runtime: correct various Go -> C function calls

Some things get converted.
Other things (too complex or too many C deps) get onM calls.
Other things (too simple) get #pragma textflag NOSPLIT.

After this CL, the offending function list is basically:
        - panic.c
        - netpoll.goc
        - mem*.c
        - race stuff
        - readgstatus
        - entersyscall/exitsyscall

LGTM=r, iant
R=golang-codereviews, r, iant
CC=dvyukov, golang-codereviews, khr
https://codereview.appspot.com/140930043

21 files changed, 297 insertions, 152 deletions

diff --git a/src/pkg/runtime/cpuprof.go b/src/pkg/runtime/cpuprof.go
index 4325d7e1c..8b1c1c632 100644
--- a/src/pkg/runtime/cpuprof.go
+++ b/src/pkg/runtime/cpuprof.go
@@ -101,7 +101,13 @@ var (
 	eod = [3]uintptr{0, 1, 0}
 )
 
-func setcpuprofilerate(int32) // proc.c
+func setcpuprofilerate_m() // proc.c
+
+func setcpuprofilerate(hz int32) {
+	g := getg()
+	g.m.scalararg[0] = uintptr(hz)
+	onM(setcpuprofilerate_m)
+}
 
 // lostProfileData is a no-op function used in profiles
 // to mark the number of profiling stack traces that were
diff --git a/src/pkg/runtime/debug.go b/src/pkg/runtime/debug.go
index ed11fb1b1..bb4bd60ed 100644
--- a/src/pkg/runtime/debug.go
+++ b/src/pkg/runtime/debug.go
@@ -24,10 +24,15 @@ func UnlockOSThread()
 // The number of logical CPUs on the local machine can be queried with NumCPU.
 // This call will go away when the scheduler improves.
 func GOMAXPROCS(n int) int {
-	return int(gomaxprocsfunc(int32(n)))
+	g := getg()
+	g.m.scalararg[0] = uintptr(n)
+	onM(gomaxprocs_m)
+	n = int(g.m.scalararg[0])
+	g.m.scalararg[0] = 0
+	return n
 }
 
-func gomaxprocsfunc(int32) int32 // proc.c
+func gomaxprocs_m() // proc.c
 
 // NumCPU returns the number of logical CPUs on the local machine.
 func NumCPU() int {
diff --git a/src/pkg/runtime/malloc.c b/src/pkg/runtime/malloc.c
index 311cc442c..c864bc93c 100644
--- a/src/pkg/runtime/malloc.c
+++ b/src/pkg/runtime/malloc.c
@@ -348,58 +348,6 @@ runtime·MHeap_SysAlloc(MHeap *h, uintptr n)
 	return p;
 }
 
-static struct
-{
-	Mutex	lock;
-	byte*	pos;
-	byte*	end;
-} persistent;
-
-enum
-{
-	PersistentAllocChunk	= 256<<10,
-	PersistentAllocMaxBlock	= 64<<10,  // VM reservation granularity is 64K on windows
-};
-
-// Wrapper around sysAlloc that can allocate small chunks.
-// There is no associated free operation.
-// Intended for things like function/type/debug-related persistent data.
-// If align is 0, uses default align (currently 8).
-void*
-runtime·persistentalloc(uintptr size, uintptr align, uint64 *stat)
-{
-	byte *p;
-
-	if(align != 0) {
-		if(align&(align-1))
-			runtime·throw("persistentalloc: align is not a power of 2");
-		if(align > PageSize)
-			runtime·throw("persistentalloc: align is too large");
-	} else
-		align = 8;
-	if(size >= PersistentAllocMaxBlock)
-		return runtime·sysAlloc(size, stat);
-	runtime·lock(&persistent.lock);
-	persistent.pos = (byte*)ROUND((uintptr)persistent.pos, align);
-	if(persistent.pos + size > persistent.end) {
-		persistent.pos = runtime·sysAlloc(PersistentAllocChunk, &mstats.other_sys);
-		if(persistent.pos == nil) {
-			runtime·unlock(&persistent.lock);
-			runtime·throw("runtime: cannot allocate memory");
-		}
-		persistent.end = persistent.pos + PersistentAllocChunk;
-	}
-	p = persistent.pos;
-	persistent.pos += size;
-	runtime·unlock(&persistent.lock);
-	if(stat != &mstats.other_sys) {
-		// reaccount the allocation against provided stat
-		runtime·xadd64(stat, size);
-		runtime·xadd64(&mstats.other_sys, -(uint64)size);
-	}
-	return p;
-}
-
 // Runtime stubs.
 
 static void*
diff --git a/src/pkg/runtime/malloc.go b/src/pkg/runtime/malloc.go
index dbe37c810..883ca0cef 100644
--- a/src/pkg/runtime/malloc.go
+++ b/src/pkg/runtime/malloc.go
@@ -431,7 +431,7 @@ func gogc(force int32) {
 	mp = acquirem()
 	mp.gcing = 1
 	releasem(mp)
-	stoptheworld()
+	onM(stoptheworld)
 	if mp != acquirem() {
 		gothrow("gogc: rescheduled")
 	}
@@ -465,7 +465,7 @@ func gogc(force int32) {
 	// all done
 	mp.gcing = 0
 	semrelease(&worldsema)
-	starttheworld()
+	onM(starttheworld)
 	releasem(mp)
 	mp = nil
 
@@ -760,3 +760,55 @@ func runfinq() {
 		}
 	}
 }
+
+var persistent struct {
+	lock mutex
+	pos  unsafe.Pointer
+	end  unsafe.Pointer
+}
+
+// Wrapper around sysAlloc that can allocate small chunks.
+// There is no associated free operation.
+// Intended for things like function/type/debug-related persistent data.
+// If align is 0, uses default align (currently 8).
+func persistentalloc(size, align uintptr, stat *uint64) unsafe.Pointer {
+	const (
+		chunk    = 256 << 10
+		maxBlock = 64 << 10 // VM reservation granularity is 64K on windows
+	)
+
+	if align != 0 {
+		if align&(align-1) != 0 {
+			gothrow("persistentalloc: align is not a power of 2")
+		}
+		if align > _PageSize {
+			gothrow("persistentalloc: align is too large")
+		}
+	} else {
+		align = 8
+	}
+
+	if size >= maxBlock {
+		return sysAlloc(size, stat)
+	}
+
+	lock(&persistent.lock)
+	persistent.pos = roundup(persistent.pos, align)
+	if uintptr(persistent.pos)+size > uintptr(persistent.end) {
+		persistent.pos = sysAlloc(chunk, &memstats.other_sys)
+		if persistent.pos == nil {
+			unlock(&persistent.lock)
+			gothrow("runtime: cannot allocate memory")
+		}
+		persistent.end = add(persistent.pos, chunk)
+	}
+	p := persistent.pos
+	persistent.pos = add(persistent.pos, size)
+	unlock(&persistent.lock)
+
+	if stat != &memstats.other_sys {
+		xadd64(stat, int64(size))
+		xadd64(&memstats.other_sys, -int64(size))
+	}
+	return p
+}
diff --git a/src/pkg/runtime/malloc.h b/src/pkg/runtime/malloc.h
index d9a2bf84b..544169194 100644
--- a/src/pkg/runtime/malloc.h
+++ b/src/pkg/runtime/malloc.h
@@ -578,7 +578,7 @@ extern bool	runtime·fingwake;
 extern FinBlock	*runtime·finq;		// list of finalizers that are to be executed
 extern FinBlock	*runtime·finc;		// cache of free blocks
 
-void	runtime·setprofilebucket(void *p, Bucket *b);
+void	runtime·setprofilebucket_m(void);
 
 bool	runtime·addfinalizer(void*, FuncVal *fn, uintptr, Type*, PtrType*);
 void	runtime·removefinalizer(void*);
diff --git a/src/pkg/runtime/mcache.c b/src/pkg/runtime/mcache.c
index 8e98890e8..bb1fc5403 100644
--- a/src/pkg/runtime/mcache.c
+++ b/src/pkg/runtime/mcache.c
@@ -52,24 +52,23 @@ freemcache(MCache *c)
 }
 
 static void
-freemcache_m(G *gp)
+freemcache_m(void)
 {
 	MCache *c;
 
 	c = g->m->ptrarg[0];
 	g->m->ptrarg[0] = nil;
 	freemcache(c);
-	runtime·gogo(&gp->sched);
 }
 
 void
 runtime·freemcache(MCache *c)
 {
-	void (*fn)(G*);
+	void (*fn)(void);
 
 	g->m->ptrarg[0] = c;
 	fn = freemcache_m;
-	runtime·mcall(&fn);
+	runtime·onM(&fn);
 }
 
 // Gets a span that has a free object in it and assigns it
diff --git a/src/pkg/runtime/mheap.c b/src/pkg/runtime/mheap.c
index 0050e9655..902a5c71a 100644
--- a/src/pkg/runtime/mheap.c
+++ b/src/pkg/runtime/mheap.c
@@ -834,9 +834,16 @@ runtime·removefinalizer(void *p)
 
 // Set the heap profile bucket associated with addr to b.
 void
-runtime·setprofilebucket(void *p, Bucket *b)
-{
+runtime·setprofilebucket_m(void)
+{	
+	void *p;
+	Bucket *b;
 	SpecialProfile *s;
+	
+	p = g->m->ptrarg[0];
+	b = g->m->ptrarg[1];
+	g->m->ptrarg[0] = nil;
+	g->m->ptrarg[1] = nil;
 
 	runtime·lock(&runtime·mheap.speciallock);
 	s = runtime·FixAlloc_Alloc(&runtime·mheap.specialprofilealloc);
diff --git a/src/pkg/runtime/mprof.go b/src/pkg/runtime/mprof.go
index b421c4f98..7177c8459 100644
--- a/src/pkg/runtime/mprof.go
+++ b/src/pkg/runtime/mprof.go
@@ -249,7 +249,14 @@ func mProf_Malloc(p unsafe.Pointer, size uintptr) {
 	setprofilebucket(p, b)
 }
 
-func setprofilebucket(p unsafe.Pointer, b *bucket) // mheap.c
+func setprofilebucket_m() // mheap.c
+
+func setprofilebucket(p unsafe.Pointer, b *bucket) {
+	g := getg()
+	g.m.ptrarg[0] = p
+	g.m.ptrarg[1] = unsafe.Pointer(b)
+	onM(setprofilebucket_m)
+}
 
 // Called when freeing a profiled block.
 func mProf_Free(b *bucket, size uintptr, freed bool) {
@@ -288,8 +295,7 @@ func SetBlockProfileRate(rate int) {
 	atomicstore64(&blockprofilerate, uint64(r))
 }
 
-func tickspersecond() int64 // runtime.c
-func fastrand1() uint32     // runtime.c
+func fastrand1() uint32     // assembly
 func readgstatus(*g) uint32 // proc.c
 
 func blockevent(cycles int64, skip int) {
@@ -531,7 +537,7 @@ func GoroutineProfile(p []StackRecord) (n int, ok bool) {
 		gp := getg()
 		semacquire(&worldsema, false)
 		gp.m.gcing = 1
-		stoptheworld()
+		onM(stoptheworld)
 
 		n = NumGoroutine()
 		if n <= len(p) {
@@ -550,7 +556,7 @@ func GoroutineProfile(p []StackRecord) (n int, ok bool) {
 
 		gp.m.gcing = 0
 		semrelease(&worldsema)
-		starttheworld()
+		onM(starttheworld)
 	}
 
 	return n, ok
@@ -576,7 +582,7 @@ func Stack(buf []byte, all bool) int {
 		semacquire(&worldsema, false)
 		mp.gcing = 1
 		releasem(mp)
-		stoptheworld()
+		onM(stoptheworld)
 		if mp != acquirem() {
 			gothrow("Stack: rescheduled")
 		}
@@ -597,7 +603,7 @@ func Stack(buf []byte, all bool) int {
 	if all {
 		mp.gcing = 0
 		semrelease(&worldsema)
-		starttheworld()
+		onM(starttheworld)
 	}
 	releasem(mp)
 	return n
diff --git a/src/pkg/runtime/os_darwin.c b/src/pkg/runtime/os_darwin.c
index bf13cdbaf..19181d60d 100644
--- a/src/pkg/runtime/os_darwin.c
+++ b/src/pkg/runtime/os_darwin.c
@@ -22,16 +22,31 @@ unimplemented(int8 *name)
 	*(int32*)1231 = 1231;
 }
 
+#pragma textflag NOSPLIT
 void
 runtime·semawakeup(M *mp)
 {
 	runtime·mach_semrelease(mp->waitsema);
 }
 
+static void
+semacreate(void)
+{
+	g->m->scalararg[0] = runtime·mach_semcreate();
+}
+
+#pragma textflag NOSPLIT
 uintptr
 runtime·semacreate(void)
 {
-	return runtime·mach_semcreate();
+	uintptr x;
+	void (*fn)(void);
+	
+	fn = semacreate;
+	runtime·onM(&fn);
+	x = g->m->scalararg[0];
+	g->m->scalararg[0] = 0;
+	return x;
 }
 
 // BSD interface for threading.
@@ -143,7 +158,6 @@ runtime·unminit(void)
 // Mach IPC, to get at semaphores
 // Definitions are in /usr/include/mach on a Mac.
 
-#pragma textflag NOSPLIT
 static void
 macherror(int32 r, int8 *fn)
 {
@@ -398,38 +412,88 @@ int32 runtime·mach_semaphore_timedwait(uint32 sema, uint32 sec, uint32 nsec);
 int32 runtime·mach_semaphore_signal(uint32 sema);
 int32 runtime·mach_semaphore_signal_all(uint32 sema);
 
-#pragma textflag NOSPLIT
-int32
-runtime·semasleep(int64 ns)
+static void
+semasleep(void)
 {
 	int32 r, secs, nsecs;
+	int64 ns;
+	
+	ns = g->m->scalararg[0] | g->m->scalararg[1]<<32;
+	g->m->scalararg[0] = 0;
+	g->m->scalararg[1] = 0;
 
 	if(ns >= 0) {
 		secs = runtime·timediv(ns, 1000000000, &nsecs);
 		r = runtime·mach_semaphore_timedwait(g->m->waitsema, secs, nsecs);
-		if(r == KERN_ABORTED || r == KERN_OPERATION_TIMED_OUT)
-			return -1;
+		if(r == KERN_ABORTED || r == KERN_OPERATION_TIMED_OUT) {
+			g->m->scalararg[0] = -1;
+			return;
+		}
 		if(r != 0)
 			macherror(r, "semaphore_wait");
-		return 0;
+		g->m->scalararg[0] = 0;
+		return;
 	}
 	while((r = runtime·mach_semaphore_wait(g->m->waitsema)) != 0) {
 		if(r == KERN_ABORTED)	// interrupted
 			continue;
 		macherror(r, "semaphore_wait");
 	}
-	return 0;
+	g->m->scalararg[0] = 0;
+	return;
 }
 
+#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 int32 mach_semrelease_errno;
+
+static void
+mach_semrelease_fail(void)
+{
+	macherror(mach_semrelease_errno, "semaphore_signal");
+}
+
+#pragma textflag NOSPLIT
 void
 runtime·mach_semrelease(uint32 sem)
 {
 	int32 r;
+	void (*fn)(void);
 
 	while((r = runtime·mach_semaphore_signal(sem)) != 0) {
 		if(r == KERN_ABORTED)	// interrupted
 			continue;
-		macherror(r, "semaphore_signal");
+		
+		// mach_semrelease must be completely nosplit,
+		// because it is called from Go code.
+		// If we're going to die, start that process on the m stack
+		// to avoid a Go stack split.
+		// Only do that if we're actually running on the g stack.
+		// We might be on the gsignal stack, and if so, onM will abort.
+		// We use the global variable instead of scalararg because
+		// we might be on the gsignal stack, having interrupted a
+		// normal call to onM. It doesn't quite matter, since the
+		// program is about to die, but better to be clean.
+		mach_semrelease_errno = r;
+		fn = mach_semrelease_fail;
+		if(g == g->m->curg)
+			runtime·onM(&fn);
+		else
+			fn();
 	}
 }
 
diff --git a/src/pkg/runtime/os_nacl.c b/src/pkg/runtime/os_nacl.c
index b3e0fc663..f859dd0d3 100644
--- a/src/pkg/runtime/os_nacl.c
+++ b/src/pkg/runtime/os_nacl.c
@@ -196,12 +196,6 @@ runtime·semawakeup(M *mp)
 	runtime·nacl_mutex_unlock(mp->waitsemalock);
 }
 
-void
-os·sigpipe(void)
-{
-	runtime·throw("too many writes on closed pipe");
-}
-
 uintptr
 runtime·memlimit(void)
 {
diff --git a/src/pkg/runtime/os_nacl.go b/src/pkg/runtime/os_nacl.go
index 2ab51b8b1..12a15aea0 100644
--- a/src/pkg/runtime/os_nacl.go
+++ b/src/pkg/runtime/os_nacl.go
@@ -24,3 +24,7 @@ func nacl_thread_create(fn, stk, tls, xx unsafe.Pointer) int32
 func nacl_nanosleep(ts, extra unsafe.Pointer) int32
 
 const stackSystem = 0
+
+func os_sigpipe() {
+	gothrow("too many writes on closed pipe")
+}
diff --git a/src/pkg/runtime/os_windows.c b/src/pkg/runtime/os_windows.c
index 172fd9228..0e53d8a31 100644
--- a/src/pkg/runtime/os_windows.c
+++ b/src/pkg/runtime/os_windows.c
@@ -588,12 +588,6 @@ runtime·resetcpuprofiler(int32 hz)
 	runtime·atomicstore((uint32*)&g->m->profilehz, hz);
 }
 
-void
-os·sigpipe(void)
-{
-	runtime·throw("too many writes on closed pipe");
-}
-
 uintptr
 runtime·memlimit(void)
 {
diff --git a/src/pkg/runtime/os_windows.go b/src/pkg/runtime/os_windows.go
index 57bd431f7..15957b329 100644
--- a/src/pkg/runtime/os_windows.go
+++ b/src/pkg/runtime/os_windows.go
@@ -23,3 +23,7 @@ func setlasterror(err uint32)
 func usleep1(usec uint32)
 
 const stackSystem = 512 * ptrSize
+
+func os_sigpipe() {
+	gothrow("too many writes on closed pipe")
+}
diff --git a/src/pkg/runtime/proc.c b/src/pkg/runtime/proc.c
index dfbc6142d..3e7b07c2c 100644
--- a/src/pkg/runtime/proc.c
+++ b/src/pkg/runtime/proc.c
@@ -761,10 +761,7 @@ runtime·stoptheworld(void)
 	// that is blocked trying to acquire the lock.
 	if(g->m->locks > 0)
 		runtime·throw("stoptheworld: holding locks");
-	// There is no evidence that stoptheworld on g0 does not work,
-	// we just don't do it today.
-	if(g == g->m->g0)
-		runtime·throw("stoptheworld: on g0");
+
 	runtime·lock(&runtime·sched.lock);
 	runtime·sched.stopwait = runtime·gomaxprocs;
 	runtime·atomicstore((uint32*)&runtime·sched.gcwaiting, 1);
@@ -2085,42 +2082,65 @@ exitsyscall0(G *gp)
 	schedule();  // Never returns.
 }
 
-// Called from syscall package before fork.
-#pragma textflag NOSPLIT
-void
-syscall·runtime_BeforeFork(void)
+static void
+beforefork(void)
 {
+	G *gp;
+	
+	gp = g->m->curg;
 	// Fork can hang if preempted with signals frequently enough (see issue 5517).
 	// Ensure that we stay on the same M where we disable profiling.
-	g->m->locks++;
-	if(g->m->profilehz != 0)
+	gp->m->locks++;
+	if(gp->m->profilehz != 0)
 		runtime·resetcpuprofiler(0);
 
 	// This function is called before fork in syscall package.
 	// Code between fork and exec must not allocate memory nor even try to grow stack.
 	// Here we spoil g->stackguard to reliably detect any attempts to grow stack.
 	// runtime_AfterFork will undo this in parent process, but not in child.
-	g->m->forkstackguard = g->stackguard;
-	g->stackguard0 = StackPreempt-1;
-	g->stackguard = StackPreempt-1;
+	gp->m->forkstackguard = gp->stackguard;
+	gp->stackguard0 = StackPreempt-1;
+	gp->stackguard = StackPreempt-1;
 }
 
-// Called from syscall package after fork in parent.
+// Called from syscall package before fork.
 #pragma textflag NOSPLIT
 void
-syscall·runtime_AfterFork(void)
+syscall·runtime_BeforeFork(void)
 {
-	int32 hz;
+	void (*fn)(void);
+	
+	fn = beforefork;
+	runtime·onM(&fn);
+}
 
+static void
+afterfork(void)
+{
+	int32 hz;
+	G *gp;
+	
+	gp = g->m->curg;
 	// See the comment in runtime_BeforeFork.
-	g->stackguard0 = g->m->forkstackguard;
-	g->stackguard = g->m->forkstackguard;
-	g->m->forkstackguard = 0;
+	gp->stackguard0 = gp->m->forkstackguard;
+	gp->stackguard = gp->m->forkstackguard;
+	gp->m->forkstackguard = 0;
 
 	hz = runtime·sched.profilehz;
 	if(hz != 0)
 		runtime·resetcpuprofiler(hz);
-	g->m->locks--;
+	gp->m->locks--;
+}
+
+// Called from syscall package after fork in parent.
+#pragma textflag NOSPLIT
+void
+syscall·runtime_AfterFork(void)
+{
+	void (*fn)(void);
+	
+	fn = afterfork;
+	runtime·onM(&fn);
 }
 
 // Hook used by runtime·malg to call runtime·stackalloc on the
@@ -2453,10 +2473,13 @@ runtime·Breakpoint(void)
 
 // Implementation of runtime.GOMAXPROCS.
 // delete when scheduler is even stronger
-int32
-runtime·gomaxprocsfunc(int32 n)
+void
+runtime·gomaxprocs_m(void)
 {
-	int32 ret;
+	int32 n, ret;
+	
+	n = g->m->scalararg[0];
+	g->m->scalararg[0] = 0;
 
 	if(n > MaxGomaxprocs)
 		n = MaxGomaxprocs;
@@ -2464,7 +2487,8 @@ runtime·gomaxprocsfunc(int32 n)
 	ret = runtime·gomaxprocs;
 	if(n <= 0 || n == ret) {
 		runtime·unlock(&runtime·sched.lock);
-		return ret;
+		g->m->scalararg[0] = ret;
+		return;
 	}
 	runtime·unlock(&runtime·sched.lock);
 
@@ -2476,7 +2500,8 @@ runtime·gomaxprocsfunc(int32 n)
 	runtime·semrelease(&runtime·worldsema);
 	runtime·starttheworld();
 
-	return ret;
+	g->m->scalararg[0] = ret;
+	return;
 }
 
 // lockOSThread is called by runtime.LockOSThread and runtime.lockOSThread below
@@ -2540,6 +2565,7 @@ runtime·lockedOSThread(void)
 	return g->lockedm != nil && g->m->lockedg != nil;
 }
 
+#pragma textflag NOSPLIT
 int32
 runtime·gcount(void)
 {
@@ -2737,8 +2763,13 @@ runtime·sigprof(uint8 *pc, uint8 *sp, uint8 *lr, G *gp, M *mp)
 
 // Arrange to call fn with a traceback hz times a second.
 void
-runtime·setcpuprofilerate(int32 hz)
+runtime·setcpuprofilerate_m(void)
 {
+	int32 hz;
+	
+	hz = g->m->scalararg[0];
+	g->m->scalararg[0] = 0;
+
 	// Force sane arguments.
 	if(hz < 0)
 		hz = 0;
diff --git a/src/pkg/runtime/runtime.c b/src/pkg/runtime/runtime.c
index b0adfb601..751181274 100644
--- a/src/pkg/runtime/runtime.c
+++ b/src/pkg/runtime/runtime.c
@@ -20,6 +20,7 @@ static uint32 traceback_cache = 2<<1;
 //	GOTRACEBACK=1   default behavior - show tracebacks but exclude runtime frames
 //	GOTRACEBACK=2   show tracebacks including runtime frames
 //	GOTRACEBACK=crash   show tracebacks including runtime frames, then crash (core dump etc)
+#pragma textflag NOSPLIT
 int32
 runtime·gotraceback(bool *crash)
 {
@@ -266,37 +267,6 @@ runtime·check(void)
 		runtime·throw("FixedStack is not power-of-2");
 }
 
-static Mutex ticksLock;
-static int64 ticks;
-
-// Note: Called by runtime/pprof in addition to runtime code.
-int64
-runtime·tickspersecond(void)
-{
-	int64 res, t0, t1, c0, c1;
-
-	res = (int64)runtime·atomicload64((uint64*)&ticks);
-	if(res != 0)
-		return ticks;
-	runtime·lock(&ticksLock);
-	res = ticks;
-	if(res == 0) {
-		t0 = runtime·nanotime();
-		c0 = runtime·cputicks();
-		runtime·usleep(100*1000);
-		t1 = runtime·nanotime();
-		c1 = runtime·cputicks();
-		if(t1 == t0)
-			t1++;
-		res = (c1-c0)*1000*1000*1000/(t1-t0);
-		if(res == 0)
-			res++;
-		runtime·atomicstore64((uint64*)&ticks, res);
-	}
-	runtime·unlock(&ticksLock);
-	return res;
-}
-
 #pragma dataflag NOPTR
 DebugVars	runtime·debug;
 
diff --git a/src/pkg/runtime/runtime.go b/src/pkg/runtime/runtime.go
new file mode 100644
index 000000000..d5b31559a
--- /dev/null
+++ b/src/pkg/runtime/runtime.go
@@ -0,0 +1,37 @@
+// 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 runtime
+
+var ticks struct {
+	lock mutex
+	val  uint64
+}
+
+// Note: Called by runtime/pprof in addition to runtime code.
+func tickspersecond() int64 {
+	r := int64(atomicload64(&ticks.val))
+	if r != 0 {
+		return r
+	}
+	lock(&ticks.lock)
+	r = int64(ticks.val)
+	if r == 0 {
+		t0 := nanotime()
+		c0 := cputicks()
+		usleep(100 * 1000)
+		t1 := nanotime()
+		c1 := cputicks()
+		if t1 == t0 {
+			t1++
+		}
+		r = (c1 - c0) * 1000 * 1000 * 1000 / (t1 - t0)
+		if r == 0 {
+			r++
+		}
+		atomicstore64(&ticks.val, uint64(r))
+	}
+	unlock(&ticks.lock)
+	return r
+}
diff --git a/src/pkg/runtime/signal_unix.c b/src/pkg/runtime/signal_unix.c
index 4d582d3f8..0e33ece49 100644
--- a/src/pkg/runtime/signal_unix.c
+++ b/src/pkg/runtime/signal_unix.c
@@ -93,7 +93,7 @@ runtime·resetcpuprofiler(int32 hz)
 }
 
 void
-os·sigpipe(void)
+runtime·sigpipe(void)
 {
 	runtime·setsig(SIGPIPE, SIG_DFL, false);
 	runtime·raise(SIGPIPE);
diff --git a/src/pkg/runtime/signal_unix.go b/src/pkg/runtime/signal_unix.go
new file mode 100644
index 000000000..ba77b6e7b
--- /dev/null
+++ b/src/pkg/runtime/signal_unix.go
@@ -0,0 +1,13 @@
+// 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.
+
+// +build darwin dragonfly freebsd linux netbsd openbsd solaris
+
+package runtime
+
+func sigpipe()
+
+func os_sigpipe() {
+	onM(sigpipe)
+}
diff --git a/src/pkg/runtime/stack_test.go b/src/pkg/runtime/stack_test.go
index b3dcbd12a..a822d73db 100644
--- a/src/pkg/runtime/stack_test.go
+++ b/src/pkg/runtime/stack_test.go
@@ -341,3 +341,12 @@ func TestStackOutput(t *testing.T) {
 		t.Errorf("Stack output should begin with \"goroutine \"")
 	}
 }
+
+func TestStackAllOutput(t *testing.T) {
+	b := make([]byte, 1024)
+	stk := string(b[:Stack(b, true)])
+	if !strings.HasPrefix(stk, "goroutine ") {
+		t.Errorf("Stack (len %d):\n%s", len(stk), stk)
+		t.Errorf("Stack output should begin with \"goroutine \"")
+	}
+}
diff --git a/src/pkg/runtime/stubs.go b/src/pkg/runtime/stubs.go
index 287b3df05..3719c7501 100644
--- a/src/pkg/runtime/stubs.go
+++ b/src/pkg/runtime/stubs.go
@@ -202,7 +202,6 @@ func reflectcall(fn, arg unsafe.Pointer, n uint32, retoffset uint32)
 func procyield(cycles uint32)
 func osyield()
 func cgocallback_gofunc(fv *funcval, frame unsafe.Pointer, framesize uintptr)
-func persistentalloc(size, align uintptr, stat *uint64) unsafe.Pointer
 func readgogc() int32
 func purgecachedstats(c *mcache)
 func gostringnocopy(b *byte) string
diff --git a/src/pkg/runtime/thunk.s b/src/pkg/runtime/thunk.s
index 048b7a723..0b5963e70 100644
--- a/src/pkg/runtime/thunk.s
+++ b/src/pkg/runtime/thunk.s
@@ -118,3 +118,6 @@ TEXT reflect·makechan(SB),NOSPLIT,$0-0
 
 TEXT reflect·rselect(SB), NOSPLIT, $0-0
 	JMP	runtime·reflect_rselect(SB)
+
+TEXT os·sigpipe(SB), NOSPLIT, $0-0
+	JMP	runtime·os_sigpipe(SB)