runtime: increase stack frame during cgo call on windows/amd64

Fixes #3945.

R=golang-dev, minux.ma
CC=golang-dev, vcc.163
https://golang.org/cl/6490056

misc/cgo/test/cgo_test.go

@@ -29,5 +29,6 @@ func TestSetEnv(t *testing.T)              { testSetEnv(t) }
 func TestHelpers(t *testing.T)             { testHelpers(t) }
 func TestLibgcc(t *testing.T)              { testLibgcc(t) }
 func Test1635(t *testing.T)                { test1635(t) }
+func TestPrintf(t *testing.T)              { testPrintf(t) }
 
 func BenchmarkCgoCall(b *testing.B) { benchCgoCall(b) }

misc/cgo/test/issue3945.go

@@ -0,0 +1,22 @@
+// Copyright 2012 The Go Authors.  All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package cgotest
+
+// Test that cgo reserves enough stack space during cgo call.
+// See http://golang.org/issue/3945 for details.
+
+// #include <stdio.h>
+//
+// void say() {
+//    printf("%s from C\n", "hello");
+// }
+//
+import "C"
+
+import "testing"
+
+func testPrintf(t *testing.T) {
+	C.say()
+}

src/pkg/runtime/asm_amd64.s

@@ -489,19 +489,21 @@ TEXT runtime·asmcgocall(SB),7,$0
 	MOVQ	(g_sched+gobuf_sp)(SI), SP
 
 	// Now on a scheduling stack (a pthread-created stack).
-	SUBQ	$48, SP
+	// Make sure we have enough room for 4 stack-backed fast-call
+	// registers as per windows amd64 calling convention.
+	SUBQ	$64, SP
 	ANDQ	$~15, SP	// alignment for gcc ABI
-	MOVQ	DI, 32(SP)	// save g
-	MOVQ	DX, 24(SP)	// save SP
+	MOVQ	DI, 48(SP)	// save g
+	MOVQ	DX, 40(SP)	// save SP
 	MOVQ	BX, DI		// DI = first argument in AMD64 ABI
 	MOVQ	BX, CX		// CX = first argument in Win64
 	CALL	AX
 
 	// Restore registers, g, stack pointer.
 	get_tls(CX)
-	MOVQ	32(SP), DI
+	MOVQ	48(SP), DI
 	MOVQ	DI, g(CX)
-	MOVQ	24(SP), SP
+	MOVQ	40(SP), SP
 	RET
 
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize)