thunderbolt: Add support for frame mode

When high-speed DMA paths are used to transfer arbitrary data over a
Thunderbolt link, DMA rings should be in frame mode instead of raw mode.
The latter is used by the control channel (ring 0). In frame mode each
data frame can hold up to 4kB payload.

This patch modifies the DMA ring code to allow configuring a ring to be
in frame mode by passing a new flag (RING_FLAG_FRAME) to the ring when
it is allocated. In addition there might be need to enable end-to-end
(E2E) workaround for the ring to prevent losing Rx frames in certain
situations. We add another flag (RING_FLAG_E2E) that can be used for
this purpose.

This code is based on the work done by Amir Levy and Michael Jamet.

Signed-off-by: Michael Jamet <michael.jamet@intel.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Yehezkel Bernat <yehezkel.bernat@intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

4 files changed, 61 insertions, 30 deletions

diff --git a/drivers/thunderbolt/ctl.c b/drivers/thunderbolt/ctl.c
index 46e393c5fd1d..05400b77dcd7 100644
--- a/drivers/thunderbolt/ctl.c
+++ b/drivers/thunderbolt/ctl.c
@@ -618,7 +618,8 @@ struct tb_ctl *tb_ctl_alloc(struct tb_nhi *nhi, event_cb cb, void *cb_data)
 	if (!ctl->tx)
 		goto err;
 
-	ctl->rx = ring_alloc_rx(nhi, 0, 10, RING_FLAG_NO_SUSPEND);
+	ctl->rx = ring_alloc_rx(nhi, 0, 10, RING_FLAG_NO_SUSPEND, 0xffff,
+				0xffff);
 	if (!ctl->rx)
 		goto err;
 
diff --git a/drivers/thunderbolt/nhi.c b/drivers/thunderbolt/nhi.c
index 8a7a3d0133f9..bebcad3d2c1f 100644
--- a/drivers/thunderbolt/nhi.c
+++ b/drivers/thunderbolt/nhi.c
@@ -22,6 +22,12 @@
 #define RING_TYPE(ring) ((ring)->is_tx ? "TX ring" : "RX ring")
 
 /*
+ * Used to enable end-to-end workaround for missing RX packets. Do not
+ * use this ring for anything else.
+ */
+#define RING_E2E_UNUSED_HOPID	2
+
+/*
  * Minimal number of vectors when we use MSI-X. Two for control channel
  * Rx/Tx and the rest four are for cross domain DMA paths.
  */
@@ -229,23 +235,6 @@ static void ring_work(struct work_struct *work)
 			frame->eof = ring->descriptors[ring->tail].eof;
 			frame->sof = ring->descriptors[ring->tail].sof;
 			frame->flags = ring->descriptors[ring->tail].flags;
-			if (frame->sof != 0)
-				dev_WARN(&ring->nhi->pdev->dev,
-					 "%s %d got unexpected SOF: %#x\n",
-					 RING_TYPE(ring), ring->hop,
-					 frame->sof);
-			/*
-			 * known flags:
-			 * raw not enabled, interupt not set: 0x2=0010
-			 * raw enabled: 0xa=1010
-			 * raw not enabled: 0xb=1011
-			 * partial frame (>MAX_FRAME_SIZE): 0xe=1110
-			 */
-			if (frame->flags != 0xa)
-				dev_WARN(&ring->nhi->pdev->dev,
-					 "%s %d got unexpected flags: %#x\n",
-					 RING_TYPE(ring), ring->hop,
-					 frame->flags);
 		}
 		ring->tail = (ring->tail + 1) % ring->size;
 	}
@@ -321,12 +310,17 @@ static void ring_release_msix(struct tb_ring *ring)
 }
 
 static struct tb_ring *ring_alloc(struct tb_nhi *nhi, u32 hop, int size,
-				  bool transmit, unsigned int flags)
+				  bool transmit, unsigned int flags,
+				  u16 sof_mask, u16 eof_mask)
 {
 	struct tb_ring *ring = NULL;
 	dev_info(&nhi->pdev->dev, "allocating %s ring %d of size %d\n",
 		 transmit ? "TX" : "RX", hop, size);
 
+	/* Tx Ring 2 is reserved for E2E workaround */
+	if (transmit && hop == RING_E2E_UNUSED_HOPID)
+		return NULL;
+
 	mutex_lock(&nhi->lock);
 	if (hop >= nhi->hop_count) {
 		dev_WARN(&nhi->pdev->dev, "invalid hop: %d\n", hop);
@@ -353,6 +347,8 @@ static struct tb_ring *ring_alloc(struct tb_nhi *nhi, u32 hop, int size,
 	ring->is_tx = transmit;
 	ring->size = size;
 	ring->flags = flags;
+	ring->sof_mask = sof_mask;
+	ring->eof_mask = eof_mask;
 	ring->head = 0;
 	ring->tail = 0;
 	ring->running = false;
@@ -384,13 +380,13 @@ err:
 struct tb_ring *ring_alloc_tx(struct tb_nhi *nhi, int hop, int size,
 			      unsigned int flags)
 {
-	return ring_alloc(nhi, hop, size, true, flags);
+	return ring_alloc(nhi, hop, size, true, flags, 0, 0);
 }
 
 struct tb_ring *ring_alloc_rx(struct tb_nhi *nhi, int hop, int size,
-			      unsigned int flags)
+			      unsigned int flags, u16 sof_mask, u16 eof_mask)
 {
-	return ring_alloc(nhi, hop, size, false, flags);
+	return ring_alloc(nhi, hop, size, false, flags, sof_mask, eof_mask);
 }
 
 /**
@@ -400,6 +396,9 @@ struct tb_ring *ring_alloc_rx(struct tb_nhi *nhi, int hop, int size,
  */
 void ring_start(struct tb_ring *ring)
 {
+	u16 frame_size;
+	u32 flags;
+
 	mutex_lock(&ring->nhi->lock);
 	mutex_lock(&ring->lock);
 	if (ring->nhi->going_away)
@@ -411,18 +410,39 @@ void ring_start(struct tb_ring *ring)
 	dev_info(&ring->nhi->pdev->dev, "starting %s %d\n",
 		 RING_TYPE(ring), ring->hop);
 
+	if (ring->flags & RING_FLAG_FRAME) {
+		/* Means 4096 */
+		frame_size = 0;
+		flags = RING_FLAG_ENABLE;
+	} else {
+		frame_size = TB_FRAME_SIZE;
+		flags = RING_FLAG_ENABLE | RING_FLAG_RAW;
+	}
+
+	if (ring->flags & RING_FLAG_E2E && !ring->is_tx) {
+		u32 hop;
+
+		/*
+		 * In order not to lose Rx packets we enable end-to-end
+		 * workaround which transfers Rx credits to an unused Tx
+		 * HopID.
+		 */
+		hop = RING_E2E_UNUSED_HOPID << REG_RX_OPTIONS_E2E_HOP_SHIFT;
+		hop &= REG_RX_OPTIONS_E2E_HOP_MASK;
+		flags |= hop | RING_FLAG_E2E_FLOW_CONTROL;
+	}
+
 	ring_iowrite64desc(ring, ring->descriptors_dma, 0);
 	if (ring->is_tx) {
 		ring_iowrite32desc(ring, ring->size, 12);
 		ring_iowrite32options(ring, 0, 4); /* time releated ? */
-		ring_iowrite32options(ring,
-				      RING_FLAG_ENABLE | RING_FLAG_RAW, 0);
+		ring_iowrite32options(ring, flags, 0);
 	} else {
-		ring_iowrite32desc(ring,
-				   (TB_FRAME_SIZE << 16) | ring->size, 12);
-		ring_iowrite32options(ring, 0xffffffff, 4); /* SOF EOF mask */
-		ring_iowrite32options(ring,
-				      RING_FLAG_ENABLE | RING_FLAG_RAW, 0);
+		u32 sof_eof_mask = ring->sof_mask << 16 | ring->eof_mask;
+
+		ring_iowrite32desc(ring, (frame_size << 16) | ring->size, 12);
+		ring_iowrite32options(ring, sof_eof_mask, 4);
+		ring_iowrite32options(ring, flags, 0);
 	}
 	ring_interrupt_active(ring, true);
 	ring->running = true;
diff --git a/drivers/thunderbolt/nhi.h b/drivers/thunderbolt/nhi.h
index 0e05828983db..4503ddbeccb3 100644
--- a/drivers/thunderbolt/nhi.h
+++ b/drivers/thunderbolt/nhi.h
@@ -56,6 +56,8 @@ struct tb_nhi {
  * @irq: MSI-X irq number if the ring uses MSI-X. %0 otherwise.
  * @vector: MSI-X vector number the ring uses (only set if @irq is > 0)
  * @flags: Ring specific flags
+ * @sof_mask: Bit mask used to detect start of frame PDF
+ * @eof_mask: Bit mask used to detect end of frame PDF
  */
 struct tb_ring {
 	struct mutex lock;
@@ -74,10 +76,16 @@ struct tb_ring {
 	int irq;
 	u8 vector;
 	unsigned int flags;
+	u16 sof_mask;
+	u16 eof_mask;
 };
 
 /* Leave ring interrupt enabled on suspend */
 #define RING_FLAG_NO_SUSPEND	BIT(0)
+/* Configure the ring to be in frame mode */
+#define RING_FLAG_FRAME		BIT(1)
+/* Enable end-to-end flow control */
+#define RING_FLAG_E2E		BIT(2)
 
 struct ring_frame;
 typedef void (*ring_cb)(struct tb_ring*, struct ring_frame*, bool canceled);
@@ -100,7 +108,7 @@ struct ring_frame {
 struct tb_ring *ring_alloc_tx(struct tb_nhi *nhi, int hop, int size,
 			      unsigned int flags);
 struct tb_ring *ring_alloc_rx(struct tb_nhi *nhi, int hop, int size,
-			      unsigned int flags);
+			      unsigned int flags, u16 sof_mask, u16 eof_mask);
 void ring_start(struct tb_ring *ring);
 void ring_stop(struct tb_ring *ring);
 void ring_free(struct tb_ring *ring);
diff --git a/drivers/thunderbolt/nhi_regs.h b/drivers/thunderbolt/nhi_regs.h
index 46eff69b19ad..491a4c0c18fc 100644
--- a/drivers/thunderbolt/nhi_regs.h
+++ b/drivers/thunderbolt/nhi_regs.h
@@ -77,6 +77,8 @@ struct ring_desc {
  * ..: unknown
  */
 #define REG_RX_OPTIONS_BASE	0x29800
+#define REG_RX_OPTIONS_E2E_HOP_MASK	GENMASK(22, 12)
+#define REG_RX_OPTIONS_E2E_HOP_SHIFT	12
 
 /*
  * three bitfields: tx, rx, rx overflow