cgo: fixes callback for windows amd64

R=rsc
CC=golang-dev
https://golang.org/cl/4826041

misc/cgo/life/Makefile

@@ -12,6 +12,20 @@ CGOFILES=\
 CGO_OFILES=\
 	c-life.o\
 	
+ifeq ($(GOOS),windows)
+ifeq ($(GOARCH),amd64)
+CGO_OFILES+=\
+	lib64_libmingwex_a-wassert.o\
+	lib64_libmingw32_a-mingw_helpers.o\
+
+lib64_libmingwex_a-wassert.o:
+	ar -x /mingw/x86_64-w64-mingw32/lib/libmingwex.a lib64_libmingwex_a-wassert.o
+
+lib64_libmingw32_a-mingw_helpers.o:
+	ar -x /mingw/x86_64-w64-mingw32/lib/libmingw32.a  lib64_libmingw32_a-mingw_helpers.o
+endif
+endif
+
 CLEANFILES+=life
 
 include ../../../src/Make.pkg

src/pkg/runtime/amd64/asm.s

@@ -448,19 +448,19 @@ TEXT runtime·asmcgocall(SB),7,$0
 	MOVQ	(g_sched+gobuf_sp)(SI), SP
 
 	// Now on a scheduling stack (a pthread-created stack).
-	SUBQ	$32, SP
+	SUBQ	$48, SP
 	ANDQ	$~15, SP	// alignment for gcc ABI
-	MOVQ	DI, 16(SP)	// save g
-	MOVQ	DX, 8(SP)	// save SP
+	MOVQ	DI, 32(SP)	// save g
+	MOVQ	DX, 24(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	16(SP), DI
+	MOVQ	32(SP), DI
 	MOVQ	DI, g(CX)
-	MOVQ	8(SP), SP
+	MOVQ	24(SP), SP
 	RET
 
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize)

src/pkg/runtime/cgo/amd64.S

@@ -32,7 +32,11 @@ EXT(crosscall_amd64):
 	pushq %r14
 	pushq %r15
 
+#if defined(_WIN64)
+	call *%rcx	/* fn */
+#else
 	call *%rdi	/* fn */
+#endif
 
 	popq %r15
 	popq %r14
@@ -58,10 +62,21 @@ EXT(crosscall2):
 	movq  %r14, 0x30(%rsp)
 	movq  %r15, 0x38(%rsp)
 
+#if defined(_WIN64)
+	// Win64 save RBX, RBP, RDI, RSI, RSP, R12, R13, R14, and R15
+	movq	%rdi, 0x40(%rsp)
+	movq	%rsi, 0x48(%rsp)
+
+	movq  %rdx, 0(%rsp)	/* arg */
+	movq  %r8, 8(%rsp)	/* argsize (includes padding) */
+	
+	call %rcx	/* fn */
+#else
 	movq  %rsi, 0(%rsp)	/* arg */
 	movq  %rdx, 8(%rsp)	/* argsize (includes padding) */
 
 	call *%rdi	/* fn */
+#endif
 
 	movq  0x10(%rsp), %rbx
 	movq  0x18(%rsp), %rbp
@@ -69,5 +84,9 @@ EXT(crosscall2):
 	movq  0x28(%rsp), %r13
 	movq  0x30(%rsp), %r14
 	movq  0x38(%rsp), %r15
+#if defined(__WIN64)
+	movq	0x40(%rsp), %rdi
+	movq	0x48(%rsp), %rsi
+#endif	
 	addq  $0x58, %rsp
 	ret