runtime: reduce frame size for runtime.cgocallback_gofunc

Tying preemption to stack splits means that we have to able to
complete the call to exitsyscall (inside cgocallbackg at least for now)
without any stack split checks, meaning that the whole sequence
has to work within 128 bytes of stack, unless we increase the size
of the red zone. This CL frees up 24 bytes along that critical path
on amd64. (The 32-bit systems have plenty of space because all
their words are smaller.)

R=dvyukov
CC=golang-dev
https://codereview.appspot.com/11676043

1 files changed, 19 insertions, 4 deletions

diff --git a/src/pkg/runtime/cgocall.c b/src/pkg/runtime/cgocall.c
index 16bc76554..a624509cf 100644
--- a/src/pkg/runtime/cgocall.c
+++ b/src/pkg/runtime/cgocall.c
@@ -228,13 +228,25 @@ runtime·cfree(void *p)
 
 static FuncVal unwindmf = {unwindm};
 
+typedef struct CallbackArgs CallbackArgs;
+struct CallbackArgs
+{
+	FuncVal *fn;
+	void *arg;
+	uintptr argsize;
+};
+
+#define CBARGS (CallbackArgs*)((byte*)m->g0->sched.sp+(3+(thechar=='5'))*sizeof(void*))
+
 void
-runtime·cgocallbackg(FuncVal *fn, void *arg, uintptr argsize)
+runtime·cgocallbackg(void)
 {
 	Defer d;
+	CallbackArgs *cb;
 
 	if(m->racecall) {
-		reflect·call(fn, arg, argsize);
+		cb = CBARGS;
+		reflect·call(cb->fn, cb->arg, cb->argsize);
 		return;
 	}
 
@@ -261,7 +273,8 @@ runtime·cgocallbackg(FuncVal *fn, void *arg, uintptr argsize)
 		runtime·raceacquire(&cgosync);
 
 	// Invoke callback.
-	reflect·call(fn, arg, argsize);
+	cb = CBARGS;
+	reflect·call(cb->fn, cb->arg, cb->argsize);
 
 	if(raceenabled)
 		runtime·racereleasemerge(&cgosync);
@@ -286,9 +299,11 @@ unwindm(void)
 		runtime·throw("runtime: unwindm not implemented");
 	case '8':
 	case '6':
-	case '5':
 		m->g0->sched.sp = *(uintptr*)m->g0->sched.sp;
 		break;
+	case '5':
+		m->g0->sched.sp = *(uintptr*)((byte*)m->g0->sched.sp + 4);
+		break;
 	}
 }