thread_watcher.cc revision 5821806d5e7f356e8fa4b058a389a808ea183019
15821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// Copyright (c) 2012 The Chromium Authors. All rights reserved.
25821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// Use of this source code is governed by a BSD-style license that can be
35821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// found in the LICENSE file.
45821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
55821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "chrome/browser/metrics/thread_watcher.h"
65821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
75821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include <math.h>  // ceil
85821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
95821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "base/bind.h"
105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "base/compiler_specific.h"
115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "base/debug/alias.h"
125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "base/lazy_instance.h"
135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "base/string_tokenizer.h"
145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "base/threading/thread_restrictions.h"
155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "build/build_config.h"
165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "chrome/browser/metrics/metrics_service.h"
175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "chrome/common/chrome_switches.h"
185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "chrome/common/chrome_version_info.h"
195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "chrome/common/logging_chrome.h"
205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#if defined(OS_WIN)
225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "base/win/windows_version.h"
235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#endif
245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)using content::BrowserThread;
265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)namespace {
285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// The following are unique function names for forcing the crash when a thread
305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// is unresponsive. This makes it possible to tell from the callstack alone what
315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// thread was unresponsive.
325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// We disable optimizations for this block of functions so the compiler doesn't
345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// merge them all together.
355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)MSVC_DISABLE_OPTIMIZE()
365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)MSVC_PUSH_DISABLE_WARNING(4748)
375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)int* NullPointer() {
395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  return reinterpret_cast<int*>(NULL);
405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)NOINLINE void ThreadUnresponsive_UI() {
435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *NullPointer() = __LINE__;
445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)NOINLINE void ThreadUnresponsive_DB() {
475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *NullPointer() = __LINE__;
485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)NOINLINE void ThreadUnresponsive_WEBKIT() {
515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *NullPointer() = __LINE__;
525821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)NOINLINE void ThreadUnresponsive_FILE() {
555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *NullPointer() = __LINE__;
565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
585821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)NOINLINE void ThreadUnresponsive_FILE_USER_BLOCKING() {
595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *NullPointer() = __LINE__;
605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)NOINLINE void ThreadUnresponsive_PROCESS_LAUNCHER() {
635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *NullPointer() = __LINE__;
645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)NOINLINE void ThreadUnresponsive_CACHE() {
675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *NullPointer() = __LINE__;
685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)NOINLINE void ThreadUnresponsive_IO() {
715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *NullPointer() = __LINE__;
725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)MSVC_POP_WARNING()
755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)MSVC_ENABLE_OPTIMIZE();
765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void CrashBecauseThreadWasUnresponsive(BrowserThread::ID thread_id) {
785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::debug::Alias(&thread_id);
795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
805821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  switch (thread_id) {
815821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    case BrowserThread::UI:
825821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      return ThreadUnresponsive_UI();
835821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    case BrowserThread::DB:
845821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      return ThreadUnresponsive_DB();
855821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    case BrowserThread::WEBKIT_DEPRECATED:
865821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      return ThreadUnresponsive_WEBKIT();
875821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    case BrowserThread::FILE:
885821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      return ThreadUnresponsive_FILE();
895821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    case BrowserThread::FILE_USER_BLOCKING:
905821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      return ThreadUnresponsive_FILE_USER_BLOCKING();
915821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    case BrowserThread::PROCESS_LAUNCHER:
925821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      return ThreadUnresponsive_PROCESS_LAUNCHER();
935821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    case BrowserThread::CACHE:
945821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      return ThreadUnresponsive_CACHE();
955821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    case BrowserThread::IO:
965821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      return ThreadUnresponsive_IO();
975821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    case BrowserThread::ID_COUNT:
985821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      CHECK(false);  // This shouldn't actually be reached!
995821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      break;
1005821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1015821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    // Omission of the default hander is intentional -- that way the compiler
1025821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    // should warn if our switch becomes outdated.
1035821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
1045821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1055821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  CHECK(false) << "Unknown thread was unresponsive.";  // Shouldn't be reached.
1065821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
1075821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1085821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}  // namespace
1095821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// ThreadWatcher methods and members.
1115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ThreadWatcher::ThreadWatcher(const WatchingParams& params)
1125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    : thread_id_(params.thread_id),
1135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      thread_name_(params.thread_name),
1145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      watched_loop_(
1155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)          BrowserThread::GetMessageLoopProxyForThread(params.thread_id)),
1165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      sleep_time_(params.sleep_time),
1175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      unresponsive_time_(params.unresponsive_time),
1185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      ping_time_(base::TimeTicks::Now()),
1195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      pong_time_(ping_time_),
1205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      ping_sequence_number_(0),
1215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      active_(false),
1225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      ping_count_(params.unresponsive_threshold),
1235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      response_time_histogram_(NULL),
1245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      unresponsive_time_histogram_(NULL),
1255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      unresponsive_count_(0),
1265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      hung_processing_complete_(false),
1275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      unresponsive_threshold_(params.unresponsive_threshold),
1285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      crash_on_hang_(params.crash_on_hang),
1295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      live_threads_threshold_(params.live_threads_threshold),
1305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      ALLOW_THIS_IN_INITIALIZER_LIST(weak_ptr_factory_(this)) {
1315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
1325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  Initialize();
1335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
1345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ThreadWatcher::~ThreadWatcher() {}
1365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
1385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::StartWatching(const WatchingParams& params) {
1395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK_GE(params.sleep_time.InMilliseconds(), 0);
1405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK_GE(params.unresponsive_time.InMilliseconds(),
1415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)            params.sleep_time.InMilliseconds());
1425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // If we are not on WatchDogThread, then post a task to call StartWatching on
1445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // WatchDogThread.
1455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!WatchDogThread::CurrentlyOnWatchDogThread()) {
1465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    WatchDogThread::PostTask(
1475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)        FROM_HERE,
1485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)        base::Bind(&ThreadWatcher::StartWatching, params));
1495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
1505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
1515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1525821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
1535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Create a new thread watcher object for the given thread and activate it.
1555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ThreadWatcher* watcher = new ThreadWatcher(params);
1565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(watcher);
1585821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // If we couldn't register the thread watcher object, we are shutting down,
1595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // then don't activate thread watching.
1605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!ThreadWatcherList::IsRegistered(params.thread_id))
1615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
1625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  watcher->ActivateThreadWatching();
1635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
1645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::ActivateThreadWatching() {
1665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
1675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (active_) return;
1685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  active_ = true;
1695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ping_count_ = unresponsive_threshold_;
1705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ResetHangCounters();
1715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  MessageLoop::current()->PostTask(
1725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      FROM_HERE,
1735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Bind(&ThreadWatcher::PostPingMessage,
1745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                 weak_ptr_factory_.GetWeakPtr()));
1755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
1765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::DeActivateThreadWatching() {
1785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
1795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  active_ = false;
1805821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ping_count_ = 0;
1815821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  weak_ptr_factory_.InvalidateWeakPtrs();
1825821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
1835821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1845821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::WakeUp() {
1855821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
1865821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // There is some user activity, PostPingMessage task of thread watcher if
1875821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // needed.
1885821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!active_) return;
1895821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1905821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Throw away the previous |unresponsive_count_| and start over again. Just
1915821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // before going to sleep, |unresponsive_count_| could be very close to
1925821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // |unresponsive_threshold_| and when user becomes active,
1935821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // |unresponsive_count_| can go over |unresponsive_threshold_| if there was no
1945821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // response for ping messages. Reset |unresponsive_count_| to start measuring
1955821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // the unresponsiveness of the threads when system becomes active.
1965821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  unresponsive_count_ = 0;
1975821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
1985821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (ping_count_ <= 0) {
1995821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    ping_count_ = unresponsive_threshold_;
2005821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    ResetHangCounters();
2015821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    PostPingMessage();
2025821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  } else {
2035821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    ping_count_ = unresponsive_threshold_;
2045821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
2055821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
2065821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2075821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::PostPingMessage() {
2085821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
2095821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // If we have stopped watching or if the user is idle, then stop sending
2105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // ping messages.
2115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!active_ || ping_count_ <= 0)
2125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
2135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Save the current time when we have sent ping message.
2155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ping_time_ = base::TimeTicks::Now();
2165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Send a ping message to the watched thread. Callback will be called on
2185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // the WatchDogThread.
2195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::Closure callback(
2205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Bind(&ThreadWatcher::OnPongMessage, weak_ptr_factory_.GetWeakPtr(),
2215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                 ping_sequence_number_));
2225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (watched_loop_->PostTask(
2235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)          FROM_HERE,
2245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)          base::Bind(&ThreadWatcher::OnPingMessage, thread_id_,
2255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                     callback))) {
2265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      // Post a task to check the responsiveness of watched thread.
2275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      MessageLoop::current()->PostDelayedTask(
2285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)          FROM_HERE,
2295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)          base::Bind(&ThreadWatcher::OnCheckResponsiveness,
2305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                     weak_ptr_factory_.GetWeakPtr(), ping_sequence_number_),
2315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)          unresponsive_time_);
2325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  } else {
2335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    // Watched thread might have gone away, stop watching it.
2345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    DeActivateThreadWatching();
2355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
2365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
2375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::OnPongMessage(uint64 ping_sequence_number) {
2395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
2405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Record watched thread's response time.
2425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::TimeTicks now = base::TimeTicks::Now();
2435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::TimeDelta response_time = now - ping_time_;
2445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  response_time_histogram_->AddTime(response_time);
2455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Save the current time when we have got pong message.
2475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  pong_time_ = now;
2485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Check if there are any extra pings in flight.
2505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK_EQ(ping_sequence_number_, ping_sequence_number);
2515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (ping_sequence_number_ != ping_sequence_number)
2525821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
2535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Increment sequence number for the next ping message to indicate watched
2555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // thread is responsive.
2565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ++ping_sequence_number_;
2575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2585821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // If we have stopped watching or if the user is idle, then stop sending
2595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // ping messages.
2605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!active_ || --ping_count_ <= 0)
2615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
2625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  MessageLoop::current()->PostDelayedTask(
2645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      FROM_HERE,
2655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Bind(&ThreadWatcher::PostPingMessage,
2665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                 weak_ptr_factory_.GetWeakPtr()),
2675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      sleep_time_);
2685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
2695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::OnCheckResponsiveness(uint64 ping_sequence_number) {
2715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
2725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // If we have stopped watching then consider thread as responding.
2735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!active_) {
2745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    responsive_ = true;
2755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
2765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
2775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // If the latest ping_sequence_number_ is not same as the ping_sequence_number
2785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // that is passed in, then we can assume OnPongMessage was called.
2795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // OnPongMessage increments ping_sequence_number_.
2805821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (ping_sequence_number_ != ping_sequence_number) {
2815821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    // Reset unresponsive_count_ to zero because we got a response from the
2825821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    // watched thread.
2835821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    ResetHangCounters();
2845821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2855821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    responsive_ = true;
2865821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
2875821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
2885821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Record that we got no response from watched thread.
2895821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  GotNoResponse();
2905821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
2915821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Post a task to check the responsiveness of watched thread.
2925821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  MessageLoop::current()->PostDelayedTask(
2935821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      FROM_HERE,
2945821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Bind(&ThreadWatcher::OnCheckResponsiveness,
2955821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                 weak_ptr_factory_.GetWeakPtr(), ping_sequence_number_),
2965821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      unresponsive_time_);
2975821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  responsive_ = false;
2985821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
2995821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3005821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::Initialize() {
3015821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
3025821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ThreadWatcherList::Register(this);
3035821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3045821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  const std::string response_time_histogram_name =
3055821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      "ThreadWatcher.ResponseTime." + thread_name_;
3065821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  response_time_histogram_ = base::Histogram::FactoryTimeGet(
3075821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      response_time_histogram_name,
3085821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::TimeDelta::FromMilliseconds(1),
3095821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::TimeDelta::FromSeconds(100), 50,
3105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Histogram::kUmaTargetedHistogramFlag);
3115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  const std::string unresponsive_time_histogram_name =
3135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      "ThreadWatcher.Unresponsive." + thread_name_;
3145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  unresponsive_time_histogram_ = base::Histogram::FactoryTimeGet(
3155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      unresponsive_time_histogram_name,
3165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::TimeDelta::FromMilliseconds(1),
3175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::TimeDelta::FromSeconds(100), 50,
3185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Histogram::kUmaTargetedHistogramFlag);
3195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  const std::string responsive_count_histogram_name =
3215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      "ThreadWatcher.ResponsiveThreads." + thread_name_;
3225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  responsive_count_histogram_ = base::LinearHistogram::FactoryGet(
3235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      responsive_count_histogram_name, 1, 10, 11,
3245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Histogram::kUmaTargetedHistogramFlag);
3255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  const std::string unresponsive_count_histogram_name =
3275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      "ThreadWatcher.UnresponsiveThreads." + thread_name_;
3285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  unresponsive_count_histogram_ = base::LinearHistogram::FactoryGet(
3295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      unresponsive_count_histogram_name, 1, 10, 11,
3305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Histogram::kUmaTargetedHistogramFlag);
3315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
3325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
3345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::OnPingMessage(const BrowserThread::ID& thread_id,
3355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                  const base::Closure& callback_task) {
3365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // This method is called on watched thread.
3375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(BrowserThread::CurrentlyOn(thread_id));
3385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  WatchDogThread::PostTask(FROM_HERE, callback_task);
3395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
3405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::ResetHangCounters() {
3425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
3435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  unresponsive_count_ = 0;
3445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  hung_processing_complete_ = false;
3455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
3465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcher::GotNoResponse() {
3485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
3495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ++unresponsive_count_;
3515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!IsVeryUnresponsive())
3525821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
3535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Record total unresponsive_time since last pong message.
3555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::TimeDelta unresponse_time = base::TimeTicks::Now() - pong_time_;
3565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  unresponsive_time_histogram_->AddTime(unresponse_time);
3575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3585821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // We have already collected stats for the non-responding watched thread.
3595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (hung_processing_complete_)
3605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
3615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Record how other threads are responding.
3635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  uint32 responding_thread_count = 0;
3645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  uint32 unresponding_thread_count = 0;
3655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ThreadWatcherList::GetStatusOfThreads(&responding_thread_count,
3665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                        &unresponding_thread_count);
3675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Record how many watched threads are responding.
3695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  responsive_count_histogram_->Add(responding_thread_count);
3705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Record how many watched threads are not responding.
3725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  unresponsive_count_histogram_->Add(unresponding_thread_count);
3735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Crash the browser if the watched thread is to be crashed on hang and if the
3755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // number of other threads responding is less than or equal to
3765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // live_threads_threshold_ and at least one other thread is responding.
3775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (crash_on_hang_ &&
3785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      responding_thread_count > 0 &&
3795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      responding_thread_count <= live_threads_threshold_) {
3805821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    static bool crashed_once = false;
3815821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    if (!crashed_once) {
3825821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      crashed_once = true;
3835821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      CrashBecauseThreadWasUnresponsive(thread_id_);
3845821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    }
3855821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
3865821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3875821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  hung_processing_complete_ = true;
3885821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
3895821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3905821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)bool ThreadWatcher::IsVeryUnresponsive() {
3915821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
3925821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  return unresponsive_count_ >= unresponsive_threshold_;
3935821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
3945821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
3955821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// ThreadWatcherList methods and members.
3965821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
3975821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
3985821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ThreadWatcherList* ThreadWatcherList::g_thread_watcher_list_ = NULL;
3995821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4005821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)const int ThreadWatcherList::kSleepSeconds = 1;
4015821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4025821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)const int ThreadWatcherList::kUnresponsiveSeconds = 2;
4035821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4045821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)const int ThreadWatcherList::kUnresponsiveCount = 9;
4055821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4065821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)const int ThreadWatcherList::kLiveThreadsThreshold = 3;
4075821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4085821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4095821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherList::StartWatchingAll(const CommandLine& command_line) {
4105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  uint32 unresponsive_threshold;
4115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  std::set<std::string> crash_on_hang_thread_names;
4125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  uint32 live_threads_threshold;
4135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ParseCommandLine(command_line,
4145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                   &unresponsive_threshold,
4155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                   &crash_on_hang_thread_names,
4165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                   &live_threads_threshold);
4175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ThreadWatcherObserver::SetupNotifications(
4195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::TimeDelta::FromSeconds(kSleepSeconds * unresponsive_threshold));
4205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  WatchDogThread::PostDelayedTask(
4225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      FROM_HERE,
4235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Bind(&ThreadWatcherList::InitializeAndStartWatching,
4245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                 unresponsive_threshold,
4255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                 crash_on_hang_thread_names,
4265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                 live_threads_threshold),
4275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::TimeDelta::FromSeconds(120));
4285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
4295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherList::StopWatchingAll() {
4325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ThreadWatcherObserver::RemoveNotifications();
4335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DeleteAll();
4345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
4355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherList::Register(ThreadWatcher* watcher) {
4385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
4395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!g_thread_watcher_list_)
4405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
4415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(!g_thread_watcher_list_->Find(watcher->thread_id()));
4425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_thread_watcher_list_->registered_[watcher->thread_id()] = watcher;
4435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
4445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)bool ThreadWatcherList::IsRegistered(const BrowserThread::ID thread_id) {
4475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
4485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  return NULL != ThreadWatcherList::Find(thread_id);
4495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
4505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4525821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherList::GetStatusOfThreads(uint32* responding_thread_count,
4535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                           uint32* unresponding_thread_count) {
4545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
4555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *responding_thread_count = 0;
4565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *unresponding_thread_count = 0;
4575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!g_thread_watcher_list_)
4585821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
4595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  for (RegistrationList::iterator it =
4615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)           g_thread_watcher_list_->registered_.begin();
4625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)       g_thread_watcher_list_->registered_.end() != it;
4635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)       ++it) {
4645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    if (it->second->IsVeryUnresponsive())
4655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      ++(*unresponding_thread_count);
4665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    else
4675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      ++(*responding_thread_count);
4685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
4695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
4705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherList::WakeUpAll() {
4735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
4745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!g_thread_watcher_list_)
4755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
4765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  for (RegistrationList::iterator it =
4785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)           g_thread_watcher_list_->registered_.begin();
4795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)       g_thread_watcher_list_->registered_.end() != it;
4805821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)       ++it)
4815821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    it->second->WakeUp();
4825821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
4835821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4845821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ThreadWatcherList::ThreadWatcherList() {
4855821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
4865821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  CHECK(!g_thread_watcher_list_);
4875821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_thread_watcher_list_ = this;
4885821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
4895821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4905821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ThreadWatcherList::~ThreadWatcherList() {
4915821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
4925821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(this == g_thread_watcher_list_);
4935821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_thread_watcher_list_ = NULL;
4945821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
4955821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
4965821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
4975821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherList::ParseCommandLine(
4985821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const CommandLine& command_line,
4995821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    uint32* unresponsive_threshold,
5005821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    std::set<std::string>* crash_on_hang_thread_names,
5015821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    uint32* live_threads_threshold) {
5025821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Determine |unresponsive_threshold| based on switches::kCrashOnHangSeconds.
5035821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *unresponsive_threshold = kUnresponsiveCount;
5045821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5055821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Increase the unresponsive_threshold on the Stable and Beta channels to
5065821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // reduce the number of crashes due to ThreadWatcher.
5075821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  chrome::VersionInfo::Channel channel = chrome::VersionInfo::GetChannel();
5085821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (channel == chrome::VersionInfo::CHANNEL_STABLE) {
5095821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    *unresponsive_threshold *= 4;
5105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  } else if (channel == chrome::VersionInfo::CHANNEL_BETA) {
5115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    *unresponsive_threshold *= 2;
5125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
5135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#if defined(OS_WIN)
5155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // For Windows XP (old systems), double the unresponsive_threshold to give
5165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // the OS a chance to schedule UI/IO threads a time slice to respond with a
5175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // pong message (to get around limitations with the OS).
5185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (base::win::GetVersion() <= base::win::VERSION_XP)
5195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    *unresponsive_threshold *= 2;
5205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#endif
5215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  std::string crash_on_hang_seconds =
5235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      command_line.GetSwitchValueASCII(switches::kCrashOnHangSeconds);
5245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!crash_on_hang_seconds.empty()) {
5255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    int crash_seconds = atoi(crash_on_hang_seconds.c_str());
5265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    if (crash_seconds > 0) {
5275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      *unresponsive_threshold = static_cast<uint32>(
5285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)          ceil(static_cast<float>(crash_seconds) / kUnresponsiveSeconds));
5295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    }
5305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
5315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  std::string crash_on_hang_threads;
5335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Default to crashing the browser if UI or IO threads are not responsive
5355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // except in stable channel.
5365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (channel == chrome::VersionInfo::CHANNEL_STABLE)
5375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    crash_on_hang_threads = "";
5385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  else
5395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    crash_on_hang_threads = "UI,IO";
5405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (command_line.HasSwitch(switches::kCrashOnHangThreads)) {
5425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    crash_on_hang_threads =
5435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)        command_line.GetSwitchValueASCII(switches::kCrashOnHangThreads);
5445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
5455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  StringTokenizer tokens(crash_on_hang_threads, ",");
5465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  while (tokens.GetNext())
5475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    crash_on_hang_thread_names->insert(tokens.token());
5485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Determine |live_threads_threshold| based on switches::kCrashOnLive.
5505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  *live_threads_threshold = kLiveThreadsThreshold;
5515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (command_line.HasSwitch(switches::kCrashOnLive)) {
5525821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    std::string live_threads =
5535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)        command_line.GetSwitchValueASCII(switches::kCrashOnLive);
5545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    *live_threads_threshold = static_cast<uint32>(atoi(live_threads.c_str()));
5555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
5565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
5575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5585821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
5595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherList::InitializeAndStartWatching(
5605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    uint32 unresponsive_threshold,
5615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const std::set<std::string>& crash_on_hang_thread_names,
5625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    uint32 live_threads_threshold) {
5635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
5645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ThreadWatcherList* thread_watcher_list = new ThreadWatcherList();
5665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  CHECK(thread_watcher_list);
5675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  BrowserThread::PostTask(
5695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      BrowserThread::UI,
5705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      FROM_HERE,
5715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Bind(&StartupTimeBomb::DisarmStartupTimeBomb));
5725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  const base::TimeDelta kSleepTime =
5745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::TimeDelta::FromSeconds(kSleepSeconds);
5755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  const base::TimeDelta kUnresponsiveTime =
5765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::TimeDelta::FromSeconds(kUnresponsiveSeconds);
5775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  StartWatching(BrowserThread::UI, "UI", kSleepTime, kUnresponsiveTime,
5795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                unresponsive_threshold, crash_on_hang_thread_names,
5805821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                live_threads_threshold);
5815821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  StartWatching(BrowserThread::IO, "IO", kSleepTime, kUnresponsiveTime,
5825821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                unresponsive_threshold, crash_on_hang_thread_names,
5835821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                live_threads_threshold);
5845821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  StartWatching(BrowserThread::DB, "DB", kSleepTime, kUnresponsiveTime,
5855821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                unresponsive_threshold, crash_on_hang_thread_names,
5865821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                live_threads_threshold);
5875821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  StartWatching(BrowserThread::FILE, "FILE", kSleepTime, kUnresponsiveTime,
5885821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                unresponsive_threshold, crash_on_hang_thread_names,
5895821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                live_threads_threshold);
5905821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  StartWatching(BrowserThread::FILE_USER_BLOCKING, "FILE_USER_BLOCKING",
5915821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                kSleepTime, kUnresponsiveTime,
5925821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                unresponsive_threshold, crash_on_hang_thread_names,
5935821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                live_threads_threshold);
5945821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  StartWatching(BrowserThread::CACHE, "CACHE", kSleepTime, kUnresponsiveTime,
5955821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                unresponsive_threshold, crash_on_hang_thread_names,
5965821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                live_threads_threshold);
5975821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
5985821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
5995821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
6005821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherList::StartWatching(
6015821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const BrowserThread::ID& thread_id,
6025821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const std::string& thread_name,
6035821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const base::TimeDelta& sleep_time,
6045821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const base::TimeDelta& unresponsive_time,
6055821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    uint32 unresponsive_threshold,
6065821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const std::set<std::string>& crash_on_hang_thread_names,
6075821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    uint32 live_threads_threshold) {
6085821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
6095821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  std::set<std::string>::const_iterator it =
6115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      crash_on_hang_thread_names.find(thread_name);
6125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  bool crash_on_hang = (it != crash_on_hang_thread_names.end());
6135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ThreadWatcher::StartWatching(
6155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      ThreadWatcher::WatchingParams(thread_id,
6165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                    thread_name,
6175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                    sleep_time,
6185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                    unresponsive_time,
6195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                    unresponsive_threshold,
6205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                    crash_on_hang,
6215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                    live_threads_threshold));
6225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
6235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
6255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherList::DeleteAll() {
6265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!WatchDogThread::CurrentlyOnWatchDogThread()) {
6275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    WatchDogThread::PostTask(
6285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)        FROM_HERE,
6295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)        base::Bind(&ThreadWatcherList::DeleteAll));
6305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
6315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
6325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
6345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!g_thread_watcher_list_)
6355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
6365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Delete all thread watcher objects.
6385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  while (!g_thread_watcher_list_->registered_.empty()) {
6395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    RegistrationList::iterator it = g_thread_watcher_list_->registered_.begin();
6405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    delete it->second;
6415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    g_thread_watcher_list_->registered_.erase(it);
6425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
6435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  delete g_thread_watcher_list_;
6455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
6465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
6485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ThreadWatcher* ThreadWatcherList::Find(const BrowserThread::ID& thread_id) {
6495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(WatchDogThread::CurrentlyOnWatchDogThread());
6505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!g_thread_watcher_list_)
6515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return NULL;
6525821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  RegistrationList::iterator it =
6535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      g_thread_watcher_list_->registered_.find(thread_id);
6545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (g_thread_watcher_list_->registered_.end() == it)
6555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return NULL;
6565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  return it->second;
6575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
6585821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// ThreadWatcherObserver methods and members.
6605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
6615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
6625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ThreadWatcherObserver* ThreadWatcherObserver::g_thread_watcher_observer_ = NULL;
6635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ThreadWatcherObserver::ThreadWatcherObserver(
6655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const base::TimeDelta& wakeup_interval)
6665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    : last_wakeup_time_(base::TimeTicks::Now()),
6675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      wakeup_interval_(wakeup_interval) {
6685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  CHECK(!g_thread_watcher_observer_);
6695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_thread_watcher_observer_ = this;
6705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
6715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ThreadWatcherObserver::~ThreadWatcherObserver() {
6735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(this == g_thread_watcher_observer_);
6745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_thread_watcher_observer_ = NULL;
6755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
6765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
6785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherObserver::SetupNotifications(
6795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const base::TimeDelta& wakeup_interval) {
6805821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
6815821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  ThreadWatcherObserver* observer = new ThreadWatcherObserver(wakeup_interval);
6825821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  MetricsService::SetUpNotifications(&observer->registrar_, observer);
6835821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
6845821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6855821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
6865821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherObserver::RemoveNotifications() {
6875821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
6885821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (!g_thread_watcher_observer_)
6895821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
6905821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_thread_watcher_observer_->registrar_.RemoveAll();
6915821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  delete g_thread_watcher_observer_;
6925821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
6935821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
6945821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ThreadWatcherObserver::Observe(
6955821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    int type,
6965821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const content::NotificationSource& source,
6975821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const content::NotificationDetails& details) {
6985821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // There is some user activity, see if thread watchers are to be awakened.
6995821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::TimeTicks now = base::TimeTicks::Now();
7005821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if ((now - last_wakeup_time_) < wakeup_interval_)
7015821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
7025821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  last_wakeup_time_ = now;
7035821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  WatchDogThread::PostTask(
7045821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      FROM_HERE,
7055821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Bind(&ThreadWatcherList::WakeUpAll));
7065821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
7075821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7085821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// WatchDogThread methods and members.
7095821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// This lock protects g_watchdog_thread.
7115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)static base::LazyInstance<base::Lock>::Leaky
7125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    g_watchdog_lock = LAZY_INSTANCE_INITIALIZER;
7135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// The singleton of this class.
7155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)static WatchDogThread* g_watchdog_thread = NULL;
7165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)WatchDogThread::WatchDogThread() : Thread("BrowserWatchdog") {
7185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
7195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)WatchDogThread::~WatchDogThread() {
7215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  Stop();
7225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
7235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
7255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)bool WatchDogThread::CurrentlyOnWatchDogThread() {
7265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::AutoLock lock(g_watchdog_lock.Get());
7275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  return g_watchdog_thread &&
7285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    g_watchdog_thread->message_loop() == MessageLoop::current();
7295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
7305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
7325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)bool WatchDogThread::PostTask(const tracked_objects::Location& from_here,
7335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                              const base::Closure& task) {
7345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  return PostTaskHelper(from_here, task, base::TimeDelta());
7355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
7365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
7385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)bool WatchDogThread::PostDelayedTask(const tracked_objects::Location& from_here,
7395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                     const base::Closure& task,
7405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                                     base::TimeDelta delay) {
7415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  return PostTaskHelper(from_here, task, delay);
7425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
7435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
7455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)bool WatchDogThread::PostTaskHelper(
7465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const tracked_objects::Location& from_here,
7475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    const base::Closure& task,
7485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    base::TimeDelta delay) {
7495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  {
7505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    base::AutoLock lock(g_watchdog_lock.Get());
7515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7525821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    MessageLoop* message_loop = g_watchdog_thread ?
7535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)        g_watchdog_thread->message_loop() : NULL;
7545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    if (message_loop) {
7555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      message_loop->PostDelayedTask(from_here, task, delay);
7565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      return true;
7575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    }
7585821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
7595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  return false;
7615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
7625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void WatchDogThread::Init() {
7645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // This thread shouldn't be allowed to perform any blocking disk I/O.
7655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::ThreadRestrictions::SetIOAllowed(false);
7665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::AutoLock lock(g_watchdog_lock.Get());
7685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  CHECK(!g_watchdog_thread);
7695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_watchdog_thread = this;
7705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
7715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void WatchDogThread::CleanUp() {
7735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::AutoLock lock(g_watchdog_lock.Get());
7745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_watchdog_thread = NULL;
7755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
7765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)namespace {
7785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// StartupWatchDogThread methods and members.
7805821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
7815821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// Class for detecting hangs during startup.
7825821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)class StartupWatchDogThread : public base::Watchdog {
7835821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles) public:
7845821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Constructor specifies how long the StartupWatchDogThread will wait before
7855821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // alarming.
7865821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  explicit StartupWatchDogThread(const base::TimeDelta& duration)
7875821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      : base::Watchdog(duration, "Startup watchdog thread", true) {
7885821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
7895821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
7905821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Alarm is called if the time expires after an Arm() without someone calling
7915821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Disarm(). When Alarm goes off, in release mode we get the crash dump
7925821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // without crashing and in debug mode we break into the debugger.
7935821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  virtual void Alarm() {
7945821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#ifndef NDEBUG
7955821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    DCHECK(false);
7965821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#else
7975821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    logging::DumpWithoutCrashing();
7985821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#endif
7995821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
8005821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8015821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DISALLOW_COPY_AND_ASSIGN(StartupWatchDogThread);
8025821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)};
8035821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8045821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// ShutdownWatchDogThread methods and members.
8055821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
8065821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// Class for detecting hangs during shutdown.
8075821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)class ShutdownWatchDogThread : public base::Watchdog {
8085821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles) public:
8095821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Constructor specifies how long the ShutdownWatchDogThread will wait before
8105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // alarming.
8115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  explicit ShutdownWatchDogThread(const base::TimeDelta& duration)
8125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      : base::Watchdog(duration, "Shutdown watchdog thread", true) {
8135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
8145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Alarm is called if the time expires after an Arm() without someone calling
8165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // Disarm(). We crash the browser if this method is called.
8175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  virtual void Alarm() {
8185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    CHECK(false);
8195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
8205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DISALLOW_COPY_AND_ASSIGN(ShutdownWatchDogThread);
8225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)};
8235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}  // namespace
8245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// StartupTimeBomb methods and members.
8265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
8275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
8285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)StartupTimeBomb* StartupTimeBomb::g_startup_timebomb_ = NULL;
8295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)StartupTimeBomb::StartupTimeBomb()
8315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    : startup_watchdog_(NULL),
8325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      thread_id_(base::PlatformThread::CurrentId()) {
8335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  CHECK(!g_startup_timebomb_);
8345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_startup_timebomb_ = this;
8355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
8365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)StartupTimeBomb::~StartupTimeBomb() {
8385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(this == g_startup_timebomb_);
8395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK_EQ(thread_id_, base::PlatformThread::CurrentId());
8405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (startup_watchdog_)
8415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    Disarm();
8425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  g_startup_timebomb_ = NULL;
8435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
8445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void StartupTimeBomb::Arm(const base::TimeDelta& duration) {
8465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK_EQ(thread_id_, base::PlatformThread::CurrentId());
8475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(!startup_watchdog_);
8485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // TODO(rtenneti): http://crbug.com/112970. Don't arm the startup timebomb
8495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // until we fix breakpad code not to crash in logging::DumpWithoutCrashing().
8505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // startup_watchdog_ = new StartupWatchDogThread(duration);
8515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // startup_watchdog_->Arm();
8525821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  return;
8535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
8545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void StartupTimeBomb::Disarm() {
8565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK_EQ(thread_id_, base::PlatformThread::CurrentId());
8575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (startup_watchdog_) {
8585821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    startup_watchdog_->Disarm();
8595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    startup_watchdog_->Cleanup();
8605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    DeleteStartupWatchdog();
8615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
8625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
8635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void StartupTimeBomb::DeleteStartupWatchdog() {
8655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK_EQ(thread_id_, base::PlatformThread::CurrentId());
8665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (startup_watchdog_->IsJoinable()) {
8675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    // Allow the watchdog thread to shutdown on UI. Watchdog thread shutdowns
8685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    // very fast.
8695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    base::ThreadRestrictions::SetIOAllowed(true);
8705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    delete startup_watchdog_;
8715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    startup_watchdog_ = NULL;
8725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    return;
8735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
8745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  MessageLoop::current()->PostDelayedTask(
8755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      FROM_HERE,
8765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::Bind(&StartupTimeBomb::DeleteStartupWatchdog,
8775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)                 base::Unretained(this)),
8785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      base::TimeDelta::FromSeconds(10));
8795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
8805821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8815821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// static
8825821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void StartupTimeBomb::DisarmStartupTimeBomb() {
8835821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(BrowserThread::CurrentlyOn(BrowserThread::UI));
8845821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (g_startup_timebomb_)
8855821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    g_startup_timebomb_->Disarm();
8865821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
8875821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8885821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// ShutdownWatcherHelper methods and members.
8895821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)//
8905821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// ShutdownWatcherHelper is a wrapper class for detecting hangs during
8915821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)// shutdown.
8925821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ShutdownWatcherHelper::ShutdownWatcherHelper()
8935821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    : shutdown_watchdog_(NULL),
8945821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)      thread_id_(base::PlatformThread::CurrentId()) {
8955821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
8965821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
8975821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)ShutdownWatcherHelper::~ShutdownWatcherHelper() {
8985821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK_EQ(thread_id_, base::PlatformThread::CurrentId());
8995821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (shutdown_watchdog_) {
9005821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    shutdown_watchdog_->Disarm();
9015821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    delete shutdown_watchdog_;
9025821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    shutdown_watchdog_ = NULL;
9035821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
9045821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
9055821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
9065821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)void ShutdownWatcherHelper::Arm(const base::TimeDelta& duration) {
9075821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK_EQ(thread_id_, base::PlatformThread::CurrentId());
9085821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  DCHECK(!shutdown_watchdog_);
9095821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  base::TimeDelta actual_duration = duration;
9105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
9115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  chrome::VersionInfo::Channel channel = chrome::VersionInfo::GetChannel();
9125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (channel == chrome::VersionInfo::CHANNEL_STABLE) {
9135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    actual_duration *= 20;
9145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  } else if (channel == chrome::VersionInfo::CHANNEL_BETA ||
9155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)             channel == chrome::VersionInfo::CHANNEL_DEV) {
9165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    actual_duration *= 10;
9175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  }
9185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
9195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#if defined(OS_WIN)
9205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  // On Windows XP, give twice the time for shutdown.
9215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if (base::win::GetVersion() <= base::win::VERSION_XP)
9225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)    actual_duration *= 2;
9235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#endif
9245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
9255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  shutdown_watchdog_ = new ShutdownWatchDogThread(actual_duration);
9265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  shutdown_watchdog_->Arm();
9275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
928