/* -*- 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 POSIX clocks. // // The "tick" unit for POSIX clocks is simply a nanosecond, as this is // the smallest unit of time representable by struct timespec. That // doesn't mean that a nanosecond is the resolution of TimeDurations // obtained with this API; see TimeDuration::Resolution; // #include #include #include #if defined(__DragonFly__) || defined(__FreeBSD__) \ || defined(__NetBSD__) || defined(__OpenBSD__) #include #include #endif #if defined(__DragonFly__) || defined(__FreeBSD__) #include #endif #if defined(__NetBSD__) #undef KERN_PROC #define KERN_PROC KERN_PROC2 #define KINFO_PROC struct kinfo_proc2 #else #define KINFO_PROC struct kinfo_proc #endif #if defined(__DragonFly__) #define KP_START_SEC kp_start.tv_sec #define KP_START_USEC kp_start.tv_usec #elif defined(__FreeBSD__) #define KP_START_SEC ki_start.tv_sec #define KP_START_USEC ki_start.tv_usec #else #define KP_START_SEC p_ustart_sec #define KP_START_USEC p_ustart_usec #endif #include "mozilla/TimeStamp.h" #include "nsCRT.h" #include "prprf.h" #include "prthread.h" #include "nsDebug.h" // Estimate of the smallest duration of time we can measure. static uint64_t sResolution; static uint64_t sResolutionSigDigs; static const uint16_t kNsPerUs = 1000; static const uint64_t kNsPerMs = 1000000; static const uint64_t kNsPerSec = 1000000000; static const double kNsPerMsd = 1000000.0; static const double kNsPerSecd = 1000000000.0; static uint64_t TimespecToNs(const struct timespec& ts) { uint64_t baseNs = uint64_t(ts.tv_sec) * kNsPerSec; return baseNs + uint64_t(ts.tv_nsec); } static uint64_t ClockTimeNs() { struct timespec ts; // this can't fail: we know &ts is valid, and TimeStamp::Startup() // checks that CLOCK_MONOTONIC is supported (and aborts if not) clock_gettime(CLOCK_MONOTONIC, &ts); // tv_sec is defined to be relative to an arbitrary point in time, // but it would be madness for that point in time to be earlier than // the Epoch. So we can safely assume that even if time_t is 32 // bits, tv_sec won't overflow while the browser is open. Revisit // this argument if we're still building with 32-bit time_t around // the year 2037. return TimespecToNs(ts); } static uint64_t ClockResolutionNs() { // NB: why not rely on clock_getres()? Two reasons: (i) it might // lie, and (ii) it might return an "ideal" resolution that while // theoretically true, could never be measured in practice. Since // clock_gettime() likely involves a system call on your platform, // the "actual" timing resolution shouldn't be lower than syscall // overhead. uint64_t start = ClockTimeNs(); uint64_t end = ClockTimeNs(); uint64_t minres = (end - start); // 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 for (int i = 0; i < 9; ++i) { start = ClockTimeNs(); end = ClockTimeNs(); uint64_t candidate = (start - end); if (candidate < minres) minres = candidate; } if (0 == minres) { // measurable resolution is either incredibly low, ~1ns, or very // high. fall back on clock_getres() struct timespec ts; if (0 == clock_getres(CLOCK_MONOTONIC, &ts)) { minres = TimespecToNs(ts); } } if (0 == minres) { // clock_getres probably failed. fall back on NSPR's resolution // assumption minres = 1 * kNsPerMs; } return minres; } namespace mozilla { double TimeDuration::ToSeconds() const { return double(mValue) / kNsPerSecd; } double TimeDuration::ToSecondsSigDigits() const { // don't report a value < mResolution ... int64_t valueSigDigs = sResolution * (mValue / sResolution); // and chop off insignificant digits valueSigDigs = sResolutionSigDigs * (valueSigDigs / sResolutionSigDigs); return double(valueSigDigs) / kNsPerSecd; } TimeDuration TimeDuration::FromMilliseconds(double aMilliseconds) { return TimeDuration::FromTicks(aMilliseconds * kNsPerMsd); } TimeDuration TimeDuration::Resolution() { return TimeDuration::FromTicks(int64_t(sResolution)); } struct TimeStampInitialization { TimeStampInitialization() { TimeStamp::Startup(); } ~TimeStampInitialization() { TimeStamp::Shutdown(); } }; static TimeStampInitialization initOnce; static bool gInitialized = false; nsresult TimeStamp::Startup() { if (gInitialized) return NS_OK; struct timespec dummy; if (0 != clock_gettime(CLOCK_MONOTONIC, &dummy)) NS_RUNTIMEABORT("CLOCK_MONOTONIC is absent!"); sResolution = ClockResolutionNs(); // find the number of significant digits in sResolution, for the // sake of ToSecondsSigDigits() for (sResolutionSigDigs = 1; !(sResolutionSigDigs == sResolution || 10*sResolutionSigDigs > sResolution); sResolutionSigDigs *= 10); gInitialized = true; sFirstTimeStamp = TimeStamp::Now(); sProcessCreation = TimeStamp(); return NS_OK; } void TimeStamp::Shutdown() { } TimeStamp TimeStamp::Now(bool aHighResolution) { return TimeStamp(ClockTimeNs()); } #if defined(LINUX) || defined(ANDROID) // Calculates the amount of jiffies that have elapsed since boot and up to the // starttime value of a specific process as found in its /proc/*/stat file. // Returns 0 if an error occurred. static uint64_t JiffiesSinceBoot(const char *aFile) { char stat[512]; FILE *f = fopen(aFile, "r"); if (!f) return 0; int n = fread(&stat, 1, sizeof(stat) - 1, f); fclose(f); if (n <= 0) return 0; stat[n] = 0; long long unsigned startTime = 0; // instead of uint64_t to keep GCC quiet char *s = strrchr(stat, ')'); if (!s) return 0; int rv = sscanf(s + 2, "%*c %*d %*d %*d %*d %*d %*u %*u %*u %*u " "%*u %*u %*u %*d %*d %*d %*d %*d %*d %llu", &startTime); if (rv != 1 || !startTime) return 0; return startTime; } // Computes the interval that has elapsed between the thread creation and the // process creation by comparing the starttime fields in the respective // /proc/*/stat files. The resulting value will be a good approximation of the // process uptime. This value will be stored at the address pointed by aTime; // if an error occurred 0 will be stored instead. static void ComputeProcessUptimeThread(void *aTime) { uint64_t *uptime = static_cast(aTime); long hz = sysconf(_SC_CLK_TCK); *uptime = 0; if (!hz) return; char threadStat[40]; sprintf(threadStat, "/proc/self/task/%d/stat", (pid_t) syscall(__NR_gettid)); uint64_t threadJiffies = JiffiesSinceBoot(threadStat); uint64_t selfJiffies = JiffiesSinceBoot("/proc/self/stat"); if (!threadJiffies || !selfJiffies) return; *uptime = ((threadJiffies - selfJiffies) * kNsPerSec) / hz; } // Computes and returns the process uptime in us on Linux & its derivatives. // Returns 0 if an error was encountered. uint64_t TimeStamp::ComputeProcessUptime() { uint64_t uptime = 0; PRThread *thread = PR_CreateThread(PR_USER_THREAD, ComputeProcessUptimeThread, &uptime, PR_PRIORITY_NORMAL, PR_LOCAL_THREAD, PR_JOINABLE_THREAD, 0); PR_JoinThread(thread); return uptime / kNsPerUs; } #elif defined(__DragonFly__) || defined(__FreeBSD__) \ || defined(__NetBSD__) || defined(__OpenBSD__) // Computes and returns the process uptime in us on various BSD flavors. // Returns 0 if an error was encountered. uint64_t TimeStamp::ComputeProcessUptime() { struct timespec ts; int rv = clock_gettime(CLOCK_REALTIME, &ts); if (rv == -1) { return 0; } int mib[] = { CTL_KERN, KERN_PROC, KERN_PROC_PID, getpid(), #if defined(__NetBSD__) || defined(__OpenBSD__) sizeof(KINFO_PROC), 1, #endif }; u_int mibLen = sizeof(mib) / sizeof(mib[0]); KINFO_PROC proc; size_t bufferSize = sizeof(proc); rv = sysctl(mib, mibLen, &proc, &bufferSize, NULL, 0); if (rv == -1) return 0; uint64_t startTime = ((uint64_t)proc.KP_START_SEC * kNsPerSec) + (proc.KP_START_USEC * kNsPerUs); uint64_t now = ((uint64_t)ts.tv_sec * kNsPerSec) + ts.tv_nsec; if (startTime > now) return 0; return (now - startTime) / kNsPerUs; } #else uint64_t TimeStamp::ComputeProcessUptime() { return 0; } #endif } // namespace mozilla