get_tls(BX)
	MOVQ	g(BX), BP
	MOVQ	g_m(BP), BP
	// restore libcall
	LEAQ	m_libcall(BP), R11
	MOVQ	72(SP), R10
	MOVQ	R10, libcall_fn(R11)
	MOVQ	80(SP), R10
	MOVQ	R10, libcall_args(R11)
	MOVQ	88(SP), R10
	MOVQ	R10, libcall_n(R11)
	MOVQ    152(SP), R10
	MOVQ    R10, libcall_r1(R11)
	MOVQ    160(SP), R10
	MOVQ    R10, libcall_r2(R11)

	// restore scratch
	LEAQ	(m_mOS+mOS_scratch)(BP), R11

	BEQ	sigtramp	// g.m == nil

	// Save m->libcall. We need to do this because we
	// might get interrupted by a signal in runtime·asmcgocall.
	MOVD	(m_libcall+libcall_fn)(R6), R7
	MOVD	R7, 96(R1)
	MOVD	(m_libcall+libcall_args)(R6), R7
	MOVD	R7, 104(R1)
	MOVD	(m_libcall+libcall_n)(R6), R7
	MOVD	R7, 112(R1)
	MOVD	(m_libcall+libcall_r1)(R6), R7
	MOVD	R7, 120(R1)
	MOVD	(m_libcall+libcall_r2)(R6), R7
	MOVD	R7, 128(R1)

	CLD
	REP; MOVSL

docall:
	// Call stdcall or cdecl function.
	// DI SI BP BX are preserved, SP is not
	CALL	libcall_fn(BX)
	MOVL	BP, SP

	// Return result.
	MOVL	fn+0(FP), BX
	MOVL	AX, libcall_r1(BX)
	MOVL	DX, libcall_r2(BX)

	// GetLastError().
	MOVL	0x34(FS), AX
	MOVL	AX, libcall_err(BX)

	RET

// faster get/set last error
TEXT runtime·getlasterror(SB),NOSPLIT,$0
	MOVL	0x34(FS), AX

_3args:
	MOVD	(2*8)(R12), R2
_2args:
	MOVD	(1*8)(R12), R1
_1args:
	MOVD	(0*8)(R12), R0
_0args:

	MOVD	libcall_fn(R19), R12	// branch to libcall->fn
	BL	(R12)

	MOVD	R20, RSP			// free stack space
	MOVD	R0, libcall_r1(R19)		// save return value to libcall->r1
	// TODO(rsc) floating point like amd64 in libcall->r2?

	// GetLastError
	MOVD	TEB_error(R18_PLATFORM), R0
	MOVD	R0, libcall_err(R19)

TEXT runtime·asmstdcall(SB),NOSPLIT,$16
	MOVQ	SP, AX
	ANDQ	$~15, SP	// alignment as per Windows requirement
	MOVQ	AX, 8(SP)
	MOVQ	CX, 0(SP)	// asmcgocall will put first argument into CX.

	MOVQ	libcall_fn(CX), AX
	MOVQ	libcall_args(CX), SI
	MOVQ	libcall_n(CX), CX

	// SetLastError(0).
	MOVQ	0x30(GS), DI
	MOVL	$0, 0x68(DI)

	SUBQ	$(const_maxArgs*8), SP	// room for args

	// Fast version, do not store args on the stack.

		mp = gp.m
	}
	if mp != nil && mp.libcallsp == 0 {
		mp.libcallg.set(gp)
		mp.libcallpc = getcallerpc()
		// sp must be the last, because once async cpu profiler finds
		// all three values to be non-zero, it will use them
		mp.libcallsp = getcallersp()
	} else {
		mp = nil
	}

	var libcall libcall
	libcall.fn = uintptr(unsafe.Pointer(fn))
	libcall.n = 2
	libcall.args = uintptr(noescape(unsafe.Pointer(&a1)))

	mp := gp.m
	resetLibcall := true
	if mp.libcallsp == 0 {
		mp.libcallg.set(gp)
		mp.libcallpc = getcallerpc()
		// sp must be the last, because once async cpu profiler finds
		// all three values to be non-zero, it will use them
		mp.libcallsp = getcallersp()
	} else {
		resetLibcall = false // See comment in sys_darwin.go:libcCall
	}

	c := libcall{
		fn:   uintptr(unsafe.Pointer(fn)),
		n:    2,

	gp := getg()
	var mp *m
	if gp != nil {
		mp = gp.m
	}
	if mp != nil && mp.libcallsp == 0 {
		mp.libcallg.set(gp)
		mp.libcallpc = getcallerpc()
		// sp must be the last, because once async cpu profiler finds
		// all three values to be non-zero, it will use them
		mp.libcallsp = getcallersp()
	} else {
		// Make sure we don't reset libcallsp. This makes
		// libcCall reentrant; We remember the g/pc/sp for the

	if resetLibcall {
		mp.libcallsp = 0
	}
	return mp.libcall.r1
}

//go:nosplit
func stdcall0(fn stdFunction) uintptr {
	mp := getg().m
	mp.libcall.n = 0
	mp.libcall.args = 0
	return stdcall(fn)
}

//go:nosplit
//go:cgo_unsafe_args
func stdcall1(fn stdFunction, a0 uintptr) uintptr {
	mp := getg().m
	mp.libcall.n = 1

//go:nosplit
//go:cgo_unsafe_args
func doMmap(addr, n, prot, flags, fd, off uintptr) (uintptr, uintptr) {
	var libcall libcall
	libcall.fn = uintptr(unsafe.Pointer(&libc_mmap))
	libcall.n = 6
	libcall.args = uintptr(noescape(unsafe.Pointer(&addr)))
	asmcgocall(unsafe.Pointer(&asmsysvicall6x), unsafe.Pointer(&libcall))
	return libcall.r1, libcall.err
}

//go:nosplit
func munmap(addr unsafe.Pointer, n uintptr) {