runtime: implement usleep in Go instead of assembly on Windows

Windows APIs are normally not arch-specific, so it's better to
implement them in Go instead of assembly.

It was previously implemented in assembly because it was the only way
to support calls without a valid g. This CL defines a new function,
stdcall_no_g, that can be used in such cases.

While here, I've also replaced the use of the deprecated syscall
NtWaitForSingleObject with WaitForSingleObject. The former may
give the illusion of being more accurate, as it takes a higher
resolution timeout, but it's not. Windows time resolution is 15.6ms,
and can be as high as 1ms when using a high resolution timer, which
WaitForSingleObject supports.

Change-Id: I903400220ade4d4ccc15685c8da47182430f8686
Reviewed-on: https://go-review.googlesource.com/c/go/+/526477
Reviewed-by: Bryan Mills <bcmills@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Alex Brainman <alex.brainman@gmail.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Run-TryBot: Quim Muntal <quimmuntal@gmail.com>

src/runtime/os_windows.go

@@ -138,7 +138,6 @@ var (
 	// Load ntdll.dll manually during startup, otherwise Mingw
 	// links wrong printf function to cgo executable (see issue
 	// 12030 for details).
-	_NtWaitForSingleObject  stdFunction
 	_RtlGetCurrentPeb       stdFunction
 	_RtlGetNtVersionNumbers stdFunction
 
@@ -269,7 +268,6 @@ func loadOptionalSyscalls() {
 	if n32 == 0 {
 		throw("ntdll.dll not found")
 	}
-	_NtWaitForSingleObject = windowsFindfunc(n32, []byte("NtWaitForSingleObject\000"))
 	_RtlGetCurrentPeb = windowsFindfunc(n32, []byte("RtlGetCurrentPeb\000"))
 	_RtlGetNtVersionNumbers = windowsFindfunc(n32, []byte("RtlGetNtVersionNumbers\000"))
 
@@ -1069,7 +1067,6 @@ func stdcall7(fn stdFunction, a0, a1, a2, a3, a4, a5, a6 uintptr) uintptr {
 }
 
 // These must run on the system stack only.
-func usleep2(dt int32)
 
 //go:nosplit
 func osyield_no_g() {
@@ -1085,23 +1082,27 @@ func osyield() {
 
 //go:nosplit
 func usleep_no_g(us uint32) {
-	dt := -10 * int32(us) // relative sleep (negative), 100ns units
-	usleep2(dt)
+	timeout := uintptr(us) / 1000 // ms units
+	args := [...]uintptr{_INVALID_HANDLE_VALUE, timeout}
+	stdcall_no_g(_WaitForSingleObject, len(args), uintptr(noescape(unsafe.Pointer(&args[0]))))
 }
 
 //go:nosplit
 func usleep(us uint32) {
 	systemstack(func() {
-		dt := -10 * int64(us) // relative sleep (negative), 100ns units
+		var h, timeout uintptr
 		// If the high-res timer is available and its handle has been allocated for this m, use it.
 		// Otherwise fall back to the low-res one, which doesn't need a handle.
 		if haveHighResTimer && getg().m.highResTimer != 0 {
-			h := getg().m.highResTimer
+			h = getg().m.highResTimer
+			dt := -10 * int64(us) // relative sleep (negative), 100ns units
 			stdcall6(_SetWaitableTimer, h, uintptr(unsafe.Pointer(&dt)), 0, 0, 0, 0)
-			stdcall3(_NtWaitForSingleObject, h, 0, 0)
+			timeout = _INFINITE
 		} else {
-			usleep2(int32(dt))
+			h = _INVALID_HANDLE_VALUE
+			timeout = uintptr(us) / 1000 // ms units
 		}
+		stdcall2(_WaitForSingleObject, h, timeout)
 	})
 }
 

src/runtime/sys_windows_386.s

@@ -234,23 +234,6 @@ TEXT runtime·setldt(SB),NOSPLIT,$0-12
 	MOVL	DX, 0(CX)(FS)
 	RET
 
-// Runs on OS stack.
-// duration (in -100ns units) is in dt+0(FP).
-// g may be nil.
-TEXT runtime·usleep2(SB),NOSPLIT,$20-4
-	MOVL	dt+0(FP), BX
-	MOVL	$-1, hi-4(SP)
-	MOVL	BX, lo-8(SP)
-	LEAL	lo-8(SP), BX
-	MOVL	BX, ptime-12(SP)
-	MOVL	$0, alertable-16(SP)
-	MOVL	$-1, handle-20(SP)
-	MOVL	SP, BP
-	MOVL	runtime·_NtWaitForSingleObject(SB), AX
-	CALL	AX
-	MOVL	BP, SP
-	RET
-
 TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
 loop:
 	MOVL	(_INTERRUPT_TIME+time_hi1), AX

src/runtime/sys_windows_amd64.s

@@ -243,25 +243,6 @@ TEXT runtime·settls(SB),NOSPLIT,$0
 	MOVQ	DI, 0(CX)(GS)
 	RET
 
-// Runs on OS stack.
-// duration (in -100ns units) is in dt+0(FP).
-// g may be nil.
-// The function leaves room for 4 syscall parameters
-// (as per windows amd64 calling convention).
-TEXT runtime·usleep2(SB),NOSPLIT,$48-4
-	MOVLQSX	dt+0(FP), BX
-	MOVQ	SP, AX
-	ANDQ	$~15, SP	// alignment as per Windows requirement
-	MOVQ	AX, 40(SP)
-	LEAQ	32(SP), R8  // ptime
-	MOVQ	BX, (R8)
-	MOVQ	$-1, CX // handle
-	MOVQ	$0, DX // alertable
-	MOVQ	runtime·_NtWaitForSingleObject(SB), AX
-	CALL	AX
-	MOVQ	40(SP), SP
-	RET
-
 TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
 	MOVQ	$_INTERRUPT_TIME, DI
 	MOVQ	time_lo(DI), AX

src/runtime/sys_windows_arm.s

@@ -195,25 +195,6 @@ TEXT runtime·tstart_stdcall(SB),NOSPLIT|NOFRAME,$0
 	MOVW	$0, R0
 	MOVM.IA.W (R13), [R4-R11, R15]		// pop {r4-r11, pc}
 
-// Runs on OS stack.
-// duration (in -100ns units) is in dt+0(FP).
-// g may be nil.
-TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$0-4
-	MOVW	dt+0(FP), R3
-	MOVM.DB.W [R4, R14], (R13)	// push {r4, lr}
-	MOVW	R13, R4			// Save SP
-	SUB	$8, R13			// R13 = R13 - 8
-	BIC	$0x7, R13		// Align SP for ABI
-	MOVW	$0, R1			// R1 = FALSE (alertable)
-	MOVW	$-1, R0			// R0 = handle
-	MOVW	R13, R2			// R2 = pTime
-	MOVW	R3, 0(R2)		// time_lo
-	MOVW	R0, 4(R2)		// time_hi
-	MOVW	runtime·_NtWaitForSingleObject(SB), R3
-	BL	(R3)
-	MOVW	R4, R13			// Restore SP
-	MOVM.IA.W (R13), [R4, R15]	// pop {R4, pc}
-
 TEXT ·publicationBarrier(SB),NOSPLIT|NOFRAME,$0-0
 	B	runtime·armPublicationBarrier(SB)
 

src/runtime/sys_windows_arm64.s

@@ -228,21 +228,6 @@ TEXT runtime·tstart_stdcall(SB),NOSPLIT,$96-0
 	MOVD	$0, R0
 	RET
 
-// Runs on OS stack.
-// duration (in -100ns units) is in dt+0(FP).
-// g may be nil.
-TEXT runtime·usleep2(SB),NOSPLIT,$32-4
-	MOVW	dt+0(FP), R0
-	MOVD	$16(RSP), R2		// R2 = pTime
-	MOVD	R0, 0(R2)		// *pTime = -dt
-	MOVD	$-1, R0			// R0 = handle
-	MOVD	$0, R1			// R1 = FALSE (alertable)
-	MOVD	runtime·_NtWaitForSingleObject(SB), R3
-	SUB	$16, RSP	// skip over saved frame pointer below RSP
-	BL	(R3)
-	ADD	$16, RSP
-	RET
-
 TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
 	MOVD	$_INTERRUPT_TIME, R3
 	MOVD	time_lo(R3), R0