• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
4 
5 #ifndef BASE_TRACKED_OBJECTS_H_
6 #define BASE_TRACKED_OBJECTS_H_
7 
8 #include <stdint.h>
9 
10 #include <map>
11 #include <set>
12 #include <stack>
13 #include <string>
14 #include <utility>
15 #include <vector>
16 
17 #include "base/atomicops.h"
18 #include "base/base_export.h"
19 #include "base/containers/hash_tables.h"
20 #include "base/gtest_prod_util.h"
21 #include "base/lazy_instance.h"
22 #include "base/location.h"
23 #include "base/macros.h"
24 #include "base/process/process_handle.h"
25 #include "base/profiler/alternate_timer.h"
26 #include "base/profiler/tracked_time.h"
27 #include "base/synchronization/lock.h"
28 #include "base/threading/thread_checker.h"
29 #include "base/threading/thread_local_storage.h"
30 
31 namespace base {
32 struct TrackingInfo;
33 }
34 
35 // TrackedObjects provides a database of stats about objects (generally Tasks)
36 // that are tracked.  Tracking means their birth, death, duration, birth thread,
37 // death thread, and birth place are recorded.  This data is carefully spread
38 // across a series of objects so that the counts and times can be rapidly
39 // updated without (usually) having to lock the data, and hence there is usually
40 // very little contention caused by the tracking.  The data can be viewed via
41 // the about:profiler URL, with a variety of sorting and filtering choices.
42 //
43 // These classes serve as the basis of a profiler of sorts for the Tasks system.
44 // As a result, design decisions were made to maximize speed, by minimizing
45 // recurring allocation/deallocation, lock contention and data copying.  In the
46 // "stable" state, which is reached relatively quickly, there is no separate
47 // marginal allocation cost associated with construction or destruction of
48 // tracked objects, no locks are generally employed, and probably the largest
49 // computational cost is associated with obtaining start and stop times for
50 // instances as they are created and destroyed.
51 //
52 // The following describes the life cycle of tracking an instance.
53 //
54 // First off, when the instance is created, the FROM_HERE macro is expanded
55 // to specify the birth place (file, line, function) where the instance was
56 // created.  That data is used to create a transient Location instance
57 // encapsulating the above triple of information.  The strings (like __FILE__)
58 // are passed around by reference, with the assumption that they are static, and
59 // will never go away.  This ensures that the strings can be dealt with as atoms
60 // with great efficiency (i.e., copying of strings is never needed, and
61 // comparisons for equality can be based on pointer comparisons).
62 //
63 // Next, a Births instance is created for use ONLY on the thread where this
64 // instance was created.  That Births instance records (in a base class
65 // BirthOnThread) references to the static data provided in a Location instance,
66 // as well as a pointer specifying the thread on which the birth takes place.
67 // Hence there is at most one Births instance for each Location on each thread.
68 // The derived Births class contains slots for recording statistics about all
69 // instances born at the same location.  Statistics currently include only the
70 // count of instances constructed.
71 //
72 // Since the base class BirthOnThread contains only constant data, it can be
73 // freely accessed by any thread at any time (i.e., only the statistic needs to
74 // be handled carefully, and stats are updated exclusively on the birth thread).
75 //
76 // For Tasks, having now either constructed or found the Births instance
77 // described above, a pointer to the Births instance is then recorded into the
78 // PendingTask structure in MessageLoop.  This fact alone is very useful in
79 // debugging, when there is a question of where an instance came from.  In
80 // addition, the birth time is also recorded and used to later evaluate the
81 // lifetime duration of the whole Task.  As a result of the above embedding, we
82 // can find out a Task's location of birth, and thread of birth, without using
83 // any locks, as all that data is constant across the life of the process.
84 //
85 // The above work *could* also be done for any other object as well by calling
86 // TallyABirthIfActive() and TallyRunOnNamedThreadIfTracking() as appropriate.
87 //
88 // The amount of memory used in the above data structures depends on how many
89 // threads there are, and how many Locations of construction there are.
90 // Fortunately, we don't use memory that is the product of those two counts, but
91 // rather we only need one Births instance for each thread that constructs an
92 // instance at a Location.  In many cases, instances are only created on one
93 // thread, so the memory utilization is actually fairly restrained.
94 //
95 // Lastly, when an instance is deleted, the final tallies of statistics are
96 // carefully accumulated.  That tallying writes into slots (members) in a
97 // collection of DeathData instances.  For each birth place Location that is
98 // destroyed on a thread, there is a DeathData instance to record the additional
99 // death count, as well as accumulate the run-time and queue-time durations for
100 // the instance as it is destroyed (dies).  By maintaining a single place to
101 // aggregate this running sum *only* for the given thread, we avoid the need to
102 // lock such DeathData instances. (i.e., these accumulated stats in a DeathData
103 // instance are exclusively updated by the singular owning thread).
104 //
105 // With the above life cycle description complete, the major remaining detail
106 // is explaining how each thread maintains a list of DeathData instances, and
107 // of Births instances, and is able to avoid additional (redundant/unnecessary)
108 // allocations.
109 //
110 // Each thread maintains a list of data items specific to that thread in a
111 // ThreadData instance (for that specific thread only).  The two critical items
112 // are lists of DeathData and Births instances.  These lists are maintained in
113 // STL maps, which are indexed by Location.  As noted earlier, we can compare
114 // locations very efficiently as we consider the underlying data (file,
115 // function, line) to be atoms, and hence pointer comparison is used rather than
116 // (slow) string comparisons.
117 //
118 // To provide a mechanism for iterating over all "known threads," which means
119 // threads that have recorded a birth or a death, we create a singly linked list
120 // of ThreadData instances.  Each such instance maintains a pointer to the next
121 // one.  A static member of ThreadData provides a pointer to the first item on
122 // this global list, and access via that all_thread_data_list_head_ item
123 // requires the use of the list_lock_.
124 // When new ThreadData instances is added to the global list, it is pre-pended,
125 // which ensures that any prior acquisition of the list is valid (i.e., the
126 // holder can iterate over it without fear of it changing, or the necessity of
127 // using an additional lock.  Iterations are actually pretty rare (used
128 // primarily for cleanup, or snapshotting data for display), so this lock has
129 // very little global performance impact.
130 //
131 // The above description tries to define the high performance (run time)
132 // portions of these classes.  After gathering statistics, calls instigated
133 // by visiting about:profiler will assemble and aggregate data for display.  The
134 // following data structures are used for producing such displays.  They are
135 // not performance critical, and their only major constraint is that they should
136 // be able to run concurrently with ongoing augmentation of the birth and death
137 // data.
138 //
139 // This header also exports collection of classes that provide "snapshotted"
140 // representations of the core tracked_objects:: classes.  These snapshotted
141 // representations are designed for safe transmission of the tracked_objects::
142 // data across process boundaries.  Each consists of:
143 // (1) a default constructor, to support the IPC serialization macros,
144 // (2) a constructor that extracts data from the type being snapshotted, and
145 // (3) the snapshotted data.
146 //
147 // For a given birth location, information about births is spread across data
148 // structures that are asynchronously changing on various threads.  For
149 // serialization and display purposes, we need to construct TaskSnapshot
150 // instances for each combination of birth thread, death thread, and location,
151 // along with the count of such lifetimes.  We gather such data into a
152 // TaskSnapshot instances, so that such instances can be sorted and
153 // aggregated (and remain frozen during our processing).
154 //
155 // Profiling consists of phases.  The concrete phase in the sequence of phases
156 // is identified by its 0-based index.
157 //
158 // The ProcessDataPhaseSnapshot struct is a serialized representation of the
159 // list of ThreadData objects for a process for a concrete profiling phase.  It
160 // holds a set of TaskSnapshots.  The statistics in a snapshot are gathered
161 // asynhcronously relative to their ongoing updates.
162 // It is possible, though highly unlikely, that stats could be incorrectly
163 // recorded by this process (all data is held in 32 bit ints, but we are not
164 // atomically collecting all data, so we could have count that does not, for
165 // example, match with the number of durations we accumulated).  The advantage
166 // to having fast (non-atomic) updates of the data outweighs the minimal risk of
167 // a singular corrupt statistic snapshot (only the snapshot could be corrupt,
168 // not the underlying and ongoing statistic).  In contrast, pointer data that
169 // is accessed during snapshotting is completely invariant, and hence is
170 // perfectly acquired (i.e., no potential corruption, and no risk of a bad
171 // memory reference).
172 //
173 // TODO(jar): We can implement a Snapshot system that *tries* to grab the
174 // snapshots on the source threads *when* they have MessageLoops available
175 // (worker threads don't have message loops generally, and hence gathering from
176 // them will continue to be asynchronous).  We had an implementation of this in
177 // the past, but the difficulty is dealing with message loops being terminated.
178 // We can *try* to spam the available threads via some message loop proxy to
179 // achieve this feat, and it *might* be valuable when we are collecting data
180 // for upload via UMA (where correctness of data may be more significant than
181 // for a single screen of about:profiler).
182 //
183 // TODO(jar): We need to store DataCollections, and provide facilities for
184 // taking the difference between two gathered DataCollections.  For now, we're
185 // just adding a hack that Reset()s to zero all counts and stats.  This is also
186 // done in a slightly thread-unsafe fashion, as the resetting is done
187 // asynchronously relative to ongoing updates (but all data is 32 bit in size).
188 // For basic profiling, this will work "most of the time," and should be
189 // sufficient... but storing away DataCollections is the "right way" to do this.
190 // We'll accomplish this via JavaScript storage of snapshots, and then we'll
191 // remove the Reset() methods.  We may also need a short-term-max value in
192 // DeathData that is reset (as synchronously as possible) during each snapshot.
193 // This will facilitate displaying a max value for each snapshot period.
194 
195 namespace tracked_objects {
196 
197 //------------------------------------------------------------------------------
198 // For a specific thread, and a specific birth place, the collection of all
199 // death info (with tallies for each death thread, to prevent access conflicts).
200 class ThreadData;
201 class BASE_EXPORT BirthOnThread {
202  public:
203   BirthOnThread(const Location& location, const ThreadData& current);
204 
location()205   const Location& location() const { return location_; }
birth_thread()206   const ThreadData* birth_thread() const { return birth_thread_; }
207 
208  private:
209   // File/lineno of birth.  This defines the essence of the task, as the context
210   // of the birth (construction) often tell what the item is for.  This field
211   // is const, and hence safe to access from any thread.
212   const Location location_;
213 
214   // The thread that records births into this object.  Only this thread is
215   // allowed to update birth_count_ (which changes over time).
216   const ThreadData* const birth_thread_;
217 
218   DISALLOW_COPY_AND_ASSIGN(BirthOnThread);
219 };
220 
221 //------------------------------------------------------------------------------
222 // A "snapshotted" representation of the BirthOnThread class.
223 
224 struct BASE_EXPORT BirthOnThreadSnapshot {
225   BirthOnThreadSnapshot();
226   explicit BirthOnThreadSnapshot(const BirthOnThread& birth);
227   ~BirthOnThreadSnapshot();
228 
229   LocationSnapshot location;
230   std::string thread_name;
231 };
232 
233 //------------------------------------------------------------------------------
234 // A class for accumulating counts of births (without bothering with a map<>).
235 
236 class BASE_EXPORT Births: public BirthOnThread {
237  public:
238   Births(const Location& location, const ThreadData& current);
239 
240   int birth_count() const;
241 
242   // When we have a birth we update the count for this birthplace.
243   void RecordBirth();
244 
245  private:
246   // The number of births on this thread for our location_.
247   int birth_count_;
248 
249   DISALLOW_COPY_AND_ASSIGN(Births);
250 };
251 
252 //------------------------------------------------------------------------------
253 // A "snapshotted" representation of the DeathData class.
254 
255 struct BASE_EXPORT DeathDataSnapshot {
256   DeathDataSnapshot();
257 
258   // Constructs the snapshot from individual values.
259   // The alternative would be taking a DeathData parameter, but this would
260   // create a loop since DeathData indirectly refers DeathDataSnapshot.  Passing
261   // a wrapper structure as a param or using an empty constructor for
262   // snapshotting DeathData would be less efficient.
263   DeathDataSnapshot(int count,
264                     int32_t run_duration_sum,
265                     int32_t run_duration_max,
266                     int32_t run_duration_sample,
267                     int32_t queue_duration_sum,
268                     int32_t queue_duration_max,
269                     int32_t queue_duration_sample);
270   ~DeathDataSnapshot();
271 
272   // Calculates and returns the delta between this snapshot and an earlier
273   // snapshot of the same task |older|.
274   DeathDataSnapshot Delta(const DeathDataSnapshot& older) const;
275 
276   int count;
277   int32_t run_duration_sum;
278   int32_t run_duration_max;
279   int32_t run_duration_sample;
280   int32_t queue_duration_sum;
281   int32_t queue_duration_max;
282   int32_t queue_duration_sample;
283 };
284 
285 //------------------------------------------------------------------------------
286 // A "snapshotted" representation of the DeathData for a particular profiling
287 // phase.  Used as an element of the list of phase snapshots owned by DeathData.
288 
289 struct DeathDataPhaseSnapshot {
290   DeathDataPhaseSnapshot(int profiling_phase,
291                          int count,
292                          int32_t run_duration_sum,
293                          int32_t run_duration_max,
294                          int32_t run_duration_sample,
295                          int32_t queue_duration_sum,
296                          int32_t queue_duration_max,
297                          int32_t queue_duration_sample,
298                          const DeathDataPhaseSnapshot* prev);
299 
300   // Profiling phase at which completion this snapshot was taken.
301   int profiling_phase;
302 
303   // Death data snapshot.
304   DeathDataSnapshot death_data;
305 
306   // Pointer to a snapshot from the previous phase.
307   const DeathDataPhaseSnapshot* prev;
308 };
309 
310 //------------------------------------------------------------------------------
311 // Information about deaths of a task on a given thread, called "death thread".
312 // Access to members of this class is never protected by a lock.  The fields
313 // are accessed in such a way that corruptions resulting from race conditions
314 // are not significant, and don't accumulate as a result of multiple accesses.
315 // All invocations of DeathData::OnProfilingPhaseCompleted and
316 // ThreadData::SnapshotMaps (which takes DeathData snapshot) in a given process
317 // must be called from the same thread. It doesn't matter what thread it is, but
318 // it's important the same thread is used as a snapshot thread during the whole
319 // process lifetime.  All fields except sample_probability_count_ can be
320 // snapshotted.
321 
322 class BASE_EXPORT DeathData {
323  public:
324   DeathData();
325   DeathData(const DeathData& other);
326   ~DeathData();
327 
328   // Update stats for a task destruction (death) that had a Run() time of
329   // |duration|, and has had a queueing delay of |queue_duration|.
330   void RecordDeath(const int32_t queue_duration,
331                    const int32_t run_duration,
332                    const uint32_t random_number);
333 
334   // Metrics and past snapshots accessors, used only for serialization and in
335   // tests.
count()336   int count() const { return base::subtle::NoBarrier_Load(&count_); }
run_duration_sum()337   int32_t run_duration_sum() const {
338     return base::subtle::NoBarrier_Load(&run_duration_sum_);
339   }
run_duration_max()340   int32_t run_duration_max() const {
341     return base::subtle::NoBarrier_Load(&run_duration_max_);
342   }
run_duration_sample()343   int32_t run_duration_sample() const {
344     return base::subtle::NoBarrier_Load(&run_duration_sample_);
345   }
queue_duration_sum()346   int32_t queue_duration_sum() const {
347     return base::subtle::NoBarrier_Load(&queue_duration_sum_);
348   }
queue_duration_max()349   int32_t queue_duration_max() const {
350     return base::subtle::NoBarrier_Load(&queue_duration_max_);
351   }
queue_duration_sample()352   int32_t queue_duration_sample() const {
353     return base::subtle::NoBarrier_Load(&queue_duration_sample_);
354   }
last_phase_snapshot()355   const DeathDataPhaseSnapshot* last_phase_snapshot() const {
356     return last_phase_snapshot_;
357   }
358 
359   // Called when the current profiling phase, identified by |profiling_phase|,
360   // ends.
361   // Must be called only on the snapshot thread.
362   void OnProfilingPhaseCompleted(int profiling_phase);
363 
364  private:
365   // Members are ordered from most regularly read and updated, to least
366   // frequently used.  This might help a bit with cache lines.
367   // Number of runs seen (divisor for calculating averages).
368   // Can be incremented only on the death thread.
369   base::subtle::Atomic32 count_;
370 
371   // Count used in determining probability of selecting exec/queue times from a
372   // recorded death as samples.
373   // Gets incremented only on the death thread, but can be set to 0 by
374   // OnProfilingPhaseCompleted() on the snapshot thread.
375   base::subtle::Atomic32 sample_probability_count_;
376 
377   // Basic tallies, used to compute averages.  Can be incremented only on the
378   // death thread.
379   base::subtle::Atomic32 run_duration_sum_;
380   base::subtle::Atomic32 queue_duration_sum_;
381   // Max values, used by local visualization routines.  These are often read,
382   // but rarely updated.  The max values get assigned only on the death thread,
383   // but these fields can be set to 0 by OnProfilingPhaseCompleted() on the
384   // snapshot thread.
385   base::subtle::Atomic32 run_duration_max_;
386   base::subtle::Atomic32 queue_duration_max_;
387   // Samples, used by crowd sourcing gatherers.  These are almost never read,
388   // and rarely updated.  They can be modified only on the death thread.
389   base::subtle::Atomic32 run_duration_sample_;
390   base::subtle::Atomic32 queue_duration_sample_;
391 
392   // Snapshot of this death data made at the last profiling phase completion, if
393   // any.  DeathData owns the whole list starting with this pointer.
394   // Can be accessed only on the snapshot thread.
395   const DeathDataPhaseSnapshot* last_phase_snapshot_;
396 
397   DISALLOW_ASSIGN(DeathData);
398 };
399 
400 //------------------------------------------------------------------------------
401 // A temporary collection of data that can be sorted and summarized.  It is
402 // gathered (carefully) from many threads.  Instances are held in arrays and
403 // processed, filtered, and rendered.
404 // The source of this data was collected on many threads, and is asynchronously
405 // changing.  The data in this instance is not asynchronously changing.
406 
407 struct BASE_EXPORT TaskSnapshot {
408   TaskSnapshot();
409   TaskSnapshot(const BirthOnThreadSnapshot& birth,
410                const DeathDataSnapshot& death_data,
411                const std::string& death_thread_name);
412   ~TaskSnapshot();
413 
414   BirthOnThreadSnapshot birth;
415   // Delta between death data for a thread for a certain profiling phase and the
416   // snapshot for the pervious phase, if any.  Otherwise, just a snapshot.
417   DeathDataSnapshot death_data;
418   std::string death_thread_name;
419 };
420 
421 //------------------------------------------------------------------------------
422 // For each thread, we have a ThreadData that stores all tracking info generated
423 // on this thread.  This prevents the need for locking as data accumulates.
424 // We use ThreadLocalStorage to quickly identfy the current ThreadData context.
425 // We also have a linked list of ThreadData instances, and that list is used to
426 // harvest data from all existing instances.
427 
428 struct ProcessDataPhaseSnapshot;
429 struct ProcessDataSnapshot;
430 class BASE_EXPORT TaskStopwatch;
431 
432 // Map from profiling phase number to the process-wide snapshotted
433 // representation of the list of ThreadData objects that died during the given
434 // phase.
435 typedef std::map<int, ProcessDataPhaseSnapshot> PhasedProcessDataSnapshotMap;
436 
437 class BASE_EXPORT ThreadData {
438  public:
439   // Current allowable states of the tracking system.  The states can vary
440   // between ACTIVE and DEACTIVATED, but can never go back to UNINITIALIZED.
441   enum Status {
442     UNINITIALIZED,         // Pristine, link-time state before running.
443     DORMANT_DURING_TESTS,  // Only used during testing.
444     DEACTIVATED,           // No longer recording profiling.
445     PROFILING_ACTIVE,      // Recording profiles.
446     STATUS_LAST = PROFILING_ACTIVE
447   };
448 
449   typedef base::hash_map<Location, Births*, Location::Hash> BirthMap;
450   typedef std::map<const Births*, DeathData> DeathMap;
451 
452   // Initialize the current thread context with a new instance of ThreadData.
453   // This is used by all threads that have names, and should be explicitly
454   // set *before* any births on the threads have taken place.  It is generally
455   // only used by the message loop, which has a well defined thread name.
456   static void InitializeThreadContext(const std::string& suggested_name);
457 
458   // Using Thread Local Store, find the current instance for collecting data.
459   // If an instance does not exist, construct one (and remember it for use on
460   // this thread.
461   // This may return NULL if the system is disabled for any reason.
462   static ThreadData* Get();
463 
464   // Fills |process_data_snapshot| with phased snapshots of all profiling
465   // phases, including the current one, identified by |current_profiling_phase|.
466   // |current_profiling_phase| is necessary because a child process can start
467   // after several phase-changing events, so it needs to receive the current
468   // phase number from the browser process to fill the correct entry for the
469   // current phase in the |process_data_snapshot| map.
470   static void Snapshot(int current_profiling_phase,
471                        ProcessDataSnapshot* process_data_snapshot);
472 
473   // Called when the current profiling phase, identified by |profiling_phase|,
474   // ends.
475   // |profiling_phase| is necessary because a child process can start after
476   // several phase-changing events, so it needs to receive the phase number from
477   // the browser process to fill the correct entry in the
478   // completed_phases_snapshots_ map.
479   static void OnProfilingPhaseCompleted(int profiling_phase);
480 
481   // Finds (or creates) a place to count births from the given location in this
482   // thread, and increment that tally.
483   // TallyABirthIfActive will returns NULL if the birth cannot be tallied.
484   static Births* TallyABirthIfActive(const Location& location);
485 
486   // Records the end of a timed run of an object.  The |completed_task| contains
487   // a pointer to a Births, the time_posted, and a delayed_start_time if any.
488   // The |start_of_run| indicates when we started to perform the run of the
489   // task.  The delayed_start_time is non-null for tasks that were posted as
490   // delayed tasks, and it indicates when the task should have run (i.e., when
491   // it should have posted out of the timer queue, and into the work queue.
492   // The |end_of_run| was just obtained by a call to Now() (just after the task
493   // finished).  It is provided as an argument to help with testing.
494   static void TallyRunOnNamedThreadIfTracking(
495       const base::TrackingInfo& completed_task,
496       const TaskStopwatch& stopwatch);
497 
498   // Record the end of a timed run of an object.  The |birth| is the record for
499   // the instance, the |time_posted| records that instant, which is presumed to
500   // be when the task was posted into a queue to run on a worker thread.
501   // The |start_of_run| is when the worker thread started to perform the run of
502   // the task.
503   // The |end_of_run| was just obtained by a call to Now() (just after the task
504   // finished).
505   static void TallyRunOnWorkerThreadIfTracking(const Births* births,
506                                                const TrackedTime& time_posted,
507                                                const TaskStopwatch& stopwatch);
508 
509   // Record the end of execution in region, generally corresponding to a scope
510   // being exited.
511   static void TallyRunInAScopedRegionIfTracking(const Births* births,
512                                                 const TaskStopwatch& stopwatch);
513 
thread_name()514   const std::string& thread_name() const { return thread_name_; }
515 
516   // Initializes all statics if needed (this initialization call should be made
517   // while we are single threaded).
518   static void Initialize();
519 
520   // Sets internal status_.
521   // If |status| is false, then status_ is set to DEACTIVATED.
522   // If |status| is true, then status_ is set to PROFILING_ACTIVE.
523   static void InitializeAndSetTrackingStatus(Status status);
524 
525   static Status status();
526 
527   // Indicate if any sort of profiling is being done (i.e., we are more than
528   // DEACTIVATED).
529   static bool TrackingStatus();
530 
531   // Enables profiler timing.
532   static void EnableProfilerTiming();
533 
534   // Provide a time function that does nothing (runs fast) when we don't have
535   // the profiler enabled.  It will generally be optimized away when it is
536   // ifdef'ed to be small enough (allowing the profiler to be "compiled out" of
537   // the code).
538   static TrackedTime Now();
539 
540   // Use the function |now| to provide current times, instead of calling the
541   // TrackedTime::Now() function.  Since this alternate function is being used,
542   // the other time arguments (used for calculating queueing delay) will be
543   // ignored.
544   static void SetAlternateTimeSource(NowFunction* now);
545 
546   // This function can be called at process termination to validate that thread
547   // cleanup routines have been called for at least some number of named
548   // threads.
549   static void EnsureCleanupWasCalled(int major_threads_shutdown_count);
550 
551  private:
552   friend class TaskStopwatch;
553   // Allow only tests to call ShutdownSingleThreadedCleanup.  We NEVER call it
554   // in production code.
555   // TODO(jar): Make this a friend in DEBUG only, so that the optimizer has a
556   // better change of optimizing (inlining? etc.) private methods (knowing that
557   // there will be no need for an external entry point).
558   friend class TrackedObjectsTest;
559   FRIEND_TEST_ALL_PREFIXES(TrackedObjectsTest, MinimalStartupShutdown);
560   FRIEND_TEST_ALL_PREFIXES(TrackedObjectsTest, TinyStartupShutdown);
561 
562   typedef std::map<const BirthOnThread*, int> BirthCountMap;
563 
564   typedef std::vector<std::pair<const Births*, DeathDataPhaseSnapshot>>
565       DeathsSnapshot;
566 
567   // Worker thread construction creates a name since there is none.
568   explicit ThreadData(int thread_number);
569 
570   // Message loop based construction should provide a name.
571   explicit ThreadData(const std::string& suggested_name);
572 
573   ~ThreadData();
574 
575   // Push this instance to the head of all_thread_data_list_head_, linking it to
576   // the previous head.  This is performed after each construction, and leaves
577   // the instance permanently on that list.
578   void PushToHeadOfList();
579 
580   // (Thread safe) Get start of list of all ThreadData instances using the lock.
581   static ThreadData* first();
582 
583   // Iterate through the null terminated list of ThreadData instances.
584   ThreadData* next() const;
585 
586 
587   // In this thread's data, record a new birth.
588   Births* TallyABirth(const Location& location);
589 
590   // Find a place to record a death on this thread.
591   void TallyADeath(const Births& births,
592                    int32_t queue_duration,
593                    const TaskStopwatch& stopwatch);
594 
595   // Snapshots (under a lock) the profiled data for the tasks for this thread
596   // and writes all of the executed tasks' data -- i.e. the data for all
597   // profiling phases (including the current one: |current_profiling_phase|) for
598   // the tasks with with entries in the death_map_ -- into |phased_snapshots|.
599   // Also updates the |birth_counts| tally for each task to keep track of the
600   // number of living instances of the task -- that is, each task maps to the
601   // number of births for the task that have not yet been balanced by a death.
602   void SnapshotExecutedTasks(int current_profiling_phase,
603                              PhasedProcessDataSnapshotMap* phased_snapshots,
604                              BirthCountMap* birth_counts);
605 
606   // Using our lock, make a copy of the specified maps.  This call may be made
607   // on  non-local threads, which necessitate the use of the lock to prevent
608   // the map(s) from being reallocated while they are copied.
609   void SnapshotMaps(int profiling_phase,
610                     BirthMap* birth_map,
611                     DeathsSnapshot* deaths);
612 
613   // Called for this thread when the current profiling phase, identified by
614   // |profiling_phase|, ends.
615   void OnProfilingPhaseCompletedOnThread(int profiling_phase);
616 
617   // This method is called by the TLS system when a thread terminates.
618   // The argument may be NULL if this thread has never tracked a birth or death.
619   static void OnThreadTermination(void* thread_data);
620 
621   // This method should be called when a worker thread terminates, so that we
622   // can save all the thread data into a cache of reusable ThreadData instances.
623   void OnThreadTerminationCleanup();
624 
625   // Cleans up data structures, and returns statics to near pristine (mostly
626   // uninitialized) state.  If there is any chance that other threads are still
627   // using the data structures, then the |leak| argument should be passed in as
628   // true, and the data structures (birth maps, death maps, ThreadData
629   // insntances, etc.) will be leaked and not deleted.  If you have joined all
630   // threads since the time that InitializeAndSetTrackingStatus() was called,
631   // then you can pass in a |leak| value of false, and this function will
632   // delete recursively all data structures, starting with the list of
633   // ThreadData instances.
634   static void ShutdownSingleThreadedCleanup(bool leak);
635 
636   // When non-null, this specifies an external function that supplies monotone
637   // increasing time functcion.
638   static NowFunction* now_function_;
639 
640   // If true, now_function_ returns values that can be used to calculate queue
641   // time.
642   static bool now_function_is_time_;
643 
644   // We use thread local store to identify which ThreadData to interact with.
645   static base::ThreadLocalStorage::StaticSlot tls_index_;
646 
647   // List of ThreadData instances for use with worker threads.  When a worker
648   // thread is done (terminated), we push it onto this list.  When a new worker
649   // thread is created, we first try to re-use a ThreadData instance from the
650   // list, and if none are available, construct a new one.
651   // This is only accessed while list_lock_ is held.
652   static ThreadData* first_retired_worker_;
653 
654   // Link to the most recently created instance (starts a null terminated list).
655   // The list is traversed by about:profiler when it needs to snapshot data.
656   // This is only accessed while list_lock_ is held.
657   static ThreadData* all_thread_data_list_head_;
658 
659   // The next available worker thread number.  This should only be accessed when
660   // the list_lock_ is held.
661   static int worker_thread_data_creation_count_;
662 
663   // The number of times TLS has called us back to cleanup a ThreadData
664   // instance.  This is only accessed while list_lock_ is held.
665   static int cleanup_count_;
666 
667   // Incarnation sequence number, indicating how many times (during unittests)
668   // we've either transitioned out of UNINITIALIZED, or into that state.  This
669   // value is only accessed while the list_lock_ is held.
670   static int incarnation_counter_;
671 
672   // Protection for access to all_thread_data_list_head_, and to
673   // unregistered_thread_data_pool_.  This lock is leaked at shutdown.
674   // The lock is very infrequently used, so we can afford to just make a lazy
675   // instance and be safe.
676   static base::LazyInstance<base::Lock>::Leaky list_lock_;
677 
678   // We set status_ to SHUTDOWN when we shut down the tracking service.
679   static base::subtle::Atomic32 status_;
680 
681   // Link to next instance (null terminated list).  Used to globally track all
682   // registered instances (corresponds to all registered threads where we keep
683   // data).
684   ThreadData* next_;
685 
686   // Pointer to another ThreadData instance for a Worker-Thread that has been
687   // retired (its thread was terminated).  This value is non-NULL only for a
688   // retired ThreadData associated with a Worker-Thread.
689   ThreadData* next_retired_worker_;
690 
691   // The name of the thread that is being recorded.  If this thread has no
692   // message_loop, then this is a worker thread, with a sequence number postfix.
693   std::string thread_name_;
694 
695   // Indicate if this is a worker thread, and the ThreadData contexts should be
696   // stored in the unregistered_thread_data_pool_ when not in use.
697   // Value is zero when it is not a worker thread.  Value is a positive integer
698   // corresponding to the created thread name if it is a worker thread.
699   int worker_thread_number_;
700 
701   // A map used on each thread to keep track of Births on this thread.
702   // This map should only be accessed on the thread it was constructed on.
703   // When a snapshot is needed, this structure can be locked in place for the
704   // duration of the snapshotting activity.
705   BirthMap birth_map_;
706 
707   // Similar to birth_map_, this records informations about death of tracked
708   // instances (i.e., when a tracked instance was destroyed on this thread).
709   // It is locked before changing, and hence other threads may access it by
710   // locking before reading it.
711   DeathMap death_map_;
712 
713   // Lock to protect *some* access to BirthMap and DeathMap.  The maps are
714   // regularly read and written on this thread, but may only be read from other
715   // threads.  To support this, we acquire this lock if we are writing from this
716   // thread, or reading from another thread.  For reading from this thread we
717   // don't need a lock, as there is no potential for a conflict since the
718   // writing is only done from this thread.
719   mutable base::Lock map_lock_;
720 
721   // A random number that we used to select decide which sample to keep as a
722   // representative sample in each DeathData instance.  We can't start off with
723   // much randomness (because we can't call RandInt() on all our threads), so
724   // we stir in more and more as we go.
725   uint32_t random_number_;
726 
727   // Record of what the incarnation_counter_ was when this instance was created.
728   // If the incarnation_counter_ has changed, then we avoid pushing into the
729   // pool (this is only critical in tests which go through multiple
730   // incarnations).
731   int incarnation_count_for_pool_;
732 
733   // Most recently started (i.e. most nested) stopwatch on the current thread,
734   // if it exists; NULL otherwise.
735   TaskStopwatch* current_stopwatch_;
736 
737   DISALLOW_COPY_AND_ASSIGN(ThreadData);
738 };
739 
740 //------------------------------------------------------------------------------
741 // Stopwatch to measure task run time or simply create a time interval that will
742 // be subtracted from the current most nested task's run time.  Stopwatches
743 // coordinate with the stopwatches in which they are nested to avoid
744 // double-counting nested tasks run times.
745 
746 class BASE_EXPORT TaskStopwatch {
747  public:
748   // Starts the stopwatch.
749   TaskStopwatch();
750   ~TaskStopwatch();
751 
752   // Starts stopwatch.
753   void Start();
754 
755   // Stops stopwatch.
756   void Stop();
757 
758   // Returns the start time.
759   TrackedTime StartTime() const;
760 
761   // Task's duration is calculated as the wallclock duration between starting
762   // and stopping this stopwatch, minus the wallclock durations of any other
763   // instances that are immediately nested in this one, started and stopped on
764   // this thread during that period.
765   int32_t RunDurationMs() const;
766 
767   // Returns tracking info for the current thread.
768   ThreadData* GetThreadData() const;
769 
770  private:
771   // Time when the stopwatch was started.
772   TrackedTime start_time_;
773 
774   // Wallclock duration of the task.
775   int32_t wallclock_duration_ms_;
776 
777   // Tracking info for the current thread.
778   ThreadData* current_thread_data_;
779 
780   // Sum of wallclock durations of all stopwatches that were directly nested in
781   // this one.
782   int32_t excluded_duration_ms_;
783 
784   // Stopwatch which was running on our thread when this stopwatch was started.
785   // That preexisting stopwatch must be adjusted to the exclude the wallclock
786   // duration of this stopwatch.
787   TaskStopwatch* parent_;
788 
789 #if DCHECK_IS_ON()
790   // State of the stopwatch.  Stopwatch is first constructed in a created state
791   // state, then is optionally started/stopped, then destructed.
792   enum { CREATED, RUNNING, STOPPED } state_;
793 
794   // Currently running stopwatch that is directly nested in this one, if such
795   // stopwatch exists.  NULL otherwise.
796   TaskStopwatch* child_;
797 #endif
798 };
799 
800 //------------------------------------------------------------------------------
801 // A snapshotted representation of the list of ThreadData objects for a process,
802 // for a single profiling phase.
803 
804 struct BASE_EXPORT ProcessDataPhaseSnapshot {
805  public:
806   ProcessDataPhaseSnapshot();
807   ~ProcessDataPhaseSnapshot();
808 
809   std::vector<TaskSnapshot> tasks;
810 };
811 
812 //------------------------------------------------------------------------------
813 // A snapshotted representation of the list of ThreadData objects for a process,
814 // for all profiling phases, including the current one.
815 
816 struct BASE_EXPORT ProcessDataSnapshot {
817  public:
818   ProcessDataSnapshot();
819   ~ProcessDataSnapshot();
820 
821   PhasedProcessDataSnapshotMap phased_snapshots;
822   base::ProcessId process_id;
823 };
824 
825 }  // namespace tracked_objects
826 
827 #endif  // BASE_TRACKED_OBJECTS_H_
828