diff --git a/src/pkg/runtime/debug.go b/src/pkg/runtime/debug.go
index 6526f16a04..b802fc63f7 100644
--- a/src/pkg/runtime/debug.go
+++ b/src/pkg/runtime/debug.go
@@ -30,7 +30,7 @@ func NumCPU() int
 func NumCgoCall() int64
 
 // NumGoroutine returns the number of goroutines that currently exist.
-func NumGoroutine() int32
+func NumGoroutine() int
 
 // MemProfileRate controls the fraction of memory allocations
 // that are recorded and reported in the memory profile.
@@ -89,15 +89,14 @@ func (r *MemProfileRecord) Stack() []uintptr {
 // of calling MemProfile directly.
 func MemProfile(p []MemProfileRecord, inuseZero bool) (n int, ok bool)
 
-// A ThreadProfileRecord describes the execution stack that
-// caused a new thread to be created.
-type ThreadProfileRecord struct {
+// A StackRecord describes a single execution stack.
+type StackRecord struct {
 	Stack0 [32]uintptr // stack trace for this record; ends at first 0 entry
 }
 
 // Stack returns the stack trace associated with the record,
 // a prefix of r.Stack0.
-func (r *ThreadProfileRecord) Stack() []uintptr {
+func (r *StackRecord) Stack() []uintptr {
 	for i, v := range r.Stack0 {
 		if v == 0 {
 			return r.Stack0[0:i]
@@ -106,13 +105,21 @@ func (r *ThreadProfileRecord) Stack() []uintptr {
 	return r.Stack0[0:]
 }
 
-// ThreadProfile returns n, the number of records in the current thread profile.
-// If len(p) >= n, ThreadProfile copies the profile into p and returns n, true.
-// If len(p) < n, ThreadProfile does not change p and returns n, false.
+// ThreadCreateProfile returns n, the number of records in the thread creation profile.
+// If len(p) >= n, ThreadCreateProfile copies the profile into p and returns n, true.
+// If len(p) < n, ThreadCreateProfile does not change p and returns n, false.
 //
 // Most clients should use the runtime/pprof package instead
-// of calling ThreadProfile directly.
-func ThreadProfile(p []ThreadProfileRecord) (n int, ok bool)
+// of calling ThreadCreateProfile directly.
+func ThreadCreateProfile(p []StackRecord) (n int, ok bool)
+
+// GoroutineProfile returns n, the number of records in the active goroutine stack profile.
+// If len(p) >= n, GoroutineProfile copies the profile into p and returns n, true.
+// If len(p) < n, GoroutineProfile does not change p and returns n, false.
+//
+// Most clients should use the runtime/pprof package instead
+// of calling GoroutineProfile directly.
+func GoroutineProfile(p []StackRecord) (n int, ok bool)
 
 // CPUProfile returns the next chunk of binary CPU profiling stack trace data,
 // blocking until data is available.  If profiling is turned off and all the profile
@@ -130,3 +137,9 @@ func CPUProfile() []byte
 // the testing package's -test.cpuprofile flag instead of calling
 // SetCPUProfileRate directly.
 func SetCPUProfileRate(hz int)
+
+// Stack formats a stack trace of the calling goroutine into buf
+// and returns the number of bytes written to buf.
+// If all is true, Stack formats stack traces of all other goroutines
+// into buf after the trace for the current goroutine.
+func Stack(buf []byte, all bool) int
diff --git a/src/pkg/runtime/malloc.h b/src/pkg/runtime/malloc.h
index 5f03693f52..d846f6810b 100644
--- a/src/pkg/runtime/malloc.h
+++ b/src/pkg/runtime/malloc.h
@@ -411,17 +411,9 @@ enum
 
 void	runtime·MProf_Malloc(void*, uintptr);
 void	runtime·MProf_Free(void*, uintptr);
+void	runtime·MProf_GC(void);
 int32	runtime·helpgc(bool*);
 void	runtime·gchelper(void);
 
-// Malloc profiling settings.
-// Must match definition in extern.go.
-enum {
-	MProf_None = 0,
-	MProf_Sample = 1,
-	MProf_All = 2,
-};
-extern int32 runtime·malloc_profile;
-
 bool	runtime·getfinalizer(void *p, bool del, void (**fn)(void*), int32 *nret);
 void	runtime·walkfintab(void (*fn)(void*));
diff --git a/src/pkg/runtime/mgc0.c b/src/pkg/runtime/mgc0.c
index fd1babfd35..e043864c19 100644
--- a/src/pkg/runtime/mgc0.c
+++ b/src/pkg/runtime/mgc0.c
@@ -52,6 +52,21 @@ enum {
 
 #define bitMask (bitBlockBoundary | bitAllocated | bitMarked | bitSpecial)
 
+// Holding worldsema grants an M the right to try to stop the world.
+// The procedure is:
+//
+//	runtime·semacquire(&runtime·worldsema);
+//	m->gcing = 1;
+//	runtime·stoptheworld();
+//
+//	... do stuff ...
+//
+//	m->gcing = 0;
+//	runtime·semrelease(&runtime·worldsema);
+//	runtime·starttheworld();
+//
+uint32 runtime·worldsema = 1;
+
 // TODO: Make these per-M.
 static uint64 nhandoff;
 
@@ -816,11 +831,6 @@ runtime·gchelper(void)
 		runtime·notewakeup(&work.alldone);
 }
 
-// Semaphore, not Lock, so that the goroutine
-// reschedules when there is contention rather
-// than spinning.
-static uint32 gcsema = 1;
-
 // Initialized from $GOGC.  GOGC=off means no gc.
 //
 // Next gc is after we've allocated an extra amount of
@@ -903,9 +913,9 @@ runtime·gc(int32 force)
 	if(gcpercent < 0)
 		return;
 
-	runtime·semacquire(&gcsema);
+	runtime·semacquire(&runtime·worldsema);
 	if(!force && mstats.heap_alloc < mstats.next_gc) {
-		runtime·semrelease(&gcsema);
+		runtime·semrelease(&runtime·worldsema);
 		return;
 	}
 
@@ -981,8 +991,9 @@ runtime·gc(int32 force)
 			mstats.nmalloc, mstats.nfree,
 			nhandoff);
 	}
-
-	runtime·semrelease(&gcsema);
+	
+	runtime·MProf_GC();
+	runtime·semrelease(&runtime·worldsema);
 
 	// If we could have used another helper proc, start one now,
 	// in the hope that it will be available next time.
@@ -1004,17 +1015,17 @@ runtime·gc(int32 force)
 void
 runtime·ReadMemStats(MStats *stats)
 {
-	// Have to acquire gcsema to stop the world,
+	// Have to acquire worldsema to stop the world,
 	// because stoptheworld can only be used by
 	// one goroutine at a time, and there might be
 	// a pending garbage collection already calling it.
-	runtime·semacquire(&gcsema);
+	runtime·semacquire(&runtime·worldsema);
 	m->gcing = 1;
 	runtime·stoptheworld();
 	cachestats();
 	*stats = mstats;
 	m->gcing = 0;
-	runtime·semrelease(&gcsema);
+	runtime·semrelease(&runtime·worldsema);
 	runtime·starttheworld(false);
 }
 
diff --git a/src/pkg/runtime/mprof.goc b/src/pkg/runtime/mprof.goc
index 70e991b8bb..0bbce85836 100644
--- a/src/pkg/runtime/mprof.goc
+++ b/src/pkg/runtime/mprof.goc
@@ -26,6 +26,10 @@ struct Bucket
 	uintptr	frees;
 	uintptr	alloc_bytes;
 	uintptr	free_bytes;
+	uintptr	recent_allocs;  // since last gc
+	uintptr	recent_frees;
+	uintptr	recent_alloc_bytes;
+	uintptr	recent_free_bytes;
 	uintptr	hash;
 	uintptr	nstk;
 	uintptr	stk[1];
@@ -39,7 +43,7 @@ static uintptr bucketmem;
 
 // Return the bucket for stk[0:nstk], allocating new bucket if needed.
 static Bucket*
-stkbucket(uintptr *stk, int32 nstk)
+stkbucket(uintptr *stk, int32 nstk, bool alloc)
 {
 	int32 i;
 	uintptr h;
@@ -66,6 +70,9 @@ stkbucket(uintptr *stk, int32 nstk)
 		   runtime·mcmp((byte*)b->stk, (byte*)stk, nstk*sizeof stk[0]) == 0)
 			return b;
 
+	if(!alloc)
+		return nil;
+
 	b = runtime·mallocgc(sizeof *b + nstk*sizeof stk[0], FlagNoProfiling, 0, 1);
 	bucketmem += sizeof *b + nstk*sizeof stk[0];
 	runtime·memmove(b->stk, stk, nstk*sizeof stk[0]);
@@ -78,6 +85,26 @@ stkbucket(uintptr *stk, int32 nstk)
 	return b;
 }
 
+// Record that a gc just happened: all the 'recent' statistics are now real.
+void
+runtime·MProf_GC(void)
+{
+	Bucket *b;
+	
+	runtime·lock(&proflock);
+	for(b=buckets; b; b=b->allnext) {
+		b->allocs += b->recent_allocs;
+		b->frees += b->recent_frees;
+		b->alloc_bytes += b->recent_alloc_bytes;
+		b->free_bytes += b->recent_free_bytes;
+		b->recent_allocs = 0;
+		b->recent_frees = 0;
+		b->recent_alloc_bytes = 0;
+		b->recent_free_bytes = 0;
+	}
+	runtime·unlock(&proflock);
+}
+
 // Map from pointer to Bucket* that allocated it.
 // Three levels:
 //	Linked-list hash table for top N-20 bits.
@@ -198,9 +225,9 @@ runtime·MProf_Malloc(void *p, uintptr size)
 	m->nomemprof++;
 	nstk = runtime·callers(1, stk, 32);
 	runtime·lock(&proflock);
-	b = stkbucket(stk, nstk);
-	b->allocs++;
-	b->alloc_bytes += size;
+	b = stkbucket(stk, nstk, true);
+	b->recent_allocs++;
+	b->recent_alloc_bytes += size;
 	setaddrbucket((uintptr)p, b);
 	runtime·unlock(&proflock);
 	m->nomemprof--;
@@ -219,8 +246,8 @@ runtime·MProf_Free(void *p, uintptr size)
 	runtime·lock(&proflock);
 	b = getaddrbucket((uintptr)p);
 	if(b != nil) {
-		b->frees++;
-		b->free_bytes += size;
+		b->recent_frees++;
+		b->recent_free_bytes += size;
 	}
 	runtime·unlock(&proflock);
 	m->nomemprof--;
@@ -274,13 +301,13 @@ func MemProfile(p Slice, include_inuse_zero bool) (n int32, ok bool) {
 	runtime·unlock(&proflock);
 }
 
-// Must match ThreadProfileRecord in debug.go.
+// Must match StackRecord in debug.go.
 typedef struct TRecord TRecord;
 struct TRecord {
 	uintptr stk[32];
 };
 
-func ThreadProfile(p Slice) (n int32, ok bool) {
+func ThreadCreateProfile(p Slice) (n int32, ok bool) {
 	TRecord *r;
 	M *first, *m;
 	
@@ -298,3 +325,80 @@ func ThreadProfile(p Slice) (n int32, ok bool) {
 		}
 	}
 }
+
+func Stack(b Slice, all bool) (n int32) {
+	byte *pc, *sp;
+	
+	sp = runtime·getcallersp(&b);
+	pc = runtime·getcallerpc(&b);
+
+	if(all) {
+		runtime·semacquire(&runtime·worldsema);
+		m->gcing = 1;
+		runtime·stoptheworld();
+	}
+
+	if(b.len == 0)
+		n = 0;
+	else{
+		g->writebuf = (byte*)b.array;
+		g->writenbuf = b.len;
+		runtime·goroutineheader(g);
+		runtime·traceback(pc, sp, 0, g);
+		if(all)
+			runtime·tracebackothers(g);
+		n = b.len - g->writenbuf;
+		g->writebuf = nil;
+		g->writenbuf = 0;
+	}
+	
+	if(all) {
+		m->gcing = 0;
+		runtime·semrelease(&runtime·worldsema);
+		runtime·starttheworld(false);
+	}
+}
+
+static void
+saveg(byte *pc, byte *sp, G *g, TRecord *r)
+{
+	int32 n;
+	
+	n = runtime·gentraceback(pc, sp, 0, g, 0, r->stk, nelem(r->stk));
+	if(n < nelem(r->stk))
+		r->stk[n] = 0;
+}
+
+func GoroutineProfile(b Slice) (n int32, ok bool) {
+	byte *pc, *sp;
+	TRecord *r;
+	G *gp;
+	
+	sp = runtime·getcallersp(&b);
+	pc = runtime·getcallerpc(&b);
+	
+	ok = false;
+	n = runtime·gcount();
+	if(n <= b.len) {
+		runtime·semacquire(&runtime·worldsema);
+		m->gcing = 1;
+		runtime·stoptheworld();
+
+		n = runtime·gcount();
+		if(n <= b.len) {
+			ok = true;
+			r = (TRecord*)b.array;
+			saveg(pc, sp, g, r++);
+			for(gp = runtime·allg; gp != nil; gp = gp->alllink) {
+				if(gp == g || gp->status == Gdead)
+					continue;
+				saveg(gp->sched.pc, gp->sched.sp, gp, r++);
+			}
+		}
+	
+		m->gcing = 0;
+		runtime·semrelease(&runtime·worldsema);
+		runtime·starttheworld(false);
+	}
+}
+
diff --git a/src/pkg/runtime/print.c b/src/pkg/runtime/print.c
index 0d8caaf912..6702c3cde7 100644
--- a/src/pkg/runtime/print.c
+++ b/src/pkg/runtime/print.c
@@ -9,6 +9,26 @@
 
 static void vprintf(int8*, byte*);
 
+// write to goroutine-local buffer if diverting output,
+// or else standard error.
+static void
+gwrite(void *v, int32 n)
+{
+	if(g == nil || g->writebuf == nil) {
+		runtime·write(2, v, n);
+		return;
+	}
+	
+	if(g->writenbuf == 0)
+		return;
+	
+	if(n > g->writenbuf)
+		n = g->writenbuf;
+	runtime·memmove(g->writebuf, v, n);
+	g->writebuf += n;
+	g->writenbuf -= n;
+}
+
 void
 runtime·dump(byte *p, int32 n)
 {
@@ -29,7 +49,7 @@ runtime·dump(byte *p, int32 n)
 void
 runtime·prints(int8 *s)
 {
-	runtime·write(2, s, runtime·findnull((byte*)s));
+	gwrite(s, runtime·findnull((byte*)s));
 }
 
 #pragma textflag 7
@@ -59,7 +79,7 @@ vprintf(int8 *s, byte *base)
 		if(*p != '%')
 			continue;
 		if(p > lp)
-			runtime·write(2, lp, p-lp);
+			gwrite(lp, p-lp);
 		p++;
 		narg = 0;
 		switch(*p) {
@@ -150,7 +170,7 @@ vprintf(int8 *s, byte *base)
 		lp = p+1;
 	}
 	if(p > lp)
-		runtime·write(2, lp, p-lp);
+		gwrite(lp, p-lp);
 
 	//runtime·unlock(&debuglock);
 }
@@ -176,10 +196,10 @@ void
 runtime·printbool(bool v)
 {
 	if(v) {
-		runtime·write(2, (byte*)"true", 4);
+		gwrite((byte*)"true", 4);
 		return;
 	}
-	runtime·write(2, (byte*)"false", 5);
+	gwrite((byte*)"false", 5);
 }
 
 void
@@ -190,15 +210,15 @@ runtime·printfloat(float64 v)
 	float64 h;
 
 	if(runtime·isNaN(v)) {
-		runtime·write(2, "NaN", 3);
+		gwrite("NaN", 3);
 		return;
 	}
 	if(runtime·isInf(v, 1)) {
-		runtime·write(2, "+Inf", 4);
+		gwrite("+Inf", 4);
 		return;
 	}
 	if(runtime·isInf(v, -1)) {
-		runtime·write(2, "-Inf", 4);
+		gwrite("-Inf", 4);
 		return;
 	}
 
@@ -257,16 +277,16 @@ runtime·printfloat(float64 v)
 	buf[n+4] = (e/100) + '0';
 	buf[n+5] = (e/10)%10 + '0';
 	buf[n+6] = (e%10) + '0';
-	runtime·write(2, buf, n+7);
+	gwrite(buf, n+7);
 }
 
 void
 runtime·printcomplex(Complex128 v)
 {
-	runtime·write(2, "(", 1);
+	gwrite("(", 1);
 	runtime·printfloat(v.real);
 	runtime·printfloat(v.imag);
-	runtime·write(2, "i)", 2);
+	gwrite("i)", 2);
 }
 
 void
@@ -281,14 +301,14 @@ runtime·printuint(uint64 v)
 			break;
 		v = v/10;
 	}
-	runtime·write(2, buf+i, nelem(buf)-i);
+	gwrite(buf+i, nelem(buf)-i);
 }
 
 void
 runtime·printint(int64 v)
 {
 	if(v < 0) {
-		runtime·write(2, "-", 1);
+		gwrite("-", 1);
 		v = -v;
 	}
 	runtime·printuint(v);
@@ -308,7 +328,7 @@ runtime·printhex(uint64 v)
 		buf[--i] = '0';
 	buf[--i] = 'x';
 	buf[--i] = '0';
-	runtime·write(2, buf+i, nelem(buf)-i);
+	gwrite(buf+i, nelem(buf)-i);
 }
 
 void
@@ -323,23 +343,23 @@ runtime·printstring(String v)
 	extern uint32 runtime·maxstring;
 
 	if(v.len > runtime·maxstring) {
-		runtime·write(2, "[invalid string]", 16);
+		gwrite("[invalid string]", 16);
 		return;
 	}
 	if(v.len > 0)
-		runtime·write(2, v.str, v.len);
+		gwrite(v.str, v.len);
 }
 
 void
 runtime·printsp(void)
 {
-	runtime·write(2, " ", 1);
+	gwrite(" ", 1);
 }
 
 void
 runtime·printnl(void)
 {
-	runtime·write(2, "\n", 1);
+	gwrite("\n", 1);
 }
 
 void
diff --git a/src/pkg/runtime/proc.c b/src/pkg/runtime/proc.c
index d09f075327..d94bec8855 100644
--- a/src/pkg/runtime/proc.c
+++ b/src/pkg/runtime/proc.c
@@ -1664,6 +1664,12 @@ runtime·NumGoroutine(int32 ret)
 	FLUSH(&ret);
 }
 
+int32
+runtime·gcount(void)
+{
+	return runtime·sched.gcount;
+}
+
 int32
 runtime·mcount(void)
 {
diff --git a/src/pkg/runtime/runtime.h b/src/pkg/runtime/runtime.h
index 910f00c8a2..1f4407a093 100644
--- a/src/pkg/runtime/runtime.h
+++ b/src/pkg/runtime/runtime.h
@@ -191,6 +191,8 @@ struct	G
 	M*	lockedm;
 	M*	idlem;
 	int32	sig;
+	int32	writenbuf;
+	byte*	writebuf;
 	uintptr	sigcode0;
 	uintptr	sigcode1;
 	uintptr	sigpc;
@@ -545,6 +547,7 @@ bool	runtime·addfinalizer(void*, void(*fn)(void*), int32);
 void	runtime·runpanic(Panic*);
 void*	runtime·getcallersp(void*);
 int32	runtime·mcount(void);
+int32	runtime·gcount(void);
 void	runtime·mcall(void(*)(G*));
 uint32	runtime·fastrand1(void);
 
@@ -585,10 +588,9 @@ int64	runtime·cputicks(void);
 #pragma	varargck	type	"s"	uint8*
 #pragma	varargck	type	"S"	String
 
-// TODO(rsc): Remove. These are only temporary,
-// for the mark and sweep collector.
 void	runtime·stoptheworld(void);
 void	runtime·starttheworld(bool);
+extern uint32 runtime·worldsema;
 
 /*
  * mutual exclusion locks.  in the uncontended case,