gecko-dev/toolkit/recordreplay/Thread.h

393 строки
14 KiB
C
Исходник Обычный вид История

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#ifndef mozilla_recordreplay_Thread_h
#define mozilla_recordreplay_Thread_h
#include "mozilla/Atomics.h"
#include "File.h"
#include "Lock.h"
#include "Monitor.h"
#include <pthread.h>
#include <setjmp.h>
namespace mozilla {
namespace recordreplay {
// Threads Overview.
//
// The main thread and each thread that is spawned when thread events are not
// passed through have their behavior recorded.
//
// While recording, each recorded thread has an associated Thread object which
// can be fetched with Thread::Current and stores the thread's ID, its file for
// storing events that occur in the thread, and some other thread local state.
// Otherwise, threads are spawned and destroyed as usual for the process.
//
// While rewinding, the same Thread structure exists for each recorded thread.
// Several additional changes are needed to facilitate rewinding and IPC:
//
// 1. All recorded threads are spawned early on, before any checkpoint has been
// reached. These threads idle until the process calls the system's thread
// creation API, and then they run with the start routine the process
// provided. After the start routine finishes they idle indefinitely,
// potentially running new start routines if their thread ID is reused. This
// allows the process to rewind itself without needing to spawn or destroy
// any threads.
//
// 2. Some additional number of threads are spawned for use by the IPC and
// memory snapshot mechanisms. These have associated Thread
// structures but are not recorded and always pass through thread events.
//
// 3. All recorded threads and must be able to enter a particular blocking
// state, under Thread::Wait, when requested by the main thread calling
// WaitForIdleThreads. For most recorded threads this happens when the
// thread attempts to take a recorded lock and blocks in Lock::Wait.
// The only exception is for JS helper threads, which never take recorded
// locks. For these threads, NotifyUnrecordedWait and
// MaybeWaitForCheckpointSave must be used to enter this state.
//
// 4. Once all recorded threads are idle, the main thread is able to record
// memory snapshots and thread stacks for later rewinding. Additional
// threads created for #2 above do not idle and do not have their state
// included in snapshots, but they are designed to avoid interfering with
// the main thread while it is taking or restoring a checkpoint.
// The ID used by the process main thread.
static const size_t MainThreadId = 1;
// The maximum ID useable by recorded threads.
static const size_t MaxRecordedThreadId = 70;
// The maximum number of threads which are not recorded but need a Thread so
// that they can participate in e.g. Wait/Notify calls.
static const size_t MaxNumNonRecordedThreads = 12;
static const size_t MaxThreadId = MaxRecordedThreadId + MaxNumNonRecordedThreads;
typedef pthread_t NativeThreadId;
// Information about the execution state of a thread.
class Thread
{
public:
// Signature for the start function of a thread.
typedef void (*Callback)(void*);
private:
// Monitor used to protect various thread information (see Thread.h) and to
// wait on or signal progress for a thread.
static Monitor* gMonitor;
// Thread ID in the recording, fixed at creation.
size_t mId;
// Whether to pass events in the thread through without recording/replaying.
// This is only used by the associated thread.
bool mPassThroughEvents;
// Whether to crash if we try to record/replay thread events. This is only
// used by the associated thread.
size_t mDisallowEvents;
// Whether execution has diverged from the recording and the thread's
// recorded events cannot be accessed.
bool mDivergedFromRecording;
// Whether this thread should diverge from the recording at the next
// opportunity. This can be set from any thread.
Atomic<bool, SequentiallyConsistent, Behavior::DontPreserve> mShouldDivergeFromRecording;
// Start routine and argument which the thread is currently executing. This
// is cleared after the routine finishes and another start routine may be
// assigned to the thread. mNeedsJoin specifies whether the thread must be
// joined before it is completely dead and can be reused. This is protected
// by the thread monitor.
Callback mStart;
void* mStartArg;
bool mNeedsJoin;
// ID for this thread used by the system.
NativeThreadId mNativeId;
// Stream with events for the thread. This is only used on the thread itself.
Stream* mEvents;
// Stack boundary of the thread, protected by the thread monitor.
uint8_t* mStackBase;
size_t mStackSize;
// File descriptor to block on when the thread is idle, fixed at creation.
FileHandle mIdlefd;
// File descriptor to notify to wake the thread up, fixed at creation.
FileHandle mNotifyfd;
// Whether the thread should attempt to idle.
Atomic<bool, SequentiallyConsistent, Behavior::DontPreserve> mShouldIdle;
// Whether the thread is waiting on idlefd.
Atomic<bool, SequentiallyConsistent, Behavior::DontPreserve> mIdle;
// Any callback which should be invoked so the thread can make progress,
// and whether the callback has been invoked yet while the main thread is
// waiting for threads to become idle. Protected by the thread monitor.
std::function<void()> mUnrecordedWaitCallback;
bool mUnrecordedWaitOnlyWhenDiverged;
bool mUnrecordedWaitNotified;
// Identifier of any atomic which this thread currently holds.
Maybe<size_t> mAtomicLockId;
public:
///////////////////////////////////////////////////////////////////////////////
// Public Routines
///////////////////////////////////////////////////////////////////////////////
// Accessors for some members that never change.
size_t Id() { return mId; }
NativeThreadId NativeId() { return mNativeId; }
Stream& Events() { return *mEvents; }
uint8_t* StackBase() { return mStackBase; }
size_t StackSize() { return mStackSize; }
inline bool IsMainThread() const { return mId == MainThreadId; }
inline bool IsRecordedThread() const { return mId <= MaxRecordedThreadId; }
inline bool IsNonMainRecordedThread() const { return IsRecordedThread() && !IsMainThread(); }
// Access the flag for whether this thread is passing events through.
void SetPassThrough(bool aPassThrough) {
MOZ_RELEASE_ASSERT(mPassThroughEvents == !aPassThrough);
mPassThroughEvents = aPassThrough;
}
bool PassThroughEvents() const {
return mPassThroughEvents;
}
// Access the counter for whether events are disallowed in this thread.
void BeginDisallowEvents() {
mDisallowEvents++;
}
void EndDisallowEvents() {
MOZ_RELEASE_ASSERT(mDisallowEvents);
mDisallowEvents--;
}
bool AreEventsDisallowed() const {
return mDisallowEvents != 0;
}
// Access the flag for whether this thread's execution has diverged from the
// recording. Once set, this is only unset by rewinding to a point where the
// flag is clear.
void DivergeFromRecording() {
mDivergedFromRecording = true;
}
bool HasDivergedFromRecording() const {
return mDivergedFromRecording;
}
// Mark this thread as needing to diverge from the recording soon, and wake
// it up in case it can make progress now. The mShouldDivergeFromRecording
// flag is separate from mDivergedFromRecording so that the thread can only
// begin diverging from the recording at calls to MaybeDivergeFromRecording.
void SetShouldDivergeFromRecording() {
MOZ_RELEASE_ASSERT(CurrentIsMainThread());
mShouldDivergeFromRecording = true;
Notify(mId);
}
bool WillDivergeFromRecordingSoon() {
MOZ_RELEASE_ASSERT(CurrentIsMainThread());
return mShouldDivergeFromRecording;
}
bool MaybeDivergeFromRecording() {
if (mShouldDivergeFromRecording) {
mDivergedFromRecording = true;
}
return mDivergedFromRecording;
}
// Return whether this thread may read or write to its recorded event stream.
bool CanAccessRecording() const {
return !PassThroughEvents() && !AreEventsDisallowed() && !HasDivergedFromRecording();
}
// The actual start routine at the root of all recorded threads, and of all
// threads when replaying.
static void ThreadMain(void* aArgument);
// Bind this Thread to the current system thread, setting Thread::Current()
// and some other basic state.
void BindToCurrent();
// Initialize thread state.
static void InitializeThreads();
// Get the current thread, or null if this is a system thread.
static Thread* Current();
// Helper to test if this is the process main thread.
static bool CurrentIsMainThread();
// Lookup a Thread by various methods.
static Thread* GetById(size_t aId);
static Thread* GetByNativeId(NativeThreadId aNativeId);
static Thread* GetByStackPointer(void* aSp);
// Spawn all non-main recorded threads used for recording/replaying.
static void SpawnAllThreads();
// Spawn the specified thread.
static void SpawnThread(Thread* aThread);
// Spawn a non-recorded thread with the specified start routine/argument.
static Thread* SpawnNonRecordedThread(Callback aStart, void* aArgument);
// Wait until a thread has initialized its stack and other state.
static void WaitUntilInitialized(Thread* aThread);
// Start an existing thread, for use when the process has called a thread
// creation system API when events were not passed through. The return value
// is the native ID of the result.
static NativeThreadId StartThread(Callback aStart, void* aArgument, bool aNeedsJoin);
// Wait until this thread finishes executing its start routine.
void Join();
// Give access to the atomic lock which the thread owns.
Maybe<size_t>& AtomicLockId() { return mAtomicLockId; }
///////////////////////////////////////////////////////////////////////////////
// Thread Coordination
///////////////////////////////////////////////////////////////////////////////
// Basic API for threads to coordinate activity with each other, for use
// during replay. Each Notify() on a thread ID will cause that thread to
// return from one call to Wait(). Thus, if a thread Wait()'s and then
// another thread Notify()'s its ID, the first thread will wake up afterward.
// Similarly, if a thread Notify()'s another thread which is not waiting,
// that second thread will return from its next Wait() without needing
// another Notify().
//
// If the main thread has called WaitForIdleThreads, then calling
// Wait() will put this thread in the desired idle state. WaitNoIdle() will
// never cause the thread to enter the idle state, and should be used
// carefully to avoid deadlocks with the main thread.
static void Wait();
static void WaitNoIdle();
static void Notify(size_t aId);
// Wait indefinitely, until the process is rewound.
static void WaitForever();
// Wait indefinitely, without allowing this thread to be rewound.
static void WaitForeverNoIdle();
// See RecordReplay.h.
void NotifyUnrecordedWait(const std::function<void()>& aCallback,
bool aOnlyWhenDiverged);
static void MaybeWaitForCheckpointSave();
// Wait for all other threads to enter the idle state necessary for saving
// or restoring a checkpoint. This may only be called on the main thread.
static void WaitForIdleThreads();
// After WaitForIdleThreads(), the main thread will call this to allow
// other threads to resume execution.
static void ResumeIdleThreads();
// Allow a single thread to resume execution.
static void ResumeSingleIdleThread(size_t aId);
// Return whether this thread is in the idle state entered after
// WaitForIdleThreads.
bool IsIdle() { return mIdle; }
};
// This uses a stack pointer instead of TLS to make sure events are passed
// through, for avoiding thorny reentrance issues.
class AutoEnsurePassThroughThreadEventsUseStackPointer
{
Thread* mThread;
bool mPassedThrough;
public:
AutoEnsurePassThroughThreadEventsUseStackPointer()
: mThread(Thread::GetByStackPointer(this))
, mPassedThrough(!mThread || mThread->PassThroughEvents())
{
if (!mPassedThrough) {
mThread->SetPassThrough(true);
}
}
~AutoEnsurePassThroughThreadEventsUseStackPointer()
{
if (!mPassedThrough) {
mThread->SetPassThrough(false);
}
}
};
// Mark a region of code where a thread's event stream can be accessed.
// This class has several properties:
//
// - When recording, all writes to the thread's event stream occur atomically
// within the class: the end of the stream cannot be hit at an intermediate
// point.
//
// - When replaying, this checks for the end of the stream, and blocks the
// thread if necessary.
//
// - When replaying, this is a point where the thread can begin diverging from
// the recording. Checks for divergence should occur after the constructor
// finishes.
class MOZ_RAII RecordingEventSection
{
Thread* mThread;
public:
explicit RecordingEventSection(Thread* aThread)
: mThread(aThread)
{
if (!aThread || !aThread->CanAccessRecording()) {
return;
}
if (IsRecording()) {
MOZ_RELEASE_ASSERT(!aThread->Events().mInRecordingEventSection);
aThread->Events().mFile->mStreamLock.ReadLock();
aThread->Events().mInRecordingEventSection = true;
} else {
while (!aThread->MaybeDivergeFromRecording() && aThread->Events().AtEnd()) {
HitEndOfRecording();
}
}
}
~RecordingEventSection() {
if (!mThread || !mThread->CanAccessRecording()) {
return;
}
if (IsRecording()) {
mThread->Events().mFile->mStreamLock.ReadUnlock();
mThread->Events().mInRecordingEventSection = false;
}
}
bool CanAccessEvents() {
if (!mThread || mThread->PassThroughEvents() || mThread->HasDivergedFromRecording()) {
return false;
}
MOZ_RELEASE_ASSERT(mThread->CanAccessRecording());
return true;
}
};
} // namespace recordreplay
} // namespace mozilla
#endif // mozilla_recordreplay_Thread_h