runtime: make more functions safe for Go

Convert no-op race functions.
Everything else is tiny and gets NOSPLITs.

After this, all that is left on darwin is sysAlloc, panic, and gothrow (all pending).
There may be system-specific calls in other builds.

LGTM=iant
R=golang-codereviews, iant
CC=dvyukov, golang-codereviews, khr, r
https://codereview.appspot.com/140240044

19 files changed, 274 insertions, 220 deletions

diff --git a/src/pkg/runtime/cgocall.go b/src/pkg/runtime/cgocall.go
index d44b832df..1037c5dc2 100644
--- a/src/pkg/runtime/cgocall.go
+++ b/src/pkg/runtime/cgocall.go
@@ -101,7 +101,7 @@ func cgocall_errno(fn, arg unsafe.Pointer) int32 {
 
 	// Create an extra M for callbacks on threads not created by Go on first cgo call.
 	if needextram == 1 && cas(&needextram, 1, 0) {
-		newextram()
+		onM(newextram)
 	}
 
 	/*
@@ -188,7 +188,7 @@ func cgocallbackg1() {
 	gp := getg()
 	if gp.m.needextram {
 		gp.m.needextram = false
-		newextram()
+		onM(newextram)
 	}
 
 	// Add entry to defer stack in case of panic.
diff --git a/src/pkg/runtime/os_darwin.c b/src/pkg/runtime/os_darwin.c
index d3890eee5..466591877 100644
--- a/src/pkg/runtime/os_darwin.c
+++ b/src/pkg/runtime/os_darwin.c
@@ -71,6 +71,7 @@ runtime·osinit(void)
 		runtime·ncpu = out;
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
diff --git a/src/pkg/runtime/os_dragonfly.c b/src/pkg/runtime/os_dragonfly.c
index a2a88e4c8..f93c02ce1 100644
--- a/src/pkg/runtime/os_dragonfly.c
+++ b/src/pkg/runtime/os_dragonfly.c
@@ -72,19 +72,40 @@ runtime·futexsleep(uint32 *addr, uint32 val, int64 ns)
 	*(int32*)0x1005 = 0x1005;
 }
 
+static void badfutexwakeup(void);
+
+#pragma textflag NOSPLIT
 void
 runtime·futexwakeup(uint32 *addr, uint32 cnt)
 {
 	int32 ret;
+	void (*fn)(void);
 
 	ret = runtime·sys_umtx_wakeup(addr, cnt);
 	if(ret >= 0)
 		return;
 
-	runtime·printf("umtx_wake addr=%p ret=%d\n", addr, ret);
+	g->m->ptrarg[0] = addr;
+	g->m->scalararg[0] = ret;
+	fn = badfutexwakeup;
+	if(g == g->m->gsignal)
+		fn();
+	else
+		runtime·onM(&fn);
 	*(int32*)0x1006 = 0x1006;
 }
 
+static void
+badfutexwakeup(void)
+{
+	void *addr;
+	int32 ret;
+	
+	addr = g->m->ptrarg[0];
+	ret = g->m->scalararg[0];
+	runtime·printf("umtx_wake addr=%p ret=%d\n", addr, ret);
+}
+
 void runtime·lwp_start(void*);
 
 void
@@ -119,6 +140,7 @@ runtime·osinit(void)
 	runtime·ncpu = getncpu();
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
diff --git a/src/pkg/runtime/os_freebsd.c b/src/pkg/runtime/os_freebsd.c
index ae959d90b..ab1edcd93 100644
--- a/src/pkg/runtime/os_freebsd.c
+++ b/src/pkg/runtime/os_freebsd.c
@@ -73,19 +73,40 @@ fail:
 	*(int32*)0x1005 = 0x1005;
 }
 
+static void badfutexwakeup(void);
+
+#pragma textflag NOSPLIT
 void
 runtime·futexwakeup(uint32 *addr, uint32 cnt)
 {
 	int32 ret;
+	void (*fn)(void);
 
 	ret = runtime·sys_umtx_op(addr, UMTX_OP_WAKE_PRIVATE, cnt, nil, nil);
 	if(ret >= 0)
 		return;
 
-	runtime·printf("umtx_wake addr=%p ret=%d\n", addr, ret);
+	g->m->ptrarg[0] = addr;
+	g->m->scalararg[0] = ret;
+	fn = badfutexwakeup;
+	if(g == g->m->gsignal)
+		fn();
+	else
+		runtime·onM(&fn);
 	*(int32*)0x1006 = 0x1006;
 }
 
+static void
+badfutexwakeup(void)
+{
+	void *addr;
+	int32 ret;
+	
+	addr = g->m->ptrarg[0];
+	ret = g->m->scalararg[0];
+	runtime·printf("umtx_wake addr=%p ret=%d\n", addr, ret);
+}
+
 void runtime·thr_start(void*);
 
 void
@@ -127,6 +148,7 @@ runtime·osinit(void)
 	runtime·ncpu = getncpu();
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
diff --git a/src/pkg/runtime/os_linux.c b/src/pkg/runtime/os_linux.c
index 34521cc86..7ab4858d9 100644
--- a/src/pkg/runtime/os_linux.c
+++ b/src/pkg/runtime/os_linux.c
@@ -55,24 +55,44 @@ runtime·futexsleep(uint32 *addr, uint32 val, int64 ns)
 	runtime·futex(addr, FUTEX_WAIT, val, &ts, nil, 0);
 }
 
+static void badfutexwakeup(void);
+
 // If any procs are sleeping on addr, wake up at most cnt.
+#pragma textflag NOSPLIT
 void
 runtime·futexwakeup(uint32 *addr, uint32 cnt)
 {
 	int64 ret;
+	void (*fn)(void);
 
 	ret = runtime·futex(addr, FUTEX_WAKE, cnt, nil, nil, 0);
-
 	if(ret >= 0)
 		return;
 
 	// I don't know that futex wakeup can return
 	// EAGAIN or EINTR, but if it does, it would be
 	// safe to loop and call futex again.
-	runtime·printf("futexwakeup addr=%p returned %D\n", addr, ret);
+	g->m->ptrarg[0] = addr;
+	g->m->scalararg[0] = (int32)ret; // truncated but fine
+	fn = badfutexwakeup;
+	if(g == g->m->gsignal)
+		fn();
+	else
+		runtime·onM(&fn);
 	*(int32*)0x1006 = 0x1006;
 }
 
+static void
+badfutexwakeup(void)
+{
+	void *addr;
+	int64 ret;
+	
+	addr = g->m->ptrarg[0];
+	ret = (int32)g->m->scalararg[0];
+	runtime·printf("futexwakeup addr=%p returned %D\n", addr, ret);
+}
+
 extern runtime·sched_getaffinity(uintptr pid, uintptr len, uintptr *buf);
 static int32
 getproccount(void)
@@ -162,6 +182,7 @@ runtime·osinit(void)
 byte*	runtime·startup_random_data;
 uint32	runtime·startup_random_data_len;
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
diff --git a/src/pkg/runtime/os_nacl.c b/src/pkg/runtime/os_nacl.c
index f859dd0d3..fba32e568 100644
--- a/src/pkg/runtime/os_nacl.c
+++ b/src/pkg/runtime/os_nacl.c
@@ -65,6 +65,7 @@ runtime·crash(void)
 	*(int32*)0 = 0;
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
@@ -112,8 +113,8 @@ runtime·newosproc(M *mp, void *stk)
 	}
 }
 
-uintptr
-runtime·semacreate(void)
+static void
+semacreate(void)
 {
 	int32 mu, cond;
 	
@@ -128,14 +129,32 @@ runtime·semacreate(void)
 		runtime·throw("semacreate");
 	}
 	g->m->waitsemalock = mu;
-	return cond; // assigned to m->waitsema
+	g->m->scalararg[0] = cond; // assigned to m->waitsema
 }
 
 #pragma textflag NOSPLIT
-int32
-runtime·semasleep(int64 ns)
+uint32
+runtime·semacreate(void)
+{
+	void (*fn)(void);
+	uint32 x;
+	
+	fn = semacreate;
+	runtime·onM(&fn);
+	x = g->m->scalararg[0];
+	g->m->scalararg[0] = 0;
+	return x;
+}
+
+static void
+semasleep(void)
 {
 	int32 ret;
+	int64 ns;
+	
+	ns = (int64)(uint32)g->m->scalararg[0] | (int64)(uint32)g->m->scalararg[1]<<32;
+	g->m->scalararg[0] = 0;
+	g->m->scalararg[1] = 0;
 	
 	ret = runtime·nacl_mutex_lock(g->m->waitsemalock);
 	if(ret < 0) {
@@ -145,7 +164,8 @@ runtime·semasleep(int64 ns)
 	if(g->m->waitsemacount > 0) {
 		g->m->waitsemacount = 0;
 		runtime·nacl_mutex_unlock(g->m->waitsemalock);
-		return 0;
+		g->m->scalararg[0] = 0;
+		return;
 	}
 
 	while(g->m->waitsemacount == 0) {
@@ -163,7 +183,8 @@ runtime·semasleep(int64 ns)
 			ret = runtime·nacl_cond_timed_wait_abs(g->m->waitsema, g->m->waitsemalock, &ts);
 			if(ret == -ETIMEDOUT) {
 				runtime·nacl_mutex_unlock(g->m->waitsemalock);
-				return -1;
+				g->m->scalararg[0] = -1;
+				return;
 			}
 			if(ret < 0) {
 				//runtime·printf("nacl_cond_timed_wait_abs: error %d\n", -ret);
@@ -174,14 +195,34 @@ runtime·semasleep(int64 ns)
 			
 	g->m->waitsemacount = 0;
 	runtime·nacl_mutex_unlock(g->m->waitsemalock);
-	return 0;
+	g->m->scalararg[0] = 0;
 }
 
-void
-runtime·semawakeup(M *mp)
+#pragma textflag NOSPLIT
+int32
+runtime·semasleep(int64 ns)
+{
+	int32 r;
+	void (*fn)(void);
+
+	g->m->scalararg[0] = (uint32)ns;
+	g->m->scalararg[1] = (uint32)(ns>>32);
+	fn = semasleep;
+	runtime·onM(&fn);
+	r = g->m->scalararg[0];
+	g->m->scalararg[0] = 0;
+	return r;
+}
+
+static void
+semawakeup(void)
 {
 	int32 ret;
+	M *mp;
 	
+	mp = g->m->ptrarg[0];
+	g->m->ptrarg[0] = nil;
+
 	ret = runtime·nacl_mutex_lock(mp->waitsemalock);
 	if(ret < 0) {
 		//runtime·printf("nacl_mutex_lock: error %d\n", -ret);
@@ -196,6 +237,17 @@ runtime·semawakeup(M *mp)
 	runtime·nacl_mutex_unlock(mp->waitsemalock);
 }
 
+#pragma textflag NOSPLIT
+void
+runtime·semawakeup(M *mp)
+{
+	void (*fn)(void);
+
+	g->m->ptrarg[0] = mp;
+	fn = semawakeup;
+	runtime·onM(&fn);
+}
+
 uintptr
 runtime·memlimit(void)
 {
diff --git a/src/pkg/runtime/os_netbsd.c b/src/pkg/runtime/os_netbsd.c
index 8567146ed..29ea0cb42 100644
--- a/src/pkg/runtime/os_netbsd.c
+++ b/src/pkg/runtime/os_netbsd.c
@@ -185,6 +185,7 @@ runtime·osinit(void)
 	runtime·ncpu = getncpu();
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
diff --git a/src/pkg/runtime/os_openbsd.c b/src/pkg/runtime/os_openbsd.c
index dbd12a128..728cb88c4 100644
--- a/src/pkg/runtime/os_openbsd.c
+++ b/src/pkg/runtime/os_openbsd.c
@@ -164,6 +164,7 @@ runtime·osinit(void)
 	runtime·ncpu = getncpu();
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
diff --git a/src/pkg/runtime/os_plan9.c b/src/pkg/runtime/os_plan9.c
index fad5b2717..efff24ea0 100644
--- a/src/pkg/runtime/os_plan9.c
+++ b/src/pkg/runtime/os_plan9.c
@@ -100,6 +100,7 @@ runtime·crash(void)
 	*(int32*)0 = 0;
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
diff --git a/src/pkg/runtime/os_solaris.c b/src/pkg/runtime/os_solaris.c
index d064384e8..c7ec3f212 100644
--- a/src/pkg/runtime/os_solaris.c
+++ b/src/pkg/runtime/os_solaris.c
@@ -143,6 +143,7 @@ runtime·newosproc(M *mp, void *stk)
 	}
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
diff --git a/src/pkg/runtime/os_windows.c b/src/pkg/runtime/os_windows.c
index 0e53d8a31..9bf6ecbc6 100644
--- a/src/pkg/runtime/os_windows.c
+++ b/src/pkg/runtime/os_windows.c
@@ -109,6 +109,7 @@ runtime·osinit(void)
 	}
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·get_random_data(byte **rnd, int32 *rnd_len)
 {
diff --git a/src/pkg/runtime/proc.c b/src/pkg/runtime/proc.c
index 45ae1bc2a..0c72e4bd0 100644
--- a/src/pkg/runtime/proc.c
+++ b/src/pkg/runtime/proc.c
@@ -2534,6 +2534,7 @@ lockOSThread(void)
 	g->lockedm = g->m;
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·LockOSThread(void)
 {
@@ -2541,6 +2542,7 @@ runtime·LockOSThread(void)
 	lockOSThread();
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·lockOSThread(void)
 {
@@ -2562,6 +2564,7 @@ unlockOSThread(void)
 	g->lockedm = nil;
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·UnlockOSThread(void)
 {
@@ -2569,15 +2572,28 @@ runtime·UnlockOSThread(void)
 	unlockOSThread();
 }
 
+static void badunlockOSThread(void);
+
+#pragma textflag NOSPLIT
 void
 runtime·unlockOSThread(void)
 {
-	if(g->m->locked < LockInternal)
-		runtime·throw("runtime: internal error: misuse of lockOSThread/unlockOSThread");
+	void (*fn)(void);
+
+	if(g->m->locked < LockInternal) {
+		fn = badunlockOSThread;
+		runtime·onM(&fn);
+	}
 	g->m->locked -= LockInternal;
 	unlockOSThread();
 }
 
+static void
+badunlockOSThread(void)
+{
+	runtime·throw("runtime: internal error: misuse of lockOSThread/unlockOSThread");
+}
+
 bool
 runtime·lockedOSThread(void)
 {
diff --git a/src/pkg/runtime/race.c b/src/pkg/runtime/race.c
index d5e77c756..125309510 100644
--- a/src/pkg/runtime/race.c
+++ b/src/pkg/runtime/race.c
@@ -80,6 +80,7 @@ void runtime·racesymbolizethunk(void*);
 void runtime·racecall(void(*f)(void), ...);
 
 // checks if the address has shadow (i.e. heap or data/bss)
+#pragma textflag NOSPLIT
 static bool
 isvalidaddr(uintptr addr)
 {
@@ -90,6 +91,7 @@ isvalidaddr(uintptr addr)
 	return false;
 }
 
+#pragma textflag NOSPLIT
 uintptr
 runtime·raceinit(void)
 {
@@ -106,12 +108,14 @@ runtime·raceinit(void)
 	return racectx;
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racefini(void)
 {
 	runtime·racecall(__tsan_fini);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racemapshadow(void *addr, uintptr size)
 {
@@ -129,6 +133,7 @@ runtime·racemalloc(void *p, uintptr sz)
 	runtime·racecall(__tsan_malloc, p, sz);
 }
 
+#pragma textflag NOSPLIT
 uintptr
 runtime·racegostart(void *pc)
 {
@@ -144,12 +149,14 @@ runtime·racegostart(void *pc)
 	return racectx;
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racegoend(void)
 {
 	runtime·racecall(__tsan_go_end, g->racectx);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racewriterangepc(void *addr, uintptr sz, void *callpc, void *pc)
 {
@@ -165,6 +172,7 @@ runtime·racewriterangepc(void *addr, uintptr sz, void *callpc, void *pc)
 		runtime·racefuncexit();
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racereadrangepc(void *addr, uintptr sz, void *callpc, void *pc)
 {
@@ -180,6 +188,7 @@ runtime·racereadrangepc(void *addr, uintptr sz, void *callpc, void *pc)
 		runtime·racefuncexit();
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racewriteobjectpc(void *addr, Type *t, void *callpc, void *pc)
 {
@@ -192,6 +201,7 @@ runtime·racewriteobjectpc(void *addr, Type *t, void *callpc, void *pc)
 		runtime·racewritepc(addr, callpc, pc);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racereadobjectpc(void *addr, Type *t, void *callpc, void *pc)
 {
@@ -204,12 +214,14 @@ runtime·racereadobjectpc(void *addr, Type *t, void *callpc, void *pc)
 		runtime·racereadpc(addr, callpc, pc);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·raceacquire(void *addr)
 {
 	runtime·raceacquireg(g, addr);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·raceacquireg(G *gp, void *addr)
 {
@@ -218,6 +230,7 @@ runtime·raceacquireg(G *gp, void *addr)
 	runtime·racecall(__tsan_acquire, gp->racectx, addr);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racerelease(void *addr)
 {
@@ -226,6 +239,7 @@ runtime·racerelease(void *addr)
 	runtime·racereleaseg(g, addr);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racereleaseg(G *gp, void *addr)
 {
@@ -234,12 +248,14 @@ runtime·racereleaseg(G *gp, void *addr)
 	runtime·racecall(__tsan_release, gp->racectx, addr);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racereleasemerge(void *addr)
 {
 	runtime·racereleasemergeg(g, addr);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racereleasemergeg(G *gp, void *addr)
 {
@@ -248,6 +264,7 @@ runtime·racereleasemergeg(G *gp, void *addr)
 	runtime·racecall(__tsan_release_merge, gp->racectx, addr);
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·racefingo(void)
 {
@@ -255,6 +272,7 @@ runtime·racefingo(void)
 }
 
 // func RaceAcquire(addr unsafe.Pointer)
+#pragma textflag NOSPLIT
 void
 runtime·RaceAcquire(void *addr)
 {
@@ -262,6 +280,7 @@ runtime·RaceAcquire(void *addr)
 }
 
 // func RaceRelease(addr unsafe.Pointer)
+#pragma textflag NOSPLIT
 void
 runtime·RaceRelease(void *addr)
 {
@@ -269,6 +288,7 @@ runtime·RaceRelease(void *addr)
 }
 
 // func RaceReleaseMerge(addr unsafe.Pointer)
+#pragma textflag NOSPLIT
 void
 runtime·RaceReleaseMerge(void *addr)
 {
@@ -276,6 +296,7 @@ runtime·RaceReleaseMerge(void *addr)
 }
 
 // func RaceDisable()
+#pragma textflag NOSPLIT
 void
 runtime·RaceDisable(void)
 {
@@ -284,43 +305,10 @@ runtime·RaceDisable(void)
 }
 
 // func RaceEnable()
+#pragma textflag NOSPLIT
 void
 runtime·RaceEnable(void)
 {
 	if(--g->raceignore == 0)
 		runtime·racecall(__tsan_go_ignore_sync_end, g->racectx);
 }
-
-typedef struct SymbolizeContext SymbolizeContext;
-struct SymbolizeContext
-{
-	uintptr	pc;
-	int8*	func;
-	int8*	file;
-	uintptr	line;
-	uintptr	off;
-	uintptr	res;
-};
-
-// Callback from C into Go, runs on g0.
-void
-runtime·racesymbolize(SymbolizeContext *ctx)
-{
-	Func *f;
-	String file;
-
-	f = runtime·findfunc(ctx->pc);
-	if(f == nil) {
-		ctx->func = "??";
-		ctx->file = "-";
-		ctx->line = 0;
-		ctx->off = ctx->pc;
-		ctx->res = 1;
-		return;
-	}
-	ctx->func = runtime·funcname(f);
-	ctx->line = runtime·funcline(f, ctx->pc, &file);
-	ctx->file = (int8*)file.str;  // assume zero-terminated
-	ctx->off = ctx->pc - f->entry;
-	ctx->res = 1;
-}
diff --git a/src/pkg/runtime/race.go b/src/pkg/runtime/race.go
index df8493e35..c7573517d 100644
--- a/src/pkg/runtime/race.go
+++ b/src/pkg/runtime/race.go
@@ -38,7 +38,7 @@ func raceReadObjectPC(t *_type, addr unsafe.Pointer, callerpc, pc uintptr) {
 	if kind == kindArray || kind == kindStruct {
 		// for composite objects we have to read every address
 		// because a write might happen to any subobject.
-		racereadrangepc(addr, int(t.size), callerpc, pc)
+		racereadrangepc(addr, t.size, callerpc, pc)
 	} else {
 		// for non-composite objects we can read just the start
 		// address, as any write must write the first byte.
@@ -51,10 +51,75 @@ func raceWriteObjectPC(t *_type, addr unsafe.Pointer, callerpc, pc uintptr) {
 	if kind == kindArray || kind == kindStruct {
 		// for composite objects we have to write every address
 		// because a write might happen to any subobject.
-		racewriterangepc(addr, int(t.size), callerpc, pc)
+		racewriterangepc(addr, t.size, callerpc, pc)
 	} else {
 		// for non-composite objects we can write just the start
 		// address, as any write must write the first byte.
 		racewritepc(addr, callerpc, pc)
 	}
 }
+
+//go:noescape
+func racereadpc(addr unsafe.Pointer, callpc, pc uintptr)
+
+//go:noescape
+func racewritepc(addr unsafe.Pointer, callpc, pc uintptr)
+
+//go:noescape
+func racereadrangepc(addr unsafe.Pointer, len uintptr, callpc, pc uintptr)
+
+//go:noescape
+func racewriterangepc(addr unsafe.Pointer, len uintptr, callpc, pc uintptr)
+
+//go:noescape
+func raceacquire(addr unsafe.Pointer)
+
+//go:noescape
+func racerelease(addr unsafe.Pointer)
+
+//go:noescape
+func raceacquireg(gp *g, addr unsafe.Pointer)
+
+//go:noescape
+func racereleaseg(gp *g, addr unsafe.Pointer)
+
+func racefingo()
+
+//go:noescape
+func racemalloc(p unsafe.Pointer, size uintptr)
+
+//go:noescape
+func racereleasemerge(addr unsafe.Pointer)
+
+type symbolizeContext struct {
+	pc   uintptr
+	fn   *byte
+	file *byte
+	line uintptr
+	off  uintptr
+	res  uintptr
+}
+
+var qq = [...]byte{'?', '?', 0}
+var dash = [...]byte{'-', 0}
+
+// Callback from C into Go, runs on g0.
+func racesymbolize(ctx *symbolizeContext) {
+	f := findfunc(ctx.pc)
+	if f == nil {
+		ctx.fn = &qq[0]
+		ctx.file = &dash[0]
+		ctx.line = 0
+		ctx.off = ctx.pc
+		ctx.res = 1
+		return
+	}
+
+	ctx.fn = funcname(f)
+	var file string
+	ctx.line = uintptr(funcline(f, ctx.pc, &file))
+	ctx.file = &bytes(file)[0] // assume NUL-terminated
+	ctx.off = ctx.pc - f.entry
+	ctx.res = 1
+	return
+}
diff --git a/src/pkg/runtime/race0.c b/src/pkg/runtime/race0.c
deleted file mode 100644
index eddb0be79..000000000
--- a/src/pkg/runtime/race0.c
+++ /dev/null
@@ -1,124 +0,0 @@
-// Copyright 2011 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// Stub implementation of the race detector API.
-// +build !race
-
-#include "runtime.h"
-
-uintptr
-runtime·raceinit(void)
-{
-	return 0;
-}
-
-void
-runtime·racefini(void)
-{
-}
-
-
-void
-runtime·racemapshadow(void *addr, uintptr size)
-{
-	USED(addr);
-	USED(size);
-}
-
-void
-runtime·racewritepc(void *addr, void *callpc, void *pc)
-{
-	USED(addr);
-	USED(callpc);
-	USED(pc);
-}
-
-void
-runtime·racereadpc(void *addr, void *callpc, void *pc)
-{
-	USED(addr);
-	USED(callpc);
-	USED(pc);
-}
-
-void
-runtime·racewriterangepc(void *addr, uintptr sz, void *callpc, void *pc)
-{
-	USED(addr);
-	USED(sz);
-	USED(callpc);
-	USED(pc);
-}
-
-void
-runtime·racereadrangepc(void *addr, uintptr sz, void *callpc, void *pc)
-{
-	USED(addr);
-	USED(sz);
-	USED(callpc);
-	USED(pc);
-}
-
-void
-runtime·raceacquire(void *addr)
-{
-	USED(addr);
-}
-
-void
-runtime·raceacquireg(G *gp, void *addr)
-{
-	USED(gp);
-	USED(addr);
-}
-
-void
-runtime·racerelease(void *addr)
-{
-	USED(addr);
-}
-
-void
-runtime·racereleaseg(G *gp, void *addr)
-{
-	USED(gp);
-	USED(addr);
-}
-
-void
-runtime·racereleasemerge(void *addr)
-{
-	USED(addr);
-}
-
-void
-runtime·racereleasemergeg(G *gp, void *addr)
-{
-	USED(gp);
-	USED(addr);
-}
-
-void
-runtime·racefingo(void)
-{
-}
-
-void
-runtime·racemalloc(void *p, uintptr sz)
-{
-	USED(p);
-	USED(sz);
-}
-
-uintptr
-runtime·racegostart(void *pc)
-{
-	USED(pc);
-	return 0;
-}
-
-void
-runtime·racegoend(void)
-{
-}
diff --git a/src/pkg/runtime/race0.go b/src/pkg/runtime/race0.go
index 2e67ae6a5..5d90cc859 100644
--- a/src/pkg/runtime/race0.go
+++ b/src/pkg/runtime/race0.go
@@ -14,7 +14,24 @@ import (
 
 const raceenabled = false
 
-func raceReadObjectPC(t *_type, addr unsafe.Pointer, callerpc, pc uintptr) {
-}
-func raceWriteObjectPC(t *_type, addr unsafe.Pointer, callerpc, pc uintptr) {
-}
+// Because raceenabled is false, none of these functions should be called.
+
+func raceReadObjectPC(t *_type, addr unsafe.Pointer, callerpc, pc uintptr)  { gothrow("race") }
+func raceWriteObjectPC(t *_type, addr unsafe.Pointer, callerpc, pc uintptr) { gothrow("race") }
+func raceinit()                                                             { gothrow("race") }
+func racefini()                                                             { gothrow("race") }
+func racemapshadow(addr unsafe.Pointer, size uintptr)                       { gothrow("race") }
+func racewritepc(addr unsafe.Pointer, callerpc, pc uintptr)                 { gothrow("race") }
+func racereadpc(addr unsafe.Pointer, callerpc, pc uintptr)                  { gothrow("race") }
+func racereadrangepc(addr unsafe.Pointer, sz, callerpc, pc uintptr)         { gothrow("race") }
+func racewriterangepc(addr unsafe.Pointer, sz, callerpc, pc uintptr)        { gothrow("race") }
+func raceacquire(addr unsafe.Pointer)                                       { gothrow("race") }
+func raceacquireg(gp *g, addr unsafe.Pointer)                               { gothrow("race") }
+func racerelease(addr unsafe.Pointer)                                       { gothrow("race") }
+func racereleaseg(gp *g, addr unsafe.Pointer)                               { gothrow("race") }
+func racereleasemerge(addr unsafe.Pointer)                                  { gothrow("race") }
+func racereleasemergeg(gp *g, addr unsafe.Pointer)                          { gothrow("race") }
+func racefingo()                                                            { gothrow("race") }
+func racemalloc(p unsafe.Pointer, sz uintptr)                               { gothrow("race") }
+func racegostart(pc uintptr) uintptr                                        { gothrow("race"); return 0 }
+func racegoend()                                                            { gothrow("race") }
diff --git a/src/pkg/runtime/slice.go b/src/pkg/runtime/slice.go
index 68a225a50..3b88927c6 100644
--- a/src/pkg/runtime/slice.go
+++ b/src/pkg/runtime/slice.go
@@ -48,7 +48,7 @@ func growslice(t *slicetype, old sliceStruct, n int64) sliceStruct {
 
 	if raceenabled {
 		callerpc := getcallerpc(unsafe.Pointer(&t))
-		racereadrangepc(old.array, old.len*int(t.elem.size), callerpc, funcPC(growslice))
+		racereadrangepc(old.array, uintptr(old.len*int(t.elem.size)), callerpc, funcPC(growslice))
 	}
 
 	et := t.elem
@@ -104,8 +104,8 @@ func slicecopy(to sliceStruct, fm sliceStruct, width uintptr) int {
 	if raceenabled {
 		callerpc := getcallerpc(unsafe.Pointer(&to))
 		pc := funcPC(slicecopy)
-		racewriterangepc(to.array, n*int(width), callerpc, pc)
-		racereadrangepc(fm.array, n*int(width), callerpc, pc)
+		racewriterangepc(to.array, uintptr(n*int(width)), callerpc, pc)
+		racereadrangepc(fm.array, uintptr(n*int(width)), callerpc, pc)
 	}
 
 	size := uintptr(n) * width
@@ -131,7 +131,7 @@ func slicestringcopy(to []byte, fm string) int {
 	if raceenabled {
 		callerpc := getcallerpc(unsafe.Pointer(&to))
 		pc := funcPC(slicestringcopy)
-		racewriterangepc(unsafe.Pointer(&to[0]), n, callerpc, pc)
+		racewriterangepc(unsafe.Pointer(&to[0]), uintptr(n), callerpc, pc)
 	}
 
 	memmove(unsafe.Pointer(&to[0]), unsafe.Pointer((*stringStruct)(unsafe.Pointer(&fm)).str), uintptr(n))
diff --git a/src/pkg/runtime/string.go b/src/pkg/runtime/string.go
index 91f33db11..da3160449 100644
--- a/src/pkg/runtime/string.go
+++ b/src/pkg/runtime/string.go
@@ -62,7 +62,7 @@ func concatstring5(a [5]string) string {
 func slicebytetostring(b []byte) string {
 	if raceenabled && len(b) > 0 {
 		racereadrangepc(unsafe.Pointer(&b[0]),
-			len(b),
+			uintptr(len(b)),
 			getcallerpc(unsafe.Pointer(&b)),
 			funcPC(slicebytetostring))
 	}
@@ -82,7 +82,7 @@ func slicebytetostringtmp(b []byte) string {
 
 	if raceenabled && len(b) > 0 {
 		racereadrangepc(unsafe.Pointer(&b[0]),
-			len(b),
+			uintptr(len(b)),
 			getcallerpc(unsafe.Pointer(&b)),
 			funcPC(slicebytetostringtmp))
 	}
@@ -119,7 +119,7 @@ func stringtoslicerune(s string) []rune {
 func slicerunetostring(a []rune) string {
 	if raceenabled && len(a) > 0 {
 		racereadrangepc(unsafe.Pointer(&a[0]),
-			len(a)*int(unsafe.Sizeof(a[0])),
+			uintptr(len(a))*unsafe.Sizeof(a[0]),
 			getcallerpc(unsafe.Pointer(&a)),
 			funcPC(slicerunetostring))
 	}
diff --git a/src/pkg/runtime/stubs.go b/src/pkg/runtime/stubs.go
index 558963376..7d59b5c19 100644
--- a/src/pkg/runtime/stubs.go
+++ b/src/pkg/runtime/stubs.go
@@ -14,35 +14,6 @@ import "unsafe"
 const ptrSize = 4 << (^uintptr(0) >> 63) // unsafe.Sizeof(uintptr(0)) but an ideal const
 const regSize = 4 << (^uintreg(0) >> 63) // unsafe.Sizeof(uintreg(0)) but an ideal const
 
-//go:noescape
-func racereadpc(addr unsafe.Pointer, callpc, pc uintptr)
-
-//go:noescape
-func racewritepc(addr unsafe.Pointer, callpc, pc uintptr)
-
-//go:noescape
-func racereadrangepc(addr unsafe.Pointer, len int, callpc, pc uintptr)
-
-//go:noescape
-func racewriterangepc(addr unsafe.Pointer, len int, callpc, pc uintptr)
-
-//go:noescape
-func raceacquire(addr unsafe.Pointer)
-
-//go:noescape
-func racerelease(addr unsafe.Pointer)
-
-//go:noescape
-func racereleasemerge(addr unsafe.Pointer)
-
-//go:noescape
-func raceacquireg(gp *g, addr unsafe.Pointer)
-
-//go:noescape
-func racereleaseg(gp *g, addr unsafe.Pointer)
-
-func racefingo()
-
 // Should be a built-in for unsafe.Pointer?
 //go:nosplit
 func add(p unsafe.Pointer, x uintptr) unsafe.Pointer {
@@ -135,8 +106,6 @@ func goexit_m()
 //go:noescape
 func memclr(ptr unsafe.Pointer, n uintptr)
 
-func racemalloc(p unsafe.Pointer, size uintptr)
-
 // memmove copies n bytes from "from" to "to".
 // in memmove_*.s
 //go:noescape