Update task tracking to not depend on message_loop_ singleton

I also did a bunch of cleanup, and transitioned to tracking
the duration of a run, rather than the time from posting
(construction of a task) to completion of the run.  It is
less interesting for now to track queueing delay, and we
need a focus on task execution time.  I left in the hook
(API) with the expectation that I'll be extending the
tracked_objects code to include this as well.

I also landed changes to run in Linux/Mac.  The fact that
I've punted on shutdown made this landing easy (all
code support was previously lost during migration to
some flavor of bind support).

r=willchan,jam,viettrungluu,ajwong
BUG=62728
Review URL: http://codereview.chromium.org/8233037

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@105694 0039d316-1c4b-4281-b951-d872f2087c98

8 files changed, 356 insertions, 396 deletions

diff --git a/base/message_loop.cc b/base/message_loop.cc
index cadb944..ce5e900 100644
--- a/base/message_loop.cc
+++ b/base/message_loop.cc
@@ -476,23 +476,25 @@ void MessageLoop::RunTask(const PendingTask& pending_task) {
   base::debug::Alias(&program_counter);
 
   HistogramEvent(kTaskRunEvent);
+
+#if defined(TRACK_ALL_TASK_OBJECTS)
+  TimeTicks start_of_run = tracked_objects::ThreadData::Now();
+#endif  // defined(TRACK_ALL_TASK_OBJECTS)
+
   FOR_EACH_OBSERVER(TaskObserver, task_observers_,
                     WillProcessTask(pending_task.time_posted));
   pending_task.task.Run();
   FOR_EACH_OBSERVER(TaskObserver, task_observers_,
                     DidProcessTask(pending_task.time_posted));
-
 #if defined(TRACK_ALL_TASK_OBJECTS)
-  tracked_objects::ThreadData::TallyADeathIfActive(
-      pending_task.post_births,
-      TimeTicks::Now() - pending_task.time_posted);
+  tracked_objects::ThreadData::TallyADeathIfActive(pending_task.post_births,
+    pending_task.time_posted, pending_task.delayed_run_time, start_of_run);
 #endif  // defined(TRACK_ALL_TASK_OBJECTS)
 
   nestable_tasks_allowed_ = true;
 }
 
-bool MessageLoop::DeferOrRunPendingTask(
-    const PendingTask& pending_task) {
+bool MessageLoop::DeferOrRunPendingTask(const PendingTask& pending_task) {
   if (pending_task.nestable || state_->run_depth == 1) {
     RunTask(pending_task);
     // Show that we ran a task (Note: a new one might arrive as a
diff --git a/base/threading/worker_pool_posix.cc b/base/threading/worker_pool_posix.cc
index 3a4408c..b6c30e5 100644
--- a/base/threading/worker_pool_posix.cc
+++ b/base/threading/worker_pool_posix.cc
@@ -87,7 +87,15 @@ void WorkerThread::ThreadMain() {
     UNSHIPPED_TRACE_EVENT2("task", "WorkerThread::ThreadMain::Run",
         "src_file", pending_task.posted_from.file_name(),
         "src_func", pending_task.posted_from.function_name());
+
+#if defined(TRACK_ALL_TASK_OBJECTS)
+    TimeTicks start_of_run = tracked_objects::ThreadData::Now();
+#endif  // defined(TRACK_ALL_TASK_OBJECTS)
     pending_task.task.Run();
+#if defined(TRACK_ALL_TASK_OBJECTS)
+    tracked_objects::ThreadData::TallyADeathIfActive(pending_task.post_births,
+        pending_task.time_posted, TimeTicks::TimeTicks(), start_of_run);
+#endif  // defined(TRACK_ALL_TASK_OBJECTS)
   }
 
   // The WorkerThread is non-joinable, so it deletes itself.
@@ -113,6 +121,10 @@ PosixDynamicThreadPool::PendingTask::PendingTask(
     const base::Closure& task)
     : posted_from(posted_from),
       task(task) {
+#if defined(TRACK_ALL_TASK_OBJECTS)
+  post_births = tracked_objects::ThreadData::TallyABirthIfActive(posted_from);
+  time_posted = tracked_objects::ThreadData::Now();
+#endif  // defined(TRACK_ALL_TASK_OBJECTS)
 }
 
 PosixDynamicThreadPool::PendingTask::~PendingTask() {
diff --git a/base/threading/worker_pool_posix.h b/base/threading/worker_pool_posix.h
index 59593e2..c0a60cc 100644
--- a/base/threading/worker_pool_posix.h
+++ b/base/threading/worker_pool_posix.h
@@ -31,11 +31,13 @@
 #include "base/basictypes.h"
 #include "base/callback.h"
 #include "base/location.h"
+#include "base/time.h"
 #include "base/memory/ref_counted.h"
 #include "base/memory/scoped_ptr.h"
 #include "base/synchronization/condition_variable.h"
 #include "base/synchronization/lock.h"
 #include "base/threading/platform_thread.h"
+#include "base/tracked_objects.h"
 
 class Task;
 
@@ -50,10 +52,14 @@ class BASE_EXPORT PosixDynamicThreadPool
     PendingTask(const tracked_objects::Location& posted_from,
                 const base::Closure& task);
     ~PendingTask();
-    // TODO(ajwong): After we figure out why Mac's ~AtExitManager dies when
-    // destructing the lock, add in extra info so we can call
-    // tracked_objects::TallyADeathIfActive() and
-    // tracked_objects::TallyABirthIfActive correctly.
+
+#if defined(TRACK_ALL_TASK_OBJECTS)
+    // Counter for location where the Closure was posted from.
+    tracked_objects::Births* post_births;
+
+    // Time the task was posted.
+    TimeTicks time_posted;
+#endif  // defined(TRACK_ALL_TASK_OBJECTS)
 
     const tracked_objects::Location posted_from;
 
diff --git a/base/threading/worker_pool_win.cc b/base/threading/worker_pool_win.cc
index b7eaad1..671e004 100644
--- a/base/threading/worker_pool_win.cc
+++ b/base/threading/worker_pool_win.cc
@@ -22,7 +22,7 @@ struct PendingTask {
         task(task) {
 #if defined(TRACK_ALL_TASK_OBJECTS)
     post_births = tracked_objects::ThreadData::TallyABirthIfActive(posted_from);
-    time_posted = TimeTicks::Now();
+    time_posted = tracked_objects::ThreadData::Now();
 #endif  // defined(TRACK_ALL_TASK_OBJECTS)
   }
 
@@ -46,12 +46,16 @@ DWORD CALLBACK WorkItemCallback(void* param) {
   UNSHIPPED_TRACE_EVENT2("task", "WorkItemCallback::Run",
                          "src_file", pending_task->posted_from.file_name(),
                          "src_func", pending_task->posted_from.function_name());
+
+#if defined(TRACK_ALL_TASK_OBJECTS)
+  TimeTicks start_of_run = tracked_objects::ThreadData::Now();
+#endif  // defined(TRACK_ALL_TASK_OBJECTS)
   pending_task->task.Run();
 #if defined(TRACK_ALL_TASK_OBJECTS)
-  tracked_objects::ThreadData::TallyADeathIfActive(
-      pending_task->post_births,
-      TimeTicks::Now() - pending_task->time_posted);
+  tracked_objects::ThreadData::TallyADeathIfActive(pending_task->post_births,
+      pending_task->time_posted, TimeTicks::TimeTicks(), start_of_run);
 #endif  // defined(TRACK_ALL_TASK_OBJECTS)
+
   delete pending_task;
   return 0;
 }
diff --git a/base/tracked_objects.cc b/base/tracked_objects.cc
index 5df017b..58a9afd 100644
--- a/base/tracked_objects.cc
+++ b/base/tracked_objects.cc
@@ -24,55 +24,57 @@ base::ThreadLocalStorage::Slot ThreadData::tls_index_(base::LINKER_INITIALIZED);
 // static
 AutoTracking::State AutoTracking::state_ = AutoTracking::kNeverBeenRun;
 
+// A locked protected counter to assign sequence number to threads.
+// static
+int ThreadData::thread_number_counter = 0;
+
 //------------------------------------------------------------------------------
 // Death data tallies durations when a death takes place.
 
-void DeathData::RecordDeath(const TimeDelta& duration) {
+void DeathData::RecordDeath(const TimeDelta& queue_duration,
+                            const TimeDelta& run_duration) {
   ++count_;
-  life_duration_ += duration;
-  int64 milliseconds = duration.InMilliseconds();
-  square_duration_ += milliseconds * milliseconds;
+  queue_duration_ += queue_duration;
+  run_duration_ += run_duration;
 }
 
-int DeathData::AverageMsDuration() const {
-  return static_cast<int>(life_duration_.InMilliseconds() / count_);
+int DeathData::AverageMsRunDuration() const {
+  return static_cast<int>(run_duration_.InMilliseconds() / count_);
 }
 
-double DeathData::StandardDeviation() const {
-  double average = AverageMsDuration();
-  double variance = static_cast<float>(square_duration_)/count_
-                    - average * average;
-  return sqrt(variance);
+int DeathData::AverageMsQueueDuration() const {
+  return static_cast<int>(queue_duration_.InMilliseconds() / count_);
 }
 
-
 void DeathData::AddDeathData(const DeathData& other) {
   count_ += other.count_;
-  life_duration_ += other.life_duration_;
-  square_duration_ += other.square_duration_;
+  queue_duration_ += other.queue_duration_;
+  run_duration_ += other.run_duration_;
 }
 
 void DeathData::Write(std::string* output) const {
   if (!count_)
     return;
-  if (1 == count_) {
-    base::StringAppendF(output, "(1)Life in %dms ", AverageMsDuration());
-  } else {
-    base::StringAppendF(output, "(%d)Lives %dms/life ",
-                        count_, AverageMsDuration());
-  }
+  base::StringAppendF(output, "%s:%d, ",
+                      (count_ == 1) ? "Life" : "Lives", count_);
+  base::StringAppendF(output, "Run:%"PRId64"ms(%dms/life) ",
+                      run_duration_.InMilliseconds(),
+                      AverageMsRunDuration());
+  base::StringAppendF(output, "Queue:%"PRId64"ms(%dms/life) ",
+                      queue_duration_.InMilliseconds(),
+                      AverageMsQueueDuration());
 }
 
 void DeathData::Clear() {
   count_ = 0;
-  life_duration_ = TimeDelta();
-  square_duration_ = 0;
+  queue_duration_ = TimeDelta();
+  run_duration_ = TimeDelta();
 }
 
 //------------------------------------------------------------------------------
 BirthOnThread::BirthOnThread(const Location& location)
     : location_(location),
-      birth_thread_(ThreadData::current()) { }
+      birth_thread_(ThreadData::Get()) { }
 
 //------------------------------------------------------------------------------
 Births::Births(const Location& location)
@@ -90,45 +92,66 @@ base::Lock ThreadData::list_lock_;
 // static
 ThreadData::Status ThreadData::status_ = ThreadData::UNINITIALIZED;
 
+ThreadData::ThreadData(const std::string& suggested_name) : next_(NULL) {
+  DCHECK_GE(suggested_name.size(), 0u);
+  thread_name_ = suggested_name;
+}
+
 ThreadData::ThreadData() : next_(NULL) {
-  // This shouldn't use the MessageLoop::current() LazyInstance since this might
-  // be used on a non-joinable thread.
-  // http://crbug.com/62728
-  base::ThreadRestrictions::ScopedAllowSingleton scoped_allow_singleton;
-  message_loop_ = MessageLoop::current();
+  int thread_number;
+  {
+    base::AutoLock lock(list_lock_);
+    thread_number = ++thread_number_counter;
+  }
+  base::StringAppendF(&thread_name_, "WorkerThread-%d", thread_number);
 }
 
 ThreadData::~ThreadData() {}
 
 // static
-ThreadData* ThreadData::current() {
+void ThreadData::InitializeThreadContext(const std::string& suggested_name) {
   if (!tls_index_.initialized())
-    return NULL;
+    return;  // For unittests only.
+  RegisterCurrentContext(new ThreadData(suggested_name));
+}
 
-  ThreadData* registry = static_cast<ThreadData*>(tls_index_.Get());
-  if (!registry) {
-    // We have to create a new registry for ThreadData.
-    bool too_late_to_create = false;
-    {
-      registry = new ThreadData;
-      base::AutoLock lock(list_lock_);
-      // Use lock to insure we have most recent status.
-      if (!IsActive()) {
-        too_late_to_create = true;
-      } else {
-        // Use lock to insert into list.
-        registry->next_ = first_;
-        first_ = registry;
-      }
-    }  // Release lock.
-    if (too_late_to_create) {
-      delete registry;
-      registry = NULL;
+// static
+ThreadData* ThreadData::Get() {
+  if (!tls_index_.initialized())
+    return NULL;  // For unittests only.
+  ThreadData* registered = static_cast<ThreadData*>(tls_index_.Get());
+  if (!registered) {
+    // We have to create a new registry entry for this ThreadData.
+    // TODO(jar): Host all unamed (Worker) threads in *one* ThreadData instance,
+    // (with locking protection on that instance) or else recycle and re-use
+    // worker thread ThreadData when the worker thread terminates.
+    registered = RegisterCurrentContext(new ThreadData());
+  }
+  return registered;
+}
+
+// static
+ThreadData* ThreadData::RegisterCurrentContext(ThreadData* unregistered) {
+  DCHECK_EQ(tls_index_.Get(), static_cast<void*>(0));
+  bool too_late_to_register = false;
+  {
+    base::AutoLock lock(list_lock_);
+    // Use lock to insure we have most recent status.
+    if (!IsActive()) {
+      too_late_to_register = true;
     } else {
-      tls_index_.Set(registry);
+      // Use list_lock_ to insert as new head of list.
+      unregistered->next_ = first_;
+      first_ = unregistered;
     }
+  }  // Release lock.
+  if (too_late_to_register) {
+    delete unregistered;
+    unregistered = NULL;
+  } else {
+    tls_index_.Set(unregistered);
   }
-  return registry;
+  return unregistered;
 }
 
 // static
@@ -136,7 +159,6 @@ void ThreadData::WriteHTML(const std::string& query, std::string* output) {
   if (!ThreadData::IsActive())
     return;  // Not yet initialized.
 
-  DCHECK(ThreadData::current());
   DataCollector collected_data;  // Gather data.
   collected_data.AddListOfLivingObjects();  // Add births that are still alive.
 
@@ -165,19 +187,26 @@ void ThreadData::WriteHTML(const std::string& query, std::string* output) {
 
   const char* help_string = "The following are the keywords that can be used to"
     " sort and aggregate the data, or to select data.<br><ul>"
-    "<li><b>count</b> Number of instances seen."
-    "<li><b>duration</b> Duration in ms from construction to descrution."
-    "<li><b>birth</b> Thread on which the task was constructed."
-    "<li><b>death</b> Thread on which the task was run and deleted."
-    "<li><b>file</b> File in which the task was contructed."
-    "<li><b>function</b> Function in which the task was constructed."
-    "<li><b>line</b> Line number of the file in which the task was constructed."
+    "<li><b>Count</b> Number of instances seen."
+    "<li><b>Duration</b> Average duration in ms of Run() time."
+    "<li><b>TotalDuration</b> Summed durations in ms of Run() times."
+    "<li><b>AverageQueueDuration</b> Average duration in ms of queueing time."
+    "<li><b>TotalQueueDuration</b> Summed durations in ms of Run() times."
+    "<li><b>Birth</b> Thread on which the task was constructed."
+    "<li><b>Death</b> Thread on which the task was run and deleted."
+    "<li><b>File</b> File in which the task was contructed."
+    "<li><b>Function</b> Function in which the task was constructed."
+    "<li><b>Line</b> Line number of the file in which the task was constructed."
     "</ul><br>"
     "As examples:<ul>"
     "<li><b>about:tracking/file</b> would sort the above data by file, and"
     " aggregate data on a per-file basis."
     "<li><b>about:tracking/file=Dns</b> would only list data for tasks"
     " constructed in a file containing the text |Dns|."
+    "<li><b>about:tracking/death/duration</b> would sort the data by death"
+    " thread(i.e., where tasks ran) and then by the average runtime for the"
+    " tasks. Form an aggregation group, one per thread, showing the results on"
+    " each thread."
     "<li><b>about:tracking/birth/death</b> would sort the above list by birth"
     " thread, and then by death thread, and would aggregate data for each pair"
     " of lifetime events."
@@ -234,15 +263,6 @@ void ThreadData::WriteHTMLTotalAndSubtotals(
 }
 
 Births* ThreadData::TallyABirth(const Location& location) {
-  {
-    // This shouldn't use the MessageLoop::current() LazyInstance since this
-    // might be used on a non-joinable thread.
-    // http://crbug.com/62728
-    base::ThreadRestrictions::ScopedAllowSingleton scoped_allow_singleton;
-    if (!message_loop_)  // In case message loop wasn't yet around...
-      message_loop_ = MessageLoop::current();  // Find it now.
-  }
-
   BirthMap::iterator it = birth_map_.find(location);
   if (it != birth_map_.end()) {
     it->second->RecordBirth();
@@ -257,43 +277,59 @@ Births* ThreadData::TallyABirth(const Location& location) {
   return tracker;
 }
 
-void ThreadData::TallyADeath(const Births& lifetimes,
-                             const TimeDelta& duration) {
-  {
-    // http://crbug.com/62728
-    base::ThreadRestrictions::ScopedAllowSingleton scoped_allow_singleton;
-    if (!message_loop_)  // In case message loop wasn't yet around...
-      message_loop_ = MessageLoop::current();  // Find it now.
-  }
-
-  DeathMap::iterator it = death_map_.find(&lifetimes);
+void ThreadData::TallyADeath(const Births& the_birth,
+                             const TimeDelta& queue_duration,
+                             const TimeDelta& run_duration) {
+  DeathMap::iterator it = death_map_.find(&the_birth);
   if (it != death_map_.end()) {
-    it->second.RecordDeath(duration);
+    it->second.RecordDeath(queue_duration, run_duration);
     return;
   }
 
   base::AutoLock lock(lock_);  // Lock since the map may get relocated now.
-  death_map_[&lifetimes].RecordDeath(duration);
+  death_map_[&the_birth].RecordDeath(queue_duration, run_duration);
 }
 
 // static
 Births* ThreadData::TallyABirthIfActive(const Location& location) {
-  if (IsActive()) {
-    ThreadData* current_thread_data = current();
-    if (current_thread_data) {
-      return current_thread_data->TallyABirth(location);
-    }
-  }
-
-  return NULL;
+#if !defined(TRACK_ALL_TASK_OBJECTS)
+  return NULL;  // Not compiled in.
+#else
+  if (!IsActive())
+    return NULL;
+  ThreadData* current_thread_data = Get();
+  if (!current_thread_data)
+    return NULL;
+  return current_thread_data->TallyABirth(location);
+#endif
 }
 
 // static
 void ThreadData::TallyADeathIfActive(const Births* the_birth,
-                                     const base::TimeDelta& duration) {
-  if (IsActive() && the_birth) {
-    current()->TallyADeath(*the_birth, duration);
-  }
+                                     const base::TimeTicks& time_posted,
+                                     const base::TimeTicks& delayed_start_time,
+                                     const base::TimeTicks& start_of_run) {
+#if !defined(TRACK_ALL_TASK_OBJECTS)
+  return;  // Not compiled in.
+#else
+  if (!IsActive() || !the_birth)
+    return;
+  ThreadData* current_thread_data = Get();
+  if (!current_thread_data)
+    return;
+
+  // To avoid conflating our stats with the delay duration in a PostDelayedTask,
+  // we identify such tasks, and replace their post_time with the time they
+  // were sechudled (requested?) to emerge from the delayed task queue. This
+  // means that queueing delay for such tasks will show how long they went
+  // unserviced, after they *could* be serviced.  This is the same stat as we
+  // have for non-delayed tasks, and we consistently call it queueing delay.
+  base::TimeTicks effective_post_time =
+      (delayed_start_time.is_null()) ? time_posted : delayed_start_time;
+  base::TimeDelta queue_duration = start_of_run - effective_post_time;
+  base::TimeDelta run_duration = Now() - start_of_run;
+  current_thread_data->TallyADeath(*the_birth, queue_duration, run_duration);
+#endif
 }
 
 // static
@@ -302,12 +338,6 @@ ThreadData* ThreadData::first() {
   return first_;
 }
 
-const std::string ThreadData::ThreadName() const {
-  if (message_loop_)
-    return message_loop_->thread_name();
-  return "ThreadWithoutMessageLoop";
-}
-
 // This may be called from another thread.
 void ThreadData::SnapshotBirthMap(BirthMap *output) const {
   base::AutoLock lock(lock_);
@@ -326,7 +356,7 @@ void ThreadData::SnapshotDeathMap(DeathMap *output) const {
 
 // static
 void ThreadData::ResetAllThreadData() {
-  ThreadData* my_list = ThreadData::current()->first();
+  ThreadData* my_list = Get()->first();
 
   for (ThreadData* thread_data = my_list;
        thread_data;
@@ -344,109 +374,11 @@ void ThreadData::Reset() {
     it->second->Clear();
 }
 
-#ifdef OS_WIN
-// A class used to count down which is accessed by several threads.  This is
-// used to make sure RunOnAllThreads() actually runs a task on the expected
-// count of threads.
-class ThreadData::ThreadSafeDownCounter {
- public:
-  // Constructor sets the count, once and for all.
-  explicit ThreadSafeDownCounter(size_t count);
-
-  // Decrement the count, and return true if we hit zero.  Also delete this
-  // instance automatically when we hit zero.
-  bool LastCaller();
-
- private:
-  size_t remaining_count_;
-  base::Lock lock_;  // protect access to remaining_count_.
-};
-
-ThreadData::ThreadSafeDownCounter::ThreadSafeDownCounter(size_t count)
-    : remaining_count_(count) {
-  DCHECK_GT(remaining_count_, 0u);
-}
-
-bool ThreadData::ThreadSafeDownCounter::LastCaller() {
-  {
-    base::AutoLock lock(lock_);
-    if (--remaining_count_)
-      return false;
-  }  // Release lock, so we can delete everything in this instance.
-  delete this;
-  return true;
-}
-
-// A Task class that runs a static method supplied, and checks to see if this
-// is the last tasks instance (on last thread) that will run the method.
-// IF this is the last run, then the supplied event is signalled.
-class ThreadData::RunTheStatic : public Task {
- public:
-  typedef void (*FunctionPointer)();
-  RunTheStatic(FunctionPointer function,
-               HANDLE completion_handle,
-               ThreadSafeDownCounter* counter);
-  // Run the supplied static method, and optionally set the event.
-  void Run();
-
- private:
-  FunctionPointer function_;
-  HANDLE completion_handle_;
-  // Make sure enough tasks are called before completion is signaled.
-  ThreadSafeDownCounter* counter_;
-
-  DISALLOW_COPY_AND_ASSIGN(RunTheStatic);
-};
-
-ThreadData::RunTheStatic::RunTheStatic(FunctionPointer function,
-                                       HANDLE completion_handle,
-                                       ThreadSafeDownCounter* counter)
-    : function_(function),
-      completion_handle_(completion_handle),
-      counter_(counter) {
-}
-
-void ThreadData::RunTheStatic::Run() {
-  function_();
-  if (counter_->LastCaller())
-    SetEvent(completion_handle_);
-}
-
-// TODO(jar): This should use condition variables, and be cross platform.
-void ThreadData::RunOnAllThreads(void (*function)()) {
-  ThreadData* list = first();  // Get existing list.
-
-  std::vector<MessageLoop*> message_loops;
-  for (ThreadData* it = list; it; it = it->next()) {
-    if (current() != it && it->message_loop())
-      message_loops.push_back(it->message_loop());
-  }
-
-  ThreadSafeDownCounter* counter =
-    new ThreadSafeDownCounter(message_loops.size() + 1);  // Extra one for us!
-
-  HANDLE completion_handle = CreateEvent(NULL, false, false, NULL);
-  // Tell all other threads to run.
-  for (size_t i = 0; i < message_loops.size(); ++i)
-    message_loops[i]->PostTask(
-        FROM_HERE, new RunTheStatic(function, completion_handle, counter));
-
-  // Also run Task on our thread.
-  RunTheStatic local_task(function, completion_handle, counter);
-  local_task.Run();
-
-  WaitForSingleObject(completion_handle, INFINITE);
-  int ret_val = CloseHandle(completion_handle);
-  DCHECK(ret_val);
-}
-#endif  // OS_WIN
-
 // static
 bool ThreadData::StartTracking(bool status) {
 #if !defined(TRACK_ALL_TASK_OBJECTS)
   return false;  // Not compiled in.
-#endif
-
+#else
   if (!status) {
     base::AutoLock lock(list_lock_);
     DCHECK(status_ == ACTIVE || status_ == SHUTDOWN);
@@ -458,6 +390,7 @@ bool ThreadData::StartTracking(bool status) {
   CHECK(tls_index_.Initialize(NULL));
   status_ = ACTIVE;
   return true;
+#endif
 }
 
 // static
@@ -465,24 +398,14 @@ bool ThreadData::IsActive() {
   return status_ == ACTIVE;
 }
 
-#ifdef OS_WIN
 // static
-void ThreadData::ShutdownMultiThreadTracking() {
-  // Using lock, guarantee that no new ThreadData instances will be created.
-  if (!StartTracking(false))
-    return;
-
-  RunOnAllThreads(ShutdownDisablingFurtherTracking);
-
-  // Now the *only* threads that might change the database are the threads with
-  // no messages loops.  They might still be adding data to their birth records,
-  // but since no objects are deleted on those threads, there will be no further
-  // access to to cross-thread data.
-  // We could do a cleanup on all threads except for the ones without
-  // MessageLoops, but we won't bother doing cleanup (destruction of data) yet.
-  return;
-}
+base::TimeTicks ThreadData::Now() {
+#if defined(TRACK_ALL_TASK_OBJECTS)
+  if (status_ == ACTIVE)
+    return base::TimeTicks::Now();
 #endif
+  return base::TimeTicks();  // Super fast when disabled, or not compiled in.
+}
 
 // static
 void ThreadData::ShutdownSingleThreadedCleanup() {
@@ -514,13 +437,6 @@ void ThreadData::ShutdownSingleThreadedCleanup() {
   status_ = UNINITIALIZED;
 }
 
-// static
-void ThreadData::ShutdownDisablingFurtherTracking() {
-  // Redundantly set status SHUTDOWN on this thread.
-  if (!StartTracking(false))
-    return;
-}
-
 //------------------------------------------------------------------------------
 // Individual 3-tuple of birth (place and thread) along with death thread, and
 // the accumulated stats for instances (DeathData).
@@ -541,15 +457,15 @@ Snapshot::Snapshot(const BirthOnThread& birth_on_thread, int count)
 
 const std::string Snapshot::DeathThreadName() const {
   if (death_thread_)
-    return death_thread_->ThreadName();
+    return death_thread_->thread_name();
   return "Still_Alive";
 }
 
 void Snapshot::Write(std::string* output) const {
   death_data_.Write(output);
   base::StringAppendF(output, "%s->%s ",
-                      birth_->birth_thread()->ThreadName().c_str(),
-                      death_thread_->ThreadName().c_str());
+                      birth_->birth_thread()->thread_name().c_str(),
+                      death_thread_->thread_name().c_str());
   birth_->location().Write(true, true, output);
 }
 
@@ -564,25 +480,14 @@ DataCollector::DataCollector() {
   DCHECK(ThreadData::IsActive());
 
   // Get an unchanging copy of a ThreadData list.
-  ThreadData* my_list = ThreadData::current()->first();
+  ThreadData* my_list = ThreadData::Get()->first();
 
-  count_of_contributing_threads_ = 0;
-  for (ThreadData* thread_data = my_list;
-       thread_data;
-       thread_data = thread_data->next()) {
-    ++count_of_contributing_threads_;
-  }
-
-  // Gather data serially.  A different constructor could be used to do in
-  // parallel, and then invoke an OnCompletion task.
+  // Gather data serially.
   // This hackish approach *can* get some slighly corrupt tallies, as we are
   // grabbing values without the protection of a lock, but it has the advantage
   // of working even with threads that don't have message loops.  If a user
   // sees any strangeness, they can always just run their stats gathering a
   // second time.
-  // TODO(jar): Provide version that gathers stats safely via PostTask in all
-  // cases where thread_data supplies a message_loop to post to.  Be careful to
-  // handle message_loops that are destroyed!?!
   for (ThreadData* thread_data = my_list;
        thread_data;
        thread_data = thread_data->next()) {
@@ -594,17 +499,12 @@ DataCollector::~DataCollector() {
 }
 
 void DataCollector::Append(const ThreadData& thread_data) {
-  // Get copy of data (which is done under ThreadData's lock).
+  // Get copy of data.
   ThreadData::BirthMap birth_map;
   thread_data.SnapshotBirthMap(&birth_map);
   ThreadData::DeathMap death_map;
   thread_data.SnapshotDeathMap(&death_map);
 
-  // Use our lock to protect our accumulation activity.
-  base::AutoLock lock(accumulation_lock_);
-
-  DCHECK(count_of_contributing_threads_);
-
   for (ThreadData::DeathMap::const_iterator it = death_map.begin();
        it != death_map.end(); ++it) {
     collection_.push_back(Snapshot(*it->first, thread_data, it->second));
@@ -615,17 +515,13 @@ void DataCollector::Append(const ThreadData& thread_data) {
        it != birth_map.end(); ++it) {
     global_birth_count_[it->second] += it->second->birth_count();
   }
-
-  --count_of_contributing_threads_;
 }
 
 DataCollector::Collection* DataCollector::collection() {
-  DCHECK(!count_of_contributing_threads_);
   return &collection_;
 }
 
 void DataCollector::AddListOfLivingObjects() {
-  DCHECK(!count_of_contributing_threads_);
   for (BirthCount::iterator it = global_birth_count_.begin();
        it != global_birth_count_.end(); ++it) {
     if (it->second > 0)
@@ -681,7 +577,7 @@ void Aggregation::Write(std::string* output) const {
                         birth_threads_.size());
   } else {
     base::StringAppendF(output, "All born on %s. ",
-                        birth_threads_.begin()->first->ThreadName().c_str());
+                        birth_threads_.begin()->first->thread_name().c_str());
   }
 
   if (death_threads_.size() > 1) {
@@ -690,7 +586,7 @@ void Aggregation::Write(std::string* output) const {
   } else {
     if (death_threads_.begin()->first) {
       base::StringAppendF(output, "All deleted on %s. ",
-                          death_threads_.begin()->first->ThreadName().c_str());
+                          death_threads_.begin()->first->thread_name().c_str());
     } else {
       output->append("All these objects are still alive.");
     }
@@ -735,10 +631,10 @@ bool Comparator::operator()(const Snapshot& left,
   switch (selector_) {
     case BIRTH_THREAD:
       if (left.birth_thread() != right.birth_thread() &&
-          left.birth_thread()->ThreadName() !=
-          right.birth_thread()->ThreadName())
-        return left.birth_thread()->ThreadName() <
-            right.birth_thread()->ThreadName();
+          left.birth_thread()->thread_name() !=
+          right.birth_thread()->thread_name())
+        return left.birth_thread()->thread_name() <
+            right.birth_thread()->thread_name();
       break;
 
     case DEATH_THREAD:
@@ -783,11 +679,32 @@ bool Comparator::operator()(const Snapshot& left,
         return left.count() > right.count();  // Sort large at front of vector.
       break;
 
-    case AVERAGE_DURATION:
+    case AVERAGE_RUN_DURATION:
+      if (!left.count() || !right.count())
+        break;
+      if (left.AverageMsRunDuration() != right.AverageMsRunDuration())
+        return left.AverageMsRunDuration() > right.AverageMsRunDuration();
+      break;
+
+    case TOTAL_RUN_DURATION:
       if (!left.count() || !right.count())
         break;
-      if (left.AverageMsDuration() != right.AverageMsDuration())
-        return left.AverageMsDuration() > right.AverageMsDuration();
+      if (left.run_duration() != right.run_duration())
+        return left.run_duration() > right.run_duration();
+      break;
+
+    case AVERAGE_QUEUE_DURATION:
+      if (!left.count() || !right.count())
+        break;
+      if (left.AverageMsQueueDuration() != right.AverageMsQueueDuration())
+        return left.AverageMsQueueDuration() > right.AverageMsQueueDuration();
+      break;
+
+    case TOTAL_QUEUE_DURATION:
+      if (!left.count() || !right.count())
+        break;
+      if (left.queue_duration() != right.queue_duration())
+        return left.queue_duration() > right.queue_duration();
       break;
 
     default:
@@ -807,8 +724,8 @@ bool Comparator::Equivalent(const Snapshot& left,
   switch (selector_) {
     case BIRTH_THREAD:
       if (left.birth_thread() != right.birth_thread() &&
-          left.birth_thread()->ThreadName() !=
-              right.birth_thread()->ThreadName())
+          left.birth_thread()->thread_name() !=
+              right.birth_thread()->thread_name())
         return false;
       break;
 
@@ -837,13 +754,11 @@ bool Comparator::Equivalent(const Snapshot& left,
       break;
 
     case COUNT:
-      if (left.count() != right.count())
-        return false;
-      break;
-
-    case AVERAGE_DURATION:
-      if (left.life_duration() != right.life_duration())
-        return false;
+    case AVERAGE_RUN_DURATION:
+    case TOTAL_RUN_DURATION:
+    case AVERAGE_QUEUE_DURATION:
+    case TOTAL_QUEUE_DURATION:
+      // We don't produce separate aggretation when only counts or times differ.
       break;
 
     default:
@@ -858,7 +773,7 @@ bool Comparator::Acceptable(const Snapshot& sample) const {
   if (required_.size()) {
     switch (selector_) {
       case BIRTH_THREAD:
-        if (sample.birth_thread()->ThreadName().find(required_) ==
+        if (sample.birth_thread()->thread_name().find(required_) ==
             std::string::npos)
           return false;
         break;
@@ -934,13 +849,16 @@ void Comparator::ParseKeyphrase(const std::string& key_phrase) {
     initialized = true;
     // Sorting and aggretation keywords, which specify how to sort the data, or
     // can specify a required match from the specified field in the record.
-    key_map["count"]    = COUNT;
-    key_map["duration"] = AVERAGE_DURATION;
-    key_map["birth"]    = BIRTH_THREAD;
-    key_map["death"]    = DEATH_THREAD;
-    key_map["file"]     = BIRTH_FILE;
-    key_map["function"] = BIRTH_FUNCTION;
-    key_map["line"]     = BIRTH_LINE;
+    key_map["count"]                = COUNT;
+    key_map["totalduration"]        = TOTAL_RUN_DURATION;
+    key_map["duration"]             = AVERAGE_RUN_DURATION;
+    key_map["totalqueueduration"]   = TOTAL_QUEUE_DURATION;
+    key_map["averagequeueduration"] = AVERAGE_QUEUE_DURATION;
+    key_map["birth"]                = BIRTH_THREAD;
+    key_map["death"]                = DEATH_THREAD;
+    key_map["file"]                 = BIRTH_FILE;
+    key_map["function"]             = BIRTH_FUNCTION;
+    key_map["line"]                 = BIRTH_LINE;
 
     // Immediate commands that do not involve setting sort order.
     key_map["reset"]     = RESET_ALL_DATA;
@@ -976,7 +894,8 @@ bool Comparator::ParseQuery(const std::string& query) {
 
   // Select subgroup ordering (if we want to display the subgroup)
   SetSubgroupTiebreaker(COUNT);
-  SetSubgroupTiebreaker(AVERAGE_DURATION);
+  SetSubgroupTiebreaker(AVERAGE_RUN_DURATION);
+  SetSubgroupTiebreaker(TOTAL_RUN_DURATION);
   SetSubgroupTiebreaker(BIRTH_THREAD);
   SetSubgroupTiebreaker(DEATH_THREAD);
   SetSubgroupTiebreaker(BIRTH_FUNCTION);
@@ -992,7 +911,7 @@ bool Comparator::WriteSortGrouping(const Snapshot& sample,
   switch (selector_) {
     case BIRTH_THREAD:
       base::StringAppendF(output, "All new on %s ",
-                          sample.birth_thread()->ThreadName().c_str());
+                          sample.birth_thread()->thread_name().c_str());
       wrote_data = true;
       break;
 
@@ -1033,7 +952,7 @@ void Comparator::WriteSnapshot(const Snapshot& sample,
       !(combined_selectors_ & DEATH_THREAD))
     base::StringAppendF(output, "%s->%s ",
                         (combined_selectors_ & BIRTH_THREAD) ? "*" :
-                          sample.birth().birth_thread()->ThreadName().c_str(),
+                          sample.birth().birth_thread()->thread_name().c_str(),
                         (combined_selectors_ & DEATH_THREAD) ? "*" :
                           sample.DeathThreadName().c_str());
   sample.birth().location().Write(!(combined_selectors_ & BIRTH_FILE),
diff --git a/base/tracked_objects.h b/base/tracked_objects.h
index 8e324a2..25b1771 100644
--- a/base/tracked_objects.h
+++ b/base/tracked_objects.h
@@ -31,9 +31,7 @@
 // marginal allocation cost associated with construction or destruction of
 // tracked objects, no locks are generally employed, and probably the largest
 // computational cost is associated with obtaining start and stop times for
-// instances as they are created and destroyed.  The introduction of worker
-// threads had a slight impact on this approach, and required use of some locks
-// when accessing data from the worker threads.
+// instances as they are created and destroyed.
 //
 // The following describes the lifecycle of tracking an instance.
 //
@@ -54,9 +52,10 @@
 // The derived Births class contains slots for recording statistics about all
 // instances born at the same location.  Statistics currently include only the
 // count of instances constructed.
+//
 // Since the base class BirthOnThread contains only constant data, it can be
 // freely accessed by any thread at any time (i.e., only the statistic needs to
-// be handled carefully, and it is ONLY read or written by the birth thread).
+// be handled carefully, and stats are updated exclusively on the birth thread).
 //
 // For Tasks, having now either constructed or found the Births instance
 // described above, a pointer to the Births instance is then recorded into the
@@ -67,7 +66,7 @@
 // can find out a Task's location of birth, and thread of birth, without using
 // any locks, as all that data is constant across the life of the process.
 //
-// This can also be done for any other object as well by calling
+// The above work *could* also be done for any other object as well by calling
 // TallyABirthIfActive() and TallyADeathIfActive() as appropriate.
 //
 // The amount of memory used in the above data structures depends on how many
@@ -81,10 +80,11 @@
 // carefully accumulated.  That tallying wrties into slots (members) in a
 // collection of DeathData instances.  For each birth place Location that is
 // destroyed on a thread, there is a DeathData instance to record the additional
-// death count, as well as accumulate the lifetime duration of the instance as
-// it is destroyed (dies).  By maintaining a single place to aggregate this
-// addition *only* for the given thread, we avoid the need to lock such
-// DeathData instances.
+// death count, as well as accumulate the run-time and queue-time durations for
+// the instance as it is destroyed (dies).  By maintaining a single place to
+// aggregate this running sum *only* for the given thread, we avoid the need to
+// lock such DeathData instances. (i.e., these accumulated stats in a DeathData
+// instance are exclusively updated by the singular owning thread).
 //
 // With the above lifecycle description complete, the major remaining detail is
 // explaining how each thread maintains a list of DeathData instances, and of
@@ -129,18 +129,26 @@
 // birth and death datastructures, but have local (frozen) copies of the actual
 // statistics (birth count, durations, etc. etc.).
 //
-// A DataCollector is a container object that holds a set of Snapshots.  A
-// DataCollector can be passed from thread to thread, and each thread
-// contributes to it by adding or updating Snapshot instances.  DataCollector
-// instances are thread safe containers which are passed to various threads to
-// accumulate all Snapshot instances.
+// A DataCollector is a container object that holds a set of Snapshots. The
+// statistics in a snapshot are gathered asynhcronously relative to their
+// ongoing updates.  It is possible, though highly unlikely, that stats such
+// as a 64bit counter could incorrectly recorded by this process. The advantage
+// to having fast (non-atomic) updates of the data outweighs the minimal risk
+// of a singular corrupt statistic snapshot (only the snapshot could be corrupt,
+// not the underlying and ongoing stistic).  In constrast, pointer data that is
+// accessed during snapshotting is completely invariant, and hence is perfectly
+// acquired (i.e., no potential corruption, and no risk of a bad memory
+// reference).
 //
 // After an array of Snapshots instances are colleted into a DataCollector, they
-// need to be sorted, and possibly aggregated (example: how many threads are in
-// a specific consecutive set of Snapshots?  What was the total birth count for
-// that set? etc.).  Aggregation instances collect running sums of any set of
-// snapshot instances, and are used to print sub-totals in an about:tracking
-// page.
+// need to be prepared for display our output. We currently implement a direct
+// renderin to HTML, but we will soon also have a JSON serialization as well.
+
+// For direct HTML display, the data must be sorted, and possibly aggregated
+// (example: how many threads are in a specific consecutive set of Snapshots?
+// What was the total birth count for that set? etc.).  Aggregation instances
+// collect running sums of any set of snapshot instances, and are used to print
+// sub-totals in an about:tracking page.
 //
 // TODO(jar): I need to store DataCollections, and provide facilities for taking
 // the difference between two gathered DataCollections.  For now, I'm just
@@ -167,7 +175,7 @@ class BASE_EXPORT BirthOnThread {
   const ThreadData* birth_thread() const { return birth_thread_; }
 
  private:
-  // File/lineno of birth.  This defines the essence of the type, as the context
+  // File/lineno of birth.  This defines the essence of the task, as the context
   // of the birth (construction) often tell what the item is for.  This field
   // is const, and hence safe to access from any thread.
   const Location location_;
@@ -213,35 +221,39 @@ class BASE_EXPORT Births: public BirthOnThread {
 class BASE_EXPORT DeathData {
  public:
   // Default initializer.
-  DeathData() : count_(0), square_duration_(0) {}
+  DeathData() : count_(0) {}
 
   // When deaths have not yet taken place, and we gather data from all the
   // threads, we create DeathData stats that tally the number of births without
   // a corrosponding death.
-  explicit DeathData(int count) : count_(count), square_duration_(0) {}
+  explicit DeathData(int count) : count_(count) {}
 
-  void RecordDeath(const base::TimeDelta& duration);
+  // Update stats for a task destruction (death) that had a Run() time of
+  // |duration|, and has had a queueing delay of |queue_duration|.
+  void RecordDeath(const base::TimeDelta& queue_duration,
+                   const base::TimeDelta& run_duration);
 
   // Metrics accessors.
   int count() const { return count_; }
-  base::TimeDelta life_duration() const { return life_duration_; }
-  int64 square_duration() const { return square_duration_; }
-  int AverageMsDuration() const;
-  double StandardDeviation() const;
+  base::TimeDelta run_duration() const { return run_duration_; }
+  int AverageMsRunDuration() const;
+  base::TimeDelta queue_duration() const { return queue_duration_; }
+  int AverageMsQueueDuration() const;
 
-  // Accumulate metrics from other into this.
+  // Accumulate metrics from other into this.  This method is never used on
+  // realtime statistics, and only used in snapshots and aggregatinos.
   void AddDeathData(const DeathData& other);
 
   // Simple print of internal state.
   void Write(std::string* output) const;
 
-  // Reset all tallies to zero.
+  // Reset all tallies to zero. This is used as a hack on realtime data.
   void Clear();
 
  private:
-  int count_;                // Number of destructions.
-  base::TimeDelta life_duration_;    // Sum of all lifetime durations.
-  int64 square_duration_;  // Sum of squares in milliseconds.
+  int count_;                       // Number of destructions.
+  base::TimeDelta run_duration_;    // Sum of all Run()time durations.
+  base::TimeDelta queue_duration_;  // Sum of all queue time durations.
 };
 
 //------------------------------------------------------------------------------
@@ -260,7 +272,6 @@ class BASE_EXPORT Snapshot {
   // When snapshotting a birth, with no death yet, use this:
   Snapshot(const BirthOnThread& birth_on_thread, int count);
 
-
   const ThreadData* birth_thread() const { return birth_->birth_thread(); }
   const Location location() const { return birth_->location(); }
   const BirthOnThread& birth() const { return *birth_; }
@@ -269,9 +280,16 @@ class BASE_EXPORT Snapshot {
   const std::string DeathThreadName() const;
 
   int count() const { return death_data_.count(); }
-  base::TimeDelta life_duration() const { return death_data_.life_duration(); }
-  int64 square_duration() const { return death_data_.square_duration(); }
-  int AverageMsDuration() const { return death_data_.AverageMsDuration(); }
+  base::TimeDelta run_duration() const { return death_data_.run_duration(); }
+  int AverageMsRunDuration() const {
+    return death_data_.AverageMsRunDuration();
+  }
+  base::TimeDelta queue_duration() const {
+    return death_data_.queue_duration();
+  }
+  int AverageMsQueueDuration() const {
+    return death_data_.AverageMsQueueDuration();
+  }
 
   void Write(std::string* output) const;
 
@@ -282,10 +300,10 @@ class BASE_EXPORT Snapshot {
   const ThreadData* death_thread_;
   DeathData death_data_;
 };
+
 //------------------------------------------------------------------------------
 // DataCollector is a container class for Snapshot and BirthOnThread count
-// items.  It protects the gathering under locks, so that it could be called via
-// Posttask on any threads, or passed to all the target threads in parallel.
+// items.
 
 class BASE_EXPORT DataCollector {
  public:
@@ -312,12 +330,6 @@ class BASE_EXPORT DataCollector {
  private:
   typedef std::map<const BirthOnThread*, int> BirthCount;
 
-  // This instance may be provided to several threads to contribute data.  The
-  // following counter tracks how many more threads will contribute.  When it is
-  // zero, then all asynchronous contributions are complete, and locked access
-  // is no longer needed.
-  int count_of_contributing_threads_;
-
   // The array that we collect data into.
   Collection collection_;
 
@@ -325,8 +337,6 @@ class BASE_EXPORT DataCollector {
   // seen a death count.
   BirthCount global_birth_count_;
 
-  base::Lock accumulation_lock_;  // Protects access during accumulation phase.
-
   DISALLOW_COPY_AND_ASSIGN(DataCollector);
 };
 
@@ -381,8 +391,10 @@ class BASE_EXPORT Comparator {
     BIRTH_FUNCTION = 8,
     BIRTH_LINE = 16,
     COUNT = 32,
-    AVERAGE_DURATION = 64,
-    TOTAL_DURATION = 128,
+    AVERAGE_RUN_DURATION = 64,
+    TOTAL_RUN_DURATION = 128,
+    AVERAGE_QUEUE_DURATION = 256,
+    TOTAL_QUEUE_DURATION = 512,
 
     // Imediate action keywords.
     RESET_ALL_DATA = -1,
@@ -418,7 +430,7 @@ class BASE_EXPORT Comparator {
   // printed line.
   bool IsGroupedBy(Selector selector) const;
 
-  // Using the tiebreakers as set above, we mostly get an ordering, which
+  // Using the tiebreakers as set above, we mostly get an ordering, with some
   // equivalent groups.  If those groups are displayed (rather than just being
   // aggregated, then the following is used to order them (within the group).
   void SetSubgroupTiebreaker(Selector selector);
@@ -462,7 +474,6 @@ class BASE_EXPORT Comparator {
   bool use_tiebreaker_for_sort_only_;
 };
 
-
 //------------------------------------------------------------------------------
 // For each thread, we have a ThreadData that stores all tracking info generated
 // on this thread.  This prevents the need for locking as data accumulates.
@@ -472,15 +483,18 @@ class BASE_EXPORT ThreadData {
   typedef std::map<Location, Births*> BirthMap;
   typedef std::map<const Births*, DeathData> DeathMap;
 
-  ThreadData();
-  ~ThreadData();
+  // Initialize the current thread context with a new instance of ThreadData.
+  // This is used by all threads that have names, and can be explicitly
+  // set *before* any births are threads have taken place.  It is generally
+  // only used by the message loop, which has a well defined name.
+  static void InitializeThreadContext(const std::string& suggested_name);
 
   // Using Thread Local Store, find the current instance for collecting data.
   // If an instance does not exist, construct one (and remember it for use on
   // this thread.
   // If shutdown has already started, and we don't yet have an instance, then
   // return null.
-  static ThreadData* current();
+  static ThreadData* Get();
 
   // For a given (unescaped) about:tracking query, develop resulting HTML, and
   // append to output.
@@ -496,22 +510,33 @@ class BASE_EXPORT ThreadData {
   Births* TallyABirth(const Location& location);
 
   // Find a place to record a death on this thread.
-  void TallyADeath(const Births& lifetimes, const base::TimeDelta& duration);
+  void TallyADeath(const Births& the_birth,
+                   const base::TimeDelta& queue_duration,
+                   const base::TimeDelta& duration);
 
   // Helper methods to only tally if the current thread has tracking active.
   //
   // TallyABirthIfActive will returns NULL if the birth cannot be tallied.
   static Births* TallyABirthIfActive(const Location& location);
-  static void TallyADeathIfActive(const Births* lifetimes,
-                                  const base::TimeDelta& duration);
+
+  // Record the end of a timed run of an object.  The |the_birth| is the record
+  // for the instance, the |time_posted| and |start_of_run| are times of posting
+  // into a message loop queue, and of starting to perform the run of the task.
+  // Implied is that the run just (Now()) ended.  The current_message_loop is
+  // optional, and only used in DEBUG mode (when supplied) to verify that the
+  // ThreadData has a thread name that does indeed match the given loop's
+  // associated thread name (in RELEASE mode, its use is compiled away).
+  static void TallyADeathIfActive(const Births* the_birth,
+                                  const base::TimeTicks& time_posted,
+                                  const base::TimeTicks& delayed_start_time,
+                                  const base::TimeTicks& start_of_run);
 
   // (Thread safe) Get start of list of instances.
   static ThreadData* first();
   // Iterate through the null terminated list of instances.
   ThreadData* next() const { return next_; }
 
-  MessageLoop* message_loop() const { return message_loop_; }
-  const std::string ThreadName() const;
+  const std::string thread_name() const { return thread_name_; }
 
   // Using our lock, make a copy of the specified maps.  These calls may arrive
   // from non-local threads, and are used to quickly scan data from all threads
@@ -528,31 +553,17 @@ class BASE_EXPORT ThreadData {
   // Using our lock to protect the iteration, Clear all birth and death data.
   void Reset();
 
-  // Using the "known list of threads" gathered during births and deaths, the
-  // following attempts to run the given function once all all such threads.
-  // Note that the function can only be run on threads which have a message
-  // loop!
-  static void RunOnAllThreads(void (*Func)());
-
   // Set internal status_ to either become ACTIVE, or later, to be SHUTDOWN,
   // based on argument being true or false respectively.
   // IF tracking is not compiled in, this function will return false.
   static bool StartTracking(bool status);
   static bool IsActive();
 
-#ifdef OS_WIN
-  // WARNING: ONLY call this function when all MessageLoops are still intact for
-  // all registered threads.  IF you call it later, you will crash.
-  // Note: You don't need to call it at all, and you can wait till you are
-  // single threaded (again) to do the cleanup via
-  // ShutdownSingleThreadedCleanup().
-  // Start the teardown (shutdown) process in a multi-thread mode by disabling
-  // further additions to thread database on all threads.  First it makes a
-  // local (locked) change to prevent any more threads from registering.  Then
-  // it Posts a Task to all registered threads to be sure they are aware that no
-  // more accumulation can take place.
-  static void ShutdownMultiThreadTracking();
-#endif
+  // Provide a time function that does nothing (runs fast) when we don't have
+  // the profiler enabled.  It will generally be optimized away when it is
+  // ifdef'ed to be small enough (allowing the profiler to be "compiled out" of
+  // the code).
+  static base::TimeTicks Now();
 
   // WARNING: ONLY call this function when you are running single threaded
   // (again) and all message loops and threads have terminated.  Until that
@@ -562,6 +573,18 @@ class BASE_EXPORT ThreadData {
   static void ShutdownSingleThreadedCleanup();
 
  private:
+  // Worker thread construction creates a name.
+  ThreadData();
+  // Message loop based construction should provide a name.
+  explicit ThreadData(const std::string& suggested_name);
+
+  ~ThreadData();
+
+  // Enter a new instance into Thread Local Store.
+  // Return the instance, or null if we can't register it (because we're
+  // shutting down).
+  static ThreadData* RegisterCurrentContext(ThreadData* unregistered);
+
   // Current allowable states of the tracking system.  The states always
   // proceed towards SHUTDOWN, and never go backwards.
   enum Status {
@@ -570,17 +593,6 @@ class BASE_EXPORT ThreadData {
     SHUTDOWN,
   };
 
-#if defined(OS_WIN)
-  class ThreadSafeDownCounter;
-  class RunTheStatic;
-#endif
-
-  // Each registered thread is called to set status_ to SHUTDOWN.
-  // This is done redundantly on every registered thread because it is not
-  // protected by a mutex.  Running on all threads guarantees we get the
-  // notification into the memory cache of all possible threads.
-  static void ShutdownDisablingFurtherTracking();
-
   // We use thread local store to identify which ThreadData to interact with.
   static base::ThreadLocalStorage::Slot tls_index_;
 
@@ -589,10 +601,7 @@ class BASE_EXPORT ThreadData {
   // Protection for access to first_.
   static base::Lock list_lock_;
 
-  // We set status_ to SHUTDOWN when we shut down the tracking service. This
-  // setting is redundantly established by all participating threads so that we
-  // are *guaranteed* (without locking) that all threads can "see" the status
-  // and avoid additional calls into the  service.
+  // We set status_ to SHUTDOWN when we shut down the tracking service.
   static Status status_;
 
   // Link to next instance (null terminated list). Used to globally track all
@@ -600,10 +609,9 @@ class BASE_EXPORT ThreadData {
   // data).
   ThreadData* next_;
 
-  // The message loop where tasks needing to access this instance's private data
-  // should be directed.  Since some threads have no message loop, some
-  // instances have data that can't be (safely) modified externally.
-  MessageLoop* message_loop_;
+  // The name of the thread that is being recorded.  If this thread has no
+  // message_loop, then this is a worker thread, with a sequence number postfix.
+  std::string thread_name_;
 
   // A map used on each thread to keep track of Births on this thread.
   // This map should only be accessed on the thread it was constructed on.
@@ -625,10 +633,13 @@ class BASE_EXPORT ThreadData {
   // writing is only done from this thread.
   mutable base::Lock lock_;
 
+  // The next available thread number.  This should only be accessed when the
+  // list_lock_ is held.
+  static int thread_number_counter;
+
   DISALLOW_COPY_AND_ASSIGN(ThreadData);
 };
 
-
 //------------------------------------------------------------------------------
 // Provide simple way to to start global tracking, and to tear down tracking
 // when done.  Note that construction and destruction of this object must be
diff --git a/base/tracked_objects_unittest.cc b/base/tracked_objects_unittest.cc
index e10659d4..c9ab9682 100644
--- a/base/tracked_objects_unittest.cc
+++ b/base/tracked_objects_unittest.cc
@@ -23,11 +23,11 @@ TEST_F(TrackedObjectsTest, MinimalStartupShutdown) {
     return;
 
   EXPECT_FALSE(ThreadData::first());  // No activity even on this thread.
-  ThreadData* data = ThreadData::current();
+  ThreadData* data = ThreadData::Get();
   EXPECT_TRUE(ThreadData::first());  // Now class was constructed.
   EXPECT_TRUE(data);
   EXPECT_TRUE(!data->next());
-  EXPECT_EQ(data, ThreadData::current());
+  EXPECT_EQ(data, ThreadData::Get());
   ThreadData::BirthMap birth_map;
   data->SnapshotBirthMap(&birth_map);
   EXPECT_EQ(0u, birth_map.size());
@@ -39,11 +39,11 @@ TEST_F(TrackedObjectsTest, MinimalStartupShutdown) {
   // Do it again, just to be sure we reset state completely.
   ThreadData::StartTracking(true);
   EXPECT_FALSE(ThreadData::first());  // No activity even on this thread.
-  data = ThreadData::current();
+  data = ThreadData::Get();
   EXPECT_TRUE(ThreadData::first());  // Now class was constructed.
   EXPECT_TRUE(data);
   EXPECT_TRUE(!data->next());
-  EXPECT_EQ(data, ThreadData::current());
+  EXPECT_EQ(data, ThreadData::Get());
   birth_map.clear();
   data->SnapshotBirthMap(&birth_map);
   EXPECT_EQ(0u, birth_map.size());
@@ -64,7 +64,7 @@ TEST_F(TrackedObjectsTest, TinyStartupShutdown) {
   const ThreadData* data = ThreadData::first();
   ASSERT_TRUE(data);
   EXPECT_TRUE(!data->next());
-  EXPECT_EQ(data, ThreadData::current());
+  EXPECT_EQ(data, ThreadData::Get());
   ThreadData::BirthMap birth_map;
   data->SnapshotBirthMap(&birth_map);
   EXPECT_EQ(1u, birth_map.size());                         // 1 birth location.
@@ -78,7 +78,9 @@ TEST_F(TrackedObjectsTest, TinyStartupShutdown) {
   const Births* second_birth = ThreadData::TallyABirthIfActive(location);
   ThreadData::TallyADeathIfActive(
       second_birth,
-      base::TimeDelta::FromSeconds(1) /* Bogus duration. */);
+      base::TimeTicks::TimeTicks(), /* Bogus post_time. */
+      base::TimeTicks::TimeTicks(), /* Bogus delayed_start_time. */
+      base::TimeTicks::TimeTicks()  /* Bogus start_run_time. */);
 
   birth_map.clear();
   data->SnapshotBirthMap(&birth_map);
diff --git a/content/browser/browser_main.cc b/content/browser/browser_main.cc
index 75e80c3..b0c9c19 100644
--- a/content/browser/browser_main.cc
+++ b/content/browser/browser_main.cc
@@ -12,6 +12,7 @@
 #include "base/metrics/histogram.h"
 #include "base/system_monitor/system_monitor.h"
 #include "base/threading/thread_restrictions.h"
+#include "base/tracked_objects.h"
 #include "content/browser/browser_thread.h"
 #include "content/browser/content_browser_client.h"
 #include "content/common/hi_res_timer_manager.h"
@@ -241,12 +242,11 @@ void BrowserMainParts::MainMessageLoopStart() {
 
   main_message_loop_.reset(new MessageLoop(MessageLoop::TYPE_UI));
 
-  // TODO(viettrungluu): should these really go before setting the thread name?
+  InitializeMainThread();
+
   system_monitor_.reset(new base::SystemMonitor);
   hi_res_timer_manager_.reset(new HighResolutionTimerManager);
 
-  InitializeMainThread();
-
   network_change_notifier_.reset(net::NetworkChangeNotifier::Create());
 
   PostMainMessageLoopStart();
@@ -274,6 +274,10 @@ void BrowserMainParts::InitializeMainThread() {
   base::PlatformThread::SetName(kThreadName);
   main_message_loop().set_thread_name(kThreadName);
 
+#if defined(TRACK_ALL_TASK_OBJECTS)
+  tracked_objects::ThreadData::InitializeThreadContext(kThreadName);
+#endif  // TRACK_ALL_TASK_OBJECTS
+
   // Register the main thread by instantiating it, but don't call any methods.
   main_thread_.reset(new BrowserThread(BrowserThread::UI,
                                        MessageLoop::current()));