Imported Upstream version 1upstream/1

19 files changed, 1006 insertions, 125 deletions

diff --git a/src/pkg/sync/Makefile b/src/pkg/sync/Makefile
deleted file mode 100644
index e8a766226..000000000
--- a/src/pkg/sync/Makefile
+++ /dev/null
@@ -1,15 +0,0 @@
-# Copyright 2009 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.
-
-include ../../Make.inc
-
-TARG=sync
-GOFILES=\
-	cond.go\
-	mutex.go\
-	once.go \
-	rwmutex.go\
-	waitgroup.go\
-
-include ../../Make.pkg
diff --git a/src/pkg/sync/atomic/Makefile b/src/pkg/sync/atomic/Makefile
deleted file mode 100644
index 38d8998c0..000000000
--- a/src/pkg/sync/atomic/Makefile
+++ /dev/null
@@ -1,18 +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.
-
-include ../../../Make.inc
-
-TARG=sync/atomic
-GOFILES=\
-	doc.go\
-
-OFILES=\
-	asm_$(GOARCH).$O\
-
-ifeq ($(GOARCH),arm)
-OFILES+=asm_$(GOOS)_$(GOARCH).$O
-endif
-
-include ../../../Make.pkg
diff --git a/src/pkg/sync/atomic/asm_386.s b/src/pkg/sync/atomic/asm_386.s
index 914d2feeb..a406852f4 100644
--- a/src/pkg/sync/atomic/asm_386.s
+++ b/src/pkg/sync/atomic/asm_386.s
@@ -18,6 +18,9 @@ TEXT ·CompareAndSwapUint32(SB),7,$0
 TEXT ·CompareAndSwapUintptr(SB),7,$0
 	JMP	·CompareAndSwapUint32(SB)
 
+TEXT ·CompareAndSwapPointer(SB),7,$0
+	JMP	·CompareAndSwapUint32(SB)
+
 TEXT ·CompareAndSwapInt64(SB),7,$0
 	JMP	·CompareAndSwapUint64(SB)
 
@@ -94,3 +97,55 @@ TEXT ·LoadUint32(SB),7,$0
 	MOVL	0(AX), AX
 	MOVL	AX, ret+4(FP)
 	RET
+
+TEXT ·LoadInt64(SB),7,$0
+	JMP	·LoadUint64(SB)
+
+TEXT ·LoadUint64(SB),7,$0
+	MOVL	addrptr+0(FP), AX
+	// MOVQ and EMMS were introduced on the Pentium MMX.
+	// MOVQ (%EAX), %MM0
+	BYTE $0x0f; BYTE $0x6f; BYTE $0x00
+	// MOVQ %MM0, 0x8(%ESP)
+	BYTE $0x0f; BYTE $0x7f; BYTE $0x44; BYTE $0x24; BYTE $0x08
+	EMMS
+	RET
+
+TEXT ·LoadUintptr(SB),7,$0
+	JMP	·LoadUint32(SB)
+
+TEXT ·LoadPointer(SB),7,$0
+	JMP	·LoadUint32(SB)
+
+TEXT ·StoreInt32(SB),7,$0
+	JMP	·StoreUint32(SB)
+
+TEXT ·StoreUint32(SB),7,$0
+	MOVL	addrptr+0(FP), BP
+	MOVL	val+4(FP), AX
+	XCHGL	AX, 0(BP)
+	RET
+
+TEXT ·StoreInt64(SB),7,$0
+	JMP	·StoreUint64(SB)
+
+TEXT ·StoreUint64(SB),7,$0
+	MOVL	addrptr+0(FP), AX
+	// MOVQ and EMMS were introduced on the Pentium MMX.
+	// MOVQ 0x8(%ESP), %MM0
+	BYTE $0x0f; BYTE $0x6f; BYTE $0x44; BYTE $0x24; BYTE $0x08
+	// MOVQ %MM0, (%EAX)
+	BYTE $0x0f; BYTE $0x7f; BYTE $0x00 
+	EMMS
+	// This is essentially a no-op, but it provides required memory fencing.
+	// It can be replaced with MFENCE, but MFENCE was introduced only on the Pentium4 (SSE2).
+	XORL	AX, AX
+	LOCK
+	XADDL	AX, (SP)
+	RET
+
+TEXT ·StoreUintptr(SB),7,$0
+	JMP	·StoreUint32(SB)
+
+TEXT ·StorePointer(SB),7,$0
+	JMP	·StoreUint32(SB)
diff --git a/src/pkg/sync/atomic/asm_amd64.s b/src/pkg/sync/atomic/asm_amd64.s
index 428295063..6f8bde068 100644
--- a/src/pkg/sync/atomic/asm_amd64.s
+++ b/src/pkg/sync/atomic/asm_amd64.s
@@ -17,6 +17,9 @@ TEXT ·CompareAndSwapUint32(SB),7,$0
 TEXT ·CompareAndSwapUintptr(SB),7,$0
 	JMP	·CompareAndSwapUint64(SB)
 
+TEXT ·CompareAndSwapPointer(SB),7,$0
+	JMP	·CompareAndSwapUint64(SB)
+
 TEXT ·CompareAndSwapInt64(SB),7,$0
 	JMP	·CompareAndSwapUint64(SB)
 
@@ -67,3 +70,47 @@ TEXT ·LoadUint32(SB),7,$0
 	MOVL	AX, ret+8(FP)
 	RET
 
+TEXT ·LoadInt64(SB),7,$0
+	JMP	·LoadUint64(SB)
+
+TEXT ·LoadUint64(SB),7,$0
+	MOVQ	addrptr+0(FP), AX
+	MOVQ	0(AX), AX
+	MOVQ	AX, ret+8(FP)
+	RET
+
+TEXT ·LoadUintptr(SB),7,$0
+	JMP	·LoadPointer(SB)
+
+TEXT ·LoadPointer(SB),7,$0
+	MOVQ	addrptr+0(FP), AX
+	MOVQ	0(AX), AX
+	MOVQ	AX, ret+8(FP)
+	RET
+
+TEXT ·StoreInt32(SB),7,$0
+	JMP	·StoreUint32(SB)
+
+TEXT ·StoreUint32(SB),7,$0
+	MOVQ	addrptr+0(FP), BP
+	MOVL	val+8(FP), AX
+	XCHGL	AX, 0(BP)
+	RET
+
+TEXT ·StoreInt64(SB),7,$0
+	JMP	·StoreUint64(SB)
+
+TEXT ·StoreUint64(SB),7,$0
+	MOVQ	addrptr+0(FP), BP
+	MOVQ	val+8(FP), AX
+	XCHGQ	AX, 0(BP)
+	RET
+
+TEXT ·StoreUintptr(SB),7,$0
+	JMP	·StorePointer(SB)
+
+TEXT ·StorePointer(SB),7,$0
+	MOVQ	addrptr+0(FP), BP
+	MOVQ	val+8(FP), AX
+	XCHGQ	AX, 0(BP)
+	RET
diff --git a/src/pkg/sync/atomic/asm_arm.s b/src/pkg/sync/atomic/asm_arm.s
index 95e2f5be4..2d10a922b 100644
--- a/src/pkg/sync/atomic/asm_arm.s
+++ b/src/pkg/sync/atomic/asm_arm.s
@@ -79,6 +79,30 @@ add64loop:
 	MOVW	R5, rethi+16(FP)
 	RET
 
+TEXT ·armLoadUint64(SB),7,$0
+	BL	fastCheck64<>(SB)
+	MOVW	addrptr+0(FP), R1
+load64loop:
+	LDREXD	(R1), R2	// loads R2 and R3
+	STREXD	R2, (R1), R0	// stores R2 and R3
+	CMP	$0, R0
+	BNE	load64loop
+	MOVW	R2, vallo+4(FP)
+	MOVW	R3, valhi+8(FP)
+	RET
+
+TEXT ·armStoreUint64(SB),7,$0
+	BL	fastCheck64<>(SB)
+	MOVW	addrptr+0(FP), R1
+	MOVW	vallo+4(FP), R2
+	MOVW	valhi+8(FP), R3
+store64loop:
+	LDREXD	(R1), R4	// loads R4 and R5
+	STREXD	R2, (R1), R0	// stores R2 and R3
+	CMP	$0, R0
+	BNE	store64loop
+	RET
+
 // Check for broken 64-bit LDREXD as found in QEMU.
 // LDREXD followed by immediate STREXD should succeed.
 // If it fails, try a few times just to be sure (maybe our thread got
diff --git a/src/pkg/sync/atomic/asm_linux_arm.s b/src/pkg/sync/atomic/asm_linux_arm.s
index 9ac411944..25dc85804 100644
--- a/src/pkg/sync/atomic/asm_linux_arm.s
+++ b/src/pkg/sync/atomic/asm_linux_arm.s
@@ -50,6 +50,9 @@ cascheck:
 TEXT ·CompareAndSwapUintptr(SB),7,$0
 	B	·CompareAndSwapUint32(SB)
 
+TEXT ·CompareAndSwapPointer(SB),7,$0
+	B	·CompareAndSwapUint32(SB)
+
 TEXT ·AddInt32(SB),7,$0
 	B	·AddUint32(SB)
 
@@ -96,3 +99,39 @@ loadloop1:
 	BCC	loadloop1
 	MOVW	R1, val+4(FP)
 	RET
+
+TEXT ·LoadInt64(SB),7,$0
+	B	·armLoadUint64(SB)
+
+TEXT ·LoadUint64(SB),7,$0
+	B	·armLoadUint64(SB)
+
+TEXT ·LoadUintptr(SB),7,$0
+	B	·LoadUint32(SB)
+
+TEXT ·LoadPointer(SB),7,$0
+	B	·LoadUint32(SB)
+
+TEXT ·StoreInt32(SB),7,$0
+	B	·StoreUint32(SB)
+
+TEXT ·StoreUint32(SB),7,$0
+	MOVW	addrptr+0(FP), R2
+	MOVW	val+4(FP), R1
+storeloop1:
+	MOVW	0(R2), R0
+	BL	cas<>(SB)
+	BCC	storeloop1
+	RET
+
+TEXT ·StoreInt64(SB),7,$0
+	B	·armStoreUint64(SB)
+
+TEXT ·StoreUint64(SB),7,$0
+	B	·armStoreUint64(SB)
+
+TEXT ·StoreUintptr(SB),7,$0
+	B	·StoreUint32(SB)
+
+TEXT ·StorePointer(SB),7,$0
+	B	·StoreUint32(SB)
diff --git a/src/pkg/sync/atomic/atomic_test.go b/src/pkg/sync/atomic/atomic_test.go
index 2229e58d0..f60d997ce 100644
--- a/src/pkg/sync/atomic/atomic_test.go
+++ b/src/pkg/sync/atomic/atomic_test.go
@@ -164,17 +164,17 @@ func TestCompareAndSwapInt32(t *testing.T) {
 	for val := int32(1); val+val > val; val += val {
 		x.i = val
 		if !CompareAndSwapInt32(&x.i, val, val+1) {
-			t.Errorf("should have swapped %#x %#x", val, val+1)
+			t.Fatalf("should have swapped %#x %#x", val, val+1)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 		x.i = val + 1
 		if CompareAndSwapInt32(&x.i, val, val+2) {
-			t.Errorf("should not have swapped %#x %#x", val, val+2)
+			t.Fatalf("should not have swapped %#x %#x", val, val+2)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 	}
 	if x.before != magic32 || x.after != magic32 {
@@ -193,17 +193,17 @@ func TestCompareAndSwapUint32(t *testing.T) {
 	for val := uint32(1); val+val > val; val += val {
 		x.i = val
 		if !CompareAndSwapUint32(&x.i, val, val+1) {
-			t.Errorf("should have swapped %#x %#x", val, val+1)
+			t.Fatalf("should have swapped %#x %#x", val, val+1)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 		x.i = val + 1
 		if CompareAndSwapUint32(&x.i, val, val+2) {
-			t.Errorf("should not have swapped %#x %#x", val, val+2)
+			t.Fatalf("should not have swapped %#x %#x", val, val+2)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 	}
 	if x.before != magic32 || x.after != magic32 {
@@ -226,17 +226,17 @@ func TestCompareAndSwapInt64(t *testing.T) {
 	for val := int64(1); val+val > val; val += val {
 		x.i = val
 		if !CompareAndSwapInt64(&x.i, val, val+1) {
-			t.Errorf("should have swapped %#x %#x", val, val+1)
+			t.Fatalf("should have swapped %#x %#x", val, val+1)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 		x.i = val + 1
 		if CompareAndSwapInt64(&x.i, val, val+2) {
-			t.Errorf("should not have swapped %#x %#x", val, val+2)
+			t.Fatalf("should not have swapped %#x %#x", val, val+2)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 	}
 	if x.before != magic64 || x.after != magic64 {
@@ -259,17 +259,17 @@ func TestCompareAndSwapUint64(t *testing.T) {
 	for val := uint64(1); val+val > val; val += val {
 		x.i = val
 		if !CompareAndSwapUint64(&x.i, val, val+1) {
-			t.Errorf("should have swapped %#x %#x", val, val+1)
+			t.Fatalf("should have swapped %#x %#x", val, val+1)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 		x.i = val + 1
 		if CompareAndSwapUint64(&x.i, val, val+2) {
-			t.Errorf("should not have swapped %#x %#x", val, val+2)
+			t.Fatalf("should not have swapped %#x %#x", val, val+2)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 	}
 	if x.before != magic64 || x.after != magic64 {
@@ -290,17 +290,48 @@ func TestCompareAndSwapUintptr(t *testing.T) {
 	for val := uintptr(1); val+val > val; val += val {
 		x.i = val
 		if !CompareAndSwapUintptr(&x.i, val, val+1) {
-			t.Errorf("should have swapped %#x %#x", val, val+1)
+			t.Fatalf("should have swapped %#x %#x", val, val+1)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 		x.i = val + 1
 		if CompareAndSwapUintptr(&x.i, val, val+2) {
-			t.Errorf("should not have swapped %#x %#x", val, val+2)
+			t.Fatalf("should not have swapped %#x %#x", val, val+2)
 		}
 		if x.i != val+1 {
-			t.Errorf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+		}
+	}
+	if x.before != magicptr || x.after != magicptr {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, magicptr, magicptr)
+	}
+}
+
+func TestCompareAndSwapPointer(t *testing.T) {
+	var x struct {
+		before uintptr
+		i      unsafe.Pointer
+		after  uintptr
+	}
+	var m uint64 = magic64
+	magicptr := uintptr(m)
+	x.before = magicptr
+	x.after = magicptr
+	for val := uintptr(1); val+val > val; val += val {
+		x.i = unsafe.Pointer(val)
+		if !CompareAndSwapPointer(&x.i, unsafe.Pointer(val), unsafe.Pointer(val+1)) {
+			t.Fatalf("should have swapped %#x %#x", val, val+1)
+		}
+		if x.i != unsafe.Pointer(val+1) {
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
+		}
+		x.i = unsafe.Pointer(val + 1)
+		if CompareAndSwapPointer(&x.i, unsafe.Pointer(val), unsafe.Pointer(val+2)) {
+			t.Fatalf("should not have swapped %#x %#x", val, val+2)
+		}
+		if x.i != unsafe.Pointer(val+1) {
+			t.Fatalf("wrong x.i after swap: x.i=%#x val+1=%#x", x.i, val+1)
 		}
 	}
 	if x.before != magicptr || x.after != magicptr {
@@ -348,6 +379,236 @@ func TestLoadUint32(t *testing.T) {
 	}
 }
 
+func TestLoadInt64(t *testing.T) {
+	if test64err != nil {
+		t.Logf("Skipping 64-bit tests: %v", test64err)
+		return
+	}
+	var x struct {
+		before int64
+		i      int64
+		after  int64
+	}
+	x.before = magic64
+	x.after = magic64
+	for delta := int64(1); delta+delta > delta; delta += delta {
+		k := LoadInt64(&x.i)
+		if k != x.i {
+			t.Fatalf("delta=%d i=%d k=%d", delta, x.i, k)
+		}
+		x.i += delta
+	}
+	if x.before != magic64 || x.after != magic64 {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, uint64(magic64), uint64(magic64))
+	}
+}
+
+func TestLoadUint64(t *testing.T) {
+	if test64err != nil {
+		t.Logf("Skipping 64-bit tests: %v", test64err)
+		return
+	}
+	var x struct {
+		before uint64
+		i      uint64
+		after  uint64
+	}
+	x.before = magic64
+	x.after = magic64
+	for delta := uint64(1); delta+delta > delta; delta += delta {
+		k := LoadUint64(&x.i)
+		if k != x.i {
+			t.Fatalf("delta=%d i=%d k=%d", delta, x.i, k)
+		}
+		x.i += delta
+	}
+	if x.before != magic64 || x.after != magic64 {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, uint64(magic64), uint64(magic64))
+	}
+}
+
+func TestLoadUintptr(t *testing.T) {
+	var x struct {
+		before uintptr
+		i      uintptr
+		after  uintptr
+	}
+	var m uint64 = magic64
+	magicptr := uintptr(m)
+	x.before = magicptr
+	x.after = magicptr
+	for delta := uintptr(1); delta+delta > delta; delta += delta {
+		k := LoadUintptr(&x.i)
+		if k != x.i {
+			t.Fatalf("delta=%d i=%d k=%d", delta, x.i, k)
+		}
+		x.i += delta
+	}
+	if x.before != magicptr || x.after != magicptr {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, magicptr, magicptr)
+	}
+}
+
+func TestLoadPointer(t *testing.T) {
+	var x struct {
+		before uintptr
+		i      unsafe.Pointer
+		after  uintptr
+	}
+	var m uint64 = magic64
+	magicptr := uintptr(m)
+	x.before = magicptr
+	x.after = magicptr
+	for delta := uintptr(1); delta+delta > delta; delta += delta {
+		k := LoadPointer(&x.i)
+		if k != x.i {
+			t.Fatalf("delta=%d i=%d k=%d", delta, x.i, k)
+		}
+		x.i = unsafe.Pointer(uintptr(x.i) + delta)
+	}
+	if x.before != magicptr || x.after != magicptr {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, magicptr, magicptr)
+	}
+}
+
+func TestStoreInt32(t *testing.T) {
+	var x struct {
+		before int32
+		i      int32
+		after  int32
+	}
+	x.before = magic32
+	x.after = magic32
+	v := int32(0)
+	for delta := int32(1); delta+delta > delta; delta += delta {
+		StoreInt32(&x.i, v)
+		if x.i != v {
+			t.Fatalf("delta=%d i=%d v=%d", delta, x.i, v)
+		}
+		v += delta
+	}
+	if x.before != magic32 || x.after != magic32 {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, magic32, magic32)
+	}
+}
+
+func TestStoreUint32(t *testing.T) {
+	var x struct {
+		before uint32
+		i      uint32
+		after  uint32
+	}
+	x.before = magic32
+	x.after = magic32
+	v := uint32(0)
+	for delta := uint32(1); delta+delta > delta; delta += delta {
+		StoreUint32(&x.i, v)
+		if x.i != v {
+			t.Fatalf("delta=%d i=%d v=%d", delta, x.i, v)
+		}
+		v += delta
+	}
+	if x.before != magic32 || x.after != magic32 {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, magic32, magic32)
+	}
+}
+
+func TestStoreInt64(t *testing.T) {
+	if test64err != nil {
+		t.Logf("Skipping 64-bit tests: %v", test64err)
+		return
+	}
+	var x struct {
+		before int64
+		i      int64
+		after  int64
+	}
+	x.before = magic64
+	x.after = magic64
+	v := int64(0)
+	for delta := int64(1); delta+delta > delta; delta += delta {
+		StoreInt64(&x.i, v)
+		if x.i != v {
+			t.Fatalf("delta=%d i=%d v=%d", delta, x.i, v)
+		}
+		v += delta
+	}
+	if x.before != magic64 || x.after != magic64 {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, uint64(magic64), uint64(magic64))
+	}
+}
+
+func TestStoreUint64(t *testing.T) {
+	if test64err != nil {
+		t.Logf("Skipping 64-bit tests: %v", test64err)
+		return
+	}
+	var x struct {
+		before uint64
+		i      uint64
+		after  uint64
+	}
+	x.before = magic64
+	x.after = magic64
+	v := uint64(0)
+	for delta := uint64(1); delta+delta > delta; delta += delta {
+		StoreUint64(&x.i, v)
+		if x.i != v {
+			t.Fatalf("delta=%d i=%d v=%d", delta, x.i, v)
+		}
+		v += delta
+	}
+	if x.before != magic64 || x.after != magic64 {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, uint64(magic64), uint64(magic64))
+	}
+}
+
+func TestStoreUintptr(t *testing.T) {
+	var x struct {
+		before uintptr
+		i      uintptr
+		after  uintptr
+	}
+	var m uint64 = magic64
+	magicptr := uintptr(m)
+	x.before = magicptr
+	x.after = magicptr
+	v := uintptr(0)
+	for delta := uintptr(1); delta+delta > delta; delta += delta {
+		StoreUintptr(&x.i, v)
+		if x.i != v {
+			t.Fatalf("delta=%d i=%d v=%d", delta, x.i, v)
+		}
+		v += delta
+	}
+	if x.before != magicptr || x.after != magicptr {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, magicptr, magicptr)
+	}
+}
+
+func TestStorePointer(t *testing.T) {
+	var x struct {
+		before uintptr
+		i      unsafe.Pointer
+		after  uintptr
+	}
+	var m uint64 = magic64
+	magicptr := uintptr(m)
+	x.before = magicptr
+	x.after = magicptr
+	v := unsafe.Pointer(uintptr(0))
+	for delta := uintptr(1); delta+delta > delta; delta += delta {
+		StorePointer(&x.i, unsafe.Pointer(v))
+		if x.i != v {
+			t.Fatalf("delta=%d i=%d v=%d", delta, x.i, v)
+		}
+		v = unsafe.Pointer(uintptr(v) + delta)
+	}
+	if x.before != magicptr || x.after != magicptr {
+		t.Fatalf("wrong magic: %#x _ %#x != %#x _ %#x", x.before, x.after, magicptr, magicptr)
+	}
+}
+
 // Tests of correct behavior, with contention.
 // (Is the function atomic?)
 //
@@ -366,6 +627,7 @@ var hammer32 = []struct {
 	{"CompareAndSwapInt32", hammerCompareAndSwapInt32},
 	{"CompareAndSwapUint32", hammerCompareAndSwapUint32},
 	{"CompareAndSwapUintptr", hammerCompareAndSwapUintptr32},
+	{"CompareAndSwapPointer", hammerCompareAndSwapPointer32},
 }
 
 func init() {
@@ -374,6 +636,7 @@ func init() {
 		// 64-bit system; clear uintptr tests
 		hammer32[2].f = nil
 		hammer32[5].f = nil
+		hammer32[6].f = nil
 	}
 }
 
@@ -436,6 +699,20 @@ func hammerCompareAndSwapUintptr32(uval *uint32, count int) {
 	}
 }
 
+func hammerCompareAndSwapPointer32(uval *uint32, count int) {
+	// only safe when uintptr is 32-bit.
+	// not called on 64-bit systems.
+	val := (*unsafe.Pointer)(unsafe.Pointer(uval))
+	for i := 0; i < count; i++ {
+		for {
+			v := *val
+			if CompareAndSwapPointer(val, v, unsafe.Pointer(uintptr(v)+1)) {
+				break
+			}
+		}
+	}
+}
+
 func TestHammer32(t *testing.T) {
 	const p = 4
 	n := 100000
@@ -460,7 +737,7 @@ func TestHammer32(t *testing.T) {
 			<-c
 		}
 		if val != uint32(n)*p {
-			t.Errorf("%s: val=%d want %d", tt.name, val, n*p)
+			t.Fatalf("%s: val=%d want %d", tt.name, val, n*p)
 		}
 	}
 }
@@ -475,6 +752,7 @@ var hammer64 = []struct {
 	{"CompareAndSwapInt64", hammerCompareAndSwapInt64},
 	{"CompareAndSwapUint64", hammerCompareAndSwapUint64},
 	{"CompareAndSwapUintptr", hammerCompareAndSwapUintptr64},
+	{"CompareAndSwapPointer", hammerCompareAndSwapPointer64},
 }
 
 func init() {
@@ -483,6 +761,7 @@ func init() {
 		// 32-bit system; clear uintptr tests
 		hammer64[2].f = nil
 		hammer64[5].f = nil
+		hammer64[6].f = nil
 	}
 }
 
@@ -545,6 +824,20 @@ func hammerCompareAndSwapUintptr64(uval *uint64, count int) {
 	}
 }
 
+func hammerCompareAndSwapPointer64(uval *uint64, count int) {
+	// only safe when uintptr is 64-bit.
+	// not called on 32-bit systems.
+	val := (*unsafe.Pointer)(unsafe.Pointer(uval))
+	for i := 0; i < count; i++ {
+		for {
+			v := *val
+			if CompareAndSwapPointer(val, v, unsafe.Pointer(uintptr(v)+1)) {
+				break
+			}
+		}
+	}
+}
+
 func TestHammer64(t *testing.T) {
 	if test64err != nil {
 		t.Logf("Skipping 64-bit tests: %v", test64err)
@@ -573,63 +866,141 @@ func TestHammer64(t *testing.T) {
 			<-c
 		}
 		if val != uint64(n)*p {
-			t.Errorf("%s: val=%d want %d", tt.name, val, n*p)
+			t.Fatalf("%s: val=%d want %d", tt.name, val, n*p)
 		}
 	}
 }
 
-func hammerLoadInt32(t *testing.T, uval *uint32) {
-	val := (*int32)(unsafe.Pointer(uval))
-	for {
-		v := LoadInt32(val)
+func hammerStoreLoadInt32(t *testing.T, valp unsafe.Pointer) {
+	val := (*int32)(valp)
+	v := LoadInt32(val)
+	vlo := v & ((1 << 16) - 1)
+	vhi := v >> 16
+	if vlo != vhi {
+		t.Fatalf("Int32: %#x != %#x", vlo, vhi)
+	}
+	new := v + 1 + 1<<16
+	if vlo == 1e4 {
+		new = 0
+	}
+	StoreInt32(val, new)
+}
+
+func hammerStoreLoadUint32(t *testing.T, valp unsafe.Pointer) {
+	val := (*uint32)(valp)
+	v := LoadUint32(val)
+	vlo := v & ((1 << 16) - 1)
+	vhi := v >> 16
+	if vlo != vhi {
+		t.Fatalf("Uint32: %#x != %#x", vlo, vhi)
+	}
+	new := v + 1 + 1<<16
+	if vlo == 1e4 {
+		new = 0
+	}
+	StoreUint32(val, new)
+}
+
+func hammerStoreLoadInt64(t *testing.T, valp unsafe.Pointer) {
+	val := (*int64)(valp)
+	v := LoadInt64(val)
+	vlo := v & ((1 << 32) - 1)
+	vhi := v >> 32
+	if vlo != vhi {
+		t.Fatalf("Int64: %#x != %#x", vlo, vhi)
+	}
+	new := v + 1 + 1<<32
+	StoreInt64(val, new)
+}
+
+func hammerStoreLoadUint64(t *testing.T, valp unsafe.Pointer) {
+	val := (*uint64)(valp)
+	v := LoadUint64(val)
+	vlo := v & ((1 << 32) - 1)
+	vhi := v >> 32
+	if vlo != vhi {
+		t.Fatalf("Uint64: %#x != %#x", vlo, vhi)
+	}
+	new := v + 1 + 1<<32
+	StoreUint64(val, new)
+}
+
+func hammerStoreLoadUintptr(t *testing.T, valp unsafe.Pointer) {
+	val := (*uintptr)(valp)
+	var test64 uint64 = 1 << 50
+	arch32 := uintptr(test64) == 0
+	v := LoadUintptr(val)
+	new := v
+	if arch32 {
 		vlo := v & ((1 << 16) - 1)
 		vhi := v >> 16
 		if vlo != vhi {
-			t.Fatalf("LoadInt32: %#x != %#x", vlo, vhi)
+			t.Fatalf("Uintptr: %#x != %#x", vlo, vhi)
 		}
-		new := v + 1 + 1<<16
+		new = v + 1 + 1<<16
 		if vlo == 1e4 {
 			new = 0
 		}
-		if CompareAndSwapInt32(val, v, new) {
-			break
+	} else {
+		vlo := v & ((1 << 32) - 1)
+		vhi := v >> 32
+		if vlo != vhi {
+			t.Fatalf("Uintptr: %#x != %#x", vlo, vhi)
 		}
+		inc := uint64(1 + 1<<32)
+		new = v + uintptr(inc)
 	}
+	StoreUintptr(val, new)
 }
 
-func hammerLoadUint32(t *testing.T, val *uint32) {
-	for {
-		v := LoadUint32(val)
+func hammerStoreLoadPointer(t *testing.T, valp unsafe.Pointer) {
+	val := (*unsafe.Pointer)(valp)
+	var test64 uint64 = 1 << 50
+	arch32 := uintptr(test64) == 0
+	v := uintptr(LoadPointer(val))
+	new := v
+	if arch32 {
 		vlo := v & ((1 << 16) - 1)
 		vhi := v >> 16
 		if vlo != vhi {
-			t.Fatalf("LoadUint32: %#x != %#x", vlo, vhi)
+			t.Fatalf("Pointer: %#x != %#x", vlo, vhi)
 		}
-		new := v + 1 + 1<<16
+		new = v + 1 + 1<<16
 		if vlo == 1e4 {
 			new = 0
 		}
-		if CompareAndSwapUint32(val, v, new) {
-			break
+	} else {
+		vlo := v & ((1 << 32) - 1)
+		vhi := v >> 32
+		if vlo != vhi {
+			t.Fatalf("Pointer: %#x != %#x", vlo, vhi)
 		}
+		inc := uint64(1 + 1<<32)
+		new = v + uintptr(inc)
 	}
+	StorePointer(val, unsafe.Pointer(new))
 }
 
-func TestHammerLoad(t *testing.T) {
-	tests := [...]func(*testing.T, *uint32){hammerLoadInt32, hammerLoadUint32}
-	n := 100000
+func TestHammerStoreLoad(t *testing.T) {
+	var tests []func(*testing.T, unsafe.Pointer)
+	tests = append(tests, hammerStoreLoadInt32, hammerStoreLoadUint32,
+		hammerStoreLoadUintptr, hammerStoreLoadPointer)
+	if test64err == nil {
+		tests = append(tests, hammerStoreLoadInt64, hammerStoreLoadUint64)
+	}
+	n := int(1e6)
 	if testing.Short() {
-		n = 10000
+		n = int(1e4)
 	}
 	const procs = 8
 	defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(procs))
 	for _, tt := range tests {
 		c := make(chan int)
-		var val uint32
+		var val uint64
 		for p := 0; p < procs; p++ {
 			go func() {
 				for i := 0; i < n; i++ {
-					tt(t, &val)
+					tt(t, unsafe.Pointer(&val))
 				}
 				c <- 1
 			}()
@@ -639,3 +1010,185 @@ func TestHammerLoad(t *testing.T) {
 		}
 	}
 }
+
+func TestStoreLoadSeqCst32(t *testing.T) {
+	if runtime.NumCPU() == 1 {
+		t.Logf("Skipping test on %v processor machine", runtime.NumCPU())
+		return
+	}
+	defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(4))
+	N := int32(1e3)
+	if testing.Short() {
+		N = int32(1e2)
+	}
+	c := make(chan bool, 2)
+	X := [2]int32{}
+	ack := [2][3]int32{{-1, -1, -1}, {-1, -1, -1}}
+	for p := 0; p < 2; p++ {
+		go func(me int) {
+			he := 1 - me
+			for i := int32(1); i < N; i++ {
+				StoreInt32(&X[me], i)
+				my := LoadInt32(&X[he])
+				StoreInt32(&ack[me][i%3], my)
+				for w := 1; LoadInt32(&ack[he][i%3]) == -1; w++ {
+					if w%1000 == 0 {
+						runtime.Gosched()
+					}
+				}
+				his := LoadInt32(&ack[he][i%3])
+				if (my != i && my != i-1) || (his != i && his != i-1) {
+					t.Fatalf("invalid values: %d/%d (%d)", my, his, i)
+				}
+				if my != i && his != i {
+					t.Fatalf("store/load are not sequentially consistent: %d/%d (%d)", my, his, i)
+				}
+				StoreInt32(&ack[me][(i-1)%3], -1)
+			}
+			c <- true
+		}(p)
+	}
+	<-c
+	<-c
+}
+
+func TestStoreLoadSeqCst64(t *testing.T) {
+	if runtime.NumCPU() == 1 {
+		t.Logf("Skipping test on %v processor machine", runtime.NumCPU())
+		return
+	}
+	if test64err != nil {
+		t.Logf("Skipping 64-bit tests: %v", test64err)
+		return
+	}
+	defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(4))
+	N := int64(1e3)
+	if testing.Short() {
+		N = int64(1e2)
+	}
+	c := make(chan bool, 2)
+	X := [2]int64{}
+	ack := [2][3]int64{{-1, -1, -1}, {-1, -1, -1}}
+	for p := 0; p < 2; p++ {
+		go func(me int) {
+			he := 1 - me
+			for i := int64(1); i < N; i++ {
+				StoreInt64(&X[me], i)
+				my := LoadInt64(&X[he])
+				StoreInt64(&ack[me][i%3], my)
+				for w := 1; LoadInt64(&ack[he][i%3]) == -1; w++ {
+					if w%1000 == 0 {
+						runtime.Gosched()
+					}
+				}
+				his := LoadInt64(&ack[he][i%3])
+				if (my != i && my != i-1) || (his != i && his != i-1) {
+					t.Fatalf("invalid values: %d/%d (%d)", my, his, i)
+				}
+				if my != i && his != i {
+					t.Fatalf("store/load are not sequentially consistent: %d/%d (%d)", my, his, i)
+				}
+				StoreInt64(&ack[me][(i-1)%3], -1)
+			}
+			c <- true
+		}(p)
+	}
+	<-c
+	<-c
+}
+
+func TestStoreLoadRelAcq32(t *testing.T) {
+	if runtime.NumCPU() == 1 {
+		t.Logf("Skipping test on %v processor machine", runtime.NumCPU())
+		return
+	}
+	defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(4))
+	N := int32(1e3)
+	if testing.Short() {
+		N = int32(1e2)
+	}
+	c := make(chan bool, 2)
+	type Data struct {
+		signal int32
+		pad1   [128]int8
+		data1  int32
+		pad2   [128]int8
+		data2  float32
+	}
+	var X Data
+	for p := int32(0); p < 2; p++ {
+		go func(p int32) {
+			for i := int32(1); i < N; i++ {
+				if (i+p)%2 == 0 {
+					X.data1 = i
+					X.data2 = float32(i)
+					StoreInt32(&X.signal, i)
+				} else {
+					for w := 1; LoadInt32(&X.signal) != i; w++ {
+						if w%1000 == 0 {
+							runtime.Gosched()
+						}
+					}
+					d1 := X.data1
+					d2 := X.data2
+					if d1 != i || d2 != float32(i) {
+						t.Fatalf("incorrect data: %d/%d (%d)", d1, d2, i)
+					}
+				}
+			}
+			c <- true
+		}(p)
+	}
+	<-c
+	<-c
+}
+
+func TestStoreLoadRelAcq64(t *testing.T) {
+	if runtime.NumCPU() == 1 {
+		t.Logf("Skipping test on %v processor machine", runtime.NumCPU())
+		return
+	}
+	if test64err != nil {
+		t.Logf("Skipping 64-bit tests: %v", test64err)
+		return
+	}
+	defer runtime.GOMAXPROCS(runtime.GOMAXPROCS(4))
+	N := int64(1e3)
+	if testing.Short() {
+		N = int64(1e2)
+	}
+	c := make(chan bool, 2)
+	type Data struct {
+		signal int64
+		pad1   [128]int8
+		data1  int64
+		pad2   [128]int8
+		data2  float64
+	}
+	var X Data
+	for p := int64(0); p < 2; p++ {
+		go func(p int64) {
+			for i := int64(1); i < N; i++ {
+				if (i+p)%2 == 0 {
+					X.data1 = i
+					X.data2 = float64(i)
+					StoreInt64(&X.signal, i)
+				} else {
+					for w := 1; LoadInt64(&X.signal) != i; w++ {
+						if w%1000 == 0 {
+							runtime.Gosched()
+						}
+					}
+					d1 := X.data1
+					d2 := X.data2
+					if d1 != i || d2 != float64(i) {
+						t.Fatalf("incorrect data: %d/%d (%d)", d1, d2, i)
+					}
+				}
+			}
+			c <- true
+		}(p)
+	}
+	<-c
+	<-c
+}
diff --git a/src/pkg/sync/atomic/doc.go b/src/pkg/sync/atomic/doc.go
index b35eb539c..ecb4808ce 100644
--- a/src/pkg/sync/atomic/doc.go
+++ b/src/pkg/sync/atomic/doc.go
@@ -22,9 +22,13 @@
 //
 package atomic
 
+import (
+	"unsafe"
+)
+
 // BUG(rsc): On ARM, the 64-bit functions use instructions unavailable before ARM 11.
 //
-// On x86-32, the 64-bit functions use instructions unavailable before the Pentium.
+// On x86-32, the 64-bit functions use instructions unavailable before the Pentium MMX.
 
 // CompareAndSwapInt32 executes the compare-and-swap operation for an int32 value.
 func CompareAndSwapInt32(val *int32, old, new int32) (swapped bool)
@@ -41,6 +45,9 @@ func CompareAndSwapUint64(val *uint64, old, new uint64) (swapped bool)
 // CompareAndSwapUintptr executes the compare-and-swap operation for a uintptr value.
 func CompareAndSwapUintptr(val *uintptr, old, new uintptr) (swapped bool)
 
+// CompareAndSwapPointer executes the compare-and-swap operation for a unsafe.Pointer value.
+func CompareAndSwapPointer(val *unsafe.Pointer, old, new unsafe.Pointer) (swapped bool)
+
 // AddInt32 atomically adds delta to *val and returns the new value.
 func AddInt32(val *int32, delta int32) (new int32)
 
@@ -59,9 +66,39 @@ func AddUintptr(val *uintptr, delta uintptr) (new uintptr)
 // LoadInt32 atomically loads *addr.
 func LoadInt32(addr *int32) (val int32)
 
+// LoadInt64 atomically loads *addr.
+func LoadInt64(addr *int64) (val int64)
+
 // LoadUint32 atomically loads *addr.
 func LoadUint32(addr *uint32) (val uint32)
 
+// LoadUint64 atomically loads *addr.
+func LoadUint64(addr *uint64) (val uint64)
+
+// LoadUintptr atomically loads *addr.
+func LoadUintptr(addr *uintptr) (val uintptr)
+
+// LoadPointer atomically loads *addr.
+func LoadPointer(addr *unsafe.Pointer) (val unsafe.Pointer)
+
+// StoreInt32 atomically stores val into *addr.
+func StoreInt32(addr *int32, val int32)
+
+// StoreInt64 atomically stores val into *addr.
+func StoreInt64(addr *int64, val int64)
+
+// StoreUint32 atomically stores val into *addr.
+func StoreUint32(addr *uint32, val uint32)
+
+// StoreUint64 atomically stores val into *addr.
+func StoreUint64(addr *uint64, val uint64)
+
+// StoreUintptr atomically stores val into *addr.
+func StoreUintptr(addr *uintptr, val uintptr)
+
+// StorePointer atomically stores val into *addr.
+func StorePointer(addr *unsafe.Pointer, val unsafe.Pointer)
+
 // Helper for ARM.  Linker will discard on other systems
 func panic64() {
 	panic("sync/atomic: broken 64-bit atomic operations (buggy QEMU)")
diff --git a/src/pkg/sync/cond.go b/src/pkg/sync/cond.go
index 75494b535..1fc3deaf1 100644
--- a/src/pkg/sync/cond.go
+++ b/src/pkg/sync/cond.go
@@ -4,8 +4,6 @@
 
 package sync
 
-import "runtime"
-
 // Cond implements a condition variable, a rendezvous point
 // for goroutines waiting for or announcing the occurrence
 // of an event.
@@ -43,9 +41,10 @@ func NewCond(l Locker) *Cond {
 
 // Wait atomically unlocks c.L and suspends execution
 // of the calling goroutine.  After later resuming execution,
-// Wait locks c.L before returning.
+// Wait locks c.L before returning.  Unlike in other systems,
+// Wait cannot return unless awoken by Broadcast or Signal.
 //
-// Because L is not locked when Wait first resumes, the caller
+// Because c.L is not locked when Wait first resumes, the caller
 // typically cannot assume that the condition is true when
 // Wait returns.  Instead, the caller should Wait in a loop:
 //
@@ -65,7 +64,7 @@ func (c *Cond) Wait() {
 	c.newWaiters++
 	c.m.Unlock()
 	c.L.Unlock()
-	runtime.Semacquire(s)
+	runtime_Semacquire(s)
 	c.L.Lock()
 }
 
@@ -84,7 +83,7 @@ func (c *Cond) Signal() {
 	}
 	if c.oldWaiters > 0 {
 		c.oldWaiters--
-		runtime.Semrelease(c.oldSema)
+		runtime_Semrelease(c.oldSema)
 	}
 	c.m.Unlock()
 }
@@ -98,13 +97,13 @@ func (c *Cond) Broadcast() {
 	// Wake both generations.
 	if c.oldWaiters > 0 {
 		for i := 0; i < c.oldWaiters; i++ {
-			runtime.Semrelease(c.oldSema)
+			runtime_Semrelease(c.oldSema)
 		}
 		c.oldWaiters = 0
 	}
 	if c.newWaiters > 0 {
 		for i := 0; i < c.newWaiters; i++ {
-			runtime.Semrelease(c.newSema)
+			runtime_Semrelease(c.newSema)
 		}
 		c.newWaiters = 0
 		c.newSema = nil
diff --git a/src/pkg/sync/example_test.go b/src/pkg/sync/example_test.go
new file mode 100644
index 000000000..156492400
--- /dev/null
+++ b/src/pkg/sync/example_test.go
@@ -0,0 +1,54 @@
+// Copyright 2012 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.
+
+package sync_test
+
+import (
+	"fmt"
+	"net/http"
+	"sync"
+)
+
+// This example fetches several URLs concurrently,
+// using a WaitGroup to block until all the fetches are complete.
+func ExampleWaitGroup() {
+	var wg sync.WaitGroup
+	var urls = []string{
+		"http://www.golang.org/",
+		"http://www.google.com/",
+		"http://www.somestupidname.com/",
+	}
+	for _, url := range urls {
+		// Increment the WaitGroup counter.
+		wg.Add(1)
+		// Launch a goroutine to fetch the URL.
+		go func(url string) {
+			// Fetch the URL.
+			http.Get(url)
+			// Decrement the counter.
+			wg.Done()
+		}(url)
+	}
+	// Wait for all HTTP fetches to complete.
+	wg.Wait()
+}
+
+func ExampleOnce() {
+	var once sync.Once
+	onceBody := func() {
+		fmt.Printf("Only once\n")
+	}
+	done := make(chan bool)
+	for i := 0; i < 10; i++ {
+		go func() {
+			once.Do(onceBody)
+			done <- true
+		}()
+	}
+	for i := 0; i < 10; i++ {
+		<-done
+	}
+	// Output:
+	// Only once
+}
diff --git a/src/pkg/sync/export_test.go b/src/pkg/sync/export_test.go
new file mode 100644
index 000000000..fa5983a2d
--- /dev/null
+++ b/src/pkg/sync/export_test.go
@@ -0,0 +1,9 @@
+// Copyright 2012 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.
+
+package sync
+
+// Export for testing.
+var Runtime_Semacquire = runtime_Semacquire
+var Runtime_Semrelease = runtime_Semrelease
diff --git a/src/pkg/sync/mutex.go b/src/pkg/sync/mutex.go
index 2d46c8994..9494cc3f8 100644
--- a/src/pkg/sync/mutex.go
+++ b/src/pkg/sync/mutex.go
@@ -6,12 +6,11 @@
 // exclusion locks.  Other than the Once and WaitGroup types, most are intended
 // for use by low-level library routines.  Higher-level synchronization is
 // better done via channels and communication.
+//
+// Values containing the types defined in this package should not be copied.
 package sync
 
-import (
-	"runtime"
-	"sync/atomic"
-)
+import "sync/atomic"
 
 // A Mutex is a mutual exclusion lock.
 // Mutexes can be created as part of other structures;
@@ -58,7 +57,7 @@ func (m *Mutex) Lock() {
 			if old&mutexLocked == 0 {
 				break
 			}
-			runtime.Semacquire(&m.sema)
+			runtime_Semacquire(&m.sema)
 			awoke = true
 		}
 	}
@@ -87,7 +86,7 @@ func (m *Mutex) Unlock() {
 		// Grab the right to wake someone.
 		new = (old - 1<<mutexWaiterShift) | mutexWoken
 		if atomic.CompareAndSwapInt32(&m.state, old, new) {
-			runtime.Semrelease(&m.sema)
+			runtime_Semrelease(&m.sema)
 			return
 		}
 		old = m.state
diff --git a/src/pkg/sync/mutex_test.go b/src/pkg/sync/mutex_test.go
index 47758844f..bf78c6f60 100644
--- a/src/pkg/sync/mutex_test.go
+++ b/src/pkg/sync/mutex_test.go
@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// GOMAXPROCS=10 gotest
+// GOMAXPROCS=10 go test
 
 package sync_test
 
@@ -15,8 +15,8 @@ import (
 
 func HammerSemaphore(s *uint32, loops int, cdone chan bool) {
 	for i := 0; i < loops; i++ {
-		runtime.Semacquire(s)
-		runtime.Semrelease(s)
+		Runtime_Semacquire(s)
+		Runtime_Semrelease(s)
 	}
 	cdone <- true
 }
diff --git a/src/pkg/sync/once_test.go b/src/pkg/sync/once_test.go
index 157a3667a..37075af17 100644
--- a/src/pkg/sync/once_test.go
+++ b/src/pkg/sync/once_test.go
@@ -5,9 +5,9 @@
 package sync_test
 
 import (
+	"runtime"
 	. "sync"
 	"sync/atomic"
-	"runtime"
 	"testing"
 )
 
diff --git a/src/pkg/sync/runtime.go b/src/pkg/sync/runtime.go
new file mode 100644
index 000000000..e99599c11
--- /dev/null
+++ b/src/pkg/sync/runtime.go
@@ -0,0 +1,18 @@
+// Copyright 2012 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.
+
+package sync
+
+// defined in package runtime
+
+// Semacquire waits until *s > 0 and then atomically decrements it.
+// It is intended as a simple sleep primitive for use by the synchronization
+// library and should not be used directly.
+func runtime_Semacquire(s *uint32)
+
+// Semrelease atomically increments *s and notifies a waiting goroutine
+// if one is blocked in Semacquire.
+// It is intended as a simple wakeup primitive for use by the synchronization
+// library and should not be used directly.
+func runtime_Semrelease(s *uint32)
diff --git a/src/pkg/sync/runtime_sema_test.go b/src/pkg/sync/runtime_sema_test.go
new file mode 100644
index 000000000..57a8dbee7
--- /dev/null
+++ b/src/pkg/sync/runtime_sema_test.go
@@ -0,0 +1,101 @@
+// Copyright 2009 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.
+
+package sync_test
+
+import (
+	"runtime"
+	. "sync"
+	"sync/atomic"
+	"testing"
+)
+
+func BenchmarkSemaUncontended(b *testing.B) {
+	type PaddedSem struct {
+		sem uint32
+		pad [32]uint32
+	}
+	const CallsPerSched = 1000
+	procs := runtime.GOMAXPROCS(-1)
+	N := int32(b.N / CallsPerSched)
+	c := make(chan bool, procs)
+	for p := 0; p < procs; p++ {
+		go func() {
+			sem := new(PaddedSem)
+			for atomic.AddInt32(&N, -1) >= 0 {
+				runtime.Gosched()
+				for g := 0; g < CallsPerSched; g++ {
+					Runtime_Semrelease(&sem.sem)
+					Runtime_Semacquire(&sem.sem)
+				}
+			}
+			c <- true
+		}()
+	}
+	for p := 0; p < procs; p++ {
+		<-c
+	}
+}
+
+func benchmarkSema(b *testing.B, block, work bool) {
+	const CallsPerSched = 1000
+	const LocalWork = 100
+	procs := runtime.GOMAXPROCS(-1)
+	N := int32(b.N / CallsPerSched)
+	c := make(chan bool, procs)
+	c2 := make(chan bool, procs/2)
+	sem := uint32(0)
+	if block {
+		for p := 0; p < procs/2; p++ {
+			go func() {
+				Runtime_Semacquire(&sem)
+				c2 <- true
+			}()
+		}
+	}
+	for p := 0; p < procs; p++ {
+		go func() {
+			foo := 0
+			for atomic.AddInt32(&N, -1) >= 0 {
+				runtime.Gosched()
+				for g := 0; g < CallsPerSched; g++ {
+					Runtime_Semrelease(&sem)
+					if work {
+						for i := 0; i < LocalWork; i++ {
+							foo *= 2
+							foo /= 2
+						}
+					}
+					Runtime_Semacquire(&sem)
+				}
+			}
+			c <- foo == 42
+			Runtime_Semrelease(&sem)
+		}()
+	}
+	if block {
+		for p := 0; p < procs/2; p++ {
+			<-c2
+		}
+	}
+	for p := 0; p < procs; p++ {
+		<-c
+	}
+}
+
+func BenchmarkSemaSyntNonblock(b *testing.B) {
+	benchmarkSema(b, false, false)
+}
+
+func BenchmarkSemaSyntBlock(b *testing.B) {
+	benchmarkSema(b, true, false)
+}
+
+func BenchmarkSemaWorkNonblock(b *testing.B) {
+	benchmarkSema(b, false, true)
+}
+
+func BenchmarkSemaWorkBlock(b *testing.B) {
+	benchmarkSema(b, true, true)
+}
diff --git a/src/pkg/sync/rwmutex.go b/src/pkg/sync/rwmutex.go
index cb1a47720..782a9c319 100644
--- a/src/pkg/sync/rwmutex.go
+++ b/src/pkg/sync/rwmutex.go
@@ -4,10 +4,7 @@
 
 package sync
 
-import (
-	"runtime"
-	"sync/atomic"
-)
+import "sync/atomic"
 
 // An RWMutex is a reader/writer mutual exclusion lock.
 // The lock can be held by an arbitrary number of readers
@@ -29,7 +26,7 @@ const rwmutexMaxReaders = 1 << 30
 func (rw *RWMutex) RLock() {
 	if atomic.AddInt32(&rw.readerCount, 1) < 0 {
 		// A writer is pending, wait for it.
-		runtime.Semacquire(&rw.readerSem)
+		runtime_Semacquire(&rw.readerSem)
 	}
 }
 
@@ -42,7 +39,7 @@ func (rw *RWMutex) RUnlock() {
 		// A writer is pending.
 		if atomic.AddInt32(&rw.readerWait, -1) == 0 {
 			// The last reader unblocks the writer.
-			runtime.Semrelease(&rw.writerSem)
+			runtime_Semrelease(&rw.writerSem)
 		}
 	}
 }
@@ -60,7 +57,7 @@ func (rw *RWMutex) Lock() {
 	r := atomic.AddInt32(&rw.readerCount, -rwmutexMaxReaders) + rwmutexMaxReaders
 	// Wait for active readers.
 	if r != 0 && atomic.AddInt32(&rw.readerWait, r) != 0 {
-		runtime.Semacquire(&rw.writerSem)
+		runtime_Semacquire(&rw.writerSem)
 	}
 }
 
@@ -75,7 +72,7 @@ func (rw *RWMutex) Unlock() {
 	r := atomic.AddInt32(&rw.readerCount, rwmutexMaxReaders)
 	// Unblock blocked readers, if any.
 	for i := 0; i < int(r); i++ {
-		runtime.Semrelease(&rw.readerSem)
+		runtime_Semrelease(&rw.readerSem)
 	}
 	// Allow other writers to proceed.
 	rw.w.Unlock()
diff --git a/src/pkg/sync/rwmutex_test.go b/src/pkg/sync/rwmutex_test.go
index dc8ce9653..39d5d6540 100644
--- a/src/pkg/sync/rwmutex_test.go
+++ b/src/pkg/sync/rwmutex_test.go
@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// GOMAXPROCS=10 gotest
+// GOMAXPROCS=10 go test
 
 package sync_test
 
diff --git a/src/pkg/sync/waitgroup.go b/src/pkg/sync/waitgroup.go
index a4c9b7e43..0165b1ffb 100644
--- a/src/pkg/sync/waitgroup.go
+++ b/src/pkg/sync/waitgroup.go
@@ -4,31 +4,13 @@
 
 package sync
 
-import (
-	"runtime"
-	"sync/atomic"
-)
+import "sync/atomic"
 
 // A WaitGroup waits for a collection of goroutines to finish.
 // The main goroutine calls Add to set the number of
 // goroutines to wait for.  Then each of the goroutines
 // runs and calls Done when finished.  At the same time,
 // Wait can be used to block until all goroutines have finished.
-//
-// For example:
-//
-//   for i := 0; i < n; i++ {
-//       if !condition(i) {
-//           continue
-//       }
-//       wg.Add(1)
-//       go func() {
-//           // Do something.
-//           wg.Done()
-//       }()
-//   }
-//   wg.Wait()
-// 
 type WaitGroup struct {
 	m       Mutex
 	counter int32
@@ -60,7 +42,7 @@ func (wg *WaitGroup) Add(delta int) {
 	}
 	wg.m.Lock()
 	for i := int32(0); i < wg.waiters; i++ {
-		runtime.Semrelease(wg.sema)
+		runtime_Semrelease(wg.sema)
 	}
 	wg.waiters = 0
 	wg.sema = nil
@@ -93,5 +75,5 @@ func (wg *WaitGroup) Wait() {
 	}
 	s := wg.sema
 	wg.m.Unlock()
-	runtime.Semacquire(s)
+	runtime_Semacquire(s)
 }