runtime/race: better handling of atomic operations

This change fixes the last known false negative of the race detector --
detection of races between mutating atomic operations and non-atomic operations.
Race runtime already has functions for precise modelling of various atomic operations,
so this change just forwards all atomic ops to race runtime
instead of poor man modeling in sync/atomic package.
Performance is also improved -- full sync/atomic tests run in 60s instead of 85s now.

LGTM=khr
R=golang-codereviews, khr
CC=golang-codereviews, rsc
https://codereview.appspot.com/111310044

11 files changed, 166 insertions, 300 deletions

diff --git a/src/pkg/runtime/race.c b/src/pkg/runtime/race.c
index 07623be1a..a773addb0 100644
--- a/src/pkg/runtime/race.c
+++ b/src/pkg/runtime/race.c
@@ -24,6 +24,8 @@ void __tsan_malloc(void);
 void __tsan_acquire(void);
 void __tsan_release(void);
 void __tsan_release_merge(void);
+void __tsan_go_ignore_sync_begin(void);
+void __tsan_go_ignore_sync_end(void);
 
 // Mimic what cmd/cgo would do.
 #pragma cgo_import_static __tsan_init
@@ -36,6 +38,8 @@ void __tsan_release_merge(void);
 #pragma cgo_import_static __tsan_acquire
 #pragma cgo_import_static __tsan_release
 #pragma cgo_import_static __tsan_release_merge
+#pragma cgo_import_static __tsan_go_ignore_sync_begin
+#pragma cgo_import_static __tsan_go_ignore_sync_end
 
 // These are called from race_amd64.s.
 #pragma cgo_import_static __tsan_read
@@ -47,6 +51,17 @@ void __tsan_release_merge(void);
 #pragma cgo_import_static __tsan_func_enter
 #pragma cgo_import_static __tsan_func_exit
 
+#pragma cgo_import_static __tsan_go_atomic32_load
+#pragma cgo_import_static __tsan_go_atomic64_load
+#pragma cgo_import_static __tsan_go_atomic32_store
+#pragma cgo_import_static __tsan_go_atomic64_store
+#pragma cgo_import_static __tsan_go_atomic32_exchange
+#pragma cgo_import_static __tsan_go_atomic64_exchange
+#pragma cgo_import_static __tsan_go_atomic32_fetch_add
+#pragma cgo_import_static __tsan_go_atomic64_fetch_add
+#pragma cgo_import_static __tsan_go_atomic32_compare_exchange
+#pragma cgo_import_static __tsan_go_atomic64_compare_exchange
+
 extern byte runtime·noptrdata[];
 extern byte runtime·enoptrbss[];
   
@@ -250,32 +265,20 @@ runtime·RaceReleaseMerge(void *addr)
 	runtime·racereleasemerge(addr);
 }
 
-// func RaceSemacquire(s *uint32)
-void
-runtime·RaceSemacquire(uint32 *s)
-{
-	runtime·semacquire(s, false);
-}
-
-// func RaceSemrelease(s *uint32)
-void
-runtime·RaceSemrelease(uint32 *s)
-{
-	runtime·semrelease(s);
-}
-
 // func RaceDisable()
 void
 runtime·RaceDisable(void)
 {
-	g->raceignore++;
+	if(g->raceignore++ == 0)
+		runtime·racecall(__tsan_go_ignore_sync_begin, g->racectx);
 }
 
 // func RaceEnable()
 void
 runtime·RaceEnable(void)
 {
-	g->raceignore--;
+	if(--g->raceignore == 0)
+		runtime·racecall(__tsan_go_ignore_sync_end, g->racectx);
 }
 
 typedef struct SymbolizeContext SymbolizeContext;
diff --git a/src/pkg/runtime/race/README b/src/pkg/runtime/race/README
index 6a4259141..7f185359f 100644
--- a/src/pkg/runtime/race/README
+++ b/src/pkg/runtime/race/README
@@ -9,4 +9,4 @@ $ ./buildgo.sh
 
 Tested with gcc 4.6.1 and 4.7.0.  On Windows it's built with 64-bit MinGW.
 
-Current runtime is built on rev 210365.
+Current runtime is built on rev 215000.
diff --git a/src/pkg/runtime/race/race_darwin_amd64.syso b/src/pkg/runtime/race/race_darwin_amd64.syso
index 9061ce0aa..81b48c6c9 100644
--- a/src/pkg/runtime/race/race_darwin_amd64.syso
+++ b/src/pkg/runtime/race/race_darwin_amd64.syso
diff --git a/src/pkg/runtime/race/race_freebsd_amd64.syso b/src/pkg/runtime/race/race_freebsd_amd64.syso
index b25d868f4..5bbe32229 100644
--- a/src/pkg/runtime/race/race_freebsd_amd64.syso
+++ b/src/pkg/runtime/race/race_freebsd_amd64.syso
diff --git a/src/pkg/runtime/race/race_linux_amd64.syso b/src/pkg/runtime/race/race_linux_amd64.syso
index 8fd77ae58..49bf08ef3 100644
--- a/src/pkg/runtime/race/race_linux_amd64.syso
+++ b/src/pkg/runtime/race/race_linux_amd64.syso
diff --git a/src/pkg/runtime/race/race_windows_amd64.syso b/src/pkg/runtime/race/race_windows_amd64.syso
index 3ea80a665..a4eae9bdd 100644
--- a/src/pkg/runtime/race/race_windows_amd64.syso
+++ b/src/pkg/runtime/race/race_windows_amd64.syso
diff --git a/src/pkg/runtime/race/testdata/atomic_test.go b/src/pkg/runtime/race/testdata/atomic_test.go
index fc569b96c..232744b3d 100644
--- a/src/pkg/runtime/race/testdata/atomic_test.go
+++ b/src/pkg/runtime/race/testdata/atomic_test.go
@@ -225,8 +225,7 @@ func TestNoRaceAtomicStoreCASUint64(t *testing.T) {
 	x = 1
 }
 
-// Races with non-atomic loads are not detected.
-func TestRaceFailingAtomicStoreLoad(t *testing.T) {
+func TestRaceAtomicStoreLoad(t *testing.T) {
 	c := make(chan bool)
 	var a uint64
 	go func() {
@@ -248,8 +247,7 @@ func TestRaceAtomicLoadStore(t *testing.T) {
 	<-c
 }
 
-// Races with non-atomic loads are not detected.
-func TestRaceFailingAtomicAddLoad(t *testing.T) {
+func TestRaceAtomicAddLoad(t *testing.T) {
 	c := make(chan bool)
 	var a uint64
 	go func() {
diff --git a/src/pkg/runtime/race_amd64.s b/src/pkg/runtime/race_amd64.s
index f4e72386d..a4a9dfeb5 100644
--- a/src/pkg/runtime/race_amd64.s
+++ b/src/pkg/runtime/race_amd64.s
@@ -79,7 +79,7 @@ TEXT	runtime·RaceWrite(SB), NOSPLIT, $0-8
 TEXT	runtime·racewritepc(SB), NOSPLIT, $0-24
 	MOVQ	addr+0(FP), RARG1
 	MOVQ	callpc+8(FP), RARG2
-	MOVQ	cp+16(FP), RARG3
+	MOVQ	pc+16(FP), RARG3
 	// void __tsan_write_pc(ThreadState *thr, void *addr, void *callpc, void *pc);
 	MOVQ	$__tsan_write_pc(SB), AX
 	JMP	racecalladdr<>(SB)
@@ -180,6 +180,141 @@ TEXT	runtime·racefuncexit(SB), NOSPLIT, $0-0
 	MOVQ	$__tsan_func_exit(SB), AX
 	JMP	racecall<>(SB)
 
+// Atomic operations for sync/atomic package.
+
+// Load
+TEXT	sync∕atomic·LoadInt32(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic32_load(SB), AX
+	CALL	racecallatomic<>(SB)
+	RET
+
+TEXT	sync∕atomic·LoadInt64(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic64_load(SB), AX
+	CALL	racecallatomic<>(SB)
+	RET
+
+TEXT	sync∕atomic·LoadUint32(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·LoadInt32(SB)
+
+TEXT	sync∕atomic·LoadUint64(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·LoadInt64(SB)
+
+TEXT	sync∕atomic·LoadUintptr(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·LoadInt64(SB)
+
+TEXT	sync∕atomic·LoadPointer(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·LoadInt64(SB)
+
+// Store
+TEXT	sync∕atomic·StoreInt32(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic32_store(SB), AX
+	CALL	racecallatomic<>(SB)
+	RET
+
+TEXT	sync∕atomic·StoreInt64(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic64_store(SB), AX
+	CALL	racecallatomic<>(SB)
+	RET
+
+TEXT	sync∕atomic·StoreUint32(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·StoreInt32(SB)
+
+TEXT	sync∕atomic·StoreUint64(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·StoreInt64(SB)
+
+TEXT	sync∕atomic·StoreUintptr(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·StoreInt64(SB)
+
+TEXT	sync∕atomic·StorePointer(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·StoreInt64(SB)
+
+// Swap
+TEXT	sync∕atomic·SwapInt32(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic32_exchange(SB), AX
+	CALL	racecallatomic<>(SB)
+	RET
+
+TEXT	sync∕atomic·SwapInt64(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic64_exchange(SB), AX
+	CALL	racecallatomic<>(SB)
+	RET
+
+TEXT	sync∕atomic·SwapUint32(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·SwapInt32(SB)
+
+TEXT	sync∕atomic·SwapUint64(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·SwapInt64(SB)
+
+TEXT	sync∕atomic·SwapUintptr(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·SwapInt64(SB)
+
+TEXT	sync∕atomic·SwapPointer(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·SwapInt64(SB)
+
+// Add
+TEXT	sync∕atomic·AddInt32(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic32_fetch_add(SB), AX
+	CALL	racecallatomic<>(SB)
+	MOVL	add+8(FP), AX	// convert fetch_add to add_fetch
+	ADDL	AX, ret+16(FP)
+	RET
+
+TEXT	sync∕atomic·AddInt64(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic64_fetch_add(SB), AX
+	CALL	racecallatomic<>(SB)
+	MOVQ	add+8(FP), AX	// convert fetch_add to add_fetch
+	ADDQ	AX, ret+16(FP)
+	RET
+
+TEXT	sync∕atomic·AddUint32(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·AddInt32(SB)
+
+TEXT	sync∕atomic·AddUint64(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·AddInt64(SB)
+
+TEXT	sync∕atomic·AddUintptr(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·AddInt64(SB)
+
+TEXT	sync∕atomic·AddPointer(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·AddInt64(SB)
+
+// CompareAndSwap
+TEXT	sync∕atomic·CompareAndSwapInt32(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic32_compare_exchange(SB), AX
+	CALL	racecallatomic<>(SB)
+	RET
+
+TEXT	sync∕atomic·CompareAndSwapInt64(SB), NOSPLIT, $0-0
+	MOVQ	$__tsan_go_atomic64_compare_exchange(SB), AX
+	CALL	racecallatomic<>(SB)
+	RET
+
+TEXT	sync∕atomic·CompareAndSwapUint32(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·CompareAndSwapInt32(SB)
+
+TEXT	sync∕atomic·CompareAndSwapUint64(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·CompareAndSwapInt64(SB)
+
+TEXT	sync∕atomic·CompareAndSwapUintptr(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·CompareAndSwapInt64(SB)
+
+TEXT	sync∕atomic·CompareAndSwapPointer(SB), NOSPLIT, $0-0
+	JMP	sync∕atomic·CompareAndSwapInt64(SB)
+
+// Generic atomic operation implementation.
+// AX already contains target function.
+TEXT	racecallatomic<>(SB), NOSPLIT, $0-0
+	// Trigger SIGSEGV early.
+	MOVQ	16(SP), R12
+	MOVL	(R12), R12
+	get_tls(R12)
+	MOVQ	g(R12), R14
+	MOVQ	g_racectx(R14), RARG0	// goroutine context
+	MOVQ	8(SP), RARG1	// caller pc
+	MOVQ	(SP), RARG2	// pc
+	LEAQ	16(SP), RARG3	// arguments
+	JMP	racecall<>(SB)
+
 // void runtime·racecall(void(*f)(...), ...)
 // Calls C function f from race runtime and passes up to 4 arguments to it.
 // The arguments are never heap-object-preserving pointers, so we pretend there are no arguments.
diff --git a/src/pkg/sync/atomic/doc.go b/src/pkg/sync/atomic/doc.go
index 17ba72fa1..10fb8c917 100644
--- a/src/pkg/sync/atomic/doc.go
+++ b/src/pkg/sync/atomic/doc.go
@@ -2,8 +2,6 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build !race
-
 // Package atomic provides low-level atomic memory primitives
 // useful for implementing synchronization algorithms.
 //
diff --git a/src/pkg/sync/atomic/race.go b/src/pkg/sync/atomic/race.go
deleted file mode 100644
index 6cbbf12cb..000000000
--- a/src/pkg/sync/atomic/race.go
+++ /dev/null
@@ -1,276 +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.
-
-// +build race
-
-package atomic
-
-import (
-	"runtime"
-	"unsafe"
-)
-
-// We use runtime.RaceRead() inside of atomic operations to catch races
-// between atomic and non-atomic operations.  It will also catch races
-// between Mutex.Lock() and mutex overwrite (mu = Mutex{}).  Since we use
-// only RaceRead() we won't catch races with non-atomic loads.
-// Otherwise (if we use RaceWrite()) we will report races
-// between atomic operations (false positives).
-
-var mtx uint32 = 1 // same for all
-
-func SwapInt32(addr *int32, new int32) (old int32) {
-	return int32(SwapUint32((*uint32)(unsafe.Pointer(addr)), uint32(new)))
-}
-
-func SwapUint32(addr *uint32, new uint32) (old uint32) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	runtime.RaceAcquire(unsafe.Pointer(addr))
-	old = *addr
-	*addr = new
-	runtime.RaceReleaseMerge(unsafe.Pointer(addr))
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func SwapInt64(addr *int64, new int64) (old int64) {
-	return int64(SwapUint64((*uint64)(unsafe.Pointer(addr)), uint64(new)))
-}
-
-func SwapUint64(addr *uint64, new uint64) (old uint64) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	runtime.RaceAcquire(unsafe.Pointer(addr))
-	old = *addr
-	*addr = new
-	runtime.RaceReleaseMerge(unsafe.Pointer(addr))
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func SwapUintptr(addr *uintptr, new uintptr) (old uintptr) {
-	return uintptr(SwapPointer((*unsafe.Pointer)(unsafe.Pointer(addr)), unsafe.Pointer(new)))
-}
-
-func SwapPointer(addr *unsafe.Pointer, new unsafe.Pointer) (old unsafe.Pointer) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	runtime.RaceAcquire(unsafe.Pointer(addr))
-	old = *addr
-	*addr = new
-	runtime.RaceReleaseMerge(unsafe.Pointer(addr))
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func CompareAndSwapInt32(val *int32, old, new int32) bool {
-	return CompareAndSwapUint32((*uint32)(unsafe.Pointer(val)), uint32(old), uint32(new))
-}
-
-func CompareAndSwapUint32(val *uint32, old, new uint32) (swapped bool) {
-	_ = *val
-	swapped = false
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(val))
-	runtime.RaceAcquire(unsafe.Pointer(val))
-	if *val == old {
-		*val = new
-		swapped = true
-		runtime.RaceReleaseMerge(unsafe.Pointer(val))
-	}
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func CompareAndSwapInt64(val *int64, old, new int64) bool {
-	return CompareAndSwapUint64((*uint64)(unsafe.Pointer(val)), uint64(old), uint64(new))
-}
-
-func CompareAndSwapUint64(val *uint64, old, new uint64) (swapped bool) {
-	_ = *val
-	swapped = false
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(val))
-	runtime.RaceAcquire(unsafe.Pointer(val))
-	if *val == old {
-		*val = new
-		swapped = true
-		runtime.RaceReleaseMerge(unsafe.Pointer(val))
-	}
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func CompareAndSwapPointer(val *unsafe.Pointer, old, new unsafe.Pointer) (swapped bool) {
-	_ = *val
-	swapped = false
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(val))
-	runtime.RaceAcquire(unsafe.Pointer(val))
-	if *val == old {
-		*val = new
-		swapped = true
-		runtime.RaceReleaseMerge(unsafe.Pointer(val))
-	}
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func CompareAndSwapUintptr(val *uintptr, old, new uintptr) (swapped bool) {
-	_ = *val
-	swapped = false
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(val))
-	runtime.RaceAcquire(unsafe.Pointer(val))
-	if *val == old {
-		*val = new
-		swapped = true
-		runtime.RaceReleaseMerge(unsafe.Pointer(val))
-	}
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func AddInt32(val *int32, delta int32) int32 {
-	return int32(AddUint32((*uint32)(unsafe.Pointer(val)), uint32(delta)))
-}
-
-func AddUint32(val *uint32, delta uint32) (new uint32) {
-	_ = *val
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(val))
-	runtime.RaceAcquire(unsafe.Pointer(val))
-	*val = *val + delta
-	new = *val
-	runtime.RaceReleaseMerge(unsafe.Pointer(val))
-	runtime.RaceSemrelease(&mtx)
-
-	return
-}
-
-func AddInt64(val *int64, delta int64) int64 {
-	return int64(AddUint64((*uint64)(unsafe.Pointer(val)), uint64(delta)))
-}
-
-func AddUint64(val *uint64, delta uint64) (new uint64) {
-	_ = *val
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(val))
-	runtime.RaceAcquire(unsafe.Pointer(val))
-	*val = *val + delta
-	new = *val
-	runtime.RaceReleaseMerge(unsafe.Pointer(val))
-	runtime.RaceSemrelease(&mtx)
-
-	return
-}
-
-func AddUintptr(val *uintptr, delta uintptr) (new uintptr) {
-	_ = *val
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(val))
-	runtime.RaceAcquire(unsafe.Pointer(val))
-	*val = *val + delta
-	new = *val
-	runtime.RaceReleaseMerge(unsafe.Pointer(val))
-	runtime.RaceSemrelease(&mtx)
-
-	return
-}
-
-func LoadInt32(addr *int32) int32 {
-	return int32(LoadUint32((*uint32)(unsafe.Pointer(addr))))
-}
-
-func LoadUint32(addr *uint32) (val uint32) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	runtime.RaceAcquire(unsafe.Pointer(addr))
-	val = *addr
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func LoadInt64(addr *int64) int64 {
-	return int64(LoadUint64((*uint64)(unsafe.Pointer(addr))))
-}
-
-func LoadUint64(addr *uint64) (val uint64) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	runtime.RaceAcquire(unsafe.Pointer(addr))
-	val = *addr
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func LoadPointer(addr *unsafe.Pointer) (val unsafe.Pointer) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	runtime.RaceAcquire(unsafe.Pointer(addr))
-	val = *addr
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func LoadUintptr(addr *uintptr) (val uintptr) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	runtime.RaceAcquire(unsafe.Pointer(addr))
-	val = *addr
-	runtime.RaceSemrelease(&mtx)
-	return
-}
-
-func StoreInt32(addr *int32, val int32) {
-	StoreUint32((*uint32)(unsafe.Pointer(addr)), uint32(val))
-}
-
-func StoreUint32(addr *uint32, val uint32) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	*addr = val
-	runtime.RaceRelease(unsafe.Pointer(addr))
-	runtime.RaceSemrelease(&mtx)
-}
-
-func StoreInt64(addr *int64, val int64) {
-	StoreUint64((*uint64)(unsafe.Pointer(addr)), uint64(val))
-}
-
-func StoreUint64(addr *uint64, val uint64) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	*addr = val
-	runtime.RaceRelease(unsafe.Pointer(addr))
-	runtime.RaceSemrelease(&mtx)
-}
-
-func StorePointer(addr *unsafe.Pointer, val unsafe.Pointer) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	*addr = val
-	runtime.RaceRelease(unsafe.Pointer(addr))
-	runtime.RaceSemrelease(&mtx)
-}
-
-func StoreUintptr(addr *uintptr, val uintptr) {
-	_ = *addr
-	runtime.RaceSemacquire(&mtx)
-	runtime.RaceRead(unsafe.Pointer(addr))
-	*addr = val
-	runtime.RaceRelease(unsafe.Pointer(addr))
-	runtime.RaceSemrelease(&mtx)
-}
diff --git a/src/pkg/sync/atomic/race.s b/src/pkg/sync/atomic/race.s
new file mode 100644
index 000000000..4dadc9ed7
--- /dev/null
+++ b/src/pkg/sync/atomic/race.s
@@ -0,0 +1,8 @@
+// Copyright 2014 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.
+
+// +build race
+
+// This file is here only to allow external functions.
+// The operations are implemented in src/pkg/runtime/race_amd64.s