NULL },
+	{ 0x00000004, "PEEPHOLE", NULL },
+	{ 0x00000005, "PFIFO", vm_pfifo_subclients },
+	{ 0x00000006, "BAR", vm_bar_subclients },
+	{ 0x00000008, "PPPP", NULL },
+	{ 0x00000009, "PBSP", NULL },
+	{ 0x0000000a, "PCRYPT", NULL },
+	{ 0x0000000b, "PCOUNTER", NULL },
+	{ 0x0000000c, "SEMAPHORE_BG", NULL },
+	{ 0x0000000d, "PCOPY", NULL },
+	{ 0x0000000e, "PDAEMON", NULL },
+	{}
+};
+
+static const struct nouveau_enum vm_fault[] = {
+	{ 0x00000000, "PT_NOT_PRESENT", NULL },
+	{ 0x00000001, "PT_TOO_SHORT", NULL },
+	{ 0x00000002, "PAGE_NOT_PRESENT", NULL },
+	{ 0x00000003, "PAGE_SYSTEM_ONLY", NULL },
+	{ 0x00000004, "PAGE_READ_ONLY", NULL },
+	{ 0x00000006, "NULL_DMAOBJ", NULL },
+	{ 0x00000007, "WRONG_MEMTYPE", NULL },
+	{ 0x0000000b, "VRAM_LIMIT", NULL },
+	{ 0x0000000f, "DMAOBJ_LIMIT", NULL },
+	{}
+};
+
+static void
+nv50_fb_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_device *device = nv_device(subdev);
+	struct nv50_fb_priv *priv = (void *)subdev;
+	const struct nouveau_enum *en, *cl;
+	u32 trap[6], idx, chan;
+	u8 st0, st1, st2, st3;
+	int i;
+
+	idx = nv_rd32(priv, 0x100c90);
+	if (!(idx & 0x80000000))
+		return;
+	idx &= 0x00ffffff;
+
+	for (i = 0; i < 6; i++) {
+		nv_wr32(priv, 0x100c90, idx | i << 24);
+		trap[i] = nv_rd32(priv, 0x100c94);
+	}
+	nv_wr32(priv, 0x100c90, idx | 0x80000000);
+
+	/* decode status bits into something more useful */
+	if (device->chipset  < 0xa3 ||
+	    device->chipset == 0xaa || device->chipset == 0xac) {
+		st0 = (trap[0] & 0x0000000f) >> 0;
+		st1 = (trap[0] & 0x000000f0) >> 4;
+		st2 = (trap[0] & 0x00000f00) >> 8;
+		st3 = (trap[0] & 0x0000f000) >> 12;
+	} else {
+		st0 = (trap[0] & 0x000000ff) >> 0;
+		st1 = (trap[0] & 0x0000ff00) >> 8;
+		st2 = (trap[0] & 0x00ff0000) >> 16;
+		st3 = (trap[0] & 0xff000000) >> 24;
+	}
+	chan = (trap[2] << 16) | trap[1];
+
+	nv_error(priv, "trapped %s at 0x%02x%04x%04x on channel 0x%08x ",
+		 (trap[5] & 0x00000100) ? "read" : "write",
+		 trap[5] & 0xff, trap[4] & 0xffff, trap[3] & 0xffff, chan);
+
+	en = nouveau_enum_find(vm_engine, st0);
+	if (en)
+		printk("%s/", en->name);
+	else
+		printk("%02x/", st0);
+
+	cl = nouveau_enum_find(vm_client, st2);
+	if (cl)
+		printk("%s/", cl->name);
+	else
+		printk("%02x/", st2);
+
+	if      (cl && cl->data) cl = nouveau_enum_find(cl->data, st3);
+	else if (en && en->data) cl = nouveau_enum_find(en->data, st3);
+	else                     cl = NULL;
+	if (cl)
+		printk("%s", cl->name);
+	else
+		printk("%02x", st3);
+
+	printk(" reason: ");
+	en = nouveau_enum_find(vm_fault, st1);
+	if (en)
+		printk("%s\n", en->name);
+	else
+		printk("0x%08x\n", st1);
+}
+
+static int
+nv50_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nv50_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->r100c08_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+	if (priv->r100c08_page) {
+		priv->r100c08 = pci_map_page(device->pdev, priv->r100c08_page,
+					     0, PAGE_SIZE,
+					     PCI_DMA_BIDIRECTIONAL);
+		if (pci_dma_mapping_error(device->pdev, priv->r100c08))
+			nv_warn(priv, "failed 0x100c08 page map\n");
+	} else {
+		nv_warn(priv, "failed 0x100c08 page alloc\n");
+	}
+
+	priv->base.memtype_valid = nv50_fb_memtype_valid;
+	priv->base.ram.init = nv50_fb_vram_init;
+	priv->base.ram.get = nv50_fb_vram_new;
+	priv->base.ram.put = nv50_fb_vram_del;
+	nv_subdev(priv)->intr = nv50_fb_intr;
+	return nouveau_fb_preinit(&priv->base);
+}
+
+static void
+nv50_fb_dtor(struct nouveau_object *object)
+{
+	struct nouveau_device *device = nv_device(object);
+	struct nv50_fb_priv *priv = (void *)object;
+
+	if (priv->r100c08_page) {
+		pci_unmap_page(device->pdev, priv->r100c08, PAGE_SIZE,
+			       PCI_DMA_BIDIRECTIONAL);
+		__free_page(priv->r100c08_page);
+	}
+
+	nouveau_fb_destroy(&priv->base);
+}
+
+static int
+nv50_fb_init(struct nouveau_object *object)
+{
+	struct nouveau_device *device = nv_device(object);
+	struct nv50_fb_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_fb_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* Not a clue what this is exactly.  Without pointing it at a
+	 * scratch page, VRAM->GART blits with M2MF (as in DDX DFS)
+	 * cause IOMMU "read from address 0" errors (rh#561267)
+	 */
+	nv_wr32(priv, 0x100c08, priv->r100c08 >> 8);
+
+	/* This is needed to get meaningful information from 100c90
+	 * on traps. No idea what these values mean exactly. */
+	switch (device->chipset) {
+	case 0x50:
+		nv_wr32(priv, 0x100c90, 0x000707ff);
+		break;
+	case 0xa3:
+	case 0xa5:
+	case 0xa8:
+		nv_wr32(priv, 0x100c90, 0x000d0fff);
+		break;
+	case 0xaf:
+		nv_wr32(priv, 0x100c90, 0x089d1fff);
+		break;
+	default:
+		nv_wr32(priv, 0x100c90, 0x001d07ff);
+		break;
+	}
+
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_fb_ctor,
+		.dtor = nv50_fb_dtor,
+		.init = nv50_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/fb/nvc0.c b/nvkm/subdev/fb/nvc0.c
new file mode 100644
index 000000000..306bdf121
--- /dev/null
+++ b/nvkm/subdev/fb/nvc0.c
@@ -0,0 +1,241 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/fb.h>
+#include <subdev/bios.h>
+
+struct nvc0_fb_priv {
+	struct nouveau_fb base;
+	struct page *r100c10_page;
+	dma_addr_t r100c10;
+};
+
+/* 0 = unsupported
+ * 1 = non-compressed
+ * 3 = compressed
+ */
+static const u8 types[256] = {
+	1, 1, 3, 3, 3, 3, 0, 3, 3, 3, 3, 0, 0, 0, 0, 0,
+	0, 1, 0, 0, 0, 0, 0, 3, 3, 3, 3, 0, 0, 0, 0, 0,
+	0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0,
+	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 3, 3,
+	3, 3, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	0, 3, 3, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0,
+	0, 0, 0, 0, 0, 0, 3, 3, 3, 3, 0, 1, 1, 1, 1, 0,
+	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	0, 0, 0, 3, 3, 3, 3, 1, 1, 1, 1, 0, 0, 0, 0, 0,
+	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3,
+	3, 3, 3, 1, 0, 0, 0, 0, 0, 0, 0, 0, 3, 3, 3, 3,
+	3, 3, 0, 0, 0, 0, 0, 0, 3, 0, 0, 3, 0, 3, 0, 3,
+	3, 0, 3, 3, 3, 3, 3, 0, 0, 3, 0, 3, 0, 3, 3, 0,
+	3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 0, 1, 1, 0
+};
+
+static bool
+nvc0_fb_memtype_valid(struct nouveau_fb *pfb, u32 tile_flags)
+{
+	u8 memtype = (tile_flags & 0x0000ff00) >> 8;
+	return likely((types[memtype] == 1));
+}
+
+static int
+nvc0_fb_vram_init(struct nouveau_fb *pfb)
+{
+	struct nouveau_bios *bios = nouveau_bios(pfb);
+	const u32 rsvd_head = ( 256 * 1024) >> 12; /* vga memory */
+	const u32 rsvd_tail = (1024 * 1024) >> 12; /* vbios etc */
+	u32 parts = nv_rd32(pfb, 0x022438);
+	u32 pmask = nv_rd32(pfb, 0x022554);
+	u32 bsize = nv_rd32(pfb, 0x10f20c);
+	u32 offset, length;
+	bool uniform = true;
+	int ret, part;
+
+	nv_debug(pfb, "0x100800: 0x%08x\n", nv_rd32(pfb, 0x100800));
+	nv_debug(pfb, "parts 0x%08x mask 0x%08x\n", parts, pmask);
+
+	pfb->ram.type = nouveau_fb_bios_memtype(bios);
+	pfb->ram.ranks = (nv_rd32(pfb, 0x10f200) & 0x00000004) ? 2 : 1;
+
+	/* read amount of vram attached to each memory controller */
+	for (part = 0; part < parts; part++) {
+		if (!(pmask & (1 << part))) {
+			u32 psize = nv_rd32(pfb, 0x11020c + (part * 0x1000));
+			if (psize != bsize) {
+				if (psize < bsize)
+					bsize = psize;
+				uniform = false;
+			}
+
+			nv_debug(pfb, "%d: mem_amount 0x%08x\n", part, psize);
+			pfb->ram.size += (u64)psize << 20;
+		}
+	}
+
+	/* if all controllers have the same amount attached, there's no holes */
+	if (uniform) {
+		offset = rsvd_head;
+		length = (pfb->ram.size >> 12) - rsvd_head - rsvd_tail;
+		return nouveau_mm_init(&pfb->vram, offset, length, 1);
+	}
+
+	/* otherwise, address lowest common amount from 0GiB */
+	ret = nouveau_mm_init(&pfb->vram, rsvd_head, (bsize << 8) * parts, 1);
+	if (ret)
+		return ret;
+
+	/* and the rest starting from (8GiB + common_size) */
+	offset = (0x0200000000ULL >> 12) + (bsize << 8);
+	length = (pfb->ram.size >> 12) - (bsize << 8) - rsvd_tail;
+
+	ret = nouveau_mm_init(&pfb->vram, offset, length, 0);
+	if (ret) {
+		nouveau_mm_fini(&pfb->vram);
+		return ret;
+	}
+
+	return 0;
+}
+
+static int
+nvc0_fb_vram_new(struct nouveau_fb *pfb, u64 size, u32 align, u32 ncmin,
+		 u32 memtype, struct nouveau_mem **pmem)
+{
+	struct nouveau_mm *mm = &pfb->vram;
+	struct nouveau_mm_node *r;
+	struct nouveau_mem *mem;
+	int type = (memtype & 0x0ff);
+	int back = (memtype & 0x800);
+	int ret;
+
+	size  >>= 12;
+	align >>= 12;
+	ncmin >>= 12;
+	if (!ncmin)
+		ncmin = size;
+
+	mem = kzalloc(sizeof(*mem), GFP_KERNEL);
+	if (!mem)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&mem->regions);
+	mem->memtype = type;
+	mem->size = size;
+
+	mutex_lock(&mm->mutex);
+	do {
+		if (back)
+			ret = nouveau_mm_tail(mm, 1, size, ncmin, align, &r);
+		else
+			ret = nouveau_mm_head(mm, 1, size, ncmin, align, &r);
+		if (ret) {
+			mutex_unlock(&mm->mutex);
+			pfb->ram.put(pfb, &mem);
+			return ret;
+		}
+
+		list_add_tail(&r->rl_entry, &mem->regions);
+		size -= r->length;
+	} while (size);
+	mutex_unlock(&mm->mutex);
+
+	r = list_first_entry(&mem->regions, struct nouveau_mm_node, rl_entry);
+	mem->offset = (u64)r->offset << 12;
+	*pmem = mem;
+	return 0;
+}
+
+static int
+nvc0_fb_init(struct nouveau_object *object)
+{
+	struct nvc0_fb_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_fb_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x100c10, priv->r100c10 >> 8);
+	return 0;
+}
+
+static void
+nvc0_fb_dtor(struct nouveau_object *object)
+{
+	struct nouveau_device *device = nv_device(object);
+	struct nvc0_fb_priv *priv = (void *)object;
+
+	if (priv->r100c10_page) {
+		pci_unmap_page(device->pdev, priv->r100c10, PAGE_SIZE,
+			       PCI_DMA_BIDIRECTIONAL);
+		__free_page(priv->r100c10_page);
+	}
+
+	nouveau_fb_destroy(&priv->base);
+}
+
+static int
+nvc0_fb_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nvc0_fb_priv *priv;
+	int ret;
+
+	ret = nouveau_fb_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.memtype_valid = nvc0_fb_memtype_valid;
+	priv->base.ram.init = nvc0_fb_vram_init;
+	priv->base.ram.get = nvc0_fb_vram_new;
+	priv->base.ram.put = nv50_fb_vram_del;
+
+	priv->r100c10_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+	if (!priv->r100c10_page)
+		return -ENOMEM;
+
+	priv->r100c10 = pci_map_page(device->pdev, priv->r100c10_page, 0,
+				     PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+	if (pci_dma_mapping_error(device->pdev, priv->r100c10))
+		return -EFAULT;
+
+	return nouveau_fb_preinit(&priv->base);
+}
+
+
+struct nouveau_oclass
+nvc0_fb_oclass = {
+	.handle = NV_SUBDEV(FB, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_fb_ctor,
+		.dtor = nvc0_fb_dtor,
+		.init = nvc0_fb_init,
+		.fini = _nouveau_fb_fini,
+	},
+};
diff --git a/nvkm/subdev/gpio/Makefile.am b/nvkm/subdev/gpio/Makefile.am
new file mode 100644
index 000000000..8a18de5bf
--- /dev/null
+++ b/nvkm/subdev/gpio/Makefile.am
@@ -0,0 +1,8 @@
+noinst_LTLIBRARIES = libgpio.la
+
+libgpio_la_SOURCES = base.c \
+		     nv10.c \
+		     nv50.c \
+		     nvd0.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/gpio/base.c b/nvkm/subdev/gpio/base.c
new file mode 100644
index 000000000..9fb0f9b92
--- /dev/null
+++ b/nvkm/subdev/gpio/base.c
@@ -0,0 +1,276 @@
+/*
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/gpio.h>
+#include <subdev/bios.h>
+#include <subdev/bios/gpio.h>
+
+static int
+nouveau_gpio_drive(struct nouveau_gpio *gpio,
+		   int idx, int line, int dir, int out)
+{
+	return gpio->drive ? gpio->drive(gpio, line, dir, out) : -ENODEV;
+}
+
+static int
+nouveau_gpio_sense(struct nouveau_gpio *gpio, int idx, int line)
+{
+	return gpio->sense ? gpio->sense(gpio, line) : -ENODEV;
+}
+
+static int
+nouveau_gpio_find(struct nouveau_gpio *gpio, int idx, u8 tag, u8 line,
+		  struct dcb_gpio_func *func)
+{
+	struct nouveau_bios *bios = nouveau_bios(gpio);
+	u8  ver, len;
+	u16 data;
+
+	if (line == 0xff && tag == 0xff)
+		return -EINVAL;
+
+	data = dcb_gpio_match(bios, idx, tag, line, &ver, &len, func);
+	if (data)
+		return 0;
+
+	/* Apple iMac G4 NV18 */
+	if (nv_device_match(nv_object(gpio), 0x0189, 0x10de, 0x0010)) {
+		if (tag == DCB_GPIO_TVDAC0) {
+			*func = (struct dcb_gpio_func) {
+				.func = DCB_GPIO_TVDAC0,
+				.line = 4,
+				.log[0] = 0,
+				.log[1] = 1,
+			};
+			return 0;
+		}
+	}
+
+	return -EINVAL;
+}
+
+static int
+nouveau_gpio_set(struct nouveau_gpio *gpio, int idx, u8 tag, u8 line, int state)
+{
+	struct dcb_gpio_func func;
+	int ret;
+
+	ret = nouveau_gpio_find(gpio, idx, tag, line, &func);
+	if (ret == 0) {
+		int dir = !!(func.log[state] & 0x02);
+		int out = !!(func.log[state] & 0x01);
+		ret = nouveau_gpio_drive(gpio, idx, func.line, dir, out);
+	}
+
+	return ret;
+}
+
+static int
+nouveau_gpio_get(struct nouveau_gpio *gpio, int idx, u8 tag, u8 line)
+{
+	struct dcb_gpio_func func;
+	int ret;
+
+	ret = nouveau_gpio_find(gpio, idx, tag, line, &func);
+	if (ret == 0) {
+		ret = nouveau_gpio_sense(gpio, idx, func.line);
+		if (ret >= 0)
+			ret = (ret == (func.log[1] & 1));
+	}
+
+	return ret;
+}
+
+static int
+nouveau_gpio_irq(struct nouveau_gpio *gpio, int idx, u8 tag, u8 line, bool on)
+{
+	struct dcb_gpio_func func;
+	int ret;
+
+	ret = nouveau_gpio_find(gpio, idx, tag, line, &func);
+	if (ret == 0) {
+		if (idx == 0 && gpio->irq_enable)
+			gpio->irq_enable(gpio, func.line, on);
+		else
+			ret = -ENODEV;
+	}
+
+	return ret;
+}
+
+struct gpio_isr {
+	struct nouveau_gpio *gpio;
+	struct list_head head;
+	struct work_struct work;
+	int idx;
+	struct dcb_gpio_func func;
+	void (*handler)(void *, int);
+	void *data;
+	bool inhibit;
+};
+
+static void
+nouveau_gpio_isr_bh(struct work_struct *work)
+{
+	struct gpio_isr *isr = container_of(work, struct gpio_isr, work);
+	struct nouveau_gpio *gpio = isr->gpio;
+	unsigned long flags;
+	int state;
+
+	state = nouveau_gpio_get(gpio, isr->idx, isr->func.func,
+						 isr->func.line);
+	if (state >= 0)
+		isr->handler(isr->data, state);
+
+	spin_lock_irqsave(&gpio->lock, flags);
+	isr->inhibit = false;
+	spin_unlock_irqrestore(&gpio->lock, flags);
+}
+
+static void
+nouveau_gpio_isr_run(struct nouveau_gpio *gpio, int idx, u32 line_mask)
+{
+	struct gpio_isr *isr;
+
+	if (idx != 0)
+		return;
+
+	spin_lock(&gpio->lock);
+	list_for_each_entry(isr, &gpio->isr, head) {
+		if (line_mask & (1 << isr->func.line)) {
+			if (isr->inhibit)
+				continue;
+			isr->inhibit = true;
+			schedule_work(&isr->work);
+		}
+	}
+	spin_unlock(&gpio->lock);
+}
+
+static int
+nouveau_gpio_isr_add(struct nouveau_gpio *gpio, int idx, u8 tag, u8 line,
+		     void (*handler)(void *, int), void *data)
+{
+	struct gpio_isr *isr;
+	unsigned long flags;
+	int ret;
+
+	isr = kzalloc(sizeof(*isr), GFP_KERNEL);
+	if (!isr)
+		return -ENOMEM;
+
+	ret = nouveau_gpio_find(gpio, idx, tag, line, &isr->func);
+	if (ret) {
+		kfree(isr);
+		return ret;
+	}
+
+	INIT_WORK(&isr->work, nouveau_gpio_isr_bh);
+	isr->gpio = gpio;
+	isr->handler = handler;
+	isr->data = data;
+	isr->idx = idx;
+
+	spin_lock_irqsave(&gpio->lock, flags);
+	list_add(&isr->head, &gpio->isr);
+	spin_unlock_irqrestore(&gpio->lock, flags);
+	return 0;
+}
+
+static void
+nouveau_gpio_isr_del(struct nouveau_gpio *gpio, int idx, u8 tag, u8 line,
+		     void (*handler)(void *, int), void *data)
+{
+	struct gpio_isr *isr, *tmp;
+	struct dcb_gpio_func func;
+	unsigned long flags;
+	LIST_HEAD(tofree);
+	int ret;
+
+	ret = nouveau_gpio_find(gpio, idx, tag, line, &func);
+	if (ret == 0) {
+		spin_lock_irqsave(&gpio->lock, flags);
+		list_for_each_entry_safe(isr, tmp, &gpio->isr, head) {
+			if (memcmp(&isr->func, &func, sizeof(func)) ||
+			    isr->idx != idx ||
+			    isr->handler != handler || isr->data != data)
+				continue;
+			list_move_tail(&isr->head, &tofree);
+		}
+		spin_unlock_irqrestore(&gpio->lock, flags);
+
+		list_for_each_entry_safe(isr, tmp, &tofree, head) {
+			flush_work(&isr->work);
+			kfree(isr);
+		}
+	}
+}
+
+int
+nouveau_gpio_create_(struct nouveau_object *parent,
+		     struct nouveau_object *engine,
+		     struct nouveau_oclass *oclass, int length, void **pobject)
+{
+	struct nouveau_gpio *gpio;
+	int ret;
+
+	ret = nouveau_subdev_create_(parent, engine, oclass, 0, "GPIO", "gpio",
+				     length, pobject);
+	gpio = *pobject;
+	if (ret)
+		return ret;
+
+	gpio->find = nouveau_gpio_find;
+	gpio->set  = nouveau_gpio_set;
+	gpio->get  = nouveau_gpio_get;
+	gpio->irq  = nouveau_gpio_irq;
+	gpio->isr_run = nouveau_gpio_isr_run;
+	gpio->isr_add = nouveau_gpio_isr_add;
+	gpio->isr_del = nouveau_gpio_isr_del;
+	INIT_LIST_HEAD(&gpio->isr);
+	spin_lock_init(&gpio->lock);
+	return 0;
+}
+
+static struct dmi_system_id gpio_reset_ids[] = {
+	{
+		.ident = "Apple Macbook 10,1",
+		.matches = {
+			DMI_MATCH(DMI_SYS_VENDOR, "Apple Inc."),
+			DMI_MATCH(DMI_PRODUCT_NAME, "MacBookPro10,1"),
+		}
+	},
+	{ }
+};
+
+int
+nouveau_gpio_init(struct nouveau_gpio *gpio)
+{
+	int ret = nouveau_subdev_init(&gpio->base);
+	if (ret == 0 && gpio->reset) {
+		if (dmi_check_system(gpio_reset_ids))
+			gpio->reset(gpio, DCB_GPIO_UNUSED);
+	}
+	return ret;
+}
diff --git a/nvkm/subdev/gpio/nv10.c b/nvkm/subdev/gpio/nv10.c
new file mode 100644
index 000000000..168d16a9a
--- /dev/null
+++ b/nvkm/subdev/gpio/nv10.c
@@ -0,0 +1,169 @@
+/*
+ * Copyright (C) 2009 Francisco Jerez.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <subdev/gpio.h>
+
+struct nv10_gpio_priv {
+	struct nouveau_gpio base;
+};
+
+static int
+nv10_gpio_sense(struct nouveau_gpio *gpio, int line)
+{
+	if (line < 2) {
+		line = line * 16;
+		line = nv_rd32(gpio, 0x600818) >> line;
+		return !!(line & 0x0100);
+	} else
+	if (line < 10) {
+		line = (line - 2) * 4;
+		line = nv_rd32(gpio, 0x60081c) >> line;
+		return !!(line & 0x04);
+	} else
+	if (line < 14) {
+		line = (line - 10) * 4;
+		line = nv_rd32(gpio, 0x600850) >> line;
+		return !!(line & 0x04);
+	}
+
+	return -EINVAL;
+}
+
+static int
+nv10_gpio_drive(struct nouveau_gpio *gpio, int line, int dir, int out)
+{
+	u32 reg, mask, data;
+
+	if (line < 2) {
+		line = line * 16;
+		reg  = 0x600818;
+		mask = 0x00000011;
+		data = (dir << 4) | out;
+	} else
+	if (line < 10) {
+		line = (line - 2) * 4;
+		reg  = 0x60081c;
+		mask = 0x00000003;
+		data = (dir << 1) | out;
+	} else
+	if (line < 14) {
+		line = (line - 10) * 4;
+		reg  = 0x600850;
+		mask = 0x00000003;
+		data = (dir << 1) | out;
+	} else {
+		return -EINVAL;
+	}
+
+	nv_mask(gpio, reg, mask << line, data << line);
+	return 0;
+}
+
+static void
+nv10_gpio_irq_enable(struct nouveau_gpio *gpio, int line, bool on)
+{
+	u32 mask = 0x00010001 << line;
+
+	nv_wr32(gpio, 0x001104, mask);
+	nv_mask(gpio, 0x001144, mask, on ? mask : 0);
+}
+
+static void
+nv10_gpio_intr(struct nouveau_subdev *subdev)
+{
+	struct nv10_gpio_priv *priv = (void *)subdev;
+	u32 intr = nv_rd32(priv, 0x001104);
+	u32 hi = (intr & 0x0000ffff) >> 0;
+	u32 lo = (intr & 0xffff0000) >> 16;
+
+	priv->base.isr_run(&priv->base, 0, hi | lo);
+
+	nv_wr32(priv, 0x001104, intr);
+}
+
+static int
+nv10_gpio_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv10_gpio_priv *priv;
+	int ret;
+
+	ret = nouveau_gpio_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.drive = nv10_gpio_drive;
+	priv->base.sense = nv10_gpio_sense;
+	priv->base.irq_enable = nv10_gpio_irq_enable;
+	nv_subdev(priv)->intr = nv10_gpio_intr;
+	return 0;
+}
+
+static void
+nv10_gpio_dtor(struct nouveau_object *object)
+{
+	struct nv10_gpio_priv *priv = (void *)object;
+	nouveau_gpio_destroy(&priv->base);
+}
+
+static int
+nv10_gpio_init(struct nouveau_object *object)
+{
+	struct nv10_gpio_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_gpio_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x001140, 0x00000000);
+	nv_wr32(priv, 0x001100, 0xffffffff);
+	nv_wr32(priv, 0x001144, 0x00000000);
+	nv_wr32(priv, 0x001104, 0xffffffff);
+	return 0;
+}
+
+static int
+nv10_gpio_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv10_gpio_priv *priv = (void *)object;
+	nv_wr32(priv, 0x001140, 0x00000000);
+	nv_wr32(priv, 0x001144, 0x00000000);
+	return nouveau_gpio_fini(&priv->base, suspend);
+}
+
+struct nouveau_oclass
+nv10_gpio_oclass = {
+	.handle = NV_SUBDEV(GPIO, 0x10),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv10_gpio_ctor,
+		.dtor = nv10_gpio_dtor,
+		.init = nv10_gpio_init,
+		.fini = nv10_gpio_fini,
+	},
+};
diff --git a/nvkm/subdev/gpio/nv50.c b/nvkm/subdev/gpio/nv50.c
new file mode 100644
index 000000000..bf13a1200
--- /dev/null
+++ b/nvkm/subdev/gpio/nv50.c
@@ -0,0 +1,195 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/gpio.h>
+
+struct nv50_gpio_priv {
+	struct nouveau_gpio base;
+};
+
+static void
+nv50_gpio_reset(struct nouveau_gpio *gpio, u8 match)
+{
+	struct nouveau_bios *bios = nouveau_bios(gpio);
+	struct nv50_gpio_priv *priv = (void *)gpio;
+	u8 ver, len;
+	u16 entry;
+	int ent = -1;
+
+	while ((entry = dcb_gpio_entry(bios, 0, ++ent, &ver, &len))) {
+		static const u32 regs[] = { 0xe100, 0xe28c };
+		u32 data = nv_ro32(bios, entry);
+		u8  line =   (data & 0x0000001f);
+		u8  func =   (data & 0x0000ff00) >> 8;
+		u8  defs = !!(data & 0x01000000);
+		u8  unk0 = !!(data & 0x02000000);
+		u8  unk1 = !!(data & 0x04000000);
+		u32 val = (unk1 << 16) | unk0;
+		u32 reg = regs[line >> 4]; line &= 0x0f;
+
+		if ( func  == DCB_GPIO_UNUSED ||
+		    (match != DCB_GPIO_UNUSED && match != func))
+			continue;
+
+		gpio->set(gpio, 0, func, line, defs);
+
+		nv_mask(priv, reg, 0x00010001 << line, val << line);
+	}
+}
+
+static int
+nv50_gpio_location(int line, u32 *reg, u32 *shift)
+{
+	const u32 nv50_gpio_reg[4] = { 0xe104, 0xe108, 0xe280, 0xe284 };
+
+	if (line >= 32)
+		return -EINVAL;
+
+	*reg = nv50_gpio_reg[line >> 3];
+	*shift = (line & 7) << 2;
+	return 0;
+}
+
+static int
+nv50_gpio_drive(struct nouveau_gpio *gpio, int line, int dir, int out)
+{
+	u32 reg, shift;
+
+	if (nv50_gpio_location(line, &reg, &shift))
+		return -EINVAL;
+
+	nv_mask(gpio, reg, 7 << shift, (((dir ^ 1) << 1) | out) << shift);
+	return 0;
+}
+
+static int
+nv50_gpio_sense(struct nouveau_gpio *gpio, int line)
+{
+	u32 reg, shift;
+
+	if (nv50_gpio_location(line, &reg, &shift))
+		return -EINVAL;
+
+	return !!(nv_rd32(gpio, reg) & (4 << shift));
+}
+
+void
+nv50_gpio_irq_enable(struct nouveau_gpio *gpio, int line, bool on)
+{
+	u32 reg  = line < 16 ? 0xe050 : 0xe070;
+	u32 mask = 0x00010001 << (line & 0xf);
+
+	nv_wr32(gpio, reg + 4, mask);
+	nv_mask(gpio, reg + 0, mask, on ? mask : 0);
+}
+
+void
+nv50_gpio_intr(struct nouveau_subdev *subdev)
+{
+	struct nv50_gpio_priv *priv = (void *)subdev;
+	u32 intr0, intr1 = 0;
+	u32 hi, lo;
+
+	intr0 = nv_rd32(priv, 0xe054) & nv_rd32(priv, 0xe050);
+	if (nv_device(priv)->chipset >= 0x90)
+		intr1 = nv_rd32(priv, 0xe074) & nv_rd32(priv, 0xe070);
+
+	hi = (intr0 & 0x0000ffff) | (intr1 << 16);
+	lo = (intr0 >> 16) | (intr1 & 0xffff0000);
+	priv->base.isr_run(&priv->base, 0, hi | lo);
+
+	nv_wr32(priv, 0xe054, intr0);
+	if (nv_device(priv)->chipset >= 0x90)
+		nv_wr32(priv, 0xe074, intr1);
+}
+
+static int
+nv50_gpio_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nv50_gpio_priv *priv;
+	int ret;
+
+	ret = nouveau_gpio_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.reset = nv50_gpio_reset;
+	priv->base.drive = nv50_gpio_drive;
+	priv->base.sense = nv50_gpio_sense;
+	priv->base.irq_enable = nv50_gpio_irq_enable;
+	nv_subdev(priv)->intr = nv50_gpio_intr;
+	return 0;
+}
+
+void
+nv50_gpio_dtor(struct nouveau_object *object)
+{
+	struct nv50_gpio_priv *priv = (void *)object;
+	nouveau_gpio_destroy(&priv->base);
+}
+
+int
+nv50_gpio_init(struct nouveau_object *object)
+{
+	struct nv50_gpio_priv *priv = (void *)object;
+	int ret;
+
+	ret = nouveau_gpio_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* disable, and ack any pending gpio interrupts */
+	nv_wr32(priv, 0xe050, 0x00000000);
+	nv_wr32(priv, 0xe054, 0xffffffff);
+	if (nv_device(priv)->chipset >= 0x90) {
+		nv_wr32(priv, 0xe070, 0x00000000);
+		nv_wr32(priv, 0xe074, 0xffffffff);
+	}
+
+	return 0;
+}
+
+int
+nv50_gpio_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_gpio_priv *priv = (void *)object;
+	nv_wr32(priv, 0xe050, 0x00000000);
+	if (nv_device(priv)->chipset >= 0x90)
+		nv_wr32(priv, 0xe070, 0x00000000);
+	return nouveau_gpio_fini(&priv->base, suspend);
+}
+
+struct nouveau_oclass
+nv50_gpio_oclass = {
+	.handle = NV_SUBDEV(GPIO, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_gpio_ctor,
+		.dtor = nv50_gpio_dtor,
+		.init = nv50_gpio_init,
+		.fini = nv50_gpio_fini,
+	},
+};
diff --git a/nvkm/subdev/gpio/nvd0.c b/nvkm/subdev/gpio/nvd0.c
new file mode 100644
index 000000000..83e8b8f16
--- /dev/null
+++ b/nvkm/subdev/gpio/nvd0.c
@@ -0,0 +1,105 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/gpio.h>
+
+struct nvd0_gpio_priv {
+	struct nouveau_gpio base;
+};
+
+static void
+nvd0_gpio_reset(struct nouveau_gpio *gpio, u8 match)
+{
+	struct nouveau_bios *bios = nouveau_bios(gpio);
+	struct nvd0_gpio_priv *priv = (void *)gpio;
+	u8 ver, len;
+	u16 entry;
+	int ent = -1;
+
+	while ((entry = dcb_gpio_entry(bios, 0, ++ent, &ver, &len))) {
+		u32 data = nv_ro32(bios, entry);
+		u8  line =   (data & 0x0000003f);
+		u8  defs = !!(data & 0x00000080);
+		u8  func =   (data & 0x0000ff00) >> 8;
+		u8  unk0 =   (data & 0x00ff0000) >> 16;
+		u8  unk1 =   (data & 0x1f000000) >> 24;
+
+		if ( func  == DCB_GPIO_UNUSED ||
+		    (match != DCB_GPIO_UNUSED && match != func))
+			continue;
+
+		gpio->set(gpio, 0, func, line, defs);
+
+		nv_mask(priv, 0x00d610 + (line * 4), 0xff, unk0);
+		if (unk1--)
+			nv_mask(priv, 0x00d740 + (unk1 * 4), 0xff, line);
+	}
+}
+
+static int
+nvd0_gpio_drive(struct nouveau_gpio *gpio, int line, int dir, int out)
+{
+	u32 data = ((dir ^ 1) << 13) | (out << 12);
+	nv_mask(gpio, 0x00d610 + (line * 4), 0x00003000, data);
+	nv_mask(gpio, 0x00d604, 0x00000001, 0x00000001); /* update? */
+	return 0;
+}
+
+static int
+nvd0_gpio_sense(struct nouveau_gpio *gpio, int line)
+{
+	return !!(nv_rd32(gpio, 0x00d610 + (line * 4)) & 0x00004000);
+}
+
+static int
+nvd0_gpio_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nvd0_gpio_priv *priv;
+	int ret;
+
+	ret = nouveau_gpio_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.reset = nvd0_gpio_reset;
+	priv->base.drive = nvd0_gpio_drive;
+	priv->base.sense = nvd0_gpio_sense;
+	priv->base.irq_enable = nv50_gpio_irq_enable;
+	nv_subdev(priv)->intr = nv50_gpio_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nvd0_gpio_oclass = {
+	.handle = NV_SUBDEV(GPIO, 0xd0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvd0_gpio_ctor,
+		.dtor = nv50_gpio_dtor,
+		.init = nv50_gpio_init,
+		.fini = nv50_gpio_fini,
+	},
+};
diff --git a/nvkm/subdev/i2c/Makefile.am b/nvkm/subdev/i2c/Makefile.am
new file mode 100644
index 000000000..46c13114f
--- /dev/null
+++ b/nvkm/subdev/i2c/Makefile.am
@@ -0,0 +1,7 @@
+noinst_LTLIBRARIES = libi2c.la
+
+libi2c_la_SOURCES = base.c \
+		    aux.c \
+		    bit.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/i2c/aux.c b/nvkm/subdev/i2c/aux.c
new file mode 100644
index 000000000..dc27e794a
--- /dev/null
+++ b/nvkm/subdev/i2c/aux.c
@@ -0,0 +1,212 @@
+/*
+ * Copyright 2009 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/i2c.h>
+
+/******************************************************************************
+ * aux channel util functions
+ *****************************************************************************/
+#define AUX_DBG(fmt, args...) nv_debug(aux, "AUXCH(%d): " fmt, ch, ##args)
+#define AUX_ERR(fmt, args...) nv_error(aux, "AUXCH(%d): " fmt, ch, ##args)
+
+static void
+auxch_fini(struct nouveau_i2c *aux, int ch)
+{
+	nv_mask(aux, 0x00e4e4 + (ch * 0x50), 0x00310000, 0x00000000);
+}
+
+static int
+auxch_init(struct nouveau_i2c *aux, int ch)
+{
+	const u32 unksel = 1; /* nfi which to use, or if it matters.. */
+	const u32 ureq = unksel ? 0x00100000 : 0x00200000;
+	const u32 urep = unksel ? 0x01000000 : 0x02000000;
+	u32 ctrl, timeout;
+
+	/* wait up to 1ms for any previous transaction to be done... */
+	timeout = 1000;
+	do {
+		ctrl = nv_rd32(aux, 0x00e4e4 + (ch * 0x50));
+		udelay(1);
+		if (!timeout--) {
+			AUX_ERR("begin idle timeout 0x%08x\n", ctrl);
+			return -EBUSY;
+		}
+	} while (ctrl & 0x03010000);
+
+	/* set some magic, and wait up to 1ms for it to appear */
+	nv_mask(aux, 0x00e4e4 + (ch * 0x50), 0x00300000, ureq);
+	timeout = 1000;
+	do {
+		ctrl = nv_rd32(aux, 0x00e4e4 + (ch * 0x50));
+		udelay(1);
+		if (!timeout--) {
+			AUX_ERR("magic wait 0x%08x\n", ctrl);
+			auxch_fini(aux, ch);
+			return -EBUSY;
+		}
+	} while ((ctrl & 0x03000000) != urep);
+
+	return 0;
+}
+
+static int
+auxch_tx(struct nouveau_i2c *aux, int ch, u8 type, u32 addr, u8 *data, u8 size)
+{
+	u32 ctrl, stat, timeout, retries;
+	u32 xbuf[4] = {};
+	int ret, i;
+
+	AUX_DBG("%d: 0x%08x %d\n", type, addr, size);
+
+	ret = auxch_init(aux, ch);
+	if (ret)
+		goto out;
+
+	stat = nv_rd32(aux, 0x00e4e8 + (ch * 0x50));
+	if (!(stat & 0x10000000)) {
+		AUX_DBG("sink not detected\n");
+		ret = -ENXIO;
+		goto out;
+	}
+
+	if (!(type & 1)) {
+		memcpy(xbuf, data, size);
+		for (i = 0; i < 16; i += 4) {
+			AUX_DBG("wr 0x%08x\n", xbuf[i / 4]);
+			nv_wr32(aux, 0x00e4c0 + (ch * 0x50) + i, xbuf[i / 4]);
+		}
+	}
+
+	ctrl  = nv_rd32(aux, 0x00e4e4 + (ch * 0x50));
+	ctrl &= ~0x0001f0ff;
+	ctrl |= type << 12;
+	ctrl |= size - 1;
+	nv_wr32(aux, 0x00e4e0 + (ch * 0x50), addr);
+
+	/* retry transaction a number of times on failure... */
+	ret = -EREMOTEIO;
+	for (retries = 0; retries < 32; retries++) {
+		/* reset, and delay a while if this is a retry */
+		nv_wr32(aux, 0x00e4e4 + (ch * 0x50), 0x80000000 | ctrl);
+		nv_wr32(aux, 0x00e4e4 + (ch * 0x50), 0x00000000 | ctrl);
+		if (retries)
+			udelay(400);
+
+		/* transaction request, wait up to 1ms for it to complete */
+		nv_wr32(aux, 0x00e4e4 + (ch * 0x50), 0x00010000 | ctrl);
+
+		timeout = 1000;
+		do {
+			ctrl = nv_rd32(aux, 0x00e4e4 + (ch * 0x50));
+			udelay(1);
+			if (!timeout--) {
+				AUX_ERR("tx req timeout 0x%08x\n", ctrl);
+				goto out;
+			}
+		} while (ctrl & 0x00010000);
+
+		/* read status, and check if transaction completed ok */
+		stat = nv_mask(aux, 0x00e4e8 + (ch * 0x50), 0, 0);
+		if (!(stat & 0x000f0f00)) {
+			ret = 0;
+			break;
+		}
+
+		AUX_DBG("%02d 0x%08x 0x%08x\n", retries, ctrl, stat);
+	}
+
+	if (type & 1) {
+		for (i = 0; i < 16; i += 4) {
+			xbuf[i / 4] = nv_rd32(aux, 0x00e4d0 + (ch * 0x50) + i);
+			AUX_DBG("rd 0x%08x\n", xbuf[i / 4]);
+		}
+		memcpy(data, xbuf, size);
+	}
+
+out:
+	auxch_fini(aux, ch);
+	return ret;
+}
+
+int
+nv_rdaux(struct nouveau_i2c_port *auxch, u32 addr, u8 *data, u8 size)
+{
+	return auxch_tx(auxch->i2c, auxch->drive, 9, addr, data, size);
+}
+
+int
+nv_wraux(struct nouveau_i2c_port *auxch, u32 addr, u8 *data, u8 size)
+{
+	return auxch_tx(auxch->i2c, auxch->drive, 8, addr, data, size);
+}
+
+static int
+aux_xfer(struct i2c_adapter *adap, struct i2c_msg *msgs, int num)
+{
+	struct nouveau_i2c_port *auxch = (struct nouveau_i2c_port *)adap;
+	struct i2c_msg *msg = msgs;
+	int ret, mcnt = num;
+
+	while (mcnt--) {
+		u8 remaining = msg->len;
+		u8 *ptr = msg->buf;
+
+		while (remaining) {
+			u8 cnt = (remaining > 16) ? 16 : remaining;
+			u8 cmd;
+
+			if (msg->flags & I2C_M_RD)
+				cmd = 1;
+			else
+				cmd = 0;
+
+			if (mcnt || remaining > 16)
+				cmd |= 4; /* MOT */
+
+			ret = auxch_tx(auxch->i2c, auxch->drive, cmd,
+				       msg->addr, ptr, cnt);
+			if (ret < 0)
+				return ret;
+
+			ptr += cnt;
+			remaining -= cnt;
+		}
+
+		msg++;
+	}
+
+	return num;
+}
+
+static u32
+aux_func(struct i2c_adapter *adap)
+{
+	return I2C_FUNC_I2C | I2C_FUNC_SMBUS_EMUL;
+}
+
+const struct i2c_algorithm nouveau_i2c_aux_algo = {
+	.master_xfer = aux_xfer,
+	.functionality = aux_func
+};
diff --git a/nvkm/subdev/i2c/base.c b/nvkm/subdev/i2c/base.c
new file mode 100644
index 000000000..dbfc2abf0
--- /dev/null
+++ b/nvkm/subdev/i2c/base.c
@@ -0,0 +1,407 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "core/option.h"
+
+#include "subdev/i2c.h"
+#include "subdev/vga.h"
+
+int
+nv_rdi2cr(struct nouveau_i2c_port *port, u8 addr, u8 reg)
+{
+	u8 val;
+	struct i2c_msg msgs[] = {
+		{ .addr = addr, .flags = 0, .len = 1, .buf = &reg },
+		{ .addr = addr, .flags = I2C_M_RD, .len = 1, .buf = &val },
+	};
+
+	int ret = i2c_transfer(&port->adapter, msgs, 2);
+	if (ret != 2)
+		return -EIO;
+
+	return val;
+}
+
+int
+nv_wri2cr(struct nouveau_i2c_port *port, u8 addr, u8 reg, u8 val)
+{
+	struct i2c_msg msgs[] = {
+		{ .addr = addr, .flags = 0, .len = 1, .buf = &reg },
+		{ .addr = addr, .flags = 0, .len = 1, .buf = &val },
+	};
+
+	int ret = i2c_transfer(&port->adapter, msgs, 2);
+	if (ret != 2)
+		return -EIO;
+
+	return 0;
+}
+
+bool
+nv_probe_i2c(struct nouveau_i2c_port *port, u8 addr)
+{
+	u8 buf[] = { 0 };
+	struct i2c_msg msgs[] = {
+		{
+			.addr = addr,
+			.flags = 0,
+			.len = 1,
+			.buf = buf,
+		},
+		{
+			.addr = addr,
+			.flags = I2C_M_RD,
+			.len = 1,
+			.buf = buf,
+		}
+	};
+
+	return i2c_transfer(&port->adapter, msgs, 2) == 2;
+}
+
+static struct nouveau_i2c_port *
+nouveau_i2c_find(struct nouveau_i2c *i2c, u8 index)
+{
+	struct nouveau_bios *bios = nouveau_bios(i2c);
+	struct nouveau_i2c_port *port;
+
+	if (index == NV_I2C_DEFAULT(0) ||
+	    index == NV_I2C_DEFAULT(1)) {
+		u8  ver, hdr, cnt, len;
+		u16 i2c = dcb_i2c_table(bios, &ver, &hdr, &cnt, &len);
+		if (i2c && ver >= 0x30) {
+			u8 auxidx = nv_ro08(bios, i2c + 4);
+			if (index == NV_I2C_DEFAULT(0))
+				index = (auxidx & 0x0f) >> 0;
+			else
+				index = (auxidx & 0xf0) >> 4;
+		} else {
+			index = 2;
+		}
+	}
+
+	list_for_each_entry(port, &i2c->ports, head) {
+		if (port->index == index)
+			break;
+	}
+
+	if (&port->head == &i2c->ports)
+		return NULL;
+
+	if (nv_device(i2c)->card_type >= NV_50 && (port->dcb & 0x00000100)) {
+		u32 reg = 0x00e500, val;
+		if (port->type == 6) {
+			reg += port->drive * 0x50;
+			val  = 0x2002;
+		} else {
+			reg += ((port->dcb & 0x1e00) >> 9) * 0x50;
+			val  = 0xe001;
+		}
+
+		/* nfi, but neither auxch or i2c work if it's 1 */
+		nv_mask(i2c, reg + 0x0c, 0x00000001, 0x00000000);
+		/* nfi, but switches auxch vs normal i2c */
+		nv_mask(i2c, reg + 0x00, 0x0000f003, val);
+	}
+
+	return port;
+}
+
+static int
+nouveau_i2c_identify(struct nouveau_i2c *i2c, int index, const char *what,
+		     struct i2c_board_info *info,
+		     bool (*match)(struct nouveau_i2c_port *,
+				   struct i2c_board_info *))
+{
+	struct nouveau_i2c_port *port = nouveau_i2c_find(i2c, index);
+	int i;
+
+	if (!port) {
+		nv_debug(i2c, "no bus when probing %s on %d\n", what, index);
+		return -ENODEV;
+	}
+
+	nv_debug(i2c, "probing %ss on bus: %d\n", what, port->index);
+	for (i = 0; info[i].addr; i++) {
+		if (nv_probe_i2c(port, info[i].addr) &&
+		    (!match || match(port, &info[i]))) {
+			nv_info(i2c, "detected %s: %s\n", what, info[i].type);
+			return i;
+		}
+	}
+
+	nv_debug(i2c, "no devices found.\n");
+	return -ENODEV;
+}
+
+void
+nouveau_i2c_drive_scl(void *data, int state)
+{
+	struct nouveau_i2c_port *port = data;
+
+	if (port->type == DCB_I2C_NV04_BIT) {
+		u8 val = nv_rdvgac(port->i2c, 0, port->drive);
+		if (state) val |= 0x20;
+		else	   val &= 0xdf;
+		nv_wrvgac(port->i2c, 0, port->drive, val | 0x01);
+	} else
+	if (port->type == DCB_I2C_NV4E_BIT) {
+		nv_mask(port->i2c, port->drive, 0x2f, state ? 0x21 : 0x01);
+	} else
+	if (port->type == DCB_I2C_NVIO_BIT) {
+		if (state) port->state |= 0x01;
+		else	   port->state &= 0xfe;
+		nv_wr32(port->i2c, port->drive, 4 | port->state);
+	}
+}
+
+void
+nouveau_i2c_drive_sda(void *data, int state)
+{
+	struct nouveau_i2c_port *port = data;
+
+	if (port->type == DCB_I2C_NV04_BIT) {
+		u8 val = nv_rdvgac(port->i2c, 0, port->drive);
+		if (state) val |= 0x10;
+		else	   val &= 0xef;
+		nv_wrvgac(port->i2c, 0, port->drive, val | 0x01);
+	} else
+	if (port->type == DCB_I2C_NV4E_BIT) {
+		nv_mask(port->i2c, port->drive, 0x1f, state ? 0x11 : 0x01);
+	} else
+	if (port->type == DCB_I2C_NVIO_BIT) {
+		if (state) port->state |= 0x02;
+		else	   port->state &= 0xfd;
+		nv_wr32(port->i2c, port->drive, 4 | port->state);
+	}
+}
+
+int
+nouveau_i2c_sense_scl(void *data)
+{
+	struct nouveau_i2c_port *port = data;
+	struct nouveau_device *device = nv_device(port->i2c);
+
+	if (port->type == DCB_I2C_NV04_BIT) {
+		return !!(nv_rdvgac(port->i2c, 0, port->sense) & 0x04);
+	} else
+	if (port->type == DCB_I2C_NV4E_BIT) {
+		return !!(nv_rd32(port->i2c, port->sense) & 0x00040000);
+	} else
+	if (port->type == DCB_I2C_NVIO_BIT) {
+		if (device->card_type < NV_D0)
+			return !!(nv_rd32(port->i2c, port->sense) & 0x01);
+		else
+			return !!(nv_rd32(port->i2c, port->sense) & 0x10);
+	}
+
+	return 0;
+}
+
+int
+nouveau_i2c_sense_sda(void *data)
+{
+	struct nouveau_i2c_port *port = data;
+	struct nouveau_device *device = nv_device(port->i2c);
+
+	if (port->type == DCB_I2C_NV04_BIT) {
+		return !!(nv_rdvgac(port->i2c, 0, port->sense) & 0x08);
+	} else
+	if (port->type == DCB_I2C_NV4E_BIT) {
+		return !!(nv_rd32(port->i2c, port->sense) & 0x00080000);
+	} else
+	if (port->type == DCB_I2C_NVIO_BIT) {
+		if (device->card_type < NV_D0)
+			return !!(nv_rd32(port->i2c, port->sense) & 0x02);
+		else
+			return !!(nv_rd32(port->i2c, port->sense) & 0x20);
+	}
+
+	return 0;
+}
+
+static const u32 nv50_i2c_port[] = {
+	0x00e138, 0x00e150, 0x00e168, 0x00e180,
+	0x00e254, 0x00e274, 0x00e764, 0x00e780,
+	0x00e79c, 0x00e7b8
+};
+
+static int
+nouveau_i2c_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		 struct nouveau_oclass *oclass, void *data, u32 size,
+		 struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nouveau_bios *bios = nouveau_bios(parent);
+	struct nouveau_i2c_port *port;
+	struct nouveau_i2c *i2c;
+	struct dcb_i2c_entry info;
+	int ret, i = -1;
+
+	ret = nouveau_subdev_create(parent, engine, oclass, 0,
+				    "I2C", "i2c", &i2c);
+	*pobject = nv_object(i2c);
+	if (ret)
+		return ret;
+
+	i2c->find = nouveau_i2c_find;
+	i2c->identify = nouveau_i2c_identify;
+	INIT_LIST_HEAD(&i2c->ports);
+
+	while (!dcb_i2c_parse(bios, ++i, &info)) {
+		if (info.type == DCB_I2C_UNUSED)
+			continue;
+
+		port = kzalloc(sizeof(*port), GFP_KERNEL);
+		if (!port) {
+			nv_error(i2c, "failed port memory alloc at %d\n", i);
+			break;
+		}
+
+		port->type = info.type;
+		switch (port->type) {
+		case DCB_I2C_NV04_BIT:
+			port->drive = info.drive;
+			port->sense = info.sense;
+			break;
+		case DCB_I2C_NV4E_BIT:
+			port->drive = 0x600800 + info.drive;
+			port->sense = port->drive;
+			break;
+		case DCB_I2C_NVIO_BIT:
+			port->drive = info.drive & 0x0f;
+			if (device->card_type < NV_D0) {
+				if (port->drive >= ARRAY_SIZE(nv50_i2c_port))
+					break;
+				port->drive = nv50_i2c_port[port->drive];
+				port->sense = port->drive;
+			} else {
+				port->drive = 0x00d014 + (port->drive * 0x20);
+				port->sense = port->drive;
+			}
+			break;
+		case DCB_I2C_NVIO_AUX:
+			port->drive = info.drive & 0x0f;
+			port->sense = port->drive;
+			port->adapter.algo = &nouveau_i2c_aux_algo;
+			break;
+		default:
+			break;
+		}
+
+		if (!port->adapter.algo && !port->drive) {
+			nv_error(i2c, "I2C%d: type %d index %x/%x unknown\n",
+				 i, port->type, port->drive, port->sense);
+			kfree(port);
+			continue;
+		}
+
+		snprintf(port->adapter.name, sizeof(port->adapter.name),
+			 "nouveau-%s-%d", device->name, i);
+		port->adapter.owner = THIS_MODULE;
+		port->adapter.dev.parent = &device->pdev->dev;
+		port->i2c = i2c;
+		port->index = i;
+		port->dcb = info.data;
+		i2c_set_adapdata(&port->adapter, i2c);
+
+		if (port->adapter.algo != &nouveau_i2c_aux_algo) {
+			nouveau_i2c_drive_scl(port, 0);
+			nouveau_i2c_drive_sda(port, 1);
+			nouveau_i2c_drive_scl(port, 1);
+
+#ifdef CONFIG_NOUVEAU_I2C_INTERNAL_DEFAULT
+			if (nouveau_boolopt(device->cfgopt, "NvI2C", true)) {
+#else
+			if (nouveau_boolopt(device->cfgopt, "NvI2C", false)) {
+#endif
+				port->adapter.algo = &nouveau_i2c_bit_algo;
+				ret = i2c_add_adapter(&port->adapter);
+			} else {
+				port->adapter.algo_data = &port->bit;
+				port->bit.udelay = 10;
+				port->bit.timeout = usecs_to_jiffies(2200);
+				port->bit.data = port;
+				port->bit.setsda = nouveau_i2c_drive_sda;
+				port->bit.setscl = nouveau_i2c_drive_scl;
+				port->bit.getsda = nouveau_i2c_sense_sda;
+				port->bit.getscl = nouveau_i2c_sense_scl;
+				ret = i2c_bit_add_bus(&port->adapter);
+			}
+		} else {
+			port->adapter.algo = &nouveau_i2c_aux_algo;
+			ret = i2c_add_adapter(&port->adapter);
+		}
+
+		if (ret) {
+			nv_error(i2c, "I2C%d: failed register: %d\n", i, ret);
+			kfree(port);
+			continue;
+		}
+
+		list_add_tail(&port->head, &i2c->ports);
+	}
+
+	return 0;
+}
+
+static void
+nouveau_i2c_dtor(struct nouveau_object *object)
+{
+	struct nouveau_i2c *i2c = (void *)object;
+	struct nouveau_i2c_port *port, *temp;
+
+	list_for_each_entry_safe(port, temp, &i2c->ports, head) {
+		i2c_del_adapter(&port->adapter);
+		list_del(&port->head);
+		kfree(port);
+	}
+
+	nouveau_subdev_destroy(&i2c->base);
+}
+
+static int
+nouveau_i2c_init(struct nouveau_object *object)
+{
+	struct nouveau_i2c *i2c = (void *)object;
+	return nouveau_subdev_init(&i2c->base);
+}
+
+static int
+nouveau_i2c_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nouveau_i2c *i2c = (void *)object;
+	return nouveau_subdev_fini(&i2c->base, suspend);
+}
+
+struct nouveau_oclass
+nouveau_i2c_oclass = {
+	.handle = NV_SUBDEV(I2C, 0x00),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nouveau_i2c_ctor,
+		.dtor = nouveau_i2c_dtor,
+		.init = nouveau_i2c_init,
+		.fini = nouveau_i2c_fini,
+	},
+};
diff --git a/nvkm/subdev/i2c/bit.c b/nvkm/subdev/i2c/bit.c
new file mode 100644
index 000000000..1c4c9a5c8
--- /dev/null
+++ b/nvkm/subdev/i2c/bit.c
@@ -0,0 +1,230 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "subdev/i2c.h"
+
+#ifdef CONFIG_NOUVEAU_I2C_INTERNAL
+#define T_TIMEOUT  2200000
+#define T_RISEFALL 1000
+#define T_HOLD     5000
+
+static inline void
+i2c_drive_scl(struct nouveau_i2c_port *port, int state)
+{
+	nouveau_i2c_drive_scl(port, state);
+}
+
+static inline void
+i2c_drive_sda(struct nouveau_i2c_port *port, int state)
+{
+	nouveau_i2c_drive_sda(port, state);
+}
+
+static inline int
+i2c_sense_scl(struct nouveau_i2c_port *port)
+{
+	return nouveau_i2c_sense_scl(port);
+}
+
+static inline int
+i2c_sense_sda(struct nouveau_i2c_port *port)
+{
+	return nouveau_i2c_sense_sda(port);
+}
+
+static void
+i2c_delay(struct nouveau_i2c_port *port, u32 nsec)
+{
+	udelay((nsec + 500) / 1000);
+}
+
+static bool
+i2c_raise_scl(struct nouveau_i2c_port *port)
+{
+	u32 timeout = T_TIMEOUT / T_RISEFALL;
+
+	i2c_drive_scl(port, 1);
+	do {
+		i2c_delay(port, T_RISEFALL);
+	} while (!i2c_sense_scl(port) && --timeout);
+
+	return timeout != 0;
+}
+
+static int
+i2c_start(struct nouveau_i2c_port *port)
+{
+	int ret = 0;
+
+	port->state  = i2c_sense_scl(port);
+	port->state |= i2c_sense_sda(port) << 1;
+	if (port->state != 3) {
+		i2c_drive_scl(port, 0);
+		i2c_drive_sda(port, 1);
+		if (!i2c_raise_scl(port))
+			ret = -EBUSY;
+	}
+
+	i2c_drive_sda(port, 0);
+	i2c_delay(port, T_HOLD);
+	i2c_drive_scl(port, 0);
+	i2c_delay(port, T_HOLD);
+	return ret;
+}
+
+static void
+i2c_stop(struct nouveau_i2c_port *port)
+{
+	i2c_drive_scl(port, 0);
+	i2c_drive_sda(port, 0);
+	i2c_delay(port, T_RISEFALL);
+
+	i2c_drive_scl(port, 1);
+	i2c_delay(port, T_HOLD);
+	i2c_drive_sda(port, 1);
+	i2c_delay(port, T_HOLD);
+}
+
+static int
+i2c_bitw(struct nouveau_i2c_port *port, int sda)
+{
+	i2c_drive_sda(port, sda);
+	i2c_delay(port, T_RISEFALL);
+
+	if (!i2c_raise_scl(port))
+		return -ETIMEDOUT;
+	i2c_delay(port, T_HOLD);
+
+	i2c_drive_scl(port, 0);
+	i2c_delay(port, T_HOLD);
+	return 0;
+}
+
+static int
+i2c_bitr(struct nouveau_i2c_port *port)
+{
+	int sda;
+
+	i2c_drive_sda(port, 1);
+	i2c_delay(port, T_RISEFALL);
+
+	if (!i2c_raise_scl(port))
+		return -ETIMEDOUT;
+	i2c_delay(port, T_HOLD);
+
+	sda = i2c_sense_sda(port);
+
+	i2c_drive_scl(port, 0);
+	i2c_delay(port, T_HOLD);
+	return sda;
+}
+
+static int
+i2c_get_byte(struct nouveau_i2c_port *port, u8 *byte, bool last)
+{
+	int i, bit;
+
+	*byte = 0;
+	for (i = 7; i >= 0; i--) {
+		bit = i2c_bitr(port);
+		if (bit < 0)
+			return bit;
+		*byte |= bit << i;
+	}
+
+	return i2c_bitw(port, last ? 1 : 0);
+}
+
+static int
+i2c_put_byte(struct nouveau_i2c_port *port, u8 byte)
+{
+	int i, ret;
+	for (i = 7; i >= 0; i--) {
+		ret = i2c_bitw(port, !!(byte & (1 << i)));
+		if (ret < 0)
+			return ret;
+	}
+
+	ret = i2c_bitr(port);
+	if (ret == 1) /* nack */
+		ret = -EIO;
+	return ret;
+}
+
+static int
+i2c_addr(struct nouveau_i2c_port *port, struct i2c_msg *msg)
+{
+	u32 addr = msg->addr << 1;
+	if (msg->flags & I2C_M_RD)
+		addr |= 1;
+	return i2c_put_byte(port, addr);
+}
+
+static int
+i2c_bit_xfer(struct i2c_adapter *adap, struct i2c_msg *msgs, int num)
+{
+	struct nouveau_i2c_port *port = (struct nouveau_i2c_port *)adap;
+	struct i2c_msg *msg = msgs;
+	int ret = 0, mcnt = num;
+
+	while (!ret && mcnt--) {
+		u8 remaining = msg->len;
+		u8 *ptr = msg->buf;
+
+		ret = i2c_start(port);
+		if (ret == 0)
+			ret = i2c_addr(port, msg);
+
+		if (msg->flags & I2C_M_RD) {
+			while (!ret && remaining--)
+				ret = i2c_get_byte(port, ptr++, !remaining);
+		} else {
+			while (!ret && remaining--)
+				ret = i2c_put_byte(port, *ptr++);
+		}
+
+		msg++;
+	}
+
+	i2c_stop(port);
+	return (ret < 0) ? ret : num;
+}
+#else
+static int
+i2c_bit_xfer(struct i2c_adapter *adap, struct i2c_msg *msgs, int num)
+{
+	return -ENODEV;
+}
+#endif
+
+static u32
+i2c_bit_func(struct i2c_adapter *adap)
+{
+	return I2C_FUNC_I2C | I2C_FUNC_SMBUS_EMUL;
+}
+
+const struct i2c_algorithm nouveau_i2c_bit_algo = {
+	.master_xfer = i2c_bit_xfer,
+	.functionality = i2c_bit_func
+};
diff --git a/nvkm/subdev/ibus/Makefile.am b/nvkm/subdev/ibus/Makefile.am
new file mode 100644
index 000000000..bca1b2d32
--- /dev/null
+++ b/nvkm/subdev/ibus/Makefile.am
@@ -0,0 +1,6 @@
+noinst_LTLIBRARIES = libibus.la
+
+libibus_la_SOURCES = nvc0.c \
+		     nve0.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/ibus/nvc0.c b/nvkm/subdev/ibus/nvc0.c
new file mode 100644
index 000000000..4e977ff27
--- /dev/null
+++ b/nvkm/subdev/ibus/nvc0.c
@@ -0,0 +1,123 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/ibus.h>
+
+struct nvc0_ibus_priv {
+	struct nouveau_ibus base;
+};
+
+static void
+nvc0_ibus_intr_hub(struct nvc0_ibus_priv *priv, int i)
+{
+	u32 addr = nv_rd32(priv, 0x122120 + (i * 0x0400));
+	u32 data = nv_rd32(priv, 0x122124 + (i * 0x0400));
+	u32 stat = nv_rd32(priv, 0x122128 + (i * 0x0400));
+	nv_error(priv, "HUB%d: 0x%06x 0x%08x (0x%08x)\n", i, addr, data, stat);
+	nv_mask(priv, 0x122128 + (i * 0x0400), 0x00000200, 0x00000000);
+}
+
+static void
+nvc0_ibus_intr_rop(struct nvc0_ibus_priv *priv, int i)
+{
+	u32 addr = nv_rd32(priv, 0x124120 + (i * 0x0400));
+	u32 data = nv_rd32(priv, 0x124124 + (i * 0x0400));
+	u32 stat = nv_rd32(priv, 0x124128 + (i * 0x0400));
+	nv_error(priv, "ROP%d: 0x%06x 0x%08x (0x%08x)\n", i, addr, data, stat);
+	nv_mask(priv, 0x124128 + (i * 0x0400), 0x00000200, 0x00000000);
+}
+
+static void
+nvc0_ibus_intr_gpc(struct nvc0_ibus_priv *priv, int i)
+{
+	u32 addr = nv_rd32(priv, 0x128120 + (i * 0x0400));
+	u32 data = nv_rd32(priv, 0x128124 + (i * 0x0400));
+	u32 stat = nv_rd32(priv, 0x128128 + (i * 0x0400));
+	nv_error(priv, "GPC%d: 0x%06x 0x%08x (0x%08x)\n", i, addr, data, stat);
+	nv_mask(priv, 0x128128 + (i * 0x0400), 0x00000200, 0x00000000);
+}
+
+static void
+nvc0_ibus_intr(struct nouveau_subdev *subdev)
+{
+	struct nvc0_ibus_priv *priv = (void *)subdev;
+	u32 intr0 = nv_rd32(priv, 0x121c58);
+	u32 intr1 = nv_rd32(priv, 0x121c5c);
+	u32 hubnr = nv_rd32(priv, 0x121c70);
+	u32 ropnr = nv_rd32(priv, 0x121c74);
+	u32 gpcnr = nv_rd32(priv, 0x121c78);
+	u32 i;
+
+	for (i = 0; (intr0 & 0x0000ff00) && i < hubnr; i++) {
+		u32 stat = 0x00000100 << i;
+		if (intr0 & stat) {
+			nvc0_ibus_intr_hub(priv, i);
+			intr0 &= ~stat;
+		}
+	}
+
+	for (i = 0; (intr0 & 0xffff0000) && i < ropnr; i++) {
+		u32 stat = 0x00010000 << i;
+		if (intr0 & stat) {
+			nvc0_ibus_intr_rop(priv, i);
+			intr0 &= ~stat;
+		}
+	}
+
+	for (i = 0; intr1 && i < gpcnr; i++) {
+		u32 stat = 0x00000001 << i;
+		if (intr1 & stat) {
+			nvc0_ibus_intr_gpc(priv, i);
+			intr1 &= ~stat;
+		}
+	}
+}
+
+static int
+nvc0_ibus_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nvc0_ibus_priv *priv;
+	int ret;
+
+	ret = nouveau_ibus_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->intr = nvc0_ibus_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_ibus_oclass = {
+	.handle = NV_SUBDEV(IBUS, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_ibus_ctor,
+		.dtor = _nouveau_ibus_dtor,
+		.init = _nouveau_ibus_init,
+		.fini = _nouveau_ibus_fini,
+	},
+};
diff --git a/nvkm/subdev/ibus/nve0.c b/nvkm/subdev/ibus/nve0.c
new file mode 100644
index 000000000..7120124dc
--- /dev/null
+++ b/nvkm/subdev/ibus/nve0.c
@@ -0,0 +1,123 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/ibus.h>
+
+struct nve0_ibus_priv {
+	struct nouveau_ibus base;
+};
+
+static void
+nve0_ibus_intr_hub(struct nve0_ibus_priv *priv, int i)
+{
+	u32 addr = nv_rd32(priv, 0x122120 + (i * 0x0800));
+	u32 data = nv_rd32(priv, 0x122124 + (i * 0x0800));
+	u32 stat = nv_rd32(priv, 0x122128 + (i * 0x0800));
+	nv_error(priv, "HUB%d: 0x%06x 0x%08x (0x%08x)\n", i, addr, data, stat);
+	nv_mask(priv, 0x122128 + (i * 0x0800), 0x00000200, 0x00000000);
+}
+
+static void
+nve0_ibus_intr_rop(struct nve0_ibus_priv *priv, int i)
+{
+	u32 addr = nv_rd32(priv, 0x124120 + (i * 0x0800));
+	u32 data = nv_rd32(priv, 0x124124 + (i * 0x0800));
+	u32 stat = nv_rd32(priv, 0x124128 + (i * 0x0800));
+	nv_error(priv, "ROP%d: 0x%06x 0x%08x (0x%08x)\n", i, addr, data, stat);
+	nv_mask(priv, 0x124128 + (i * 0x0800), 0x00000200, 0x00000000);
+}
+
+static void
+nve0_ibus_intr_gpc(struct nve0_ibus_priv *priv, int i)
+{
+	u32 addr = nv_rd32(priv, 0x128120 + (i * 0x0800));
+	u32 data = nv_rd32(priv, 0x128124 + (i * 0x0800));
+	u32 stat = nv_rd32(priv, 0x128128 + (i * 0x0800));
+	nv_error(priv, "GPC%d: 0x%06x 0x%08x (0x%08x)\n", i, addr, data, stat);
+	nv_mask(priv, 0x128128 + (i * 0x0800), 0x00000200, 0x00000000);
+}
+
+static void
+nve0_ibus_intr(struct nouveau_subdev *subdev)
+{
+	struct nve0_ibus_priv *priv = (void *)subdev;
+	u32 intr0 = nv_rd32(priv, 0x120058);
+	u32 intr1 = nv_rd32(priv, 0x12005c);
+	u32 hubnr = nv_rd32(priv, 0x120070);
+	u32 ropnr = nv_rd32(priv, 0x120074);
+	u32 gpcnr = nv_rd32(priv, 0x120078);
+	u32 i;
+
+	for (i = 0; (intr0 & 0x0000ff00) && i < hubnr; i++) {
+		u32 stat = 0x00000100 << i;
+		if (intr0 & stat) {
+			nve0_ibus_intr_hub(priv, i);
+			intr0 &= ~stat;
+		}
+	}
+
+	for (i = 0; (intr0 & 0xffff0000) && i < ropnr; i++) {
+		u32 stat = 0x00010000 << i;
+		if (intr0 & stat) {
+			nve0_ibus_intr_rop(priv, i);
+			intr0 &= ~stat;
+		}
+	}
+
+	for (i = 0; intr1 && i < gpcnr; i++) {
+		u32 stat = 0x00000001 << i;
+		if (intr1 & stat) {
+			nve0_ibus_intr_gpc(priv, i);
+			intr1 &= ~stat;
+		}
+	}
+}
+
+static int
+nve0_ibus_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nve0_ibus_priv *priv;
+	int ret;
+
+	ret = nouveau_ibus_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->intr = nve0_ibus_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nve0_ibus_oclass = {
+	.handle = NV_SUBDEV(IBUS, 0xe0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nve0_ibus_ctor,
+		.dtor = _nouveau_ibus_dtor,
+		.init = _nouveau_ibus_init,
+		.fini = _nouveau_ibus_fini,
+	},
+};
diff --git a/nvkm/subdev/instmem/Makefile.am b/nvkm/subdev/instmem/Makefile.am
new file mode 100644
index 000000000..c014032c9
--- /dev/null
+++ b/nvkm/subdev/instmem/Makefile.am
@@ -0,0 +1,8 @@
+noinst_LTLIBRARIES = libinstmem.la
+
+libinstmem_la_SOURCES = base.c \
+			nv04.c \
+			nv40.c \
+			nv50.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/instmem/base.c b/nvkm/subdev/instmem/base.c
new file mode 100644
index 000000000..1188227ca
--- /dev/null
+++ b/nvkm/subdev/instmem/base.c
@@ -0,0 +1,135 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/instmem.h>
+
+int
+nouveau_instobj_create_(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass,
+			int length, void **pobject)
+{
+	struct nouveau_instmem *imem = (void *)engine;
+	struct nouveau_instobj *iobj;
+	int ret;
+
+	ret = nouveau_object_create_(parent, engine, oclass, NV_MEMOBJ_CLASS,
+				     length, pobject);
+	iobj = *pobject;
+	if (ret)
+		return ret;
+
+	list_add(&iobj->head, &imem->list);
+	return 0;
+}
+
+void
+nouveau_instobj_destroy(struct nouveau_instobj *iobj)
+{
+	if (iobj->head.prev)
+		list_del(&iobj->head);
+	return nouveau_object_destroy(&iobj->base);
+}
+
+void
+_nouveau_instobj_dtor(struct nouveau_object *object)
+{
+	struct nouveau_instobj *iobj = (void *)object;
+	return nouveau_instobj_destroy(iobj);
+}
+
+int
+nouveau_instmem_create_(struct nouveau_object *parent,
+			struct nouveau_object *engine,
+			struct nouveau_oclass *oclass,
+			int length, void **pobject)
+{
+	struct nouveau_instmem *imem;
+	int ret;
+
+	ret = nouveau_subdev_create_(parent, engine, oclass, 0,
+				     "INSTMEM", "instmem", length, pobject);
+	imem = *pobject;
+	if (ret)
+		return ret;
+
+	INIT_LIST_HEAD(&imem->list);
+	return 0;
+}
+
+int
+nouveau_instmem_init(struct nouveau_instmem *imem)
+{
+	struct nouveau_instobj *iobj;
+	int ret, i;
+
+	ret = nouveau_subdev_init(&imem->base);
+	if (ret)
+		return ret;
+
+	list_for_each_entry(iobj, &imem->list, head) {
+		if (iobj->suspend) {
+			for (i = 0; i < iobj->size; i += 4)
+				nv_wo32(iobj, i, iobj->suspend[i / 4]);
+			vfree(iobj->suspend);
+			iobj->suspend = NULL;
+		}
+	}
+
+	return 0;
+}
+
+int
+nouveau_instmem_fini(struct nouveau_instmem *imem, bool suspend)
+{
+	struct nouveau_instobj *iobj;
+	int i;
+
+	if (suspend) {
+		list_for_each_entry(iobj, &imem->list, head) {
+			iobj->suspend = vmalloc(iobj->size);
+			if (iobj->suspend) {
+				for (i = 0; i < iobj->size; i += 4)
+					iobj->suspend[i / 4] = nv_ro32(iobj, i);
+			} else
+				return -ENOMEM;
+		}
+	}
+
+	return nouveau_subdev_fini(&imem->base, suspend);
+}
+
+int
+_nouveau_instmem_init(struct nouveau_object *object)
+{
+	struct nouveau_instmem *imem = (void *)object;
+	return nouveau_instmem_init(imem);
+}
+
+int
+_nouveau_instmem_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nouveau_instmem *imem = (void *)object;
+	return nouveau_instmem_fini(imem, suspend);
+}
diff --git a/nvkm/subdev/instmem/nv04.c b/nvkm/subdev/instmem/nv04.c
new file mode 100644
index 000000000..f5bbd3834
--- /dev/null
+++ b/nvkm/subdev/instmem/nv04.c
@@ -0,0 +1,198 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/fb.h>
+
+#include "nv04.h"
+
+static int
+nv04_instobj_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv04_instmem_priv *priv = (void *)engine;
+	struct nv04_instobj_priv *node;
+	int ret, align;
+
+	align = (unsigned long)data;
+	if (!align)
+		align = 1;
+
+	ret = nouveau_instobj_create(parent, engine, oclass, &node);
+	*pobject = nv_object(node);
+	if (ret)
+		return ret;
+
+	ret = nouveau_mm_head(&priv->heap, 1, size, size, align, &node->mem);
+	if (ret)
+		return ret;
+
+	node->base.addr = node->mem->offset;
+	node->base.size = node->mem->length;
+	return 0;
+}
+
+static void
+nv04_instobj_dtor(struct nouveau_object *object)
+{
+	struct nv04_instmem_priv *priv = (void *)object->engine;
+	struct nv04_instobj_priv *node = (void *)object;
+	nouveau_mm_free(&priv->heap, &node->mem);
+	nouveau_instobj_destroy(&node->base);
+}
+
+static u32
+nv04_instobj_rd32(struct nouveau_object *object, u64 addr)
+{
+	struct nv04_instobj_priv *node = (void *)object;
+	return nv_ro32(object->engine, node->mem->offset + addr);
+}
+
+static void
+nv04_instobj_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	struct nv04_instobj_priv *node = (void *)object;
+	nv_wo32(object->engine, node->mem->offset + addr, data);
+}
+
+static struct nouveau_oclass
+nv04_instobj_oclass = {
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_instobj_ctor,
+		.dtor = nv04_instobj_dtor,
+		.init = _nouveau_instobj_init,
+		.fini = _nouveau_instobj_fini,
+		.rd32 = nv04_instobj_rd32,
+		.wr32 = nv04_instobj_wr32,
+	},
+};
+
+int
+nv04_instmem_alloc(struct nouveau_instmem *imem, struct nouveau_object *parent,
+		   u32 size, u32 align, struct nouveau_object **pobject)
+{
+	struct nouveau_object *engine = nv_object(imem);
+	struct nv04_instmem_priv *priv = (void *)(imem);
+	int ret;
+
+	ret = nouveau_object_ctor(parent, engine, &nv04_instobj_oclass,
+				  (void *)(unsigned long)align, size, pobject);
+	if (ret)
+		return ret;
+
+	/* INSTMEM itself creates objects to reserve (and preserve across
+	 * suspend/resume) various fixed data locations, each one of these
+	 * takes a reference on INSTMEM itself, causing it to never be
+	 * freed.  We drop all the self-references here to avoid this.
+	 */
+	if (unlikely(!priv->created))
+		atomic_dec(&engine->refcount);
+
+	return 0;
+}
+
+static int
+nv04_instmem_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv04_instmem_priv *priv;
+	int ret;
+
+	ret = nouveau_instmem_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	/* PRAMIN aperture maps over the end of VRAM, reserve it */
+	priv->base.reserved = 512 * 1024;
+	priv->base.alloc    = nv04_instmem_alloc;
+
+	ret = nouveau_mm_init(&priv->heap, 0, priv->base.reserved, 1);
+	if (ret)
+		return ret;
+
+	/* 0x00000-0x10000: reserve for probable vbios image */
+	ret = nouveau_gpuobj_new(parent, NULL, 0x10000, 0, 0, &priv->vbios);
+	if (ret)
+		return ret;
+
+	/* 0x10000-0x18000: reserve for RAMHT */
+	ret = nouveau_ramht_new(parent, NULL, 0x08000, 0, &priv->ramht);
+	if (ret)
+		return ret;
+
+	/* 0x18000-0x18800: reserve for RAMFC (enough for 32 nv30 channels) */
+	ret = nouveau_gpuobj_new(parent, NULL, 0x00800, 0,
+				 NVOBJ_FLAG_ZERO_ALLOC, &priv->ramfc);
+	if (ret)
+		return ret;
+
+	/* 0x18800-0x18a00: reserve for RAMRO */
+	ret = nouveau_gpuobj_new(parent, NULL, 0x00200, 0, 0, &priv->ramro);
+	if (ret)
+		return ret;
+
+	priv->created = true;
+	return 0;
+}
+
+void
+nv04_instmem_dtor(struct nouveau_object *object)
+{
+	struct nv04_instmem_priv *priv = (void *)object;
+	nouveau_gpuobj_ref(NULL, &priv->ramfc);
+	nouveau_gpuobj_ref(NULL, &priv->ramro);
+	nouveau_ramht_ref(NULL, &priv->ramht);
+	nouveau_gpuobj_ref(NULL, &priv->vbios);
+	nouveau_mm_fini(&priv->heap);
+	if (priv->iomem)
+		iounmap(priv->iomem);
+	nouveau_instmem_destroy(&priv->base);
+}
+
+static u32
+nv04_instmem_rd32(struct nouveau_object *object, u64 addr)
+{
+	return nv_rd32(object, 0x700000 + addr);
+}
+
+static void
+nv04_instmem_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	return nv_wr32(object, 0x700000 + addr, data);
+}
+
+struct nouveau_oclass
+nv04_instmem_oclass = {
+	.handle = NV_SUBDEV(INSTMEM, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_instmem_ctor,
+		.dtor = nv04_instmem_dtor,
+		.init = _nouveau_instmem_init,
+		.fini = _nouveau_instmem_fini,
+		.rd32 = nv04_instmem_rd32,
+		.wr32 = nv04_instmem_wr32,
+	},
+};
diff --git a/nvkm/subdev/instmem/nv04.h b/nvkm/subdev/instmem/nv04.h
new file mode 100644
index 000000000..7983d8d9b
--- /dev/null
+++ b/nvkm/subdev/instmem/nv04.h
@@ -0,0 +1,39 @@
+#ifndef __NV04_INSTMEM_H__
+#define __NV04_INSTMEM_H__
+
+#include <core/gpuobj.h>
+#include <core/ramht.h>
+#include <core/mm.h>
+
+#include <subdev/instmem.h>
+
+struct nv04_instmem_priv {
+	struct nouveau_instmem base;
+	bool created;
+
+	void __iomem *iomem;
+	struct nouveau_mm heap;
+
+	struct nouveau_gpuobj *vbios;
+	struct nouveau_ramht  *ramht;
+	struct nouveau_gpuobj *ramro;
+	struct nouveau_gpuobj *ramfc;
+};
+
+static inline struct nv04_instmem_priv *
+nv04_instmem(void *obj)
+{
+	return (void *)nouveau_instmem(obj);
+}
+
+struct nv04_instobj_priv {
+	struct nouveau_instobj base;
+	struct nouveau_mm_node *mem;
+};
+
+void nv04_instmem_dtor(struct nouveau_object *);
+
+int nv04_instmem_alloc(struct nouveau_instmem *, struct nouveau_object *,
+		       u32 size, u32 align, struct nouveau_object **pobject);
+
+#endif
diff --git a/nvkm/subdev/instmem/nv40.c b/nvkm/subdev/instmem/nv40.c
new file mode 100644
index 000000000..da6425320
--- /dev/null
+++ b/nvkm/subdev/instmem/nv40.c
@@ -0,0 +1,138 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "nv04.h"
+
+static inline int
+nv44_graph_class(struct nv04_instmem_priv *priv)
+{
+	if ((nv_device(priv)->chipset & 0xf0) == 0x60)
+		return 1;
+	return !(0x0baf & (1 << (nv_device(priv)->chipset & 0x0f)));
+}
+
+static int
+nv40_instmem_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct pci_dev *pdev = device->pdev;
+	struct nv04_instmem_priv *priv;
+	int ret, bar, vs;
+
+	ret = nouveau_instmem_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	/* map bar */
+	if (pci_resource_len(pdev, 2))
+		bar = 2;
+	else
+		bar = 3;
+
+	priv->iomem = ioremap(pci_resource_start(pdev, bar),
+			      pci_resource_len(pdev, bar));
+	if (!priv->iomem) {
+		nv_error(priv, "unable to map PRAMIN BAR\n");
+		return -EFAULT;
+	}
+
+	/* PRAMIN aperture maps over the end of vram, reserve enough space
+	 * to fit graphics contexts for every channel, the magics come
+	 * from engine/graph/nv40.c
+	 */
+	vs = hweight8((nv_rd32(priv, 0x001540) & 0x0000ff00) >> 8);
+	if      (device->chipset == 0x40) priv->base.reserved = 0x6aa0 * vs;
+	else if (device->chipset  < 0x43) priv->base.reserved = 0x4f00 * vs;
+	else if (nv44_graph_class(priv))  priv->base.reserved = 0x4980 * vs;
+	else				  priv->base.reserved = 0x4a40 * vs;
+	priv->base.reserved += 16 * 1024;
+	priv->base.reserved *= 32;		/* per-channel */
+	priv->base.reserved += 512 * 1024;	/* pci(e)gart table */
+	priv->base.reserved += 512 * 1024;	/* object storage */
+
+	priv->base.reserved = round_up(priv->base.reserved, 4096);
+	priv->base.alloc    = nv04_instmem_alloc;
+
+	ret = nouveau_mm_init(&priv->heap, 0, priv->base.reserved, 1);
+	if (ret)
+		return ret;
+
+	/* 0x00000-0x10000: reserve for probable vbios image */
+	ret = nouveau_gpuobj_new(parent, NULL, 0x10000, 0, 0, &priv->vbios);
+	if (ret)
+		return ret;
+
+	/* 0x10000-0x18000: reserve for RAMHT */
+	ret = nouveau_ramht_new(parent, NULL, 0x08000, 0, &priv->ramht);
+	if (ret)
+		return ret;
+
+	/* 0x18000-0x18200: reserve for RAMRO
+	 * 0x18200-0x20000: padding
+	 */
+	ret = nouveau_gpuobj_new(parent, NULL, 0x08000, 0, 0, &priv->ramro);
+	if (ret)
+		return ret;
+
+	/* 0x20000-0x21000: reserve for RAMFC
+	 * 0x21000-0x40000: padding and some unknown crap
+	 */
+	ret = nouveau_gpuobj_new(parent, NULL, 0x20000, 0,
+				 NVOBJ_FLAG_ZERO_ALLOC, &priv->ramfc);
+	if (ret)
+		return ret;
+
+	priv->created = true;
+	return 0;
+}
+
+static u32
+nv40_instmem_rd32(struct nouveau_object *object, u64 addr)
+{
+	struct nv04_instmem_priv *priv = (void *)object;
+	return ioread32_native(priv->iomem + addr);
+}
+
+static void
+nv40_instmem_wr32(struct nouveau_object *object, u64 addr, u32 data)
+{
+	struct nv04_instmem_priv *priv = (void *)object;
+	iowrite32_native(data, priv->iomem + addr);
+}
+
+struct nouveau_oclass
+nv40_instmem_oclass = {
+	.handle = NV_SUBDEV(INSTMEM, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv40_instmem_ctor,
+		.dtor = nv04_instmem_dtor,
+		.init = _nouveau_instmem_init,
+		.fini = _nouveau_instmem_fini,
+		.rd32 = nv40_instmem_rd32,
+		.wr32 = nv40_instmem_wr32,
+	},
+};
diff --git a/nvkm/subdev/instmem/nv50.c b/nvkm/subdev/instmem/nv50.c
new file mode 100644
index 000000000..cfc7e3146
--- /dev/null
+++ b/nvkm/subdev/instmem/nv50.c
@@ -0,0 +1,172 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/instmem.h>
+#include <subdev/fb.h>
+
+#include <core/mm.h>
+
+struct nv50_instmem_priv {
+	struct nouveau_instmem base;
+	spinlock_t lock;
+	u64 addr;
+};
+
+struct nv50_instobj_priv {
+	struct nouveau_instobj base;
+	struct nouveau_mem *mem;
+};
+
+static int
+nv50_instobj_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nouveau_fb *pfb = nouveau_fb(parent);
+	struct nv50_instobj_priv *node;
+	u32 align = (unsigned long)data;
+	int ret;
+
+	size  = max((size  + 4095) & ~4095, (u32)4096);
+	align = max((align + 4095) & ~4095, (u32)4096);
+
+	ret = nouveau_instobj_create(parent, engine, oclass, &node);
+	*pobject = nv_object(node);
+	if (ret)
+		return ret;
+
+	ret = pfb->ram.get(pfb, size, align, 0, 0x800, &node->mem);
+	if (ret)
+		return ret;
+
+	node->base.addr = node->mem->offset;
+	node->base.size = node->mem->size << 12;
+	node->mem->page_shift = 12;
+	return 0;
+}
+
+static void
+nv50_instobj_dtor(struct nouveau_object *object)
+{
+	struct nv50_instobj_priv *node = (void *)object;
+	struct nouveau_fb *pfb = nouveau_fb(object);
+	pfb->ram.put(pfb, &node->mem);
+	nouveau_instobj_destroy(&node->base);
+}
+
+static u32
+nv50_instobj_rd32(struct nouveau_object *object, u64 offset)
+{
+	struct nv50_instmem_priv *priv = (void *)object->engine;
+	struct nv50_instobj_priv *node = (void *)object;
+	unsigned long flags;
+	u64 base = (node->mem->offset + offset) & 0xffffff00000ULL;
+	u64 addr = (node->mem->offset + offset) & 0x000000fffffULL;
+	u32 data;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	if (unlikely(priv->addr != base)) {
+		nv_wr32(priv, 0x001700, base >> 16);
+		priv->addr = base;
+	}
+	data = nv_rd32(priv, 0x700000 + addr);
+	spin_unlock_irqrestore(&priv->lock, flags);
+	return data;
+}
+
+static void
+nv50_instobj_wr32(struct nouveau_object *object, u64 offset, u32 data)
+{
+	struct nv50_instmem_priv *priv = (void *)object->engine;
+	struct nv50_instobj_priv *node = (void *)object;
+	unsigned long flags;
+	u64 base = (node->mem->offset + offset) & 0xffffff00000ULL;
+	u64 addr = (node->mem->offset + offset) & 0x000000fffffULL;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	if (unlikely(priv->addr != base)) {
+		nv_wr32(priv, 0x001700, base >> 16);
+		priv->addr = base;
+	}
+	nv_wr32(priv, 0x700000 + addr, data);
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+static struct nouveau_oclass
+nv50_instobj_oclass = {
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_instobj_ctor,
+		.dtor = nv50_instobj_dtor,
+		.init = _nouveau_instobj_init,
+		.fini = _nouveau_instobj_fini,
+		.rd32 = nv50_instobj_rd32,
+		.wr32 = nv50_instobj_wr32,
+	},
+};
+
+static int
+nv50_instmem_alloc(struct nouveau_instmem *imem, struct nouveau_object *parent,
+		   u32 size, u32 align, struct nouveau_object **pobject)
+{
+	struct nouveau_object *engine = nv_object(imem);
+	return nouveau_object_ctor(parent, engine, &nv50_instobj_oclass,
+				   (void *)(unsigned long)align, size, pobject);
+}
+
+static int
+nv50_instmem_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		  struct nouveau_oclass *oclass, void *data, u32 size,
+		  struct nouveau_object **pobject)
+{
+	struct nv50_instmem_priv *priv;
+	int ret;
+
+	ret = nouveau_instmem_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	spin_lock_init(&priv->lock);
+	priv->base.alloc = nv50_instmem_alloc;
+	return 0;
+}
+
+static int
+nv50_instmem_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv50_instmem_priv *priv = (void *)object;
+	priv->addr = ~0ULL;
+	return nouveau_instmem_fini(&priv->base, suspend);
+}
+
+struct nouveau_oclass
+nv50_instmem_oclass = {
+	.handle = NV_SUBDEV(INSTMEM, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_instmem_ctor,
+		.dtor = _nouveau_instmem_dtor,
+		.init = _nouveau_instmem_init,
+		.fini = nv50_instmem_fini,
+	},
+};
diff --git a/nvkm/subdev/ltcg/Makefile.am b/nvkm/subdev/ltcg/Makefile.am
new file mode 100644
index 000000000..986c677ef
--- /dev/null
+++ b/nvkm/subdev/ltcg/Makefile.am
@@ -0,0 +1,5 @@
+noinst_LTLIBRARIES = libltcg.la
+
+libltcg_la_SOURCES = nvc0.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/ltcg/nvc0.c b/nvkm/subdev/ltcg/nvc0.c
new file mode 100644
index 000000000..078a2b9d6
--- /dev/null
+++ b/nvkm/subdev/ltcg/nvc0.c
@@ -0,0 +1,93 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/ltcg.h>
+
+struct nvc0_ltcg_priv {
+	struct nouveau_ltcg base;
+	u32 subp_nr;
+};
+
+static void
+nvc0_ltcg_subp_isr(struct nvc0_ltcg_priv *priv, int unit, int subp)
+{
+	u32 subp_base = 0x141000 + (unit * 0x2000) + (subp * 0x400);
+	u32 stat = nv_rd32(priv, subp_base + 0x020);
+
+	if (stat) {
+		nv_info(priv, "LTC%d_LTS%d: 0x%08x\n", unit, subp, stat);
+		nv_wr32(priv, subp_base + 0x020, stat);
+	}
+}
+
+static void
+nvc0_ltcg_intr(struct nouveau_subdev *subdev)
+{
+	struct nvc0_ltcg_priv *priv = (void *)subdev;
+	u32 units;
+
+	units = nv_rd32(priv, 0x00017c);
+	while (units) {
+		u32 subp, unit = ffs(units) - 1;
+		for (subp = 0; subp < priv->subp_nr; subp++)
+			nvc0_ltcg_subp_isr(priv, unit, subp);
+		units &= ~(1 << unit);
+	}
+
+	/* we do something horribly wrong and upset PMFB a lot, so mask off
+	 * interrupts from it after the first one until it's fixed
+	 */
+	nv_mask(priv, 0x000640, 0x02000000, 0x00000000);
+}
+
+static int
+nvc0_ltcg_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	       struct nouveau_oclass *oclass, void *data, u32 size,
+	       struct nouveau_object **pobject)
+{
+	struct nvc0_ltcg_priv *priv;
+	int ret;
+
+	ret = nouveau_ltcg_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->subp_nr = nv_rd32(priv, 0x17e8dc) >> 24;
+	nv_mask(priv, 0x17e820, 0x00100000, 0x00000000); /* INTR_EN &= ~0x10 */
+
+	nv_subdev(priv)->intr = nvc0_ltcg_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_ltcg_oclass = {
+	.handle = NV_SUBDEV(LTCG, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_ltcg_ctor,
+		.dtor = _nouveau_ltcg_dtor,
+		.init = _nouveau_ltcg_init,
+		.fini = _nouveau_ltcg_fini,
+	},
+};
diff --git a/nvkm/subdev/mc/Makefile.am b/nvkm/subdev/mc/Makefile.am
new file mode 100644
index 000000000..96d1aa790
--- /dev/null
+++ b/nvkm/subdev/mc/Makefile.am
@@ -0,0 +1,10 @@
+noinst_LTLIBRARIES = libmc.la
+
+libmc_la_SOURCES = base.c \
+		   nv04.c \
+		   nv44.c \
+		   nv50.c \
+		   nv98.c \
+		   nvc0.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/mc/base.c b/nvkm/subdev/mc/base.c
new file mode 100644
index 000000000..8379aafa6
--- /dev/null
+++ b/nvkm/subdev/mc/base.c
@@ -0,0 +1,49 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/mc.h>
+
+void
+nouveau_mc_intr(struct nouveau_subdev *subdev)
+{
+	struct nouveau_mc *pmc = nouveau_mc(subdev);
+	const struct nouveau_mc_intr *map = pmc->intr_map;
+	struct nouveau_subdev *unit;
+	u32 stat, intr;
+
+	intr = stat = nv_rd32(pmc, 0x000100);
+	while (stat && map->stat) {
+		if (stat & map->stat) {
+			unit = nouveau_subdev(subdev, map->unit);
+			if (unit && unit->intr)
+				unit->intr(unit);
+			intr &= ~map->stat;
+		}
+		map++;
+	}
+
+	if (intr) {
+		nv_error(pmc, "unknown intr 0x%08x\n", stat);
+	}
+}
diff --git a/nvkm/subdev/mc/nv04.c b/nvkm/subdev/mc/nv04.c
new file mode 100644
index 000000000..23ebe477a
--- /dev/null
+++ b/nvkm/subdev/mc/nv04.c
@@ -0,0 +1,83 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/mc.h>
+
+struct nv04_mc_priv {
+	struct nouveau_mc base;
+};
+
+const struct nouveau_mc_intr
+nv04_mc_intr[] = {
+	{ 0x00000001, NVDEV_ENGINE_MPEG },	/* NV17- MPEG/ME */
+	{ 0x00000100, NVDEV_ENGINE_FIFO },
+	{ 0x00001000, NVDEV_ENGINE_GR },
+	{ 0x00020000, NVDEV_ENGINE_VP },	/* NV40- */
+	{ 0x00100000, NVDEV_SUBDEV_TIMER },
+	{ 0x01000000, NVDEV_ENGINE_DISP },	/* NV04- PCRTC0 */
+	{ 0x02000000, NVDEV_ENGINE_DISP },	/* NV11- PCRTC1 */
+	{ 0x10000000, NVDEV_SUBDEV_GPIO },	/* PBUS */
+	{ 0x80000000, NVDEV_ENGINE_SW },
+	{}
+};
+
+static int
+nv04_mc_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv04_mc_priv *priv;
+	int ret;
+
+	ret = nouveau_mc_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->intr = nouveau_mc_intr;
+	priv->base.intr_map = nv04_mc_intr;
+	return 0;
+}
+
+int
+nv04_mc_init(struct nouveau_object *object)
+{
+	struct nv04_mc_priv *priv = (void *)object;
+
+	nv_wr32(priv, 0x000200, 0xffffffff); /* everything enabled */
+	nv_wr32(priv, 0x001850, 0x00000001); /* disable rom access */
+
+	return nouveau_mc_init(&priv->base);
+}
+
+struct nouveau_oclass
+nv04_mc_oclass = {
+	.handle = NV_SUBDEV(MC, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_mc_ctor,
+		.dtor = _nouveau_mc_dtor,
+		.init = nv04_mc_init,
+		.fini = _nouveau_mc_fini,
+	},
+};
diff --git a/nvkm/subdev/mc/nv44.c b/nvkm/subdev/mc/nv44.c
new file mode 100644
index 000000000..397d86835
--- /dev/null
+++ b/nvkm/subdev/mc/nv44.c
@@ -0,0 +1,74 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/mc.h>
+
+struct nv44_mc_priv {
+	struct nouveau_mc base;
+};
+
+static int
+nv44_mc_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv44_mc_priv *priv;
+	int ret;
+
+	ret = nouveau_mc_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->intr = nouveau_mc_intr;
+	priv->base.intr_map = nv04_mc_intr;
+	return 0;
+}
+
+static int
+nv44_mc_init(struct nouveau_object *object)
+{
+	struct nv44_mc_priv *priv = (void *)object;
+	u32 tmp = nv_rd32(priv, 0x10020c);
+
+	nv_wr32(priv, 0x000200, 0xffffffff); /* everything enabled */
+
+	nv_wr32(priv, 0x001700, tmp);
+	nv_wr32(priv, 0x001704, 0);
+	nv_wr32(priv, 0x001708, 0);
+	nv_wr32(priv, 0x00170c, tmp);
+
+	return nouveau_mc_init(&priv->base);
+}
+
+struct nouveau_oclass
+nv44_mc_oclass = {
+	.handle = NV_SUBDEV(MC, 0x44),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv44_mc_ctor,
+		.dtor = _nouveau_mc_dtor,
+		.init = nv44_mc_init,
+		.fini = _nouveau_mc_fini,
+	},
+};
diff --git a/nvkm/subdev/mc/nv50.c b/nvkm/subdev/mc/nv50.c
new file mode 100644
index 000000000..8d759f830
--- /dev/null
+++ b/nvkm/subdev/mc/nv50.c
@@ -0,0 +1,81 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/mc.h>
+
+struct nv50_mc_priv {
+	struct nouveau_mc base;
+};
+
+static const struct nouveau_mc_intr
+nv50_mc_intr[] = {
+	{ 0x00000001, NVDEV_ENGINE_MPEG },
+	{ 0x00000100, NVDEV_ENGINE_FIFO },
+	{ 0x00001000, NVDEV_ENGINE_GR },
+	{ 0x00004000, NVDEV_ENGINE_CRYPT },	/* NV84- */
+	{ 0x00008000, NVDEV_ENGINE_BSP },	/* NV84- */
+	{ 0x00100000, NVDEV_SUBDEV_TIMER },
+	{ 0x00200000, NVDEV_SUBDEV_GPIO },
+	{ 0x04000000, NVDEV_ENGINE_DISP },
+	{ 0x80000000, NVDEV_ENGINE_SW },
+	{ 0x0000d101, NVDEV_SUBDEV_FB },
+	{},
+};
+
+static int
+nv50_mc_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv50_mc_priv *priv;
+	int ret;
+
+	ret = nouveau_mc_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->intr = nouveau_mc_intr;
+	priv->base.intr_map = nv50_mc_intr;
+	return 0;
+}
+
+int
+nv50_mc_init(struct nouveau_object *object)
+{
+	struct nv50_mc_priv *priv = (void *)object;
+	nv_wr32(priv, 0x000200, 0xffffffff); /* everything on */
+	return nouveau_mc_init(&priv->base);
+}
+
+struct nouveau_oclass
+nv50_mc_oclass = {
+	.handle = NV_SUBDEV(MC, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_mc_ctor,
+		.dtor = _nouveau_mc_dtor,
+		.init = nv50_mc_init,
+		.fini = _nouveau_mc_fini,
+	},
+};
diff --git a/nvkm/subdev/mc/nv98.c b/nvkm/subdev/mc/nv98.c
new file mode 100644
index 000000000..ceb5c83f9
--- /dev/null
+++ b/nvkm/subdev/mc/nv98.c
@@ -0,0 +1,74 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/mc.h>
+
+struct nv98_mc_priv {
+	struct nouveau_mc base;
+};
+
+static const struct nouveau_mc_intr
+nv98_mc_intr[] = {
+	{ 0x00000001, NVDEV_ENGINE_PPP },
+	{ 0x00000100, NVDEV_ENGINE_FIFO },
+	{ 0x00001000, NVDEV_ENGINE_GR },
+	{ 0x00004000, NVDEV_ENGINE_CRYPT },	/* NV84:NVA3 */
+	{ 0x00008000, NVDEV_ENGINE_BSP },
+	{ 0x00100000, NVDEV_SUBDEV_TIMER },
+	{ 0x00200000, NVDEV_SUBDEV_GPIO },
+	{ 0x00400000, NVDEV_ENGINE_COPY0 },	/* NVA3-     */
+	{ 0x04000000, NVDEV_ENGINE_DISP },
+	{ 0x80000000, NVDEV_ENGINE_SW },
+	{ 0x0040d101, NVDEV_SUBDEV_FB },
+	{},
+};
+
+static int
+nv98_mc_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nv98_mc_priv *priv;
+	int ret;
+
+	ret = nouveau_mc_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->intr = nouveau_mc_intr;
+	priv->base.intr_map = nv98_mc_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nv98_mc_oclass = {
+	.handle = NV_SUBDEV(MC, 0x98),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv98_mc_ctor,
+		.dtor = _nouveau_mc_dtor,
+		.init = nv50_mc_init,
+		.fini = _nouveau_mc_fini,
+	},
+};
diff --git a/nvkm/subdev/mc/nvc0.c b/nvkm/subdev/mc/nvc0.c
new file mode 100644
index 000000000..927966827
--- /dev/null
+++ b/nvkm/subdev/mc/nvc0.c
@@ -0,0 +1,76 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/mc.h>
+
+struct nvc0_mc_priv {
+	struct nouveau_mc base;
+};
+
+static const struct nouveau_mc_intr
+nvc0_mc_intr[] = {
+	{ 0x00000001, NVDEV_ENGINE_PPP },
+	{ 0x00000020, NVDEV_ENGINE_COPY0 },
+	{ 0x00000040, NVDEV_ENGINE_COPY1 },
+	{ 0x00000100, NVDEV_ENGINE_FIFO },
+	{ 0x00001000, NVDEV_ENGINE_GR },
+	{ 0x00008000, NVDEV_ENGINE_BSP },
+	{ 0x00020000, NVDEV_ENGINE_VP },
+	{ 0x00100000, NVDEV_SUBDEV_TIMER },
+	{ 0x00200000, NVDEV_SUBDEV_GPIO },
+	{ 0x02000000, NVDEV_SUBDEV_LTCG },
+	{ 0x04000000, NVDEV_ENGINE_DISP },
+	{ 0x40000000, NVDEV_SUBDEV_IBUS },
+	{ 0x80000000, NVDEV_ENGINE_SW },
+	{},
+};
+
+static int
+nvc0_mc_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	     struct nouveau_oclass *oclass, void *data, u32 size,
+	     struct nouveau_object **pobject)
+{
+	struct nvc0_mc_priv *priv;
+	int ret;
+
+	ret = nouveau_mc_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	nv_subdev(priv)->intr = nouveau_mc_intr;
+	priv->base.intr_map = nvc0_mc_intr;
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_mc_oclass = {
+	.handle = NV_SUBDEV(MC, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_mc_ctor,
+		.dtor = _nouveau_mc_dtor,
+		.init = nv50_mc_init,
+		.fini = _nouveau_mc_fini,
+	},
+};
diff --git a/nvkm/subdev/mxm/Makefile.am b/nvkm/subdev/mxm/Makefile.am
new file mode 100644
index 000000000..2ec56b33b
--- /dev/null
+++ b/nvkm/subdev/mxm/Makefile.am
@@ -0,0 +1,7 @@
+noinst_LTLIBRARIES = libmxm.la
+
+libmxm_la_SOURCES = base.c \
+		    mxms.c \
+		    nv50.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/mxm/base.c b/nvkm/subdev/mxm/base.c
new file mode 100644
index 000000000..e286e132c
--- /dev/null
+++ b/nvkm/subdev/mxm/base.c
@@ -0,0 +1,290 @@
+/*
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/option.h>
+
+#include <subdev/i2c.h>
+#include <subdev/mxm.h>
+#include <subdev/bios.h>
+#include <subdev/bios/mxm.h>
+
+#include "mxms.h"
+
+static bool
+mxm_shadow_rom_fetch(struct nouveau_i2c_port *i2c, u8 addr,
+		     u8 offset, u8 size, u8 *data)
+{
+	struct i2c_msg msgs[] = {
+		{ .addr = addr, .flags = 0, .len = 1, .buf = &offset },
+		{ .addr = addr, .flags = I2C_M_RD, .len = size, .buf = data, },
+	};
+
+	return i2c_transfer(&i2c->adapter, msgs, 2) == 2;
+}
+
+static bool
+mxm_shadow_rom(struct nouveau_mxm *mxm, u8 version)
+{
+	struct nouveau_bios *bios = nouveau_bios(mxm);
+	struct nouveau_i2c *i2c = nouveau_i2c(mxm);
+	struct nouveau_i2c_port *port = NULL;
+	u8 i2cidx, mxms[6], addr, size;
+
+	i2cidx = mxm_ddc_map(bios, 1 /* LVDS_DDC */) & 0x0f;
+	if (i2cidx < 0x0f)
+		port = i2c->find(i2c, i2cidx);
+	if (!port)
+		return false;
+
+	addr = 0x54;
+	if (!mxm_shadow_rom_fetch(port, addr, 0, 6, mxms)) {
+		addr = 0x56;
+		if (!mxm_shadow_rom_fetch(port, addr, 0, 6, mxms))
+			return false;
+	}
+
+	mxm->mxms = mxms;
+	size = mxms_headerlen(mxm) + mxms_structlen(mxm);
+	mxm->mxms = kmalloc(size, GFP_KERNEL);
+
+	if (mxm->mxms &&
+	    mxm_shadow_rom_fetch(port, addr, 0, size, mxm->mxms))
+		return true;
+
+	kfree(mxm->mxms);
+	mxm->mxms = NULL;
+	return false;
+}
+
+#if defined(CONFIG_ACPI)
+static bool
+mxm_shadow_dsm(struct nouveau_mxm *mxm, u8 version)
+{
+	struct nouveau_device *device = nv_device(mxm);
+	static char muid[] = {
+		0x00, 0xA4, 0x04, 0x40, 0x7D, 0x91, 0xF2, 0x4C,
+		0xB8, 0x9C, 0x79, 0xB6, 0x2F, 0xD5, 0x56, 0x65
+	};
+	u32 mxms_args[] = { 0x00000000 };
+	union acpi_object args[4] = {
+		/* _DSM MUID */
+		{ .buffer.type = 3,
+		  .buffer.length = sizeof(muid),
+		  .buffer.pointer = muid,
+		},
+		/* spec says this can be zero to mean "highest revision", but
+		 * of course there's at least one bios out there which fails
+		 * unless you pass in exactly the version it supports..
+		 */
+		{ .integer.type = ACPI_TYPE_INTEGER,
+		  .integer.value = (version & 0xf0) << 4 | (version & 0x0f),
+		},
+		/* MXMS function */
+		{ .integer.type = ACPI_TYPE_INTEGER,
+		  .integer.value = 0x00000010,
+		},
+		/* Pointer to MXMS arguments */
+		{ .buffer.type = ACPI_TYPE_BUFFER,
+		  .buffer.length = sizeof(mxms_args),
+		  .buffer.pointer = (char *)mxms_args,
+		},
+	};
+	struct acpi_object_list list = { ARRAY_SIZE(args), args };
+	struct acpi_buffer retn = { ACPI_ALLOCATE_BUFFER, NULL };
+	union acpi_object *obj;
+	acpi_handle handle;
+	int ret;
+
+	handle = DEVICE_ACPI_HANDLE(&device->pdev->dev);
+	if (!handle)
+		return false;
+
+	ret = acpi_evaluate_object(handle, "_DSM", &list, &retn);
+	if (ret) {
+		nv_debug(mxm, "DSM MXMS failed: %d\n", ret);
+		return false;
+	}
+
+	obj = retn.pointer;
+	if (obj->type == ACPI_TYPE_BUFFER) {
+		mxm->mxms = kmemdup(obj->buffer.pointer,
+					 obj->buffer.length, GFP_KERNEL);
+	} else
+	if (obj->type == ACPI_TYPE_INTEGER) {
+		nv_debug(mxm, "DSM MXMS returned 0x%llx\n", obj->integer.value);
+	}
+
+	kfree(obj);
+	return mxm->mxms != NULL;
+}
+#endif
+
+#if defined(CONFIG_ACPI_WMI) || defined(CONFIG_ACPI_WMI_MODULE)
+
+#define WMI_WMMX_GUID "F6CB5C3C-9CAE-4EBD-B577-931EA32A2CC0"
+
+static u8
+wmi_wmmx_mxmi(struct nouveau_mxm *mxm, u8 version)
+{
+	u32 mxmi_args[] = { 0x494D584D /* MXMI */, version, 0 };
+	struct acpi_buffer args = { sizeof(mxmi_args), mxmi_args };
+	struct acpi_buffer retn = { ACPI_ALLOCATE_BUFFER, NULL };
+	union acpi_object *obj;
+	acpi_status status;
+
+	status = wmi_evaluate_method(WMI_WMMX_GUID, 0, 0, &args, &retn);
+	if (ACPI_FAILURE(status)) {
+		nv_debug(mxm, "WMMX MXMI returned %d\n", status);
+		return 0x00;
+	}
+
+	obj = retn.pointer;
+	if (obj->type == ACPI_TYPE_INTEGER) {
+		version = obj->integer.value;
+		nv_debug(mxm, "WMMX MXMI version %d.%d\n",
+			     (version >> 4), version & 0x0f);
+	} else {
+		version = 0;
+		nv_debug(mxm, "WMMX MXMI returned non-integer\n");
+	}
+
+	kfree(obj);
+	return version;
+}
+
+static bool
+mxm_shadow_wmi(struct nouveau_mxm *mxm, u8 version)
+{
+	u32 mxms_args[] = { 0x534D584D /* MXMS */, version, 0 };
+	struct acpi_buffer args = { sizeof(mxms_args), mxms_args };
+	struct acpi_buffer retn = { ACPI_ALLOCATE_BUFFER, NULL };
+	union acpi_object *obj;
+	acpi_status status;
+
+	if (!wmi_has_guid(WMI_WMMX_GUID)) {
+		nv_debug(mxm, "WMMX GUID not found\n");
+		return false;
+	}
+
+	mxms_args[1] = wmi_wmmx_mxmi(mxm, 0x00);
+	if (!mxms_args[1])
+		mxms_args[1] = wmi_wmmx_mxmi(mxm, version);
+	if (!mxms_args[1])
+		return false;
+
+	status = wmi_evaluate_method(WMI_WMMX_GUID, 0, 0, &args, &retn);
+	if (ACPI_FAILURE(status)) {
+		nv_debug(mxm, "WMMX MXMS returned %d\n", status);
+		return false;
+	}
+
+	obj = retn.pointer;
+	if (obj->type == ACPI_TYPE_BUFFER) {
+		mxm->mxms = kmemdup(obj->buffer.pointer,
+					 obj->buffer.length, GFP_KERNEL);
+	}
+
+	kfree(obj);
+	return mxm->mxms != NULL;
+}
+#endif
+
+static struct mxm_shadow_h {
+	const char *name;
+	bool (*exec)(struct nouveau_mxm *, u8 version);
+} _mxm_shadow[] = {
+	{ "ROM", mxm_shadow_rom },
+#if defined(CONFIG_ACPI)
+	{ "DSM", mxm_shadow_dsm },
+#endif
+#if defined(CONFIG_ACPI_WMI) || defined(CONFIG_ACPI_WMI_MODULE)
+	{ "WMI", mxm_shadow_wmi },
+#endif
+	{}
+};
+
+static int
+mxm_shadow(struct nouveau_mxm *mxm, u8 version)
+{
+	struct mxm_shadow_h *shadow = _mxm_shadow;
+	do {
+		nv_debug(mxm, "checking %s\n", shadow->name);
+		if (shadow->exec(mxm, version)) {
+			if (mxms_valid(mxm))
+				return 0;
+			kfree(mxm->mxms);
+			mxm->mxms = NULL;
+		}
+	} while ((++shadow)->name);
+	return -ENOENT;
+}
+
+int
+nouveau_mxm_create_(struct nouveau_object *parent,
+		    struct nouveau_object *engine,
+		    struct nouveau_oclass *oclass, int length, void **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nouveau_bios *bios = nouveau_bios(device);
+	struct nouveau_mxm *mxm;
+	u8  ver, len;
+	u16 data;
+	int ret;
+
+	ret = nouveau_subdev_create_(parent, engine, oclass, 0, "MXM", "mxm",
+				     length, pobject);
+	mxm = *pobject;
+	if (ret)
+		return ret;
+
+	data = mxm_table(bios, &ver, &len);
+	if (!data || !(ver = nv_ro08(bios, data))) {
+		nv_debug(mxm, "no VBIOS data, nothing to do\n");
+		return 0;
+	}
+
+	nv_info(mxm, "BIOS version %d.%d\n", ver >> 4, ver & 0x0f);
+
+	if (mxm_shadow(mxm, ver)) {
+		nv_info(mxm, "failed to locate valid SIS\n");
+#if 0
+		/* we should, perhaps, fall back to some kind of limited
+		 * mode here if the x86 vbios hasn't already done the
+		 * work for us (so we prevent loading with completely
+		 * whacked vbios tables).
+		 */
+		return -EINVAL;
+#else
+		return 0;
+#endif
+	}
+
+	nv_info(mxm, "MXMS Version %d.%d\n",
+		mxms_version(mxm) >> 8, mxms_version(mxm) & 0xff);
+	mxms_foreach(mxm, 0, NULL, NULL);
+
+	if (nouveau_boolopt(device->cfgopt, "NvMXMDCB", true))
+		mxm->action |= MXM_SANITISE_DCB;
+	return 0;
+}
diff --git a/nvkm/subdev/mxm/mxms.c b/nvkm/subdev/mxm/mxms.c
new file mode 100644
index 000000000..839ca1edc
--- /dev/null
+++ b/nvkm/subdev/mxm/mxms.c
@@ -0,0 +1,193 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/mxm.h>
+#include "mxms.h"
+
+#define ROM16(x) le16_to_cpu(*(u16 *)&(x))
+#define ROM32(x) le32_to_cpu(*(u32 *)&(x))
+
+static u8 *
+mxms_data(struct nouveau_mxm *mxm)
+{
+	return mxm->mxms;
+
+}
+
+u16
+mxms_version(struct nouveau_mxm *mxm)
+{
+	u8 *mxms = mxms_data(mxm);
+	u16 version = (mxms[4] << 8) | mxms[5];
+	switch (version ) {
+	case 0x0200:
+	case 0x0201:
+	case 0x0300:
+		return version;
+	default:
+		break;
+	}
+
+	nv_debug(mxm, "unknown version %d.%d\n", mxms[4], mxms[5]);
+	return 0x0000;
+}
+
+u16
+mxms_headerlen(struct nouveau_mxm *mxm)
+{
+	return 8;
+}
+
+u16
+mxms_structlen(struct nouveau_mxm *mxm)
+{
+	return *(u16 *)&mxms_data(mxm)[6];
+}
+
+bool
+mxms_checksum(struct nouveau_mxm *mxm)
+{
+	u16 size = mxms_headerlen(mxm) + mxms_structlen(mxm);
+	u8 *mxms = mxms_data(mxm), sum = 0;
+	while (size--)
+		sum += *mxms++;
+	if (sum) {
+		nv_debug(mxm, "checksum invalid\n");
+		return false;
+	}
+	return true;
+}
+
+bool
+mxms_valid(struct nouveau_mxm *mxm)
+{
+	u8 *mxms = mxms_data(mxm);
+	if (*(u32 *)mxms != 0x5f4d584d) {
+		nv_debug(mxm, "signature invalid\n");
+		return false;
+	}
+
+	if (!mxms_version(mxm) || !mxms_checksum(mxm))
+		return false;
+
+	return true;
+}
+
+bool
+mxms_foreach(struct nouveau_mxm *mxm, u8 types,
+	     bool (*exec)(struct nouveau_mxm *, u8 *, void *), void *info)
+{
+	u8 *mxms = mxms_data(mxm);
+	u8 *desc = mxms + mxms_headerlen(mxm);
+	u8 *fini = desc + mxms_structlen(mxm) - 1;
+	while (desc < fini) {
+		u8 type = desc[0] & 0x0f;
+		u8 headerlen = 0;
+		u8 recordlen = 0;
+		u8 entries = 0;
+
+		switch (type) {
+		case 0: /* Output Device Structure */
+			if (mxms_version(mxm) >= 0x0300)
+				headerlen = 8;
+			else
+				headerlen = 6;
+			break;
+		case 1: /* System Cooling Capability Structure */
+		case 2: /* Thermal Structure */
+		case 3: /* Input Power Structure */
+			headerlen = 4;
+			break;
+		case 4: /* GPIO Device Structure */
+			headerlen = 4;
+			recordlen = 2;
+			entries   = (ROM32(desc[0]) & 0x01f00000) >> 20;
+			break;
+		case 5: /* Vendor Specific Structure */
+			headerlen = 8;
+			break;
+		case 6: /* Backlight Control Structure */
+			if (mxms_version(mxm) >= 0x0300) {
+				headerlen = 4;
+				recordlen = 8;
+				entries   = (desc[1] & 0xf0) >> 4;
+			} else {
+				headerlen = 8;
+			}
+			break;
+		case 7: /* Fan Control Structure */
+			headerlen = 8;
+			recordlen = 4;
+			entries   = desc[1] & 0x07;
+			break;
+		default:
+			nv_debug(mxm, "unknown descriptor type %d\n", type);
+			return false;
+		}
+
+		if (nv_subdev(mxm)->debug >= NV_DBG_DEBUG && (exec == NULL)) {
+			static const char * mxms_desc_name[] = {
+				"ODS", "SCCS", "TS", "IPS",
+				"GSD", "VSS", "BCS", "FCS",
+			};
+			u8 *dump = desc;
+			int i, j;
+
+			nv_debug(mxm, "%4s: ", mxms_desc_name[type]);
+			for (j = headerlen - 1; j >= 0; j--)
+				printk("%02x", dump[j]);
+			printk("\n");
+			dump += headerlen;
+
+			for (i = 0; i < entries; i++, dump += recordlen) {
+				nv_debug(mxm, "      ");
+				for (j = recordlen - 1; j >= 0; j--)
+					printk("%02x", dump[j]);
+				printk("\n");
+			}
+		}
+
+		if (types & (1 << type)) {
+			if (!exec(mxm, desc, info))
+				return false;
+		}
+
+		desc += headerlen + (entries * recordlen);
+	}
+
+	return true;
+}
+
+void
+mxms_output_device(struct nouveau_mxm *mxm, u8 *pdata, struct mxms_odev *desc)
+{
+	u64 data = ROM32(pdata[0]);
+	if (mxms_version(mxm) >= 0x0300)
+		data |= (u64)ROM16(pdata[4]) << 32;
+
+	desc->outp_type = (data & 0x00000000000000f0ULL) >> 4;
+	desc->ddc_port  = (data & 0x0000000000000f00ULL) >> 8;
+	desc->conn_type = (data & 0x000000000001f000ULL) >> 12;
+	desc->dig_conn  = (data & 0x0000000000780000ULL) >> 19;
+}
diff --git a/nvkm/subdev/mxm/mxms.h b/nvkm/subdev/mxm/mxms.h
new file mode 100644
index 000000000..5e0be0c59
--- /dev/null
+++ b/nvkm/subdev/mxm/mxms.h
@@ -0,0 +1,22 @@
+#ifndef __NVMXM_MXMS_H__
+#define __NVMXM_MXMS_H__
+
+struct mxms_odev {
+	u8 outp_type;
+	u8 conn_type;
+	u8 ddc_port;
+	u8 dig_conn;
+};
+
+void mxms_output_device(struct nouveau_mxm *, u8 *, struct mxms_odev *);
+
+u16  mxms_version(struct nouveau_mxm *);
+u16  mxms_headerlen(struct nouveau_mxm *);
+u16  mxms_structlen(struct nouveau_mxm *);
+bool mxms_checksum(struct nouveau_mxm *);
+bool mxms_valid(struct nouveau_mxm *);
+
+bool mxms_foreach(struct nouveau_mxm *, u8,
+		  bool (*)(struct nouveau_mxm *, u8 *, void *), void *);
+
+#endif
diff --git a/nvkm/subdev/mxm/nv50.c b/nvkm/subdev/mxm/nv50.c
new file mode 100644
index 000000000..af129c2e8
--- /dev/null
+++ b/nvkm/subdev/mxm/nv50.c
@@ -0,0 +1,233 @@
+/*
+ * Copyright 2011 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/mxm.h>
+#include <subdev/bios.h>
+#include <subdev/bios/conn.h>
+#include <subdev/bios/dcb.h>
+#include <subdev/bios/mxm.h>
+
+#include "mxms.h"
+
+struct nv50_mxm_priv {
+	struct nouveau_mxm base;
+};
+
+struct context {
+	u32 *outp;
+	struct mxms_odev desc;
+};
+
+static bool
+mxm_match_tmds_partner(struct nouveau_mxm *mxm, u8 *data, void *info)
+{
+	struct context *ctx = info;
+	struct mxms_odev desc;
+
+	mxms_output_device(mxm, data, &desc);
+	if (desc.outp_type == 2 &&
+	    desc.dig_conn == ctx->desc.dig_conn)
+		return false;
+	return true;
+}
+
+static bool
+mxm_match_dcb(struct nouveau_mxm *mxm, u8 *data, void *info)
+{
+	struct nouveau_bios *bios = nouveau_bios(mxm);
+	struct context *ctx = info;
+	u64 desc = *(u64 *)data;
+
+	mxms_output_device(mxm, data, &ctx->desc);
+
+	/* match dcb encoder type to mxm-ods device type */
+	if ((ctx->outp[0] & 0x0000000f) != ctx->desc.outp_type)
+		return true;
+
+	/* digital output, have some extra stuff to match here, there's a
+	 * table in the vbios that provides a mapping from the mxm digital
+	 * connection enum values to SOR/link
+	 */
+	if ((desc & 0x00000000000000f0) >= 0x20) {
+		/* check against sor index */
+		u8 link = mxm_sor_map(bios, ctx->desc.dig_conn);
+		if ((ctx->outp[0] & 0x0f000000) != (link & 0x0f) << 24)
+			return true;
+
+		/* check dcb entry has a compatible link field */
+		link = (link & 0x30) >> 4;
+		if ((link & ((ctx->outp[1] & 0x00000030) >> 4)) != link)
+			return true;
+	}
+
+	/* mark this descriptor accounted for by setting invalid device type,
+	 * except of course some manufactures don't follow specs properly and
+	 * we need to avoid killing off the TMDS function on DP connectors
+	 * if MXM-SIS is missing an entry for it.
+	 */
+	data[0] &= ~0xf0;
+	if (ctx->desc.outp_type == 6 && ctx->desc.conn_type == 6 &&
+	    mxms_foreach(mxm, 0x01, mxm_match_tmds_partner, ctx)) {
+		data[0] |= 0x20; /* modify descriptor to match TMDS now */
+	} else {
+		data[0] |= 0xf0;
+	}
+
+	return false;
+}
+
+static int
+mxm_dcb_sanitise_entry(struct nouveau_bios *bios, void *data, int idx, u16 pdcb)
+{
+	struct nouveau_mxm *mxm = nouveau_mxm(bios);
+	struct context ctx = { .outp = (u32 *)(bios->data + pdcb) };
+	u8 type, i2cidx, link, ver, len;
+	u8 *conn;
+
+	/* look for an output device structure that matches this dcb entry.
+	 * if one isn't found, disable it.
+	 */
+	if (mxms_foreach(mxm, 0x01, mxm_match_dcb, &ctx)) {
+		nv_debug(mxm, "disable %d: 0x%08x 0x%08x\n",
+			idx, ctx.outp[0], ctx.outp[1]);
+		ctx.outp[0] |= 0x0000000f;
+		return 0;
+	}
+
+	/* modify the output's ddc/aux port, there's a pointer to a table
+	 * with the mapping from mxm ddc/aux port to dcb i2c_index in the
+	 * vbios mxm table
+	 */
+	i2cidx = mxm_ddc_map(bios, ctx.desc.ddc_port);
+	if ((ctx.outp[0] & 0x0000000f) != DCB_OUTPUT_DP)
+		i2cidx = (i2cidx & 0x0f) << 4;
+	else
+		i2cidx = (i2cidx & 0xf0);
+
+	if (i2cidx != 0xf0) {
+		ctx.outp[0] &= ~0x000000f0;
+		ctx.outp[0] |= i2cidx;
+	}
+
+	/* override dcb sorconf.link, based on what mxm data says */
+	switch (ctx.desc.outp_type) {
+	case 0x00: /* Analog CRT */
+	case 0x01: /* Analog TV/HDTV */
+		break;
+	default:
+		link = mxm_sor_map(bios, ctx.desc.dig_conn) & 0x30;
+		ctx.outp[1] &= ~0x00000030;
+		ctx.outp[1] |= link;
+		break;
+	}
+
+	/* we may need to fixup various other vbios tables based on what
+	 * the descriptor says the connector type should be.
+	 *
+	 * in a lot of cases, the vbios tables will claim DVI-I is possible,
+	 * and the mxm data says the connector is really HDMI.  another
+	 * common example is DP->eDP.
+	 */
+	conn  = bios->data;
+	conn += dcb_conn(bios, (ctx.outp[0] & 0x0000f000) >> 12, &ver, &len);
+	type  = conn[0];
+	switch (ctx.desc.conn_type) {
+	case 0x01: /* LVDS */
+		ctx.outp[1] |= 0x00000004; /* use_power_scripts */
+		/* XXX: modify default link width in LVDS table */
+		break;
+	case 0x02: /* HDMI */
+		type = DCB_CONNECTOR_HDMI_1;
+		break;
+	case 0x03: /* DVI-D */
+		type = DCB_CONNECTOR_DVI_D;
+		break;
+	case 0x0e: /* eDP, falls through to DPint */
+		ctx.outp[1] |= 0x00010000;
+	case 0x07: /* DP internal, wtf is this?? HP8670w */
+		ctx.outp[1] |= 0x00000004; /* use_power_scripts? */
+		type = DCB_CONNECTOR_eDP;
+		break;
+	default:
+		break;
+	}
+
+	if (mxms_version(mxm) >= 0x0300)
+		conn[0] = type;
+
+	return 0;
+}
+
+static bool
+mxm_show_unmatched(struct nouveau_mxm *mxm, u8 *data, void *info)
+{
+	u64 desc = *(u64 *)data;
+	if ((desc & 0xf0) != 0xf0)
+	nv_info(mxm, "unmatched output device 0x%016llx\n", desc);
+	return true;
+}
+
+static void
+mxm_dcb_sanitise(struct nouveau_mxm *mxm)
+{
+	struct nouveau_bios *bios = nouveau_bios(mxm);
+	u8  ver, hdr, cnt, len;
+	u16 dcb = dcb_table(bios, &ver, &hdr, &cnt, &len);
+	if (dcb == 0x0000 || ver != 0x40) {
+		nv_debug(mxm, "unsupported DCB version\n");
+		return;
+	}
+
+	dcb_outp_foreach(bios, NULL, mxm_dcb_sanitise_entry);
+	mxms_foreach(mxm, 0x01, mxm_show_unmatched, NULL);
+}
+
+static int
+nv50_mxm_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+	      struct nouveau_oclass *oclass, void *data, u32 size,
+	      struct nouveau_object **pobject)
+{
+	struct nv50_mxm_priv *priv;
+	int ret;
+
+	ret = nouveau_mxm_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	if (priv->base.action & MXM_SANITISE_DCB)
+		mxm_dcb_sanitise(&priv->base);
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_mxm_oclass = {
+	.handle = NV_SUBDEV(MXM, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_mxm_ctor,
+		.dtor = _nouveau_mxm_dtor,
+		.init = _nouveau_mxm_init,
+		.fini = _nouveau_mxm_fini,
+	},
+};
diff --git a/nvkm/subdev/therm/Makefile.am b/nvkm/subdev/therm/Makefile.am
new file mode 100644
index 000000000..bea5008c2
--- /dev/null
+++ b/nvkm/subdev/therm/Makefile.am
@@ -0,0 +1,10 @@
+noinst_LTLIBRARIES = libtherm.la
+
+libtherm_la_SOURCES = base.c \
+		      fan.c \
+		      ic.c \
+		      temp.c \
+		      nv40.c \
+		      nv50.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/therm/base.c b/nvkm/subdev/therm/base.c
new file mode 100644
index 000000000..1674c74a7
--- /dev/null
+++ b/nvkm/subdev/therm/base.c
@@ -0,0 +1,144 @@
+/*
+ * Copyright 2012 The Nouveau community
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Martin Peres
+ */
+
+#include <core/object.h>
+#include <core/device.h>
+
+#include <subdev/bios.h>
+
+#include "priv.h"
+
+int
+nouveau_therm_attr_get(struct nouveau_therm *therm,
+		       enum nouveau_therm_attr_type type)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+
+	switch (type) {
+	case NOUVEAU_THERM_ATTR_FAN_MIN_DUTY:
+		return priv->bios_fan.min_duty;
+	case NOUVEAU_THERM_ATTR_FAN_MAX_DUTY:
+		return priv->bios_fan.max_duty;
+	case NOUVEAU_THERM_ATTR_FAN_MODE:
+		return priv->fan.mode;
+	case NOUVEAU_THERM_ATTR_THRS_FAN_BOOST:
+		return priv->bios_sensor.thrs_fan_boost.temp;
+	case NOUVEAU_THERM_ATTR_THRS_FAN_BOOST_HYST:
+		return priv->bios_sensor.thrs_fan_boost.hysteresis;
+	case NOUVEAU_THERM_ATTR_THRS_DOWN_CLK:
+		return priv->bios_sensor.thrs_down_clock.temp;
+	case NOUVEAU_THERM_ATTR_THRS_DOWN_CLK_HYST:
+		return priv->bios_sensor.thrs_down_clock.hysteresis;
+	case NOUVEAU_THERM_ATTR_THRS_CRITICAL:
+		return priv->bios_sensor.thrs_critical.temp;
+	case NOUVEAU_THERM_ATTR_THRS_CRITICAL_HYST:
+		return priv->bios_sensor.thrs_critical.hysteresis;
+	case NOUVEAU_THERM_ATTR_THRS_SHUTDOWN:
+		return priv->bios_sensor.thrs_shutdown.temp;
+	case NOUVEAU_THERM_ATTR_THRS_SHUTDOWN_HYST:
+		return priv->bios_sensor.thrs_shutdown.hysteresis;
+	}
+
+	return -EINVAL;
+}
+
+int
+nouveau_therm_attr_set(struct nouveau_therm *therm,
+		       enum nouveau_therm_attr_type type, int value)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+
+	switch (type) {
+	case NOUVEAU_THERM_ATTR_FAN_MIN_DUTY:
+		if (value < 0)
+			value = 0;
+		if (value > priv->bios_fan.max_duty)
+			value = priv->bios_fan.max_duty;
+		priv->bios_fan.min_duty = value;
+		return 0;
+	case NOUVEAU_THERM_ATTR_FAN_MAX_DUTY:
+		if (value < 0)
+			value = 0;
+		if (value < priv->bios_fan.min_duty)
+			value = priv->bios_fan.min_duty;
+		priv->bios_fan.max_duty = value;
+		return 0;
+	case NOUVEAU_THERM_ATTR_FAN_MODE:
+		return nouveau_therm_fan_set_mode(therm, value);
+	case NOUVEAU_THERM_ATTR_THRS_FAN_BOOST:
+		priv->bios_sensor.thrs_fan_boost.temp = value;
+		return 0;
+	case NOUVEAU_THERM_ATTR_THRS_FAN_BOOST_HYST:
+		priv->bios_sensor.thrs_fan_boost.hysteresis = value;
+		return 0;
+	case NOUVEAU_THERM_ATTR_THRS_DOWN_CLK:
+		priv->bios_sensor.thrs_down_clock.temp = value;
+		return 0;
+	case NOUVEAU_THERM_ATTR_THRS_DOWN_CLK_HYST:
+		priv->bios_sensor.thrs_down_clock.hysteresis = value;
+		return 0;
+	case NOUVEAU_THERM_ATTR_THRS_CRITICAL:
+		priv->bios_sensor.thrs_critical.temp = value;
+		return 0;
+	case NOUVEAU_THERM_ATTR_THRS_CRITICAL_HYST:
+		priv->bios_sensor.thrs_critical.hysteresis = value;
+		return 0;
+	case NOUVEAU_THERM_ATTR_THRS_SHUTDOWN:
+		priv->bios_sensor.thrs_shutdown.temp = value;
+		return 0;
+	case NOUVEAU_THERM_ATTR_THRS_SHUTDOWN_HYST:
+		priv->bios_sensor.thrs_shutdown.hysteresis = value;
+		return 0;
+	}
+
+	return -EINVAL;
+}
+
+int
+nouveau_therm_init(struct nouveau_object *object)
+{
+	struct nouveau_therm *therm = (void *)object;
+	struct nouveau_therm_priv *priv = (void *)therm;
+	int ret;
+
+	ret = nouveau_subdev_init(&therm->base);
+	if (ret)
+		return ret;
+
+	if (priv->fan.percent >= 0)
+		therm->fan_set(therm, priv->fan.percent);
+
+	return 0;
+}
+
+int
+nouveau_therm_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nouveau_therm *therm = (void *)object;
+	struct nouveau_therm_priv *priv = (void *)therm;
+
+	priv->fan.percent = therm->fan_get(therm);
+
+	return nouveau_subdev_fini(&therm->base, suspend);
+}
diff --git a/nvkm/subdev/therm/fan.c b/nvkm/subdev/therm/fan.c
new file mode 100644
index 000000000..523178685
--- /dev/null
+++ b/nvkm/subdev/therm/fan.c
@@ -0,0 +1,234 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ * 	    Martin Peres
+ */
+
+#include "priv.h"
+
+#include <core/object.h>
+#include <core/device.h>
+#include <subdev/gpio.h>
+#include <subdev/timer.h>
+
+int
+nouveau_therm_fan_get(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	struct nouveau_gpio *gpio = nouveau_gpio(therm);
+	struct dcb_gpio_func func;
+	int card_type = nv_device(therm)->card_type;
+	u32 divs, duty;
+	int ret;
+
+	if (!priv->fan.pwm_get)
+		return -ENODEV;
+
+	ret = gpio->find(gpio, 0, DCB_GPIO_PWM_FAN, 0xff, &func);
+	if (ret == 0) {
+		ret = priv->fan.pwm_get(therm, func.line, &divs, &duty);
+		if (ret == 0 && divs) {
+			divs = max(divs, duty);
+			if (card_type <= NV_40 || (func.log[0] & 1))
+				duty = divs - duty;
+			return (duty * 100) / divs;
+		}
+
+		return gpio->get(gpio, 0, func.func, func.line) * 100;
+	}
+
+	return -ENODEV;
+}
+
+int
+nouveau_therm_fan_set(struct nouveau_therm *therm, int percent)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	struct nouveau_gpio *gpio = nouveau_gpio(therm);
+	struct dcb_gpio_func func;
+	int card_type = nv_device(therm)->card_type;
+	u32 divs, duty;
+	int ret;
+
+	if (priv->fan.mode == FAN_CONTROL_NONE)
+		return -EINVAL;
+
+	if (!priv->fan.pwm_set)
+		return -ENODEV;
+
+	if (percent < priv->bios_fan.min_duty)
+		percent = priv->bios_fan.min_duty;
+	if (percent > priv->bios_fan.max_duty)
+		percent = priv->bios_fan.max_duty;
+
+	ret = gpio->find(gpio, 0, DCB_GPIO_PWM_FAN, 0xff, &func);
+	if (ret == 0) {
+		divs = priv->bios_perf_fan.pwm_divisor;
+		if (priv->bios_fan.pwm_freq) {
+			divs = 1;
+			if (priv->fan.pwm_clock)
+				divs = priv->fan.pwm_clock(therm);
+			divs /= priv->bios_fan.pwm_freq;
+		}
+
+		duty = ((divs * percent) + 99) / 100;
+		if (card_type <= NV_40 || (func.log[0] & 1))
+			duty = divs - duty;
+
+		ret = priv->fan.pwm_set(therm, func.line, divs, duty);
+		return ret;
+	}
+
+	return -ENODEV;
+}
+
+int
+nouveau_therm_fan_sense(struct nouveau_therm *therm)
+{
+	struct nouveau_timer *ptimer = nouveau_timer(therm);
+	struct nouveau_gpio *gpio = nouveau_gpio(therm);
+	struct dcb_gpio_func func;
+	u32 cycles, cur, prev;
+	u64 start, end, tach;
+
+	if (gpio->find(gpio, 0, DCB_GPIO_FAN_SENSE, 0xff, &func))
+		return -ENODEV;
+
+	/* Time a complete rotation and extrapolate to RPM:
+	 * When the fan spins, it changes the value of GPIO FAN_SENSE.
+	 * We get 4 changes (0 -> 1 -> 0 -> 1) per complete rotation.
+	 */
+	start = ptimer->read(ptimer);
+	prev = gpio->get(gpio, 0, func.func, func.line);
+	cycles = 0;
+	do {
+		usleep_range(500, 1000); /* supports 0 < rpm < 7500 */
+
+		cur = gpio->get(gpio, 0, func.func, func.line);
+		if (prev != cur) {
+			if (!start)
+				start = ptimer->read(ptimer);
+			cycles++;
+			prev = cur;
+		}
+	} while (cycles < 5 && ptimer->read(ptimer) - start < 250000000);
+	end = ptimer->read(ptimer);
+
+	if (cycles == 5) {
+		tach = (u64)60000000000ULL;
+		do_div(tach, (end - start));
+		return tach;
+	} else
+		return 0;
+}
+
+int
+nouveau_therm_fan_set_mode(struct nouveau_therm *therm,
+			   enum nouveau_therm_fan_mode mode)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+
+	if (priv->fan.mode == mode)
+		return 0;
+
+	if (mode < FAN_CONTROL_NONE || mode >= FAN_CONTROL_NR)
+		return -EINVAL;
+
+	switch (mode)
+	{
+	case FAN_CONTROL_NONE:
+		nv_info(therm, "switch fan to no-control mode\n");
+		break;
+	case FAN_CONTROL_MANUAL:
+		nv_info(therm, "switch fan to manual mode\n");
+		break;
+	case FAN_CONTROL_NR:
+		break;
+	}
+
+	priv->fan.mode = mode;
+	return 0;
+}
+
+int
+nouveau_therm_fan_user_get(struct nouveau_therm *therm)
+{
+	return nouveau_therm_fan_get(therm);
+}
+
+int
+nouveau_therm_fan_user_set(struct nouveau_therm *therm, int percent)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+
+	if (priv->fan.mode != FAN_CONTROL_MANUAL)
+		return -EINVAL;
+
+	return nouveau_therm_fan_set(therm, percent);
+}
+
+void
+nouveau_therm_fan_set_defaults(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+
+	priv->bios_fan.pwm_freq = 0;
+	priv->bios_fan.min_duty = 0;
+	priv->bios_fan.max_duty = 100;
+}
+
+
+static void
+nouveau_therm_fan_safety_checks(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+
+	if (priv->bios_fan.min_duty > 100)
+		priv->bios_fan.min_duty = 100;
+	if (priv->bios_fan.max_duty > 100)
+		priv->bios_fan.max_duty = 100;
+
+	if (priv->bios_fan.min_duty > priv->bios_fan.max_duty)
+		priv->bios_fan.min_duty = priv->bios_fan.max_duty;
+}
+
+int nouveau_fan_pwm_clock_dummy(struct nouveau_therm *therm)
+{
+	return 1;
+}
+
+int
+nouveau_therm_fan_ctor(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	struct nouveau_bios *bios = nouveau_bios(therm);
+
+	nouveau_therm_fan_set_defaults(therm);
+	nvbios_perf_fan_parse(bios, &priv->bios_perf_fan);
+	if (nvbios_therm_fan_parse(bios, &priv->bios_fan))
+		nv_error(therm, "parsing the thermal table failed\n");
+	nouveau_therm_fan_safety_checks(therm);
+
+	nouveau_therm_fan_set_mode(therm, FAN_CONTROL_NONE);
+
+	return 0;
+}
diff --git a/nvkm/subdev/therm/ic.c b/nvkm/subdev/therm/ic.c
new file mode 100644
index 000000000..e512ff0aa
--- /dev/null
+++ b/nvkm/subdev/therm/ic.c
@@ -0,0 +1,116 @@
+/*
+ * Copyright 2012 Nouveau community
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Martin Peres
+ */
+
+#include "priv.h"
+
+#include <subdev/i2c.h>
+#include <subdev/bios/extdev.h>
+
+static bool
+probe_monitoring_device(struct nouveau_i2c_port *i2c,
+			struct i2c_board_info *info)
+{
+	struct nouveau_therm_priv *priv = (void *)nouveau_therm(i2c->i2c);
+	struct i2c_client *client;
+
+	request_module("%s%s", I2C_MODULE_PREFIX, info->type);
+
+	client = i2c_new_device(&i2c->adapter, info);
+	if (!client)
+		return false;
+
+	if (!client->driver || client->driver->detect(client, info)) {
+		i2c_unregister_device(client);
+		return false;
+	}
+
+	nv_info(priv,
+		"Found an %s at address 0x%x (controlled by lm_sensors)\n",
+		info->type, info->addr);
+	priv->ic = client;
+
+	return true;
+}
+
+void
+nouveau_therm_ic_ctor(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	struct nouveau_bios *bios = nouveau_bios(therm);
+	struct nouveau_i2c *i2c = nouveau_i2c(therm);
+	struct nvbios_extdev_func extdev_entry;
+	struct i2c_board_info info[] = {
+		{ I2C_BOARD_INFO("w83l785ts", 0x2d) },
+		{ I2C_BOARD_INFO("w83781d", 0x2d) },
+		{ I2C_BOARD_INFO("adt7473", 0x2e) },
+		{ I2C_BOARD_INFO("adt7473", 0x2d) },
+		{ I2C_BOARD_INFO("adt7473", 0x2c) },
+		{ I2C_BOARD_INFO("f75375", 0x2e) },
+		{ I2C_BOARD_INFO("lm99", 0x4c) },
+		{ I2C_BOARD_INFO("lm90", 0x4c) },
+		{ I2C_BOARD_INFO("lm90", 0x4d) },
+		{ I2C_BOARD_INFO("adm1021", 0x18) },
+		{ I2C_BOARD_INFO("adm1021", 0x19) },
+		{ I2C_BOARD_INFO("adm1021", 0x1a) },
+		{ I2C_BOARD_INFO("adm1021", 0x29) },
+		{ I2C_BOARD_INFO("adm1021", 0x2a) },
+		{ I2C_BOARD_INFO("adm1021", 0x2b) },
+		{ I2C_BOARD_INFO("adm1021", 0x4c) },
+		{ I2C_BOARD_INFO("adm1021", 0x4d) },
+		{ I2C_BOARD_INFO("adm1021", 0x4e) },
+		{ I2C_BOARD_INFO("lm63", 0x18) },
+		{ I2C_BOARD_INFO("lm63", 0x4e) },
+		{ }
+	};
+
+	if (!nvbios_extdev_find(bios, NVBIOS_EXTDEV_LM89, &extdev_entry)) {
+		struct i2c_board_info board[] = {
+			{ I2C_BOARD_INFO("lm90", extdev_entry.addr >> 1) },
+			{ }
+		};
+
+		i2c->identify(i2c, NV_I2C_DEFAULT(0), "monitoring device",
+				  board, probe_monitoring_device);
+		if (priv->ic)
+			return;
+	}
+
+	if (!nvbios_extdev_find(bios, NVBIOS_EXTDEV_ADT7473, &extdev_entry)) {
+		struct i2c_board_info board[] = {
+			{ I2C_BOARD_INFO("adt7473", extdev_entry.addr >> 1) },
+			{ }
+		};
+
+		i2c->identify(i2c, NV_I2C_DEFAULT(0), "monitoring device",
+				  board, probe_monitoring_device);
+		if (priv->ic)
+			return;
+	}
+
+	/* The vbios doesn't provide the address of an exisiting monitoring
+	   device. Let's try our static list.
+	 */
+	i2c->identify(i2c, NV_I2C_DEFAULT(0), "monitoring device", info,
+		      probe_monitoring_device);
+}
diff --git a/nvkm/subdev/therm/nv40.c b/nvkm/subdev/therm/nv40.c
new file mode 100644
index 000000000..fcf2cfe73
--- /dev/null
+++ b/nvkm/subdev/therm/nv40.c
@@ -0,0 +1,163 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ * 	    Martin Peres
+ */
+
+#include "priv.h"
+
+static int
+nv40_sensor_setup(struct nouveau_therm *therm)
+{
+	struct nouveau_device *device = nv_device(therm);
+
+	/* enable ADC readout and disable the ALARM threshold */
+	if (device->chipset >= 0x46) {
+		nv_mask(therm, 0x15b8, 0x80000000, 0);
+		nv_wr32(therm, 0x15b0, 0x80003fff);
+		return nv_rd32(therm, 0x15b4) & 0x3fff;
+	} else {
+		nv_wr32(therm, 0x15b0, 0xff);
+		return nv_rd32(therm, 0x15b4) & 0xff;
+	}
+}
+
+static int
+nv40_temp_get(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	struct nouveau_device *device = nv_device(therm);
+	struct nvbios_therm_sensor *sensor = &priv->bios_sensor;
+	int core_temp;
+
+	if (device->chipset >= 0x46) {
+		nv_wr32(therm, 0x15b0, 0x80003fff);
+		core_temp = nv_rd32(therm, 0x15b4) & 0x3fff;
+	} else {
+		nv_wr32(therm, 0x15b0, 0xff);
+		core_temp = nv_rd32(therm, 0x15b4) & 0xff;
+	}
+
+	/* Setup the sensor if the temperature is 0 */
+	if (core_temp == 0)
+		core_temp = nv40_sensor_setup(therm);
+
+	if (sensor->slope_div == 0)
+		sensor->slope_div = 1;
+	if (sensor->offset_den == 0)
+		sensor->offset_den = 1;
+	if (sensor->slope_mult < 1)
+		sensor->slope_mult = 1;
+
+	core_temp = core_temp * sensor->slope_mult / sensor->slope_div;
+	core_temp = core_temp + sensor->offset_num / sensor->offset_den;
+	core_temp = core_temp + sensor->offset_constant - 8;
+
+	return core_temp;
+}
+
+int
+nv40_fan_pwm_get(struct nouveau_therm *therm, int line, u32 *divs, u32 *duty)
+{
+	if (line == 2) {
+		u32 reg = nv_rd32(therm, 0x0010f0);
+		if (reg & 0x80000000) {
+			*duty = (reg & 0x7fff0000) >> 16;
+			*divs = (reg & 0x00007fff);
+			return 0;
+		}
+	} else
+	if (line == 9) {
+		u32 reg = nv_rd32(therm, 0x0015f4);
+		if (reg & 0x80000000) {
+			*divs = nv_rd32(therm, 0x0015f8);
+			*duty = (reg & 0x7fffffff);
+			return 0;
+		}
+	} else {
+		nv_error(therm, "unknown pwm ctrl for gpio %d\n", line);
+		return -ENODEV;
+	}
+
+	return -EINVAL;
+}
+
+int
+nv40_fan_pwm_set(struct nouveau_therm *therm, int line, u32 divs, u32 duty)
+{
+	if (line == 2) {
+		nv_wr32(therm, 0x0010f0, 0x80000000 | (duty << 16) | divs);
+	} else
+	if (line == 9) {
+		nv_wr32(therm, 0x0015f8, divs);
+		nv_wr32(therm, 0x0015f4, duty | 0x80000000);
+	} else {
+		nv_error(therm, "unknown pwm ctrl for gpio %d\n", line);
+		return -ENODEV;
+	}
+
+	return 0;
+}
+
+static int
+nv40_therm_ctor(struct nouveau_object *parent,
+		   struct nouveau_object *engine,
+		   struct nouveau_oclass *oclass, void *data, u32 size,
+		   struct nouveau_object **pobject)
+{
+	struct nouveau_therm_priv *priv;
+	struct nouveau_therm *therm;
+	int ret;
+
+	ret = nouveau_therm_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	therm = (void *) priv;
+	if (ret)
+		return ret;
+
+	nouveau_therm_ic_ctor(therm);
+	nouveau_therm_sensor_ctor(therm);
+	nouveau_therm_fan_ctor(therm);
+
+	priv->fan.pwm_get = nv40_fan_pwm_get;
+	priv->fan.pwm_set = nv40_fan_pwm_set;
+
+	therm->temp_get = nv40_temp_get;
+	therm->fan_get = nouveau_therm_fan_user_get;
+	therm->fan_set = nouveau_therm_fan_user_set;
+	therm->fan_sense = nouveau_therm_fan_sense;
+	therm->attr_get = nouveau_therm_attr_get;
+	therm->attr_set = nouveau_therm_attr_set;
+
+	return 0;
+}
+
+struct nouveau_oclass
+nv40_therm_oclass = {
+	.handle = NV_SUBDEV(THERM, 0x40),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv40_therm_ctor,
+		.dtor = _nouveau_therm_dtor,
+		.init = nouveau_therm_init,
+		.fini = nouveau_therm_fini,
+	},
+};
\ No newline at end of file
diff --git a/nvkm/subdev/therm/nv50.c b/nvkm/subdev/therm/nv50.c
new file mode 100644
index 000000000..9360ddd46
--- /dev/null
+++ b/nvkm/subdev/therm/nv50.c
@@ -0,0 +1,157 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ * 	    Martin Peres
+ */
+
+#include "priv.h"
+
+static int
+pwm_info(struct nouveau_therm *therm, int *line, int *ctrl, int *indx)
+{
+	if (*line == 0x04) {
+		*ctrl = 0x00e100;
+		*line = 4;
+		*indx = 0;
+	} else
+	if (*line == 0x09) {
+		*ctrl = 0x00e100;
+		*line = 9;
+		*indx = 1;
+	} else
+	if (*line == 0x10) {
+		*ctrl = 0x00e28c;
+		*line = 0;
+		*indx = 0;
+	} else {
+		nv_error(therm, "unknown pwm ctrl for gpio %d\n", *line);
+		return -ENODEV;
+	}
+
+	return 0;
+}
+
+int
+nv50_fan_pwm_get(struct nouveau_therm *therm, int line, u32 *divs, u32 *duty)
+{
+	int ctrl, id, ret = pwm_info(therm, &line, &ctrl, &id);
+	if (ret)
+		return ret;
+
+	if (nv_rd32(therm, ctrl) & (1 << line)) {
+		*divs = nv_rd32(therm, 0x00e114 + (id * 8));
+		*duty = nv_rd32(therm, 0x00e118 + (id * 8));
+		return 0;
+	}
+
+	return -EINVAL;
+}
+
+int
+nv50_fan_pwm_set(struct nouveau_therm *therm, int line, u32 divs, u32 duty)
+{
+	int ctrl, id, ret = pwm_info(therm, &line, &ctrl, &id);
+	if (ret)
+		return ret;
+
+	nv_mask(therm, ctrl, 0x00010001 << line, 0x00000001 << line);
+	nv_wr32(therm, 0x00e114 + (id * 8), divs);
+	nv_wr32(therm, 0x00e118 + (id * 8), duty | 0x80000000);
+	return 0;
+}
+
+int
+nv50_fan_pwm_clock(struct nouveau_therm *therm)
+{
+	int chipset = nv_device(therm)->chipset;
+	int crystal = nv_device(therm)->crystal;
+	int pwm_clock;
+
+	/* determine the PWM source clock */
+	if (chipset > 0x50 && chipset < 0x94) {
+		u8 pwm_div = nv_rd32(therm, 0x410c);
+		if (nv_rd32(therm, 0xc040) & 0x800000) {
+			/* Use the HOST clock (100 MHz)
+			* Where does this constant(2.4) comes from? */
+			pwm_clock = (100000000 >> pwm_div) * 10 / 24;
+		} else {
+			/* Where does this constant(20) comes from? */
+			pwm_clock = (crystal * 1000) >> pwm_div;
+			pwm_clock /= 20;
+		}
+	} else {
+		pwm_clock = (crystal * 1000) / 20;
+	}
+
+	return pwm_clock;
+}
+
+int
+nv50_temp_get(struct nouveau_therm *therm)
+{
+	return nv_rd32(therm, 0x20400);
+}
+
+static int
+nv50_therm_ctor(struct nouveau_object *parent,
+		   struct nouveau_object *engine,
+		   struct nouveau_oclass *oclass, void *data, u32 size,
+		   struct nouveau_object **pobject)
+{
+	struct nouveau_therm_priv *priv;
+	struct nouveau_therm *therm;
+	int ret;
+
+	ret = nouveau_therm_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	therm = (void *) priv;
+	if (ret)
+		return ret;
+
+	nouveau_therm_ic_ctor(therm);
+	nouveau_therm_sensor_ctor(therm);
+	nouveau_therm_fan_ctor(therm);
+
+	priv->fan.pwm_get = nv50_fan_pwm_get;
+	priv->fan.pwm_set = nv50_fan_pwm_set;
+	priv->fan.pwm_clock = nv50_fan_pwm_clock;
+
+	therm->temp_get = nv50_temp_get;
+	therm->fan_get = nouveau_therm_fan_user_get;
+	therm->fan_set = nouveau_therm_fan_user_set;
+	therm->fan_sense = nouveau_therm_fan_sense;
+	therm->attr_get = nouveau_therm_attr_get;
+	therm->attr_set = nouveau_therm_attr_set;
+
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_therm_oclass = {
+	.handle = NV_SUBDEV(THERM, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_therm_ctor,
+		.dtor = _nouveau_therm_dtor,
+		.init = nouveau_therm_init,
+		.fini = nouveau_therm_fini,
+	},
+};
diff --git a/nvkm/subdev/therm/priv.h b/nvkm/subdev/therm/priv.h
new file mode 100644
index 000000000..1c3cd6abc
--- /dev/null
+++ b/nvkm/subdev/therm/priv.h
@@ -0,0 +1,73 @@
+/*
+ * Copyright 2012 The Nouveau community
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Martin Peres
+ */
+
+#include <subdev/therm.h>
+
+#include <subdev/bios/extdev.h>
+#include <subdev/bios/perf.h>
+#include <subdev/bios/therm.h>
+
+struct nouveau_therm_priv {
+	struct nouveau_therm base;
+
+	/* bios */
+	struct nvbios_therm_sensor bios_sensor;
+	struct nvbios_therm_fan bios_fan;
+	struct nvbios_perf_fan bios_perf_fan;
+
+	/* fan priv */
+	struct {
+		enum nouveau_therm_fan_mode mode;
+		int percent;
+
+		int (*pwm_get)(struct nouveau_therm *, int line, u32*, u32*);
+		int (*pwm_set)(struct nouveau_therm *, int line, u32, u32);
+		int (*pwm_clock)(struct nouveau_therm *);
+	} fan;
+
+	/* ic */
+	struct i2c_client *ic;
+};
+
+int nouveau_therm_init(struct nouveau_object *object);
+int nouveau_therm_fini(struct nouveau_object *object, bool suspend);
+int nouveau_therm_attr_get(struct nouveau_therm *therm,
+		       enum nouveau_therm_attr_type type);
+int nouveau_therm_attr_set(struct nouveau_therm *therm,
+		       enum nouveau_therm_attr_type type, int value);
+
+void nouveau_therm_ic_ctor(struct nouveau_therm *therm);
+
+int nouveau_therm_sensor_ctor(struct nouveau_therm *therm);
+
+int nouveau_therm_fan_ctor(struct nouveau_therm *therm);
+int nouveau_therm_fan_get(struct nouveau_therm *therm);
+int nouveau_therm_fan_set(struct nouveau_therm *therm, int percent);
+int nouveau_therm_fan_user_get(struct nouveau_therm *therm);
+int nouveau_therm_fan_user_set(struct nouveau_therm *therm, int percent);
+int nouveau_therm_fan_set_mode(struct nouveau_therm *therm,
+			   enum nouveau_therm_fan_mode mode);
+
+
+int nouveau_therm_fan_sense(struct nouveau_therm *therm);
diff --git a/nvkm/subdev/therm/temp.c b/nvkm/subdev/therm/temp.c
new file mode 100644
index 000000000..204282301
--- /dev/null
+++ b/nvkm/subdev/therm/temp.c
@@ -0,0 +1,81 @@
+/*
+ * Copyright 2012 The Nouveau community
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Martin Peres
+ */
+
+#include "priv.h"
+
+#include <core/object.h>
+#include <core/device.h>
+
+#include <subdev/bios.h>
+
+static void
+nouveau_therm_temp_set_defaults(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+
+	priv->bios_sensor.slope_mult = 1;
+	priv->bios_sensor.slope_div = 1;
+	priv->bios_sensor.offset_num = 0;
+	priv->bios_sensor.offset_den = 1;
+	priv->bios_sensor.offset_constant = 0;
+
+	priv->bios_sensor.thrs_fan_boost.temp = 90;
+	priv->bios_sensor.thrs_fan_boost.hysteresis = 3;
+
+	priv->bios_sensor.thrs_down_clock.temp = 95;
+	priv->bios_sensor.thrs_down_clock.hysteresis = 3;
+
+	priv->bios_sensor.thrs_critical.temp = 105;
+	priv->bios_sensor.thrs_critical.hysteresis = 5;
+
+	priv->bios_sensor.thrs_shutdown.temp = 135;
+	priv->bios_sensor.thrs_shutdown.hysteresis = 5; /*not that it matters */
+}
+
+
+static void
+nouveau_therm_temp_safety_checks(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+
+	if (!priv->bios_sensor.slope_div)
+		priv->bios_sensor.slope_div = 1;
+	if (!priv->bios_sensor.offset_den)
+		priv->bios_sensor.offset_den = 1;
+}
+
+int
+nouveau_therm_sensor_ctor(struct nouveau_therm *therm)
+{
+	struct nouveau_therm_priv *priv = (void *)therm;
+	struct nouveau_bios *bios = nouveau_bios(therm);
+
+	nouveau_therm_temp_set_defaults(therm);
+	if (nvbios_therm_sensor_parse(bios, NVBIOS_THERM_DOMAIN_CORE,
+				      &priv->bios_sensor))
+		nv_error(therm, "nvbios_therm_sensor_parse failed\n");
+	nouveau_therm_temp_safety_checks(therm);
+
+	return 0;
+}
diff --git a/nvkm/subdev/timer/Makefile.am b/nvkm/subdev/timer/Makefile.am
new file mode 100644
index 000000000..fd98fdbcb
--- /dev/null
+++ b/nvkm/subdev/timer/Makefile.am
@@ -0,0 +1,6 @@
+noinst_LTLIBRARIES = libtimer.la
+
+libtimer_la_SOURCES = base.c \
+		      nv04.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/timer/base.c b/nvkm/subdev/timer/base.c
new file mode 100644
index 000000000..5d417cc99
--- /dev/null
+++ b/nvkm/subdev/timer/base.c
@@ -0,0 +1,87 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "subdev/timer.h"
+
+bool
+nouveau_timer_wait_eq(void *obj, u64 nsec, u32 addr, u32 mask, u32 data)
+{
+	struct nouveau_timer *ptimer = nouveau_timer(obj);
+	u64 time0;
+
+	time0 = ptimer->read(ptimer);
+	do {
+		if (nv_iclass(obj, NV_SUBDEV_CLASS)) {
+			if ((nv_rd32(obj, addr) & mask) == data)
+				return true;
+		} else {
+			if ((nv_ro32(obj, addr) & mask) == data)
+				return true;
+		}
+	} while (ptimer->read(ptimer) - time0 < nsec);
+
+	return false;
+}
+
+bool
+nouveau_timer_wait_ne(void *obj, u64 nsec, u32 addr, u32 mask, u32 data)
+{
+	struct nouveau_timer *ptimer = nouveau_timer(obj);
+	u64 time0;
+
+	time0 = ptimer->read(ptimer);
+	do {
+		if (nv_iclass(obj, NV_SUBDEV_CLASS)) {
+			if ((nv_rd32(obj, addr) & mask) != data)
+				return true;
+		} else {
+			if ((nv_ro32(obj, addr) & mask) != data)
+				return true;
+		}
+	} while (ptimer->read(ptimer) - time0 < nsec);
+
+	return false;
+}
+
+bool
+nouveau_timer_wait_cb(void *obj, u64 nsec, bool (*func)(void *), void *data)
+{
+	struct nouveau_timer *ptimer = nouveau_timer(obj);
+	u64 time0;
+
+	time0 = ptimer->read(ptimer);
+	do {
+		if (func(data) == true)
+			return true;
+	} while (ptimer->read(ptimer) - time0 < nsec);
+
+	return false;
+}
+
+void
+nouveau_timer_alarm(void *obj, u32 nsec, struct nouveau_alarm *alarm)
+{
+	struct nouveau_timer *ptimer = nouveau_timer(obj);
+	ptimer->alarm(ptimer, nsec, alarm);
+}
diff --git a/nvkm/subdev/timer/nv04.c b/nvkm/subdev/timer/nv04.c
new file mode 100644
index 000000000..c26ca9bef
--- /dev/null
+++ b/nvkm/subdev/timer/nv04.c
@@ -0,0 +1,249 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <subdev/timer.h>
+
+#define NV04_PTIMER_INTR_0      0x009100
+#define NV04_PTIMER_INTR_EN_0   0x009140
+#define NV04_PTIMER_NUMERATOR   0x009200
+#define NV04_PTIMER_DENOMINATOR 0x009210
+#define NV04_PTIMER_TIME_0      0x009400
+#define NV04_PTIMER_TIME_1      0x009410
+#define NV04_PTIMER_ALARM_0     0x009420
+
+struct nv04_timer_priv {
+	struct nouveau_timer base;
+	struct list_head alarms;
+	spinlock_t lock;
+};
+
+static u64
+nv04_timer_read(struct nouveau_timer *ptimer)
+{
+	struct nv04_timer_priv *priv = (void *)ptimer;
+	u32 hi, lo;
+
+	do {
+		hi = nv_rd32(priv, NV04_PTIMER_TIME_1);
+		lo = nv_rd32(priv, NV04_PTIMER_TIME_0);
+	} while (hi != nv_rd32(priv, NV04_PTIMER_TIME_1));
+
+	return ((u64)hi << 32 | lo);
+}
+
+static void
+nv04_timer_alarm_trigger(struct nouveau_timer *ptimer)
+{
+	struct nv04_timer_priv *priv = (void *)ptimer;
+	struct nouveau_alarm *alarm, *atemp;
+	unsigned long flags;
+	LIST_HEAD(exec);
+
+	/* move any due alarms off the pending list */
+	spin_lock_irqsave(&priv->lock, flags);
+	list_for_each_entry_safe(alarm, atemp, &priv->alarms, head) {
+		if (alarm->timestamp <= ptimer->read(ptimer))
+			list_move_tail(&alarm->head, &exec);
+	}
+
+	/* reschedule interrupt for next alarm time */
+	if (!list_empty(&priv->alarms)) {
+		alarm = list_first_entry(&priv->alarms, typeof(*alarm), head);
+		nv_wr32(priv, NV04_PTIMER_ALARM_0, alarm->timestamp);
+		nv_wr32(priv, NV04_PTIMER_INTR_EN_0, 0x00000001);
+	} else {
+		nv_wr32(priv, NV04_PTIMER_INTR_EN_0, 0x00000000);
+	}
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	/* execute any pending alarm handlers */
+	list_for_each_entry_safe(alarm, atemp, &exec, head) {
+		list_del(&alarm->head);
+		alarm->func(alarm);
+	}
+}
+
+static void
+nv04_timer_alarm(struct nouveau_timer *ptimer, u64 time,
+		 struct nouveau_alarm *alarm)
+{
+	struct nv04_timer_priv *priv = (void *)ptimer;
+	struct nouveau_alarm *list;
+	unsigned long flags;
+
+	alarm->timestamp = ptimer->read(ptimer) + time;
+
+	/* append new alarm to list, in soonest-alarm-first order */
+	spin_lock_irqsave(&priv->lock, flags);
+	list_for_each_entry(list, &priv->alarms, head) {
+		if (list->timestamp > alarm->timestamp)
+			break;
+	}
+	list_add_tail(&alarm->head, &list->head);
+	spin_unlock_irqrestore(&priv->lock, flags);
+
+	/* process pending alarms */
+	nv04_timer_alarm_trigger(ptimer);
+}
+
+static void
+nv04_timer_intr(struct nouveau_subdev *subdev)
+{
+	struct nv04_timer_priv *priv = (void *)subdev;
+	u32 stat = nv_rd32(priv, NV04_PTIMER_INTR_0);
+
+	if (stat & 0x00000001) {
+		nv04_timer_alarm_trigger(&priv->base);
+		nv_wr32(priv, NV04_PTIMER_INTR_0, 0x00000001);
+		stat &= ~0x00000001;
+	}
+
+	if (stat) {
+		nv_error(priv, "unknown stat 0x%08x\n", stat);
+		nv_wr32(priv, NV04_PTIMER_INTR_0, stat);
+	}
+}
+
+static int
+nv04_timer_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv04_timer_priv *priv;
+	int ret;
+
+	ret = nouveau_timer_create(parent, engine, oclass, &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.base.intr = nv04_timer_intr;
+	priv->base.read = nv04_timer_read;
+	priv->base.alarm = nv04_timer_alarm;
+
+	INIT_LIST_HEAD(&priv->alarms);
+	spin_lock_init(&priv->lock);
+	return 0;
+}
+
+static void
+nv04_timer_dtor(struct nouveau_object *object)
+{
+	struct nv04_timer_priv *priv = (void *)object;
+	return nouveau_timer_destroy(&priv->base);
+}
+
+static int
+nv04_timer_init(struct nouveau_object *object)
+{
+	struct nouveau_device *device = nv_device(object);
+	struct nv04_timer_priv *priv = (void *)object;
+	u32 m = 1, f, n, d;
+	int ret;
+
+	ret = nouveau_timer_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* aim for 31.25MHz, which gives us nanosecond timestamps */
+	d = 1000000 / 32;
+
+	/* determine base clock for timer source */
+#if 0 /*XXX*/
+	if (device->chipset < 0x40) {
+		n = nouveau_hw_get_clock(device, PLL_CORE);
+	} else
+#endif
+	if (device->chipset <= 0x40) {
+		/*XXX: figure this out */
+		f = -1;
+		n = 0;
+	} else {
+		f = device->crystal;
+		n = f;
+		while (n < (d * 2)) {
+			n += (n / m);
+			m++;
+		}
+
+		nv_wr32(priv, 0x009220, m - 1);
+	}
+
+	if (!n) {
+		nv_warn(priv, "unknown input clock freq\n");
+		if (!nv_rd32(priv, NV04_PTIMER_NUMERATOR) ||
+		    !nv_rd32(priv, NV04_PTIMER_DENOMINATOR)) {
+			nv_wr32(priv, NV04_PTIMER_NUMERATOR, 1);
+			nv_wr32(priv, NV04_PTIMER_DENOMINATOR, 1);
+		}
+		return 0;
+	}
+
+	/* reduce ratio to acceptable values */
+	while (((n % 5) == 0) && ((d % 5) == 0)) {
+		n /= 5;
+		d /= 5;
+	}
+
+	while (((n % 2) == 0) && ((d % 2) == 0)) {
+		n /= 2;
+		d /= 2;
+	}
+
+	while (n > 0xffff || d > 0xffff) {
+		n >>= 1;
+		d >>= 1;
+	}
+
+	nv_debug(priv, "input frequency : %dHz\n", f);
+	nv_debug(priv, "input multiplier: %d\n", m);
+	nv_debug(priv, "numerator       : 0x%08x\n", n);
+	nv_debug(priv, "denominator     : 0x%08x\n", d);
+	nv_debug(priv, "timer frequency : %dHz\n", (f * m) * d / n);
+
+	nv_wr32(priv, NV04_PTIMER_NUMERATOR, n);
+	nv_wr32(priv, NV04_PTIMER_DENOMINATOR, d);
+	nv_wr32(priv, NV04_PTIMER_INTR_0, 0xffffffff);
+	nv_wr32(priv, NV04_PTIMER_INTR_EN_0, 0x00000000);
+	return 0;
+}
+
+static int
+nv04_timer_fini(struct nouveau_object *object, bool suspend)
+{
+	struct nv04_timer_priv *priv = (void *)object;
+	nv_wr32(priv, NV04_PTIMER_INTR_EN_0, 0x00000000);
+	return nouveau_timer_fini(&priv->base, suspend);
+}
+
+struct nouveau_oclass
+nv04_timer_oclass = {
+	.handle = NV_SUBDEV(TIMER, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_timer_ctor,
+		.dtor = nv04_timer_dtor,
+		.init = nv04_timer_init,
+		.fini = nv04_timer_fini,
+	}
+};
diff --git a/nvkm/subdev/vm/Makefile.am b/nvkm/subdev/vm/Makefile.am
new file mode 100644
index 000000000..1f5b6fb76
--- /dev/null
+++ b/nvkm/subdev/vm/Makefile.am
@@ -0,0 +1,10 @@
+noinst_LTLIBRARIES = libvm.la
+
+libvm_la_SOURCES = base.c \
+		   nv04.c \
+		   nv41.c \
+		   nv44.c \
+		   nv50.c \
+		   nvc0.c
+
+include $(srcdir)/../Makefile.subdev
diff --git a/nvkm/subdev/vm/base.c b/nvkm/subdev/vm/base.c
new file mode 100644
index 000000000..082c11b75
--- /dev/null
+++ b/nvkm/subdev/vm/base.c
@@ -0,0 +1,478 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/gpuobj.h>
+#include <core/mm.h>
+
+#include <subdev/fb.h>
+#include <subdev/vm.h>
+
+void
+nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_mem *node)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_vmmgr *vmm = vm->vmm;
+	struct nouveau_mm_node *r;
+	int big = vma->node->type != vmm->spg_shift;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 pde  = (offset >> vmm->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (vmm->pgt_bits - bits);
+	u32 end, len;
+
+	delta = 0;
+	list_for_each_entry(r, &node->regions, rl_entry) {
+		u64 phys = (u64)r->offset << 12;
+		u32 num  = r->length >> bits;
+
+		while (num) {
+			struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
+
+			end = (pte + num);
+			if (unlikely(end >= max))
+				end = max;
+			len = end - pte;
+
+			vmm->map(vma, pgt, node, pte, len, phys, delta);
+
+			num -= len;
+			pte += len;
+			if (unlikely(end >= max)) {
+				phys += len << (bits + 12);
+				pde++;
+				pte = 0;
+			}
+
+			delta += (u64)len << vma->node->type;
+		}
+	}
+
+	vmm->flush(vm);
+}
+
+void
+nouveau_vm_map(struct nouveau_vma *vma, struct nouveau_mem *node)
+{
+	nouveau_vm_map_at(vma, 0, node);
+}
+
+void
+nouveau_vm_map_sg_table(struct nouveau_vma *vma, u64 delta, u64 length,
+			struct nouveau_mem *mem)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_vmmgr *vmm = vm->vmm;
+	int big = vma->node->type != vmm->spg_shift;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 num  = length >> vma->node->type;
+	u32 pde  = (offset >> vmm->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (vmm->pgt_bits - bits);
+	unsigned m, sglen;
+	u32 end, len;
+	int i;
+	struct scatterlist *sg;
+
+	for_each_sg(mem->sg->sgl, sg, mem->sg->nents, i) {
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
+		sglen = sg_dma_len(sg) >> PAGE_SHIFT;
+
+		end = pte + sglen;
+		if (unlikely(end >= max))
+			end = max;
+		len = end - pte;
+
+		for (m = 0; m < len; m++) {
+			dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
+
+			vmm->map_sg(vma, pgt, mem, pte, 1, &addr);
+			num--;
+			pte++;
+
+			if (num == 0)
+				goto finish;
+		}
+		if (unlikely(end >= max)) {
+			pde++;
+			pte = 0;
+		}
+		if (m < sglen) {
+			for (; m < sglen; m++) {
+				dma_addr_t addr = sg_dma_address(sg) + (m << PAGE_SHIFT);
+
+				vmm->map_sg(vma, pgt, mem, pte, 1, &addr);
+				num--;
+				pte++;
+				if (num == 0)
+					goto finish;
+			}
+		}
+
+	}
+finish:
+	vmm->flush(vm);
+}
+
+void
+nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length,
+		  struct nouveau_mem *mem)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_vmmgr *vmm = vm->vmm;
+	dma_addr_t *list = mem->pages;
+	int big = vma->node->type != vmm->spg_shift;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 num  = length >> vma->node->type;
+	u32 pde  = (offset >> vmm->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (vmm->pgt_bits - bits);
+	u32 end, len;
+
+	while (num) {
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
+
+		end = (pte + num);
+		if (unlikely(end >= max))
+			end = max;
+		len = end - pte;
+
+		vmm->map_sg(vma, pgt, mem, pte, len, list);
+
+		num  -= len;
+		pte  += len;
+		list += len;
+		if (unlikely(end >= max)) {
+			pde++;
+			pte = 0;
+		}
+	}
+
+	vmm->flush(vm);
+}
+
+void
+nouveau_vm_unmap_at(struct nouveau_vma *vma, u64 delta, u64 length)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_vmmgr *vmm = vm->vmm;
+	int big = vma->node->type != vmm->spg_shift;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 num  = length >> vma->node->type;
+	u32 pde  = (offset >> vmm->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << vmm->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (vmm->pgt_bits - bits);
+	u32 end, len;
+
+	while (num) {
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj[big];
+
+		end = (pte + num);
+		if (unlikely(end >= max))
+			end = max;
+		len = end - pte;
+
+		vmm->unmap(pgt, pte, len);
+
+		num -= len;
+		pte += len;
+		if (unlikely(end >= max)) {
+			pde++;
+			pte = 0;
+		}
+	}
+
+	vmm->flush(vm);
+}
+
+void
+nouveau_vm_unmap(struct nouveau_vma *vma)
+{
+	nouveau_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
+}
+
+static void
+nouveau_vm_unmap_pgt(struct nouveau_vm *vm, int big, u32 fpde, u32 lpde)
+{
+	struct nouveau_vmmgr *vmm = vm->vmm;
+	struct nouveau_vm_pgd *vpgd;
+	struct nouveau_vm_pgt *vpgt;
+	struct nouveau_gpuobj *pgt;
+	u32 pde;
+
+	for (pde = fpde; pde <= lpde; pde++) {
+		vpgt = &vm->pgt[pde - vm->fpde];
+		if (--vpgt->refcount[big])
+			continue;
+
+		pgt = vpgt->obj[big];
+		vpgt->obj[big] = NULL;
+
+		list_for_each_entry(vpgd, &vm->pgd_list, head) {
+			vmm->map_pgt(vpgd->obj, pde, vpgt->obj);
+		}
+
+		mutex_unlock(&vm->mm.mutex);
+		nouveau_gpuobj_ref(NULL, &pgt);
+		mutex_lock(&vm->mm.mutex);
+	}
+}
+
+static int
+nouveau_vm_map_pgt(struct nouveau_vm *vm, u32 pde, u32 type)
+{
+	struct nouveau_vmmgr *vmm = vm->vmm;
+	struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
+	struct nouveau_vm_pgd *vpgd;
+	struct nouveau_gpuobj *pgt;
+	int big = (type != vmm->spg_shift);
+	u32 pgt_size;
+	int ret;
+
+	pgt_size  = (1 << (vmm->pgt_bits + 12)) >> type;
+	pgt_size *= 8;
+
+	mutex_unlock(&vm->mm.mutex);
+	ret = nouveau_gpuobj_new(nv_object(vm->vmm), NULL, pgt_size, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC, &pgt);
+	mutex_lock(&vm->mm.mutex);
+	if (unlikely(ret))
+		return ret;
+
+	/* someone beat us to filling the PDE while we didn't have the lock */
+	if (unlikely(vpgt->refcount[big]++)) {
+		mutex_unlock(&vm->mm.mutex);
+		nouveau_gpuobj_ref(NULL, &pgt);
+		mutex_lock(&vm->mm.mutex);
+		return 0;
+	}
+
+	vpgt->obj[big] = pgt;
+	list_for_each_entry(vpgd, &vm->pgd_list, head) {
+		vmm->map_pgt(vpgd->obj, pde, vpgt->obj);
+	}
+
+	return 0;
+}
+
+int
+nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
+	       u32 access, struct nouveau_vma *vma)
+{
+	struct nouveau_vmmgr *vmm = vm->vmm;
+	u32 align = (1 << page_shift) >> 12;
+	u32 msize = size >> 12;
+	u32 fpde, lpde, pde;
+	int ret;
+
+	mutex_lock(&vm->mm.mutex);
+	ret = nouveau_mm_head(&vm->mm, page_shift, msize, msize, align,
+			     &vma->node);
+	if (unlikely(ret != 0)) {
+		mutex_unlock(&vm->mm.mutex);
+		return ret;
+	}
+
+	fpde = (vma->node->offset >> vmm->pgt_bits);
+	lpde = (vma->node->offset + vma->node->length - 1) >> vmm->pgt_bits;
+
+	for (pde = fpde; pde <= lpde; pde++) {
+		struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
+		int big = (vma->node->type != vmm->spg_shift);
+
+		if (likely(vpgt->refcount[big])) {
+			vpgt->refcount[big]++;
+			continue;
+		}
+
+		ret = nouveau_vm_map_pgt(vm, pde, vma->node->type);
+		if (ret) {
+			if (pde != fpde)
+				nouveau_vm_unmap_pgt(vm, big, fpde, pde - 1);
+			nouveau_mm_free(&vm->mm, &vma->node);
+			mutex_unlock(&vm->mm.mutex);
+			return ret;
+		}
+	}
+	mutex_unlock(&vm->mm.mutex);
+
+	vma->vm     = vm;
+	vma->offset = (u64)vma->node->offset << 12;
+	vma->access = access;
+	return 0;
+}
+
+void
+nouveau_vm_put(struct nouveau_vma *vma)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_vmmgr *vmm = vm->vmm;
+	u32 fpde, lpde;
+
+	if (unlikely(vma->node == NULL))
+		return;
+	fpde = (vma->node->offset >> vmm->pgt_bits);
+	lpde = (vma->node->offset + vma->node->length - 1) >> vmm->pgt_bits;
+
+	mutex_lock(&vm->mm.mutex);
+	nouveau_vm_unmap_pgt(vm, vma->node->type != vmm->spg_shift, fpde, lpde);
+	nouveau_mm_free(&vm->mm, &vma->node);
+	mutex_unlock(&vm->mm.mutex);
+}
+
+int
+nouveau_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64 length,
+		  u64 mm_offset, u32 block, struct nouveau_vm **pvm)
+{
+	struct nouveau_vm *vm;
+	u64 mm_length = (offset + length) - mm_offset;
+	int ret;
+
+	vm = *pvm = kzalloc(sizeof(*vm), GFP_KERNEL);
+	if (!vm)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&vm->pgd_list);
+	vm->vmm = vmm;
+	vm->refcount = 1;
+	vm->fpde = offset >> (vmm->pgt_bits + 12);
+	vm->lpde = (offset + length - 1) >> (vmm->pgt_bits + 12);
+
+	vm->pgt  = kcalloc(vm->lpde - vm->fpde + 1, sizeof(*vm->pgt), GFP_KERNEL);
+	if (!vm->pgt) {
+		kfree(vm);
+		return -ENOMEM;
+	}
+
+	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
+			      block >> 12);
+	if (ret) {
+		kfree(vm->pgt);
+		kfree(vm);
+		return ret;
+	}
+
+	return 0;
+}
+
+int
+nouveau_vm_new(struct nouveau_device *device, u64 offset, u64 length,
+	       u64 mm_offset, struct nouveau_vm **pvm)
+{
+	struct nouveau_vmmgr *vmm = nouveau_vmmgr(device);
+	return vmm->create(vmm, offset, length, mm_offset, pvm);
+}
+
+static int
+nouveau_vm_link(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
+{
+	struct nouveau_vmmgr *vmm = vm->vmm;
+	struct nouveau_vm_pgd *vpgd;
+	int i;
+
+	if (!pgd)
+		return 0;
+
+	vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL);
+	if (!vpgd)
+		return -ENOMEM;
+
+	nouveau_gpuobj_ref(pgd, &vpgd->obj);
+
+	mutex_lock(&vm->mm.mutex);
+	for (i = vm->fpde; i <= vm->lpde; i++)
+		vmm->map_pgt(pgd, i, vm->pgt[i - vm->fpde].obj);
+	list_add(&vpgd->head, &vm->pgd_list);
+	mutex_unlock(&vm->mm.mutex);
+	return 0;
+}
+
+static void
+nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *mpgd)
+{
+	struct nouveau_vm_pgd *vpgd, *tmp;
+	struct nouveau_gpuobj *pgd = NULL;
+
+	if (!mpgd)
+		return;
+
+	mutex_lock(&vm->mm.mutex);
+	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
+		if (vpgd->obj == mpgd) {
+			pgd = vpgd->obj;
+			list_del(&vpgd->head);
+			kfree(vpgd);
+			break;
+		}
+	}
+	mutex_unlock(&vm->mm.mutex);
+
+	nouveau_gpuobj_ref(NULL, &pgd);
+}
+
+static void
+nouveau_vm_del(struct nouveau_vm *vm)
+{
+	struct nouveau_vm_pgd *vpgd, *tmp;
+
+	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
+		nouveau_vm_unlink(vm, vpgd->obj);
+	}
+
+	nouveau_mm_fini(&vm->mm);
+	kfree(vm->pgt);
+	kfree(vm);
+}
+
+int
+nouveau_vm_ref(struct nouveau_vm *ref, struct nouveau_vm **ptr,
+	       struct nouveau_gpuobj *pgd)
+{
+	struct nouveau_vm *vm;
+	int ret;
+
+	vm = ref;
+	if (vm) {
+		ret = nouveau_vm_link(vm, pgd);
+		if (ret)
+			return ret;
+
+		vm->refcount++;
+	}
+
+	vm = *ptr;
+	*ptr = ref;
+
+	if (vm) {
+		nouveau_vm_unlink(vm, pgd);
+
+		if (--vm->refcount == 0)
+			nouveau_vm_del(vm);
+	}
+
+	return 0;
+}
diff --git a/nvkm/subdev/vm/nv04.c b/nvkm/subdev/vm/nv04.c
new file mode 100644
index 000000000..6adbbc9cc
--- /dev/null
+++ b/nvkm/subdev/vm/nv04.c
@@ -0,0 +1,151 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/gpuobj.h>
+
+#include "nv04.h"
+
+#define NV04_PDMA_SIZE (128 * 1024 * 1024)
+#define NV04_PDMA_PAGE (  4 * 1024)
+
+/*******************************************************************************
+ * VM map/unmap callbacks
+ ******************************************************************************/
+
+static void
+nv04_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	pte = 0x00008 + (pte * 4);
+	while (cnt) {
+		u32 page = PAGE_SIZE / NV04_PDMA_PAGE;
+		u32 phys = (u32)*list++;
+		while (cnt && page--) {
+			nv_wo32(pgt, pte, phys | 3);
+			phys += NV04_PDMA_PAGE;
+			pte += 4;
+			cnt -= 1;
+		}
+	}
+}
+
+static void
+nv04_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte = 0x00008 + (pte * 4);
+	while (cnt--) {
+		nv_wo32(pgt, pte, 0x00000000);
+		pte += 4;
+	}
+}
+
+static void
+nv04_vm_flush(struct nouveau_vm *vm)
+{
+}
+
+/*******************************************************************************
+ * VM object
+ ******************************************************************************/
+
+int
+nv04_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64 length, u64 mmstart,
+	       struct nouveau_vm **pvm)
+{
+	return -EINVAL;
+}
+
+/*******************************************************************************
+ * VMMGR subdev
+ ******************************************************************************/
+
+static int
+nv04_vmmgr_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv04_vmmgr_priv *priv;
+	struct nouveau_gpuobj *dma;
+	int ret;
+
+	ret = nouveau_vmmgr_create(parent, engine, oclass, "PCIGART",
+				   "pcigart", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.create = nv04_vm_create;
+	priv->base.limit = NV04_PDMA_SIZE;
+	priv->base.dma_bits = 32;
+	priv->base.pgt_bits = 32 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 12;
+	priv->base.map_sg = nv04_vm_map_sg;
+	priv->base.unmap = nv04_vm_unmap;
+	priv->base.flush = nv04_vm_flush;
+
+	ret = nouveau_vm_create(&priv->base, 0, NV04_PDMA_SIZE, 0, 4096,
+				&priv->vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL,
+				 (NV04_PDMA_SIZE / NV04_PDMA_PAGE) * 4 +
+				 8, 16, NVOBJ_FLAG_ZERO_ALLOC,
+				 &priv->vm->pgt[0].obj[0]);
+	dma = priv->vm->pgt[0].obj[0];
+	priv->vm->pgt[0].refcount[0] = 1;
+	if (ret)
+		return ret;
+
+	nv_wo32(dma, 0x00000, 0x0002103d); /* PCI, RW, PT, !LN */
+	nv_wo32(dma, 0x00004, NV04_PDMA_SIZE - 1);
+	return 0;
+}
+
+void
+nv04_vmmgr_dtor(struct nouveau_object *object)
+{
+	struct nv04_vmmgr_priv *priv = (void *)object;
+	if (priv->vm) {
+		nouveau_gpuobj_ref(NULL, &priv->vm->pgt[0].obj[0]);
+		nouveau_vm_ref(NULL, &priv->vm, NULL);
+	}
+	if (priv->nullp) {
+		pci_free_consistent(nv_device(priv)->pdev, 16 * 1024,
+				    priv->nullp, priv->null);
+	}
+	nouveau_vmmgr_destroy(&priv->base);
+}
+
+struct nouveau_oclass
+nv04_vmmgr_oclass = {
+	.handle = NV_SUBDEV(VM, 0x04),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv04_vmmgr_ctor,
+		.dtor = nv04_vmmgr_dtor,
+		.init = _nouveau_vmmgr_init,
+		.fini = _nouveau_vmmgr_fini,
+	},
+};
diff --git a/nvkm/subdev/vm/nv04.h b/nvkm/subdev/vm/nv04.h
new file mode 100644
index 000000000..ec42d4bc8
--- /dev/null
+++ b/nvkm/subdev/vm/nv04.h
@@ -0,0 +1,19 @@
+#ifndef __NV04_VMMGR_PRIV__
+#define __NV04_VMMGR_PRIV__
+
+#include <subdev/vm.h>
+
+struct nv04_vmmgr_priv {
+	struct nouveau_vmmgr base;
+	struct nouveau_vm *vm;
+	dma_addr_t null;
+	void *nullp;
+};
+
+static inline struct nv04_vmmgr_priv *
+nv04_vmmgr(void *obj)
+{
+	return (void *)nouveau_vmmgr(obj);
+}
+
+#endif
diff --git a/nvkm/subdev/vm/nv41.c b/nvkm/subdev/vm/nv41.c
new file mode 100644
index 000000000..9474cfca6
--- /dev/null
+++ b/nvkm/subdev/vm/nv41.c
@@ -0,0 +1,159 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/gpuobj.h>
+#include <core/option.h>
+
+#include <subdev/timer.h>
+#include <subdev/vm.h>
+
+#include "nv04.h"
+
+#define NV41_GART_SIZE (512 * 1024 * 1024)
+#define NV41_GART_PAGE (  4 * 1024)
+
+/*******************************************************************************
+ * VM map/unmap callbacks
+ ******************************************************************************/
+
+static void
+nv41_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	pte = pte * 4;
+	while (cnt) {
+		u32 page = PAGE_SIZE / NV41_GART_PAGE;
+		u64 phys = (u64)*list++;
+		while (cnt && page--) {
+			nv_wo32(pgt, pte, (phys >> 7) | 1);
+			phys += NV41_GART_PAGE;
+			pte += 4;
+			cnt -= 1;
+		}
+	}
+}
+
+static void
+nv41_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte = pte * 4;
+	while (cnt--) {
+		nv_wo32(pgt, pte, 0x00000000);
+		pte += 4;
+	}
+}
+
+static void
+nv41_vm_flush(struct nouveau_vm *vm)
+{
+	struct nv04_vmmgr_priv *priv = (void *)vm->vmm;
+
+	mutex_lock(&nv_subdev(priv)->mutex);
+	nv_wr32(priv, 0x100810, 0x00000022);
+	if (!nv_wait(priv, 0x100810, 0x00000020, 0x00000020)) {
+		nv_warn(priv, "flush timeout, 0x%08x\n",
+			nv_rd32(priv, 0x100810));
+	}
+	nv_wr32(priv, 0x100810, 0x00000000);
+	mutex_unlock(&nv_subdev(priv)->mutex);
+}
+
+/*******************************************************************************
+ * VMMGR subdev
+ ******************************************************************************/
+
+static int
+nv41_vmmgr_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nv04_vmmgr_priv *priv;
+	int ret;
+
+	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
+	    !nouveau_boolopt(device->cfgopt, "NvPCIE", true)) {
+		return nouveau_object_ctor(parent, engine, &nv04_vmmgr_oclass,
+					   data, size, pobject);
+	}
+
+	ret = nouveau_vmmgr_create(parent, engine, oclass, "PCIEGART",
+				   "pciegart", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.create = nv04_vm_create;
+	priv->base.limit = NV41_GART_SIZE;
+	priv->base.dma_bits = 39;
+	priv->base.pgt_bits = 32 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 12;
+	priv->base.map_sg = nv41_vm_map_sg;
+	priv->base.unmap = nv41_vm_unmap;
+	priv->base.flush = nv41_vm_flush;
+
+	ret = nouveau_vm_create(&priv->base, 0, NV41_GART_SIZE, 0, 4096,
+				&priv->vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL,
+				(NV41_GART_SIZE / NV41_GART_PAGE) * 4,
+				 16, NVOBJ_FLAG_ZERO_ALLOC,
+				 &priv->vm->pgt[0].obj[0]);
+	priv->vm->pgt[0].refcount[0] = 1;
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int
+nv41_vmmgr_init(struct nouveau_object *object)
+{
+	struct nv04_vmmgr_priv *priv = (void *)object;
+	struct nouveau_gpuobj *dma = priv->vm->pgt[0].obj[0];
+	int ret;
+
+	ret = nouveau_vmmgr_init(&priv->base);
+	if (ret)
+		return ret;
+
+	nv_wr32(priv, 0x100800, dma->addr | 0x00000002);
+	nv_mask(priv, 0x10008c, 0x00000100, 0x00000100);
+	nv_wr32(priv, 0x100820, 0x00000000);
+	return 0;
+}
+
+struct nouveau_oclass
+nv41_vmmgr_oclass = {
+	.handle = NV_SUBDEV(VM, 0x41),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv41_vmmgr_ctor,
+		.dtor = nv04_vmmgr_dtor,
+		.init = nv41_vmmgr_init,
+		.fini = _nouveau_vmmgr_fini,
+	},
+};
diff --git a/nvkm/subdev/vm/nv44.c b/nvkm/subdev/vm/nv44.c
new file mode 100644
index 000000000..aa8131436
--- /dev/null
+++ b/nvkm/subdev/vm/nv44.c
@@ -0,0 +1,249 @@
+/*
+ * Copyright 2012 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/gpuobj.h>
+#include <core/option.h>
+
+#include <subdev/timer.h>
+#include <subdev/vm.h>
+
+#include "nv04.h"
+
+#define NV44_GART_SIZE (512 * 1024 * 1024)
+#define NV44_GART_PAGE (  4 * 1024)
+
+/*******************************************************************************
+ * VM map/unmap callbacks
+ ******************************************************************************/
+
+static void
+nv44_vm_fill(struct nouveau_gpuobj *pgt, dma_addr_t null,
+	     dma_addr_t *list, u32 pte, u32 cnt)
+{
+	u32 base = (pte << 2) & ~0x0000000f;
+	u32 tmp[4];
+
+	tmp[0] = nv_ro32(pgt, base + 0x0);
+	tmp[1] = nv_ro32(pgt, base + 0x4);
+	tmp[2] = nv_ro32(pgt, base + 0x8);
+	tmp[3] = nv_ro32(pgt, base + 0xc);
+
+	while (cnt--) {
+		u32 addr = list ? (*list++ >> 12) : (null >> 12);
+		switch (pte++ & 0x3) {
+		case 0:
+			tmp[0] &= ~0x07ffffff;
+			tmp[0] |= addr;
+			break;
+		case 1:
+			tmp[0] &= ~0xf8000000;
+			tmp[0] |= addr << 27;
+			tmp[1] &= ~0x003fffff;
+			tmp[1] |= addr >> 5;
+			break;
+		case 2:
+			tmp[1] &= ~0xffc00000;
+			tmp[1] |= addr << 22;
+			tmp[2] &= ~0x0001ffff;
+			tmp[2] |= addr >> 10;
+			break;
+		case 3:
+			tmp[2] &= ~0xfffe0000;
+			tmp[2] |= addr << 17;
+			tmp[3] &= ~0x00000fff;
+			tmp[3] |= addr >> 15;
+			break;
+		}
+	}
+
+	nv_wo32(pgt, base + 0x0, tmp[0]);
+	nv_wo32(pgt, base + 0x4, tmp[1]);
+	nv_wo32(pgt, base + 0x8, tmp[2]);
+	nv_wo32(pgt, base + 0xc, tmp[3] | 0x40000000);
+}
+
+static void
+nv44_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	struct nv04_vmmgr_priv *priv = (void *)vma->vm->vmm;
+	u32 tmp[4];
+	int i;
+
+	if (pte & 3) {
+		u32  max = 4 - (pte & 3);
+		u32 part = (cnt > max) ? max : cnt;
+		nv44_vm_fill(pgt, priv->null, list, pte, part);
+		pte  += part;
+		list += part;
+		cnt  -= part;
+	}
+
+	while (cnt >= 4) {
+		for (i = 0; i < 4; i++)
+			tmp[i] = *list++ >> 12;
+		nv_wo32(pgt, pte++ * 4, tmp[0] >>  0 | tmp[1] << 27);
+		nv_wo32(pgt, pte++ * 4, tmp[1] >>  5 | tmp[2] << 22);
+		nv_wo32(pgt, pte++ * 4, tmp[2] >> 10 | tmp[3] << 17);
+		nv_wo32(pgt, pte++ * 4, tmp[3] >> 15 | 0x40000000);
+		cnt -= 4;
+	}
+
+	if (cnt)
+		nv44_vm_fill(pgt, priv->null, list, pte, cnt);
+}
+
+static void
+nv44_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	struct nv04_vmmgr_priv *priv = (void *)nouveau_vmmgr(pgt);
+
+	if (pte & 3) {
+		u32  max = 4 - (pte & 3);
+		u32 part = (cnt > max) ? max : cnt;
+		nv44_vm_fill(pgt, priv->null, NULL, pte, part);
+		pte  += part;
+		cnt  -= part;
+	}
+
+	while (cnt >= 4) {
+		nv_wo32(pgt, pte++ * 4, 0x00000000);
+		nv_wo32(pgt, pte++ * 4, 0x00000000);
+		nv_wo32(pgt, pte++ * 4, 0x00000000);
+		nv_wo32(pgt, pte++ * 4, 0x00000000);
+		cnt -= 4;
+	}
+
+	if (cnt)
+		nv44_vm_fill(pgt, priv->null, NULL, pte, cnt);
+}
+
+static void
+nv44_vm_flush(struct nouveau_vm *vm)
+{
+	struct nv04_vmmgr_priv *priv = (void *)vm->vmm;
+	nv_wr32(priv, 0x100814, priv->base.limit - NV44_GART_PAGE);
+	nv_wr32(priv, 0x100808, 0x00000020);
+	if (!nv_wait(priv, 0x100808, 0x00000001, 0x00000001))
+		nv_error(priv, "timeout: 0x%08x\n", nv_rd32(priv, 0x100808));
+	nv_wr32(priv, 0x100808, 0x00000000);
+}
+
+/*******************************************************************************
+ * VMMGR subdev
+ ******************************************************************************/
+
+static int
+nv44_vmmgr_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nouveau_device *device = nv_device(parent);
+	struct nv04_vmmgr_priv *priv;
+	int ret;
+
+	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
+	    !nouveau_boolopt(device->cfgopt, "NvPCIE", true)) {
+		return nouveau_object_ctor(parent, engine, &nv04_vmmgr_oclass,
+					   data, size, pobject);
+	}
+
+	ret = nouveau_vmmgr_create(parent, engine, oclass, "PCIEGART",
+				   "pciegart", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.create = nv04_vm_create;
+	priv->base.limit = NV44_GART_SIZE;
+	priv->base.dma_bits = 39;
+	priv->base.pgt_bits = 32 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 12;
+	priv->base.map_sg = nv44_vm_map_sg;
+	priv->base.unmap = nv44_vm_unmap;
+	priv->base.flush = nv44_vm_flush;
+
+	priv->nullp = pci_alloc_consistent(device->pdev, 16 * 1024, &priv->null);
+	if (!priv->nullp) {
+		nv_error(priv, "unable to allocate dummy pages\n");
+		return -ENOMEM;
+	}
+
+	ret = nouveau_vm_create(&priv->base, 0, NV44_GART_SIZE, 0, 4096,
+				&priv->vm);
+	if (ret)
+		return ret;
+
+	ret = nouveau_gpuobj_new(parent, NULL,
+				(NV44_GART_SIZE / NV44_GART_PAGE) * 4,
+				 512 * 1024, NVOBJ_FLAG_ZERO_ALLOC,
+				 &priv->vm->pgt[0].obj[0]);
+	priv->vm->pgt[0].refcount[0] = 1;
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static int
+nv44_vmmgr_init(struct nouveau_object *object)
+{
+	struct nv04_vmmgr_priv *priv = (void *)object;
+	struct nouveau_gpuobj *gart = priv->vm->pgt[0].obj[0];
+	u32 addr;
+	int ret;
+
+	ret = nouveau_vmmgr_init(&priv->base);
+	if (ret)
+		return ret;
+
+	/* calculate vram address of this PRAMIN block, object must be
+	 * allocated on 512KiB alignment, and not exceed a total size
+	 * of 512KiB for this to work correctly
+	 */
+	addr  = nv_rd32(priv, 0x10020c);
+	addr -= ((gart->addr >> 19) + 1) << 19;
+
+	nv_wr32(priv, 0x100850, 0x80000000);
+	nv_wr32(priv, 0x100818, priv->null);
+	nv_wr32(priv, 0x100804, NV44_GART_SIZE);
+	nv_wr32(priv, 0x100850, 0x00008000);
+	nv_mask(priv, 0x10008c, 0x00000200, 0x00000200);
+	nv_wr32(priv, 0x100820, 0x00000000);
+	nv_wr32(priv, 0x10082c, 0x00000001);
+	nv_wr32(priv, 0x100800, addr | 0x00000010);
+	return 0;
+}
+
+struct nouveau_oclass
+nv44_vmmgr_oclass = {
+	.handle = NV_SUBDEV(VM, 0x44),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv44_vmmgr_ctor,
+		.dtor = nv04_vmmgr_dtor,
+		.init = nv44_vmmgr_init,
+		.fini = _nouveau_vmmgr_fini,
+	},
+};
diff --git a/nvkm/subdev/vm/nv50.c b/nvkm/subdev/vm/nv50.c
new file mode 100644
index 000000000..e067f81c9
--- /dev/null
+++ b/nvkm/subdev/vm/nv50.c
@@ -0,0 +1,227 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/device.h>
+#include <core/gpuobj.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/vm.h>
+
+struct nv50_vmmgr_priv {
+	struct nouveau_vmmgr base;
+	spinlock_t lock;
+};
+
+static void
+nv50_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 pde,
+		struct nouveau_gpuobj *pgt[2])
+{
+	u64 phys = 0xdeadcafe00000000ULL;
+	u32 coverage = 0;
+
+	if (pgt[0]) {
+		phys = 0x00000003 | pgt[0]->addr; /* present, 4KiB pages */
+		coverage = (pgt[0]->size >> 3) << 12;
+	} else
+	if (pgt[1]) {
+		phys = 0x00000001 | pgt[1]->addr; /* present */
+		coverage = (pgt[1]->size >> 3) << 16;
+	}
+
+	if (phys & 1) {
+		if (coverage <= 32 * 1024 * 1024)
+			phys |= 0x60;
+		else if (coverage <= 64 * 1024 * 1024)
+			phys |= 0x40;
+		else if (coverage <= 128 * 1024 * 1024)
+			phys |= 0x20;
+	}
+
+	nv_wo32(pgd, (pde * 8) + 0, lower_32_bits(phys));
+	nv_wo32(pgd, (pde * 8) + 4, upper_32_bits(phys));
+}
+
+static inline u64
+vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target)
+{
+	phys |= 1; /* present */
+	phys |= (u64)memtype << 40;
+	phys |= target << 4;
+	if (vma->access & NV_MEM_ACCESS_SYS)
+		phys |= (1 << 6);
+	if (!(vma->access & NV_MEM_ACCESS_WO))
+		phys |= (1 << 3);
+	return phys;
+}
+
+static void
+nv50_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	    struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
+{
+	u32 comp = (mem->memtype & 0x180) >> 7;
+	u32 block, target;
+	int i;
+
+	/* IGPs don't have real VRAM, re-target to stolen system memory */
+	target = 0;
+	if (nouveau_fb(vma->vm->vmm)->ram.stolen) {
+		phys += nouveau_fb(vma->vm->vmm)->ram.stolen;
+		target = 3;
+	}
+
+	phys  = vm_addr(vma, phys, mem->memtype, target);
+	pte <<= 3;
+	cnt <<= 3;
+
+	while (cnt) {
+		u32 offset_h = upper_32_bits(phys);
+		u32 offset_l = lower_32_bits(phys);
+
+		for (i = 7; i >= 0; i--) {
+			block = 1 << (i + 3);
+			if (cnt >= block && !(pte & (block - 1)))
+				break;
+		}
+		offset_l |= (i << 7);
+
+		phys += block << (vma->node->type - 3);
+		cnt  -= block;
+		if (comp) {
+			u32 tag = mem->tag->offset + ((delta >> 16) * comp);
+			offset_h |= (tag << 17);
+			delta    += block << (vma->node->type - 3);
+		}
+
+		while (block) {
+			nv_wo32(pgt, pte + 0, offset_l);
+			nv_wo32(pgt, pte + 4, offset_h);
+			pte += 8;
+			block -= 8;
+		}
+	}
+}
+
+static void
+nv50_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 3 : 2;
+	pte <<= 3;
+	while (cnt--) {
+		u64 phys = vm_addr(vma, (u64)*list++, mem->memtype, target);
+		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
+		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
+		pte += 8;
+	}
+}
+
+static void
+nv50_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte <<= 3;
+	while (cnt--) {
+		nv_wo32(pgt, pte + 0, 0x00000000);
+		nv_wo32(pgt, pte + 4, 0x00000000);
+		pte += 8;
+	}
+}
+
+static void
+nv50_vm_flush(struct nouveau_vm *vm)
+{
+	struct nouveau_engine *engine;
+	int i;
+
+	for (i = 0; i < NVDEV_SUBDEV_NR; i++) {
+		if (atomic_read(&vm->engref[i])) {
+			engine = nouveau_engine(vm->vmm, i);
+			if (engine && engine->tlb_flush)
+				engine->tlb_flush(engine);
+		}
+	}
+}
+
+void
+nv50_vm_flush_engine(struct nouveau_subdev *subdev, int engine)
+{
+	struct nv50_vmmgr_priv *priv = (void *)nouveau_vmmgr(subdev);
+	unsigned long flags;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	nv_wr32(subdev, 0x100c80, (engine << 16) | 1);
+	if (!nv_wait(subdev, 0x100c80, 0x00000001, 0x00000000))
+		nv_error(subdev, "vm flush timeout: engine %d\n", engine);
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+static int
+nv50_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64 length,
+	       u64 mm_offset, struct nouveau_vm **pvm)
+{
+	u32 block = (1 << (vmm->pgt_bits + 12));
+	if (block > length)
+		block = length;
+
+	return nouveau_vm_create(vmm, offset, length, mm_offset, block, pvm);
+}
+
+static int
+nv50_vmmgr_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nv50_vmmgr_priv *priv;
+	int ret;
+
+	ret = nouveau_vmmgr_create(parent, engine, oclass, "VM", "vm", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.limit = 1ULL << 40;
+	priv->base.dma_bits = 40;
+	priv->base.pgt_bits  = 29 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 16;
+	priv->base.create = nv50_vm_create;
+	priv->base.map_pgt = nv50_vm_map_pgt;
+	priv->base.map = nv50_vm_map;
+	priv->base.map_sg = nv50_vm_map_sg;
+	priv->base.unmap = nv50_vm_unmap;
+	priv->base.flush = nv50_vm_flush;
+	spin_lock_init(&priv->lock);
+	return 0;
+}
+
+struct nouveau_oclass
+nv50_vmmgr_oclass = {
+	.handle = NV_SUBDEV(VM, 0x50),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nv50_vmmgr_ctor,
+		.dtor = _nouveau_vmmgr_dtor,
+		.init = _nouveau_vmmgr_init,
+		.fini = _nouveau_vmmgr_fini,
+	},
+};
diff --git a/nvkm/subdev/vm/nvc0.c b/nvkm/subdev/vm/nvc0.c
new file mode 100644
index 000000000..30c61e6c2
--- /dev/null
+++ b/nvkm/subdev/vm/nvc0.c
@@ -0,0 +1,189 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include <core/device.h>
+#include <core/gpuobj.h>
+
+#include <subdev/timer.h>
+#include <subdev/fb.h>
+#include <subdev/vm.h>
+
+struct nvc0_vmmgr_priv {
+	struct nouveau_vmmgr base;
+	spinlock_t lock;
+};
+
+static void
+nvc0_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 index,
+		struct nouveau_gpuobj *pgt[2])
+{
+	u32 pde[2] = { 0, 0 };
+
+	if (pgt[0])
+		pde[1] = 0x00000001 | (pgt[0]->addr >> 8);
+	if (pgt[1])
+		pde[0] = 0x00000001 | (pgt[1]->addr >> 8);
+
+	nv_wo32(pgd, (index * 8) + 0, pde[0]);
+	nv_wo32(pgd, (index * 8) + 4, pde[1]);
+}
+
+static inline u64
+nvc0_vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target)
+{
+	phys >>= 8;
+
+	phys |= 0x00000001; /* present */
+	if (vma->access & NV_MEM_ACCESS_SYS)
+		phys |= 0x00000002;
+
+	phys |= ((u64)target  << 32);
+	phys |= ((u64)memtype << 36);
+
+	return phys;
+}
+
+static void
+nvc0_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	    struct nouveau_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
+{
+	u32 next = 1 << (vma->node->type - 8);
+
+	phys  = nvc0_vm_addr(vma, phys, mem->memtype, 0);
+	pte <<= 3;
+	while (cnt--) {
+		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
+		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
+		phys += next;
+		pte  += 8;
+	}
+}
+
+static void
+nvc0_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       struct nouveau_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
+{
+	u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 7 : 5;
+
+	pte <<= 3;
+	while (cnt--) {
+		u64 phys = nvc0_vm_addr(vma, *list++, mem->memtype, target);
+		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
+		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
+		pte += 8;
+	}
+}
+
+static void
+nvc0_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte <<= 3;
+	while (cnt--) {
+		nv_wo32(pgt, pte + 0, 0x00000000);
+		nv_wo32(pgt, pte + 4, 0x00000000);
+		pte += 8;
+	}
+}
+
+void
+nvc0_vm_flush_engine(struct nouveau_subdev *subdev, u64 addr, int type)
+{
+	struct nvc0_vmmgr_priv *priv = (void *)nouveau_vmmgr(subdev);
+	unsigned long flags;
+
+	/* looks like maybe a "free flush slots" counter, the
+	 * faster you write to 0x100cbc to more it decreases
+	 */
+	spin_lock_irqsave(&priv->lock, flags);
+	if (!nv_wait_ne(subdev, 0x100c80, 0x00ff0000, 0x00000000)) {
+		nv_error(subdev, "vm timeout 0: 0x%08x %d\n",
+			 nv_rd32(subdev, 0x100c80), type);
+	}
+
+	nv_wr32(subdev, 0x100cb8, addr >> 8);
+	nv_wr32(subdev, 0x100cbc, 0x80000000 | type);
+
+	/* wait for flush to be queued? */
+	if (!nv_wait(subdev, 0x100c80, 0x00008000, 0x00008000)) {
+		nv_error(subdev, "vm timeout 1: 0x%08x %d\n",
+			 nv_rd32(subdev, 0x100c80), type);
+	}
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+static void
+nvc0_vm_flush(struct nouveau_vm *vm)
+{
+	struct nouveau_vm_pgd *vpgd;
+
+	list_for_each_entry(vpgd, &vm->pgd_list, head) {
+		nvc0_vm_flush_engine(nv_subdev(vm->vmm), vpgd->obj->addr, 1);
+	}
+}
+
+static int
+nvc0_vm_create(struct nouveau_vmmgr *vmm, u64 offset, u64 length,
+	       u64 mm_offset, struct nouveau_vm **pvm)
+{
+	return nouveau_vm_create(vmm, offset, length, mm_offset, 4096, pvm);
+}
+
+static int
+nvc0_vmmgr_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
+		struct nouveau_oclass *oclass, void *data, u32 size,
+		struct nouveau_object **pobject)
+{
+	struct nvc0_vmmgr_priv *priv;
+	int ret;
+
+	ret = nouveau_vmmgr_create(parent, engine, oclass, "VM", "vm", &priv);
+	*pobject = nv_object(priv);
+	if (ret)
+		return ret;
+
+	priv->base.limit = 1ULL << 40;
+	priv->base.dma_bits = 40;
+	priv->base.pgt_bits  = 27 - 12;
+	priv->base.spg_shift = 12;
+	priv->base.lpg_shift = 17;
+	priv->base.create = nvc0_vm_create;
+	priv->base.map_pgt = nvc0_vm_map_pgt;
+	priv->base.map = nvc0_vm_map;
+	priv->base.map_sg = nvc0_vm_map_sg;
+	priv->base.unmap = nvc0_vm_unmap;
+	priv->base.flush = nvc0_vm_flush;
+	spin_lock_init(&priv->lock);
+	return 0;
+}
+
+struct nouveau_oclass
+nvc0_vmmgr_oclass = {
+	.handle = NV_SUBDEV(VM, 0xc0),
+	.ofuncs = &(struct nouveau_ofuncs) {
+		.ctor = nvc0_vmmgr_ctor,
+		.dtor = _nouveau_vmmgr_dtor,
+		.init = _nouveau_vmmgr_init,
+		.fini = _nouveau_vmmgr_fini,
+	},
+};