예제 #1
1
void
LateWriteObserver::Observe(IOInterposeObserver::Observation& aOb)
{
#ifdef OBSERVE_LATE_WRITES
  // Crash if that is the shutdown check mode
  if (gShutdownChecks == SCM_CRASH) {
    MOZ_CRASH();
  }

  // If we have shutdown mode SCM_NOTHING or we can't record then abort
  if (gShutdownChecks == SCM_NOTHING || !Telemetry::CanRecordExtended()) {
    return;
  }

  // Write the stack and loaded libraries to a file. We can get here
  // concurrently from many writes, so we use multiple temporary files.
  std::vector<uintptr_t> rawStack;

  NS_StackWalk(RecordStackWalker, /* skipFrames */ 0, /* maxFrames */ 0,
               reinterpret_cast<void*>(&rawStack), 0, nullptr);
  Telemetry::ProcessedStack stack = Telemetry::GetStackAndModules(rawStack);

  nsPrintfCString nameAux("%s%s%s", mProfileDirectory,
                          NS_SLASH, "Telemetry.LateWriteTmpXXXXXX");
  char* name;
  nameAux.GetMutableData(&name);

  // We want the sha1 of the entire file, so please don't write to fd
  // directly; use sha1Stream.
  FILE* stream;
#ifdef XP_WIN
  HANDLE hFile;
  do {
    // mkstemp isn't supported so keep trying until we get a file
    int result = _mktemp_s(name, strlen(name) + 1);
    hFile = CreateFileA(name, GENERIC_WRITE, 0, nullptr, CREATE_NEW,
                        FILE_ATTRIBUTE_NORMAL, nullptr);
  } while (GetLastError() == ERROR_FILE_EXISTS);

  if (hFile == INVALID_HANDLE_VALUE) {
    NS_RUNTIMEABORT("Um, how did we get here?");
  }

  // http://support.microsoft.com/kb/139640
  int fd = _open_osfhandle((intptr_t)hFile, _O_APPEND);
  if (fd == -1) {
    NS_RUNTIMEABORT("Um, how did we get here?");
  }

  stream = _fdopen(fd, "w");
#else
  int fd = mkstemp(name);
  stream = fdopen(fd, "w");
#endif

  SHA1Stream sha1Stream(stream);

  size_t numModules = stack.GetNumModules();
  sha1Stream.Printf("%u\n", (unsigned)numModules);
  for (size_t i = 0; i < numModules; ++i) {
    Telemetry::ProcessedStack::Module module = stack.GetModule(i);
    sha1Stream.Printf("%s %s\n", module.mBreakpadId.c_str(),
                      module.mName.c_str());
  }

  size_t numFrames = stack.GetStackSize();
  sha1Stream.Printf("%u\n", (unsigned)numFrames);
  for (size_t i = 0; i < numFrames; ++i) {
    const Telemetry::ProcessedStack::Frame& frame = stack.GetFrame(i);
    // NOTE: We write the offsets, while the atos tool expects a value with
    // the virtual address added. For example, running otool -l on the the firefox
    // binary shows
    //      cmd LC_SEGMENT_64
    //      cmdsize 632
    //      segname __TEXT
    //      vmaddr 0x0000000100000000
    // so to print the line matching the offset 123 one has to run
    // atos -o firefox 0x100000123.
    sha1Stream.Printf("%d %x\n", frame.mModIndex, (unsigned)frame.mOffset);
  }

  SHA1Sum::Hash sha1;
  sha1Stream.Finish(sha1);

  // Note: These files should be deleted by telemetry once it reads them. If
  // there were no telemetry runs by the time we shut down, we just add files
  // to the existing ones instead of replacing them. Given that each of these
  // files is a bug to be fixed, that is probably the right thing to do.

  // We append the sha1 of the contents to the file name. This provides a simple
  // client side deduplication.
  nsPrintfCString finalName("%s%s", mProfileDirectory,
                            "/Telemetry.LateWriteFinal-");
  for (int i = 0; i < 20; ++i) {
    finalName.AppendPrintf("%02x", sha1[i]);
  }
  PR_Delete(finalName.get());
  PR_Rename(name, finalName.get());
#endif
}
예제 #2
0
void
ThreadMain(void*)
{
  PR_SetCurrentThreadName("Hang Monitor");

  MonitorAutoLock lock(*gMonitor);

  // In order to avoid issues with the hang monitor incorrectly triggering
  // during a general system stop such as sleeping, the monitor thread must
  // run twice to trigger hang protection.
  PRIntervalTime lastTimestamp = 0;
  int waitCount = 0;

#ifdef REPORT_CHROME_HANGS
  Telemetry::ProcessedStack stack;
  int32_t systemUptime = -1;
  int32_t firefoxUptime = -1;
  auto annotations = MakeUnique<ChromeHangAnnotations>();
#endif

  while (true) {
    if (gShutdown) {
      return; // Exit the thread
    }

    // avoid rereading the volatile value in this loop
    PRIntervalTime timestamp = gTimestamp;

    PRIntervalTime now = PR_IntervalNow();

    if (timestamp != PR_INTERVAL_NO_WAIT &&
        now < timestamp) {
      // 32-bit overflow, reset for another waiting period
      timestamp = 1; // lowest legal PRInterval value
    }

    if (timestamp != PR_INTERVAL_NO_WAIT &&
        timestamp == lastTimestamp &&
        gTimeout > 0) {
      ++waitCount;
#ifdef REPORT_CHROME_HANGS
      // Capture the chrome-hang stack + Firefox & system uptimes after
      // the minimum hang duration has been reached (not when the hang ends)
      if (waitCount == 2) {
        GetChromeHangReport(stack, systemUptime, firefoxUptime);
        ChromeHangAnnotatorCallout(*annotations);
      }
#else
      // This is the crash-on-hang feature.
      // See bug 867313 for the quirk in the waitCount comparison
      if (waitCount >= 2) {
        int32_t delay =
          int32_t(PR_IntervalToSeconds(now - timestamp));
        if (delay >= gTimeout) {
          MonitorAutoUnlock unlock(*gMonitor);
          Crash();
        }
      }
#endif
    } else {
#ifdef REPORT_CHROME_HANGS
      if (waitCount >= 2) {
        uint32_t hangDuration = PR_IntervalToSeconds(now - lastTimestamp);
        Telemetry::RecordChromeHang(hangDuration, stack, systemUptime,
                                    firefoxUptime, Move(annotations));
        stack.Clear();
        annotations = MakeUnique<ChromeHangAnnotations>();
      }
#endif
      lastTimestamp = timestamp;
      waitCount = 0;
    }

    PRIntervalTime timeout;
    if (gTimeout <= 0) {
      timeout = PR_INTERVAL_NO_TIMEOUT;
    } else {
      timeout = PR_MillisecondsToInterval(gTimeout * 500);
    }
    lock.Wait(timeout);
  }
}
예제 #3
0
void
ThreadMain(void*)
{
  PR_SetCurrentThreadName("Hang Monitor");

  MonitorAutoLock lock(*gMonitor);

  // In order to avoid issues with the hang monitor incorrectly triggering
  // during a general system stop such as sleeping, the monitor thread must
  // run twice to trigger hang protection.
  PRIntervalTime lastTimestamp = 0;
  int waitCount = 0;

#ifdef REPORT_CHROME_HANGS
  Telemetry::ProcessedStack stack;
#endif

  while (true) {
    if (gShutdown) {
      return; // Exit the thread
    }

    // avoid rereading the volatile value in this loop
    PRIntervalTime timestamp = gTimestamp;

    PRIntervalTime now = PR_IntervalNow();

    if (timestamp != PR_INTERVAL_NO_WAIT &&
        now < timestamp) {
      // 32-bit overflow, reset for another waiting period
      timestamp = 1; // lowest legal PRInterval value
    }

    if (timestamp != PR_INTERVAL_NO_WAIT &&
        timestamp == lastTimestamp &&
        gTimeout > 0) {
      ++waitCount;
      if (waitCount == 2) {
#ifdef REPORT_CHROME_HANGS
        GetChromeHangReport(stack);
#else
        int32_t delay =
          int32_t(PR_IntervalToSeconds(now - timestamp));
        if (delay > gTimeout) {
          MonitorAutoUnlock unlock(*gMonitor);
          Crash();
        }
#endif
      }
    }
    else {
#ifdef REPORT_CHROME_HANGS
      if (waitCount >= 2) {
        uint32_t hangDuration = PR_IntervalToSeconds(now - lastTimestamp);
        Telemetry::RecordChromeHang(hangDuration, stack);
        stack.Clear();
      }
#endif
      lastTimestamp = timestamp;
      waitCount = 0;
    }

    PRIntervalTime timeout;
    if (gTimeout <= 0) {
      timeout = PR_INTERVAL_NO_TIMEOUT;
    }
    else {
      timeout = PR_MillisecondsToInterval(gTimeout * 500);
    }
    lock.Wait(timeout);
  }
}