runtime: make time correctly update on Wine

Implemented low-level time system for windows on hardware (software),
which does not support memory mapped _KSYSTEM_TIME page update.

In particular this problem exists on Wine where _KSYSTEM_TIME
only contains time at the start, and is never modified.

On start we try to detect Wine and if it's so we fallback to
GetSystemTimeAsFileTime() for current time and a monotonic
timer based on QueryPerformanceCounter family of syscalls:
https://msdn.microsoft.com/en-us/library/windows/desktop/dn553408(v=vs.85).aspx

Fixes #18537

Change-Id: I269d22467ed9b0afb62056974d23e731b80c83ed
Reviewed-on: https://go-review.googlesource.com/35710
Reviewed-by: Alex Brainman <[email protected]>
Run-TryBot: Alex Brainman <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>

Author: bioothod

src/runtime/os_windows.go

@@ -73,9 +73,12 @@ var (
 	_GetQueuedCompletionStatus,
 	_GetStdHandle,
 	_GetSystemInfo,
+	_GetSystemTimeAsFileTime,
 	_GetThreadContext,
 	_LoadLibraryW,
 	_LoadLibraryA,
+	_QueryPerformanceCounter,
+	_QueryPerformanceFrequency,
 	_ResumeThread,
 	_SetConsoleCtrlHandler,
 	_SetErrorMode,
@@ -188,6 +191,11 @@ func loadOptionalSyscalls() {
 		throw("ntdll.dll not found")
 	}
 	_NtWaitForSingleObject = windowsFindfunc(n32, []byte("NtWaitForSingleObject\000"))
+
+	if windowsFindfunc(n32, []byte("wine_get_version\000")) != nil {
+		// running on Wine
+		initWine(k32)
+	}
 }
 
 //go:nosplit
@@ -292,6 +300,79 @@ func osinit() {
 	stdcall2(_SetProcessPriorityBoost, currentProcess, 1)
 }
 
+func nanotime() int64
+
+// useQPCTime controls whether time.now and nanotime use QueryPerformanceCounter.
+// This is only set to 1 when running under Wine.
+var useQPCTime uint8
+
+var qpcStartCounter int64
+var qpcMultiplier int64
+
+//go:nosplit
+func nanotimeQPC() int64 {
+	var counter int64 = 0
+	stdcall1(_QueryPerformanceCounter, uintptr(unsafe.Pointer(&counter)))
+
+	// returns number of nanoseconds
+	return (counter - qpcStartCounter) * qpcMultiplier
+}
+
+//go:nosplit
+func nowQPC() (sec int64, nsec int32, mono int64) {
+	var ft int64
+	stdcall1(_GetSystemTimeAsFileTime, uintptr(unsafe.Pointer(&ft)))
+
+	t := (ft - 116444736000000000) * 100
+
+	sec = t / 1000000000
+	nsec = int32(t - sec*1000000000)
+
+	mono = nanotimeQPC()
+	return
+}
+
+func initWine(k32 uintptr) {
+	_GetSystemTimeAsFileTime = windowsFindfunc(k32, []byte("GetSystemTimeAsFileTime\000"))
+	if _GetSystemTimeAsFileTime == nil {
+		throw("could not find GetSystemTimeAsFileTime() syscall")
+	}
+
+	_QueryPerformanceCounter = windowsFindfunc(k32, []byte("QueryPerformanceCounter\000"))
+	_QueryPerformanceFrequency = windowsFindfunc(k32, []byte("QueryPerformanceFrequency\000"))
+	if _QueryPerformanceCounter == nil || _QueryPerformanceFrequency == nil {
+		throw("could not find QPC syscalls")
+	}
+
+	// We can not simply fallback to GetSystemTimeAsFileTime() syscall, since its time is not monotonic,
+	// instead we use QueryPerformanceCounter family of syscalls to implement monotonic timer
+	// https://msdn.microsoft.com/en-us/library/windows/desktop/dn553408(v=vs.85).aspx
+
+	var tmp int64
+	stdcall1(_QueryPerformanceFrequency, uintptr(unsafe.Pointer(&tmp)))
+	if tmp == 0 {
+		throw("QueryPerformanceFrequency syscall returned zero, running on unsupported hardware")
+	}
+
+	// This should not overflow, it is a number of ticks of the performance counter per second,
+	// its resolution is at most 10 per usecond (on Wine, even smaller on real hardware), so it will be at most 10 millions here,
+	// panic if overflows.
+	if tmp > (1<<31 - 1) {
+		throw("QueryPerformanceFrequency overflow 32 bit divider, check nosplit discussion to proceed")
+	}
+	qpcFrequency := int32(tmp)
+	stdcall1(_QueryPerformanceCounter, uintptr(unsafe.Pointer(&qpcStartCounter)))
+
+	// Since we are supposed to run this time calls only on Wine, it does not lose precision,
+	// since Wine's timer is kind of emulated at 10 Mhz, so it will be a nice round multiplier of 100
+	// but for general purpose system (like 3.3 Mhz timer on i7) it will not be very precise.
+	// We have to do it this way (or similar), since multiplying QPC counter by 100 millions overflows
+	// int64 and resulted time will always be invalid.
+	qpcMultiplier = int64(timediv(1000000000, qpcFrequency, nil))
+
+	useQPCTime = 1
+}
+
 //go:nosplit
 func getRandomData(r []byte) {
 	n := 0
@@ -578,8 +659,6 @@ func unminit() {
 	*tp = 0
 }
 
-func nanotime() int64
-
 // Calling stdcall on os stack.
 // May run during STW, so write barriers are not allowed.
 //go:nowritebarrier

src/runtime/sys_windows_386.s

@@ -441,6 +441,8 @@ TEXT runtime·switchtothread(SB),NOSPLIT,$0
 #define time_hi2 8
 
 TEXT runtime·nanotime(SB),NOSPLIT,$0-8
+	CMPB	runtime·useQPCTime(SB), $0
+	JNE	useQPC
 loop:
 	MOVL	(_INTERRUPT_TIME+time_hi1), AX
 	MOVL	(_INTERRUPT_TIME+time_lo), CX
@@ -459,8 +461,13 @@ loop:
 	MOVL	AX, ret_lo+0(FP)
 	MOVL	DX, ret_hi+4(FP)
 	RET
+useQPC:
+	JMP	runtime·nanotimeQPC(SB)
+	RET
 
 TEXT time·now(SB),NOSPLIT,$0-20
+	CMPB	runtime·useQPCTime(SB), $0
+	JNE	useQPC
 loop:
 	MOVL	(_INTERRUPT_TIME+time_hi1), AX
 	MOVL	(_INTERRUPT_TIME+time_lo), CX
@@ -477,7 +484,7 @@ loop:
 	// w*100 = DX:AX
 	// subtract startNano and save for return
 	SUBL	runtime·startNano+0(SB), AX
-	SBBL runtime·startNano+4(SB), DX
+	SBBL	runtime·startNano+4(SB), DX
 	MOVL	AX, mono+12(FP)
 	MOVL	DX, mono+16(FP)
 
@@ -532,3 +539,6 @@ wall:
 	MOVL	AX, sec+0(FP)
 	MOVL	DX, sec+4(FP)
 	RET
+useQPC:
+	JMP	runtime·nowQPC(SB)
+	RET

src/runtime/sys_windows_amd64.s

@@ -474,6 +474,8 @@ TEXT runtime·switchtothread(SB),NOSPLIT|NOFRAME,$0
 #define time_hi2 8
 
 TEXT runtime·nanotime(SB),NOSPLIT,$0-8
+	CMPB	runtime·useQPCTime(SB), $0
+	JNE	useQPC
 	MOVQ	$_INTERRUPT_TIME, DI
 loop:
 	MOVL	time_hi1(DI), AX
@@ -487,8 +489,13 @@ loop:
 	SUBQ	runtime·startNano(SB), CX
 	MOVQ	CX, ret+0(FP)
 	RET
+useQPC:
+	JMP	runtime·nanotimeQPC(SB)
+	RET
 
 TEXT time·now(SB),NOSPLIT,$0-24
+	CMPB	runtime·useQPCTime(SB), $0
+	JNE	useQPC
 	MOVQ	$_INTERRUPT_TIME, DI
 loop:
 	MOVL	time_hi1(DI), AX
@@ -529,3 +536,6 @@ wall:
 	SUBQ	DX, CX
 	MOVL	CX, nsec+8(FP)
 	RET
+useQPC:
+	JMP	runtime·nowQPC(SB)
+	RET