gecko-dev/xpcom/ds/TimeStamp_windows.cpp
Brian Birtles 9d4bf99681 Bug 1039924 part 3 - Templatize TimeDuration so it can support different behaviors with regards to tick count arithmetic; r=froydnj
This patch prepares the way for having a separate StickyTimeDuration class
by factoring TimeDuration into a templated base class: BaseTimeDuration.
BaseTimeDuration takes a templated parameter, ValueCalculator, which is a helper
object that defines how various arithmetic operations are performed on its
mValue member (an int64_t count of ticks).

This patch does not actually define or use the ValueCalculator parameter yet but
simply performs the renaming and templatization.

With regards to the templatization, arithmetic operators are defined to take
objects with the same ValueCalculator template parameter (so that we don't, for
example, apply non-safe arithmetic to a StickyTimeDuration).
However, comparison operators are defined to also operate on objects with
a different ValueCalculator template parameter since comparison should be
independent of the type of arithmetic used.

Likewise, the constructor and assignment operator are defined to operate on
objects with a different ValueCalculator template parameter so that objects can
be converted from TimeDuration to StickyTimeDuration and vice-versa.
The constructor is marked as explicit, however, so that we don't silently
convert a StickyTimeDuration to a TimeDuration and unwittingly apply
non-safe arithmetic to a StickyTimeDuration.

TimeDuration is defined as a specialization of BaseTimeDuration that uses
TimeDurationValueCalculator as its ValueCalculator type.
TimeDurationValueCalculator is filled-in in a subsequent patch.
2014-09-25 14:25:49 +09:00

574 lines
17 KiB
C++

/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim:set ts=2 sw=2 sts=2 et cindent: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
// Implement TimeStamp::Now() with QueryPerformanceCounter() controlled with
// values of GetTickCount().
#include "mozilla/MathAlgorithms.h"
#include "mozilla/Mutex.h"
#include "mozilla/TimeStamp.h"
#include "nsWindowsHelpers.h"
#include <windows.h>
#include "nsCRT.h"
#include "prlog.h"
#include "prprf.h"
#include <stdio.h>
#include <intrin.h>
#if defined(PR_LOGGING)
// Log module for mozilla::TimeStamp for Windows logging...
//
// To enable logging (see prlog.h for full details):
//
// set NSPR_LOG_MODULES=TimeStampWindows:5
// set NSPR_LOG_FILE=nspr.log
//
// this enables PR_LOG_DEBUG level information and places all output in
// the file nspr.log
static PRLogModuleInfo*
GetTimeStampLog()
{
static PRLogModuleInfo* sLog;
if (!sLog) {
sLog = PR_NewLogModule("TimeStampWindows");
}
return sLog;
}
#define LOG(x) PR_LOG(GetTimeStampLog(), PR_LOG_DEBUG, x)
#else
#define LOG(x)
#endif /* PR_LOGGING */
// Estimate of the smallest duration of time we can measure.
static volatile ULONGLONG sResolution;
static volatile ULONGLONG sResolutionSigDigs;
static const double kNsPerSecd = 1000000000.0;
static const LONGLONG kNsPerSec = 1000000000;
static const LONGLONG kNsPerMillisec = 1000000;
// ----------------------------------------------------------------------------
// Global constants
// ----------------------------------------------------------------------------
// Tolerance to failures settings.
//
// What is the interval we want to have failure free.
// in [ms]
static const uint32_t kFailureFreeInterval = 5000;
// How many failures we are willing to tolerate in the interval.
static const uint32_t kMaxFailuresPerInterval = 4;
// What is the threshold to treat fluctuations as actual failures.
// in [ms]
static const uint32_t kFailureThreshold = 50;
// If we are not able to get the value of GTC time increment, use this value
// which is the most usual increment.
static const DWORD kDefaultTimeIncrement = 156001;
// ----------------------------------------------------------------------------
// Global variables, not changing at runtime
// ----------------------------------------------------------------------------
/**
* The [mt] unit:
*
* Many values are kept in ticks of the Performance Coutner x 1000,
* further just referred as [mt], meaning milli-ticks.
*
* This is needed to preserve maximum precision of the performance frequency
* representation. GetTickCount values in milliseconds are multiplied with
* frequency per second. Therefor we need to multiply QPC value by 1000 to
* have the same units to allow simple arithmentic with both QPC and GTC.
*/
#define ms2mt(x) ((x) * sFrequencyPerSec)
#define mt2ms(x) ((x) / sFrequencyPerSec)
#define mt2ms_f(x) (double(x) / sFrequencyPerSec)
// Result of QueryPerformanceFrequency
static LONGLONG sFrequencyPerSec = 0;
// How much we are tolerant to GTC occasional loose of resoltion.
// This number says how many multiples of the minimal GTC resolution
// detected on the system are acceptable. This number is empirical.
static const LONGLONG kGTCTickLeapTolerance = 4;
// Base tolerance (more: "inability of detection" range) threshold is calculated
// dynamically, and kept in sGTCResulutionThreshold.
//
// Schematically, QPC worked "100%" correctly if ((GTC_now - GTC_epoch) -
// (QPC_now - QPC_epoch)) was in [-sGTCResulutionThreshold, sGTCResulutionThreshold]
// interval every time we'd compared two time stamps.
// If not, then we check the overflow behind this basic threshold
// is in kFailureThreshold. If not, we condider it as a QPC failure. If too many
// failures in short time are detected, QPC is considered faulty and disabled.
//
// Kept in [mt]
static LONGLONG sGTCResulutionThreshold;
// If QPC is found faulty for two stamps in this interval, we engage
// the fault detection algorithm. For duration larger then this limit
// we bypass using durations calculated from QPC when jitter is detected,
// but don't touch the sUseQPC flag.
//
// Value is in [ms].
static const uint32_t kHardFailureLimit = 2000;
// Conversion to [mt]
static LONGLONG sHardFailureLimit;
// Conversion of kFailureFreeInterval and kFailureThreshold to [mt]
static LONGLONG sFailureFreeInterval;
static LONGLONG sFailureThreshold;
// ----------------------------------------------------------------------------
// Systemm status flags
// ----------------------------------------------------------------------------
// Flag for stable TSC that indicates platform where QPC is stable.
static bool sHasStableTSC = false;
// ----------------------------------------------------------------------------
// Global state variables, changing at runtime
// ----------------------------------------------------------------------------
// Initially true, set to false when QPC is found unstable and never
// returns back to true since that time.
static bool volatile sUseQPC = true;
// ----------------------------------------------------------------------------
// Global lock
// ----------------------------------------------------------------------------
// Thread spin count before entering the full wait state for sTimeStampLock.
// Inspired by Rob Arnold's work on PRMJ_Now().
static const DWORD kLockSpinCount = 4096;
// Common mutex (thanks the relative complexity of the logic, this is better
// then using CMPXCHG8B.)
// It is protecting the globals bellow.
static CRITICAL_SECTION sTimeStampLock;
// ----------------------------------------------------------------------------
// Global lock protected variables
// ----------------------------------------------------------------------------
// Timestamp in future until QPC must behave correctly.
// Set to now + kFailureFreeInterval on first QPC failure detection.
// Set to now + E * kFailureFreeInterval on following errors,
// where E is number of errors detected during last kFailureFreeInterval
// milliseconds, calculated simply as:
// E = (sFaultIntoleranceCheckpoint - now) / kFailureFreeInterval + 1.
// When E > kMaxFailuresPerInterval -> disable QPC.
//
// Kept in [mt]
static ULONGLONG sFaultIntoleranceCheckpoint = 0;
// Used only when GetTickCount64 is not available on the platform.
// Last result of GetTickCount call.
//
// Kept in [ms]
static DWORD sLastGTCResult = 0;
// Higher part of the 64-bit value of MozGetTickCount64,
// incremented atomically.
static DWORD sLastGTCRollover = 0;
namespace mozilla {
typedef ULONGLONG (WINAPI* GetTickCount64_t)();
static GetTickCount64_t sGetTickCount64 = nullptr;
// Function protecting GetTickCount result from rolling over,
// result is in [ms]
static ULONGLONG WINAPI
MozGetTickCount64()
{
DWORD GTC = ::GetTickCount();
// Cheaper then CMPXCHG8B
AutoCriticalSection lock(&sTimeStampLock);
// Pull the rollover counter forward only if new value of GTC goes way
// down under the last saved result
if ((sLastGTCResult > GTC) && ((sLastGTCResult - GTC) > (1UL << 30))) {
++sLastGTCRollover;
}
sLastGTCResult = GTC;
return ULONGLONG(sLastGTCRollover) << 32 | sLastGTCResult;
}
// Result is in [mt]
static inline ULONGLONG
PerformanceCounter()
{
LARGE_INTEGER pc;
::QueryPerformanceCounter(&pc);
return pc.QuadPart * 1000ULL;
}
static void
InitThresholds()
{
DWORD timeAdjustment = 0, timeIncrement = 0;
BOOL timeAdjustmentDisabled;
GetSystemTimeAdjustment(&timeAdjustment,
&timeIncrement,
&timeAdjustmentDisabled);
LOG(("TimeStamp: timeIncrement=%d [100ns]", timeIncrement));
if (!timeIncrement) {
timeIncrement = kDefaultTimeIncrement;
}
// Ceiling to a millisecond
// Example values: 156001, 210000
DWORD timeIncrementCeil = timeIncrement;
// Don't want to round up if already rounded, values will be: 156000, 209999
timeIncrementCeil -= 1;
// Convert to ms, values will be: 15, 20
timeIncrementCeil /= 10000;
// Round up, values will be: 16, 21
timeIncrementCeil += 1;
// Convert back to 100ns, values will be: 160000, 210000
timeIncrementCeil *= 10000;
// How many milli-ticks has the interval rounded up
LONGLONG ticksPerGetTickCountResolutionCeiling =
(int64_t(timeIncrementCeil) * sFrequencyPerSec) / 10000LL;
// GTC may jump by 32 (2*16) ms in two steps, therefor use the ceiling value.
sGTCResulutionThreshold =
LONGLONG(kGTCTickLeapTolerance * ticksPerGetTickCountResolutionCeiling);
sHardFailureLimit = ms2mt(kHardFailureLimit);
sFailureFreeInterval = ms2mt(kFailureFreeInterval);
sFailureThreshold = ms2mt(kFailureThreshold);
}
static void
InitResolution()
{
// 10 total trials is arbitrary: what we're trying to avoid by
// looping is getting unlucky and being interrupted by a context
// switch or signal, or being bitten by paging/cache effects
ULONGLONG minres = ~0ULL;
int loops = 10;
do {
ULONGLONG start = PerformanceCounter();
ULONGLONG end = PerformanceCounter();
ULONGLONG candidate = (end - start);
if (candidate < minres) {
minres = candidate;
}
} while (--loops && minres);
if (0 == minres) {
minres = 1;
}
// Converting minres that is in [mt] to nanosecods, multiplicating
// the argument to preserve resolution.
ULONGLONG result = mt2ms(minres * kNsPerMillisec);
if (0 == result) {
result = 1;
}
sResolution = result;
// find the number of significant digits in mResolution, for the
// sake of ToSecondsSigDigits()
ULONGLONG sigDigs;
for (sigDigs = 1;
!(sigDigs == result || 10 * sigDigs > result);
sigDigs *= 10);
sResolutionSigDigs = sigDigs;
}
// ----------------------------------------------------------------------------
// TimeStampValue implementation
// ----------------------------------------------------------------------------
TimeStampValue::TimeStampValue(ULONGLONG aGTC, ULONGLONG aQPC, bool aHasQPC)
: mGTC(aGTC)
, mQPC(aQPC)
, mHasQPC(aHasQPC)
, mIsNull(false)
{
}
TimeStampValue&
TimeStampValue::operator+=(const int64_t aOther)
{
mGTC += aOther;
mQPC += aOther;
return *this;
}
TimeStampValue&
TimeStampValue::operator-=(const int64_t aOther)
{
mGTC -= aOther;
mQPC -= aOther;
return *this;
}
// If the duration is less then two seconds, perform check of QPC stability
// by comparing both GTC and QPC calculated durations of this and aOther.
uint64_t
TimeStampValue::CheckQPC(const TimeStampValue& aOther) const
{
uint64_t deltaGTC = mGTC - aOther.mGTC;
if (!mHasQPC || !aOther.mHasQPC) { // Both not holding QPC
return deltaGTC;
}
uint64_t deltaQPC = mQPC - aOther.mQPC;
if (sHasStableTSC) { // For stable TSC there is no need to check
return deltaQPC;
}
// Check QPC is sane before using it.
int64_t diff = DeprecatedAbs(int64_t(deltaQPC) - int64_t(deltaGTC));
if (diff <= sGTCResulutionThreshold) {
return deltaQPC;
}
// Treat absolutely for calibration purposes
int64_t duration = DeprecatedAbs(int64_t(deltaGTC));
int64_t overflow = diff - sGTCResulutionThreshold;
LOG(("TimeStamp: QPC check after %llums with overflow %1.4fms",
mt2ms(duration), mt2ms_f(overflow)));
if (overflow <= sFailureThreshold) { // We are in the limit, let go.
return deltaQPC;
}
// QPC deviates, don't use it, since now this method may only return deltaGTC.
if (!sUseQPC) { // QPC already disabled, no need to run the fault tolerance algorithm.
return deltaGTC;
}
LOG(("TimeStamp: QPC jittered over failure threshold"));
if (duration < sHardFailureLimit) {
// Interval between the two time stamps is very short, consider
// QPC as unstable and record a failure.
uint64_t now = ms2mt(sGetTickCount64());
AutoCriticalSection lock(&sTimeStampLock);
if (sFaultIntoleranceCheckpoint && sFaultIntoleranceCheckpoint > now) {
// There's already been an error in the last fault intollerant interval.
// Time since now to the checkpoint actually holds information on how many
// failures there were in the failure free interval we have defined.
uint64_t failureCount =
(sFaultIntoleranceCheckpoint - now + sFailureFreeInterval - 1) /
sFailureFreeInterval;
if (failureCount > kMaxFailuresPerInterval) {
sUseQPC = false;
LOG(("TimeStamp: QPC disabled"));
} else {
// Move the fault intolerance checkpoint more to the future, prolong it
// to reflect the number of detected failures.
++failureCount;
sFaultIntoleranceCheckpoint = now + failureCount * sFailureFreeInterval;
LOG(("TimeStamp: recording %dth QPC failure", failureCount));
}
} else {
// Setup fault intolerance checkpoint in the future for first detected error.
sFaultIntoleranceCheckpoint = now + sFailureFreeInterval;
LOG(("TimeStamp: recording 1st QPC failure"));
}
}
return deltaGTC;
}
uint64_t
TimeStampValue::operator-(const TimeStampValue& aOther) const
{
if (mIsNull && aOther.mIsNull) {
return uint64_t(0);
}
return CheckQPC(aOther);
}
// ----------------------------------------------------------------------------
// TimeDuration and TimeStamp implementation
// ----------------------------------------------------------------------------
double
BaseTimeDurationPlatformUtils::ToSeconds(int64_t aTicks)
{
// Converting before arithmetic avoids blocked store forward
return double(aTicks) / (double(sFrequencyPerSec) * 1000.0);
}
double
BaseTimeDurationPlatformUtils::ToSecondsSigDigits(int64_t aTicks)
{
// don't report a value < mResolution ...
LONGLONG resolution = sResolution;
LONGLONG resolutionSigDigs = sResolutionSigDigs;
LONGLONG valueSigDigs = resolution * (aTicks / resolution);
// and chop off insignificant digits
valueSigDigs = resolutionSigDigs * (valueSigDigs / resolutionSigDigs);
return double(valueSigDigs) / kNsPerSecd;
}
int64_t
BaseTimeDurationPlatformUtils::TicksFromMilliseconds(double aMilliseconds)
{
return ms2mt(aMilliseconds);
}
int64_t
BaseTimeDurationPlatformUtils::ResolutionInTicks()
{
return static_cast<int64_t>(sResolution);
}
static bool
HasStableTSC()
{
union
{
int regs[4];
struct
{
int nIds;
char cpuString[12];
};
} cpuInfo;
__cpuid(cpuInfo.regs, 0);
// Only allow Intel CPUs for now
// The order of the registers is reg[1], reg[3], reg[2]. We just adjust the
// string so that we can compare in one go.
if (_strnicmp(cpuInfo.cpuString, "GenuntelineI",
sizeof(cpuInfo.cpuString))) {
return false;
}
int regs[4];
// detect if the Advanced Power Management feature is supported
__cpuid(regs, 0x80000000);
if (regs[0] < 0x80000007) {
return false;
}
__cpuid(regs, 0x80000007);
// if bit 8 is set than TSC will run at a constant rate
// in all ACPI P-state, C-states and T-states
return regs[3] & (1 << 8);
}
nsresult
TimeStamp::Startup()
{
// Decide which implementation to use for the high-performance timer.
HMODULE kernelDLL = GetModuleHandleW(L"kernel32.dll");
sGetTickCount64 = reinterpret_cast<GetTickCount64_t>(
GetProcAddress(kernelDLL, "GetTickCount64"));
if (!sGetTickCount64) {
// If the platform does not support the GetTickCount64 (Windows XP doesn't),
// then use our fallback implementation based on GetTickCount.
sGetTickCount64 = MozGetTickCount64;
}
InitializeCriticalSectionAndSpinCount(&sTimeStampLock, kLockSpinCount);
sHasStableTSC = HasStableTSC();
LOG(("TimeStamp: HasStableTSC=%d", sHasStableTSC));
LARGE_INTEGER freq;
sUseQPC = ::QueryPerformanceFrequency(&freq);
if (!sUseQPC) {
// No Performance Counter. Fall back to use GetTickCount.
InitResolution();
LOG(("TimeStamp: using GetTickCount"));
return NS_OK;
}
sFrequencyPerSec = freq.QuadPart;
LOG(("TimeStamp: QPC frequency=%llu", sFrequencyPerSec));
InitThresholds();
InitResolution();
return NS_OK;
}
void
TimeStamp::Shutdown()
{
DeleteCriticalSection(&sTimeStampLock);
}
TimeStamp
TimeStamp::Now(bool aHighResolution)
{
// sUseQPC is volatile
bool useQPC = (aHighResolution && sUseQPC);
// Both values are in [mt] units.
ULONGLONG QPC = useQPC ? PerformanceCounter() : uint64_t(0);
ULONGLONG GTC = ms2mt(sGetTickCount64());
return TimeStamp(TimeStampValue(GTC, QPC, useQPC));
}
// Computes and returns the process uptime in microseconds.
// Returns 0 if an error was encountered.
uint64_t
TimeStamp::ComputeProcessUptime()
{
SYSTEMTIME nowSys;
GetSystemTime(&nowSys);
FILETIME now;
bool success = SystemTimeToFileTime(&nowSys, &now);
if (!success) {
return 0;
}
FILETIME start, foo, bar, baz;
success = GetProcessTimes(GetCurrentProcess(), &start, &foo, &bar, &baz);
if (!success) {
return 0;
}
ULARGE_INTEGER startUsec = {
start.dwLowDateTime,
start.dwHighDateTime
};
ULARGE_INTEGER nowUsec = {
now.dwLowDateTime,
now.dwHighDateTime
};
return (nowUsec.QuadPart - startUsec.QuadPart) / 10ULL;
}
} // namespace mozilla