runtime: take the callback return value from the stack

R=brainman, lxn, rsc
CC=golang-dev
https://golang.org/cl/4126056

src/pkg/runtime/cgocall.c

@@ -53,13 +53,12 @@ runtime·cgocall(void (*fn)(void*), void *arg)
 // (arg/argsize) on to the stack, calls the function, copies the
 // arguments back where they came from, and finally returns to the old
 // stack.
-uintptr
+void
 runtime·cgocallback(void (*fn)(void), void *arg, int32 argsize)
 {
 	Gobuf oldsched, oldg1sched;
 	G *g1;
 	void *sp;
-	uintptr ret;
 
 	if(g != m->g0)
 		runtime·throw("bad g in cgocallback");
@@ -71,11 +70,11 @@ runtime·cgocallback(void (*fn)(void), void *arg, int32 argsize)
 	runtime·startcgocallback(g1);
 
 	sp = g1->sched.sp - argsize;
-	if(sp < g1->stackguard - StackGuard + 4) // +4 for return address
+	if(sp < g1->stackguard - StackGuard + 8) // +8 for return address
 		runtime·throw("g stack overflow in cgocallback");
 	runtime·mcpy(sp, arg, argsize);
 
-	ret = runtime·runcgocallback(g1, sp, fn);
+	runtime·runcgocallback(g1, sp, fn);
 
 	runtime·mcpy(arg, sp, argsize);
 
@@ -83,8 +82,6 @@ runtime·cgocallback(void (*fn)(void), void *arg, int32 argsize)
 
 	m->sched = oldsched;
 	g1->sched = oldg1sched;
-
-	return ret;
 }
 
 void

src/pkg/runtime/cgocall.h

@@ -7,6 +7,6 @@
  */
 
 void runtime·cgocall(void (*fn)(void*), void*);
-uintptr runtime·cgocallback(void (*fn)(void), void*, int32);
+void runtime·cgocallback(void (*fn)(void), void*, int32);
 void *runtime·cmalloc(uintptr);
 void runtime·cfree(void*);

src/pkg/runtime/runtime.h

@@ -443,7 +443,7 @@ void	runtime·breakpoint(void);
 void	runtime·gosched(void);
 void	runtime·goexit(void);
 void	runtime·runcgo(void (*fn)(void*), void*);
-uintptr	runtime·runcgocallback(G*, void*, void (*fn)());
+void	runtime·runcgocallback(G*, void*, void (*fn)());
 void	runtime·entersyscall(void);
 void	runtime·exitsyscall(void);
 void	runtime·startcgocallback(G*);

src/pkg/runtime/windows/386/sys.s

@@ -107,7 +107,11 @@ sigdone:
 // DX = total size of arguments
 //
 TEXT runtime·callbackasm+0(SB),7,$0
+	// preserve whatever's at the memory location that
+	// the callback will use to store the return value
 	LEAL	8(SP), CX
+	PUSHL	0(CX)(DX*1)
+	ADDL	$4, DX			// extend argsize by size of return value
 
 	// save registers as required for windows callback
 	PUSHL	0(FS)
@@ -129,7 +133,7 @@ TEXT runtime·callbackasm+0(SB),7,$0
 	CALL	runtime·cgocallback(SB)
 
 	// restore registers as required for windows callback
-	POPL	CX
+	POPL	AX
 	POPL	CX
 	POPL	DX
 	POPL	BX
@@ -139,6 +143,8 @@ TEXT runtime·callbackasm+0(SB),7,$0
 	POPL	0(FS)
 	CLD
 
+	MOVL	-4(CX)(DX*1), AX
+	POPL	-4(CX)(DX*1)
 	RET
 
 // void tstart(M *newm);