diff --git a/src/pkg/runtime/lock_futex.c b/src/pkg/runtime/lock_futex.c
index 5309a21a13a..95d590bae96 100644
--- a/src/pkg/runtime/lock_futex.c
+++ b/src/pkg/runtime/lock_futex.c
@@ -5,6 +5,7 @@
 // +build freebsd linux
 
 #include "runtime.h"
+#include "stack.h"
 
 // This implementation depends on OS-specific implementations of
 //
@@ -99,6 +100,8 @@ runtime·unlock(Lock *l)
 
 	if(--m->locks < 0)
 		runtime·throw("runtime·unlock: lock count");
+	if(m->locks == 0 && g->preempt)  // restore the preemption request in case we've cleared it in newstack
+		g->stackguard0 = StackPreempt;
 }
 
 // One-time notifications.
diff --git a/src/pkg/runtime/lock_sema.c b/src/pkg/runtime/lock_sema.c
index be4d306d1f4..069b8c1ad38 100644
--- a/src/pkg/runtime/lock_sema.c
+++ b/src/pkg/runtime/lock_sema.c
@@ -5,6 +5,7 @@
 // +build darwin netbsd openbsd plan9 windows
 
 #include "runtime.h"
+#include "stack.h"
 
 // This implementation depends on OS-specific implementations of
 //
@@ -112,6 +113,8 @@ runtime·unlock(Lock *l)
 
 	if(--m->locks < 0)
 		runtime·throw("runtime·unlock: lock count");
+	if(m->locks == 0 && g->preempt)  // restore the preemption request in case we've cleared it in newstack
+		g->stackguard0 = StackPreempt;
 }
 
 // One-time notifications.
diff --git a/src/pkg/runtime/malloc.goc b/src/pkg/runtime/malloc.goc
index 0347b90c914..78535c61f0f 100644
--- a/src/pkg/runtime/malloc.goc
+++ b/src/pkg/runtime/malloc.goc
@@ -13,6 +13,7 @@ package runtime
 #include "type.h"
 #include "typekind.h"
 #include "race.h"
+#include "stack.h"
 
 MHeap runtime·mheap;
 
@@ -94,6 +95,8 @@ runtime·mallocgc(uintptr size, uint32 flag, int32 dogc, int32 zeroed)
 		*(uintptr*)((uintptr)v+size-sizeof(uintptr)) = 0;
 
 	m->mallocing = 0;
+	if(g->preempt)  // restore the preemption request in case we've cleared it in newstack
+		g->stackguard0 = StackPreempt;
 
 	if(!(flag & FlagNoProfiling) && (rate = runtime·MemProfileRate) > 0) {
 		if(size >= rate)
diff --git a/src/pkg/runtime/mgc0.c b/src/pkg/runtime/mgc0.c
index 1349aa7726f..44262abbba9 100644
--- a/src/pkg/runtime/mgc0.c
+++ b/src/pkg/runtime/mgc0.c
@@ -2014,6 +2014,7 @@ runtime·gc(int32 force)
 	}
 
 	// all done
+	m->gcing = 0;
 	runtime·semrelease(&runtime·worldsema);
 	runtime·starttheworld();
 
@@ -2031,6 +2032,8 @@ runtime·gc(int32 force)
 		// give the queued finalizers, if any, a chance to run
 		runtime·gosched();
 	}
+	if(g->preempt)  // restore the preemption request in case we've cleared it in newstack
+		g->stackguard0 = StackPreempt;
 }
 
 static void
@@ -2115,7 +2118,6 @@ gc(struct gc_args *args)
 
 	cachestats();
 	mstats.next_gc = mstats.heap_alloc+mstats.heap_alloc*gcpercent/100;
-	m->gcing = 0;
 
 	t4 = runtime·nanotime();
 	mstats.last_gc = t4;
diff --git a/src/pkg/runtime/proc.c b/src/pkg/runtime/proc.c
index fffd04b7fb0..3ce281fc775 100644
--- a/src/pkg/runtime/proc.c
+++ b/src/pkg/runtime/proc.c
@@ -294,6 +294,8 @@ runtime·ready(G *gp)
 	if(runtime·atomicload(&runtime·sched.npidle) != 0 && runtime·atomicload(&runtime·sched.nmspinning) == 0)  // TODO: fast atomic
 		wakep();
 	m->locks--;
+	if(m->locks == 0 && g->preempt)  // restore the preemption request in case we've cleared it in newstack
+		g->stackguard0 = StackPreempt;
 }
 
 int32
@@ -475,6 +477,8 @@ runtime·starttheworld(void)
 		newm(mhelpgc, nil);
 	}
 	m->locks--;
+	if(m->locks == 0 && g->preempt)  // restore the preemption request in case we've cleared it in newstack
+		g->stackguard0 = StackPreempt;
 }
 
 // Called to start an M.
@@ -564,6 +568,8 @@ runtime·allocm(P *p)
 	if(p == m->p)
 		releasep();
 	m->locks--;
+	if(m->locks == 0 && g->preempt)  // restore the preemption request in case we've cleared it in newstack
+		g->stackguard0 = StackPreempt;
 
 	return mp;
 }
@@ -1008,6 +1014,7 @@ execute(G *gp)
 		runtime·throw("execute: bad g status");
 	}
 	gp->status = Grunning;
+	gp->preempt = false;
 	gp->stackguard0 = gp->stackguard;
 	m->p->tick++;
 	m->curg = gp;
@@ -1433,6 +1440,8 @@ runtime·exitsyscall(void)
 		// so okay to clear gcstack and gcsp.
 		g->gcstack = (uintptr)nil;
 		g->gcsp = (uintptr)nil;
+		if(g->preempt)  // restore the preemption request in case we've cleared it in newstack
+			g->stackguard0 = StackPreempt;
 		return;
 	}
 
@@ -1450,6 +1459,8 @@ runtime·exitsyscall(void)
 			g->status = Grunning;
 			g->gcstack = (uintptr)nil;
 			g->gcsp = (uintptr)nil;
+			if(g->preempt)  // restore the preemption request in case we've cleared it in newstack
+				g->stackguard0 = StackPreempt;
 			return;
 		}
 	}
@@ -1620,6 +1631,8 @@ runtime·newproc1(FuncVal *fn, byte *argp, int32 narg, int32 nret, void *callerp
 	if(runtime·atomicload(&runtime·sched.npidle) != 0 && runtime·atomicload(&runtime·sched.nmspinning) == 0 && fn->fn != runtime·main)  // TODO: fast atomic
 		wakep();
 	m->locks--;
+	if(m->locks == 0 && g->preempt)  // restore the preemption request in case we've cleared it in newstack
+		g->stackguard0 = StackPreempt;
 	return newg;
 }
 
@@ -2174,6 +2187,7 @@ if(1) return;
 	gp = mp->curg;
 	if(gp == nil || gp == mp->g0)
 		return;
+	gp->preempt = true;
 	gp->stackguard0 = StackPreempt;
 }
 
diff --git a/src/pkg/runtime/runtime.h b/src/pkg/runtime/runtime.h
index ce451b0105a..34d9541e984 100644
--- a/src/pkg/runtime/runtime.h
+++ b/src/pkg/runtime/runtime.h
@@ -253,6 +253,7 @@ struct	G
 	bool	issystem;	// do not output in stack dump
 	bool	isbackground;	// ignore in deadlock detector
 	bool	blockingsyscall;	// hint that the next syscall will block
+	bool	preempt;	// preemption signal, duplicates stackguard0 = StackPreempt
 	int8	raceignore;	// ignore race detection events
 	M*	m;		// for debuggers, but offset not hard-coded
 	M*	lockedm;
diff --git a/src/pkg/runtime/stack.c b/src/pkg/runtime/stack.c
index 9de692bba43..2ba29956b1e 100644
--- a/src/pkg/runtime/stack.c
+++ b/src/pkg/runtime/stack.c
@@ -250,7 +250,7 @@ runtime·newstack(void)
 		// We are interested in preempting user Go code, not runtime code.
 		if(oldstatus != Grunning || m->locks || m->mallocing || m->gcing) {
 			// Let the goroutine keep running for now.
-			// TODO(dvyukov): remember but delay the preemption.
+			// gp->preempt is set, so it will be preempted next time.
 			gp->stackguard0 = gp->stackguard;
 			gp->status = oldstatus;
 			runtime·gogo(&gp->sched);	// never return