diff --git a/src/pkg/runtime/asm_386.s b/src/pkg/runtime/asm_386.s
index 1e47274278..05b929f340 100644
--- a/src/pkg/runtime/asm_386.s
+++ b/src/pkg/runtime/asm_386.s
@@ -498,8 +498,22 @@ TEXT runtime·asmcgocall(SB),7,$0
 	RET
 
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
-// See cgocall.c for more details.
+// Turn the fn into a Go func (by taking its address) and call
+// cgocallback_gofunc.
 TEXT runtime·cgocallback(SB),7,$12
+	LEAL	fn+0(FP), AX
+	MOVL	AX, 0(SP)
+	MOVL	frame+4(FP), AX
+	MOVL	AX, 4(SP)
+	MOVL	framesize+8(FP), AX
+	MOVL	AX, 8(SP)
+	MOVL	$runtime·cgocallback_gofunc(SB), AX
+	CALL	AX
+	RET
+
+// cgocallback_gofunc(FuncVal*, void *frame, uintptr framesize)
+// See cgocall.c for more details.
+TEXT runtime·cgocallback_gofunc(SB),7,$12
 	// If m is nil, Go did not create the current thread.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
diff --git a/src/pkg/runtime/asm_amd64.s b/src/pkg/runtime/asm_amd64.s
index ae8470708b..9591437eec 100644
--- a/src/pkg/runtime/asm_amd64.s
+++ b/src/pkg/runtime/asm_amd64.s
@@ -531,8 +531,22 @@ TEXT runtime·asmcgocall(SB),7,$0
 	RET
 
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
-// See cgocall.c for more details.
+// Turn the fn into a Go func (by taking its address) and call
+// cgocallback_gofunc.
 TEXT runtime·cgocallback(SB),7,$24
+	LEAQ	fn+0(FP), AX
+	MOVQ	AX, 0(SP)
+	MOVQ	frame+8(FP), AX
+	MOVQ	AX, 8(SP)
+	MOVQ	framesize+16(FP), AX
+	MOVQ	AX, 16(SP)
+	MOVQ	$runtime·cgocallback_gofunc(SB), AX
+	CALL	AX
+	RET
+
+// cgocallback_gofunc(FuncVal*, void *frame, uintptr framesize)
+// See cgocall.c for more details.
+TEXT runtime·cgocallback_gofunc(SB),7,$24
 	// If m is nil, Go did not create the current thread.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
diff --git a/src/pkg/runtime/asm_arm.s b/src/pkg/runtime/asm_arm.s
index a10c91c654..35fb537331 100644
--- a/src/pkg/runtime/asm_arm.s
+++ b/src/pkg/runtime/asm_arm.s
@@ -310,8 +310,22 @@ TEXT	runtime·asmcgocall(SB),7,$0
 	RET
 
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
+// Turn the fn into a Go func (by taking its address) and call
+// cgocallback_gofunc.
+TEXT runtime·cgocallback(SB),7,$12
+	MOVW	$fn+0(FP), R0
+	MOVW	R0, 4(R13)
+	MOVW	frame+4(FP), R0
+	MOVW	R0, 8(R13)
+	MOVW	framesize+8(FP), R0
+	MOVW	R0, 12(R13)
+	MOVL	$runtime·cgocallback_gofunc(SB), R0
+	BL	(R0)
+	RET
+
+// cgocallback_gofunc(void (*fn)(void*), void *frame, uintptr framesize)
 // See cgocall.c for more details.
-TEXT	runtime·cgocallback(SB),7,$16
+TEXT	runtime·cgocallback_gofunc(SB),7,$16
 	// Load m and g from thread-local storage.
 	MOVW	cgo_load_gm(SB), R0
 	CMP	$0, R0
diff --git a/src/pkg/runtime/callback_windows_386.c b/src/pkg/runtime/callback_windows_386.c
index 159b8508e2..880588da63 100644
--- a/src/pkg/runtime/callback_windows_386.c
+++ b/src/pkg/runtime/callback_windows_386.c
@@ -80,7 +80,7 @@ runtime·compilecallback(Eface fn, bool cleanstack)
 
 	// MOVL fn, AX
 	*p++ = 0xb8;
-	*(uint32*)p = (uint32)(*(byte**)fn.data);
+	*(uint32*)p = (uint32)(fn.data);
 	p += 4;
 
 	// MOVL argsize, DX
diff --git a/src/pkg/runtime/callback_windows_amd64.c b/src/pkg/runtime/callback_windows_amd64.c
index 03a4cef136..1a47792915 100644
--- a/src/pkg/runtime/callback_windows_amd64.c
+++ b/src/pkg/runtime/callback_windows_amd64.c
@@ -78,7 +78,7 @@ runtime·compilecallback(Eface fn, bool /*cleanstack*/)
 	// MOVQ fn, AX
 	*p++ = 0x48;
 	*p++ = 0xb8;
-	*(uint64*)p = (uint64)(*(byte**)fn.data);
+	*(uint64*)p = (uint64)(fn.data);
 	p += 8;
 	// PUSH AX
 	*p++ = 0x50;
diff --git a/src/pkg/runtime/cgocall.c b/src/pkg/runtime/cgocall.c
index f89ac4684f..7848437a23 100644
--- a/src/pkg/runtime/cgocall.c
+++ b/src/pkg/runtime/cgocall.c
@@ -206,14 +206,12 @@ runtime·cfree(void *p)
 static FuncVal unwindmf = {unwindm};
 
 void
-runtime·cgocallbackg(void (*fn)(void), void *arg, uintptr argsize)
+runtime·cgocallbackg(FuncVal *fn, void *arg, uintptr argsize)
 {
 	Defer d;
-	FuncVal fv;
 
-	fv.fn = fn;
 	if(m->racecall) {
-		reflect·call(&fv, arg, argsize);
+		reflect·call(fn, arg, argsize);
 		return;
 	}
 
@@ -240,7 +238,7 @@ runtime·cgocallbackg(void (*fn)(void), void *arg, uintptr argsize)
 		runtime·raceacquire(&cgosync);
 
 	// Invoke callback.
-	reflect·call(&fv, arg, argsize);
+	reflect·call(fn, arg, argsize);
 
 	if(raceenabled)
 		runtime·racereleasemerge(&cgosync);
diff --git a/src/pkg/runtime/sys_windows_386.s b/src/pkg/runtime/sys_windows_386.s
index dbc4352e2e..a4ac7463a3 100644
--- a/src/pkg/runtime/sys_windows_386.s
+++ b/src/pkg/runtime/sys_windows_386.s
@@ -216,7 +216,7 @@ TEXT runtime·callbackasm+0(SB),7,$0
 
 	CLD
 
-	CALL	runtime·cgocallback(SB)
+	CALL	runtime·cgocallback_gofunc(SB)
 
 	POPL	AX
 	POPL	CX
diff --git a/src/pkg/runtime/sys_windows_amd64.s b/src/pkg/runtime/sys_windows_amd64.s
index 33ec33640b..fe88f3b754 100644
--- a/src/pkg/runtime/sys_windows_amd64.s
+++ b/src/pkg/runtime/sys_windows_amd64.s
@@ -272,13 +272,13 @@ TEXT runtime·callbackasm(SB),7,$0
 	MOVQ	R15, 0(SP)
 
 	// prepare call stack.  use SUBQ to hide from stack frame checks
-	// cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
+	// cgocallback(Go func, void *frame, uintptr framesize)
 	SUBQ	$24, SP
 	MOVQ	DX, 16(SP)	// uintptr framesize
 	MOVQ	CX, 8(SP)   // void *frame
-	MOVQ	AX, 0(SP)    // void (*fn)(void*)
+	MOVQ	AX, 0(SP)    // Go func
 	CLD
-	CALL  runtime·cgocallback(SB)
+	CALL  runtime·cgocallback_gofunc(SB)
 	MOVQ	0(SP), AX
 	MOVQ	8(SP), CX
 	MOVQ	16(SP), DX