runtime: use cgo runtime functions to call windows syscalls

R=rsc
CC=golang-dev, jp, vcc.163
https://golang.org/cl/4926042

src/pkg/runtime/cgocall.c

@@ -94,7 +94,7 @@ runtime·cgocall(void (*fn)(void*), void *arg)
 {
 	Defer d;
 
-	if(!runtime·iscgo)
+	if(!runtime·iscgo && !Windows)
 		runtime·throw("cgocall unavailable");
 
 	if(fn == 0)

src/pkg/runtime/mkasmh.sh

@@ -96,6 +96,7 @@ awk '
 /^aggr G$/ { aggr="g" }
 /^aggr M$/ { aggr = "m" }
 /^aggr Gobuf$/ { aggr = "gobuf" }
+/^aggr WinCall$/ { aggr = "wincall" }
 /^}/ { aggr = "" }
 
 #	Gobuf 24 sched;

src/pkg/runtime/runtime.h

@@ -70,6 +70,7 @@ typedef	struct	Hmap		Hmap;
 typedef	struct	Hchan		Hchan;
 typedef	struct	Complex64	Complex64;
 typedef	struct	Complex128	Complex128;
+typedef	struct	WinCall		WinCall;
 
 /*
  * per-cpu declaration.
@@ -250,11 +251,6 @@ struct	M
 	uint32	freglo[16];	// D[i] lsb and F[i]
 	uint32	freghi[16];	// D[i] msb and F[i+16]
 	uint32	fflag;		// floating point compare flags
-#ifdef __WINDOWS__
-#ifdef _64BIT	
-	void*	gostack;
-#endif
-#endif
 };
 
 struct	Stktop
@@ -306,6 +302,15 @@ struct	Func
 	int32	locals;	// number of 32-bit locals
 };
 
+struct	WinCall
+{
+	void	(*fn)(void*);
+	uintptr	n;	// number of parameters
+	void*	args;	// parameters
+	uintptr	r;	// return value
+	uintptr	err;	// error number
+};
+
 #ifdef __WINDOWS__
 enum {
    Windows = 1

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

@@ -4,44 +4,34 @@
 
 #include "386/asm.h"
 
-// void *stdcall_raw(void *fn, int32 count, uintptr *args)
-TEXT runtime·stdcall_raw(SB),7,$0
-	// Copy arguments from stack.
-	MOVL	fn+0(FP), AX
-	MOVL	count+4(FP), CX		// words
-	MOVL	args+8(FP), BP
+// void runtime·asmstdcall(void *c);
+TEXT runtime·asmstdcall(SB),7,$0
+	MOVL	c+0(FP), DX
 
-	// Switch to m->g0 if needed.
-	get_tls(DI)
-	MOVL	m(DI), DX
-	MOVL	m_g0(DX), SI
-	CMPL	g(DI), SI
-	MOVL	SP, BX
-	JEQ	2(PC)
-	MOVL	(g_sched+gobuf_sp)(SI), SP
-	PUSHL	BX
-	PUSHL	g(DI)
-	MOVL	SI, g(DI)
+	// SetLastError(0).
+	MOVL	$0, 0x34(FS)
 
-	// Copy args to new stack.
+	// Copy args to the stack.
+	MOVL	wincall_n(DX), CX	// words
 	MOVL	CX, BX
 	SALL	$2, BX
 	SUBL	BX, SP			// room for args
 	MOVL	SP, DI
-	MOVL	BP, SI
+	MOVL	wincall_args(DX), SI
 	CLD
 	REP; MOVSL
 
 	// Call stdcall function.
+	MOVL	wincall_fn(DX), AX
 	CALL	AX
 
-	// Restore original SP, g.
-	get_tls(DI)
-	POPL	g(DI)
-	POPL	SP
+	// Return result.
+	MOVL	c+0(FP), DX
+	MOVL	AX, wincall_r(DX)
 
-	// Someday the convention will be D is always cleared.
-	CLD
+	// GetLastError().
+	MOVL	0x34(FS), BX
+	MOVL	BX, wincall_err(DX)
 
 	RET
 

src/pkg/runtime/windows/amd64/sys.s

@@ -4,46 +4,57 @@
 
 #include "amd64/asm.h"
 
-// void *stdcall_raw(void *fn, uintptr nargs, void *args)
-TEXT runtime·stdcall_raw(SB),7,$8
-	MOVQ	fn+0(FP), AX
-	MOVQ	nargs+8(FP), CX
-	MOVQ	args+16(FP), R11
+#define maxargs 12
 
-	// Switch to m->g0 if needed.
-	get_tls(DI)
-	MOVQ	m(DI), DX
-	MOVQ	g(DI), SI
-	MOVQ	SI, 0(SP)		// save g
-	MOVQ	SP, m_gostack(DX)	// save SP
-	MOVQ	m_g0(DX), SI
-	CMPQ	g(DI), SI
-	JEQ 3(PC)
-	MOVQ	(g_sched+gobuf_sp)(SI), SP
-	ANDQ	$~15, SP
-	MOVQ	SI, g(DI)
-	
-	SUBQ	$0x60, SP
-	
-	// Copy args to new stack.
+// void runtime·asmstdcall(void *c);
+TEXT runtime·asmstdcall(SB),7,$0
+	// asmcgocall will put first argument into CX.
+	PUSHQ	CX			// save for later
+	MOVQ	wincall_fn(CX), AX
+	MOVQ	wincall_args(CX), SI
+	MOVQ	wincall_n(CX), CX
+
+	// SetLastError(0).
+	MOVQ	0x30(GS), DI
+	MOVL	$0, 0x68(DI)
+
+	SUBQ	$(maxargs*8), SP	// room for args
+
+	// Fast version, do not store args on the stack.
+	CMPL	CX, $4
+	JLE	loadregs
+
+	// Check we have enough room for args.
+	CMPL	CX, $maxargs
+	JLE	2(PC)
+	INT	$3			// not enough room -> crash
+
+	// Copy args to the stack.
 	MOVQ	SP, DI
-	MOVQ	R11, SI
 	CLD
 	REP; MOVSQ
-	MOVQ	0(R11), CX
-	MOVQ	8(R11), DX
-	MOVQ	16(R11), R8
-	MOVQ	24(R11), R9
+	MOVQ	SP, SI
+
+loadregs:
+	// Load first 4 args into correspondent registers.
+	MOVQ	0(SI), CX
+	MOVQ	8(SI), DX
+	MOVQ	16(SI), R8
+	MOVQ	24(SI), R9
 
 	// Call stdcall function.
 	CALL	AX
-	
-	// Restore original SP, g.
-	get_tls(DI)
-	MOVQ	m(DI), DX
-	MOVQ	m_gostack(DX), SP	// restore SP
-	MOVQ	0(SP), SI		// restore g
-	MOVQ	SI, g(DI)
+
+	ADDQ	$(maxargs*8), SP
+
+	// Return result.
+	POPQ	CX
+	MOVQ	AX, wincall_r(CX)
+
+	// GetLastError().
+	MOVQ	0x30(GS), DI
+	MOVL	0x68(DI), AX
+	MOVQ	AX, wincall_err(CX)
 
 	RET
 
@@ -70,7 +81,7 @@ TEXT runtime·ctrlhandler(SB),7,$0
 	MOVQ	SP, BX
 
 	// setup dummy m, g
-	SUBQ	$(m_gostack+8), SP	// at least space for m_gostack
+	SUBQ	$(m_fflag+4), SP	// at least space for m_fflag
 	LEAQ	m_tls(SP), CX
 	MOVQ	CX, 0x58(GS)
 	MOVQ	SP, m(CX)

src/pkg/runtime/windows/os.h

@@ -10,7 +10,7 @@ extern void *runtime·GetProcAddress;
 #pragma	varargck	countpos	runtime·stdcall	2
 #pragma	varargck	type		runtime·stdcall	void*
 #pragma	varargck	type		runtime·stdcall	uintptr
-void *runtime·stdcall_raw(void *fn, uintptr nargs, void *args);
+void runtime·asmstdcall(void *c);
 void *runtime·stdcall(void *fn, int32 count, ...);
 uintptr runtime·syscall(void *fn, uintptr nargs, void *args, uintptr *err);
 

src/pkg/runtime/windows/thread.c

@@ -6,6 +6,7 @@
 #include "type.h"
 #include "defs.h"
 #include "os.h"
+#include "cgocall.h"
 
 #pragma dynimport runtime·CloseHandle CloseHandle "kernel32.dll"
 #pragma dynimport runtime·CreateEvent CreateEventA "kernel32.dll"
@@ -221,34 +222,27 @@ runtime·gettime(int64 *sec, int32 *usec)
 void *
 runtime·stdcall(void *fn, int32 count, ...)
 {
-	return runtime·stdcall_raw(fn, count, (uintptr*)&count + 1);
+	WinCall c;
+
+	c.fn = fn;
+	c.n = count;
+	c.args = (uintptr*)&count + 1;
+	runtime·asmcgocall(runtime·asmstdcall, &c);
+	return (void*)c.r;
 }
 
 uintptr
 runtime·syscall(void *fn, uintptr nargs, void *args, uintptr *err)
 {
-	G *oldlock;
-	uintptr ret;
+	WinCall c;
 
-	/*
-	 * Lock g to m to ensure we stay on the same stack if we do a callback.
-	 */
-	oldlock = m->lockedg;
-	m->lockedg = g;
-	g->lockedm = m;
-
-	runtime·entersyscall();
-	runtime·setlasterror(0);
-	ret = (uintptr)runtime·stdcall_raw(fn, nargs, args);
+	c.fn = fn;
+	c.n = nargs;
+	c.args = args;
+	runtime·cgocall(runtime·asmstdcall, &c);
 	if(err)
-		*err = runtime·getlasterror();
-	runtime·exitsyscall();
-
-	m->lockedg = oldlock;
-	if(oldlock == nil)
-		g->lockedm = nil;
-
-	return ret;
+		*err = c.err;
+	return c.r;
 }
 
 uint32