gecko-dev/xpcom/threads/nsThread.cpp

1439 строки
41 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: */
2012-05-21 15:12:37 +04:00
/* 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/. */
1999-04-02 13:20:44 +04:00
#include "nsThread.h"
#include "base/message_loop.h"
#include "base/platform_thread.h"
// Chromium's logging can sometimes leak through...
#ifdef LOG
#undef LOG
#endif
#include "mozilla/ReentrantMonitor.h"
#include "nsMemoryPressure.h"
#include "nsThreadManager.h"
#include "nsIClassInfoImpl.h"
#include "nsAutoPtr.h"
#include "nsCOMPtr.h"
#include "nsQueryObject.h"
#include "pratom.h"
#include "mozilla/BackgroundHangMonitor.h"
#include "mozilla/CycleCollectedJSContext.h"
#include "mozilla/Logging.h"
#include "nsIObserverService.h"
#include "mozilla/IOInterposer.h"
#include "mozilla/ipc/MessageChannel.h"
#include "mozilla/ipc/BackgroundChild.h"
#include "mozilla/Preferences.h"
#include "mozilla/Scheduler.h"
#include "mozilla/SchedulerGroup.h"
#include "mozilla/Services.h"
#include "mozilla/StaticPrefs.h"
#include "mozilla/SystemGroup.h"
#include "nsXPCOMPrivate.h"
#include "mozilla/ChaosMode.h"
#include "mozilla/Telemetry.h"
#include "mozilla/TimeStamp.h"
#include "mozilla/Unused.h"
#include "mozilla/dom/ScriptSettings.h"
#include "nsICrashReporter.h"
#include "nsThreadSyncDispatch.h"
#include "nsServiceManagerUtils.h"
#include "GeckoProfiler.h"
#include "InputEventStatistics.h"
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
#include "ThreadEventTarget.h"
#include "mozilla/dom/ContentChild.h"
#ifdef XP_LINUX
#ifdef __GLIBC__
#include <gnu/libc-version.h>
#endif
#include <sys/mman.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <sched.h>
#include <stdio.h>
#endif
#ifdef XP_WIN
#include "mozilla/DynamicallyLinkedFunctionPtr.h"
#include <Winbase.h>
using GetCurrentThreadStackLimitsFn = void (WINAPI*)(
PULONG_PTR LowLimit, PULONG_PTR HighLimit);
#endif
#define HAVE_UALARM _BSD_SOURCE || (_XOPEN_SOURCE >= 500 || \
_XOPEN_SOURCE && _XOPEN_SOURCE_EXTENDED) && \
!(_POSIX_C_SOURCE >= 200809L || _XOPEN_SOURCE >= 700)
#if defined(XP_LINUX) && !defined(ANDROID) && defined(_GNU_SOURCE)
#define HAVE_SCHED_SETAFFINITY
#endif
#ifdef XP_MACOSX
#include <mach/mach.h>
#include <mach/thread_policy.h>
#endif
#ifdef MOZ_CANARY
# include <unistd.h>
# include <execinfo.h>
# include <signal.h>
# include <fcntl.h>
# include "nsXULAppAPI.h"
#endif
#if defined(NS_FUNCTION_TIMER) && defined(_MSC_VER)
#include "nsTimerImpl.h"
#include "mozilla/StackWalk.h"
#endif
#ifdef NS_FUNCTION_TIMER
#include "nsCRT.h"
#endif
#ifdef MOZ_TASK_TRACER
#include "GeckoTaskTracer.h"
#include "TracedTaskCommon.h"
using namespace mozilla::tasktracer;
#endif
using namespace mozilla;
static LazyLogModule sThreadLog("nsThread");
#ifdef LOG
#undef LOG
#endif
#define LOG(args) MOZ_LOG(sThreadLog, mozilla::LogLevel::Debug, args)
1999-04-02 13:20:44 +04:00
NS_DECL_CI_INTERFACE_GETTER(nsThread)
1999-04-02 13:20:44 +04:00
Array<char, nsThread::kRunnableNameBufSize> nsThread::sMainThreadRunnableName;
//-----------------------------------------------------------------------------
// Because we do not have our own nsIFactory, we have to implement nsIClassInfo
// somewhat manually.
class nsThreadClassInfo : public nsIClassInfo
{
public:
NS_DECL_ISUPPORTS_INHERITED // no mRefCnt
NS_DECL_NSICLASSINFO
nsThreadClassInfo()
{
}
};
1999-04-02 13:20:44 +04:00
NS_IMETHODIMP_(MozExternalRefCountType)
nsThreadClassInfo::AddRef()
{
return 2;
}
NS_IMETHODIMP_(MozExternalRefCountType)
nsThreadClassInfo::Release()
{
return 1;
}
NS_IMPL_QUERY_INTERFACE(nsThreadClassInfo, nsIClassInfo)
NS_IMETHODIMP
nsThreadClassInfo::GetInterfaces(uint32_t* aCount, nsIID*** aArray)
{
return NS_CI_INTERFACE_GETTER_NAME(nsThread)(aCount, aArray);
1999-04-02 13:20:44 +04:00
}
NS_IMETHODIMP
nsThreadClassInfo::GetScriptableHelper(nsIXPCScriptable** aResult)
{
*aResult = nullptr;
return NS_OK;
}
1999-04-02 13:20:44 +04:00
NS_IMETHODIMP
nsThreadClassInfo::GetContractID(nsACString& aResult)
1999-04-02 13:20:44 +04:00
{
aResult.SetIsVoid(true);
return NS_OK;
1999-04-02 13:20:44 +04:00
}
NS_IMETHODIMP
nsThreadClassInfo::GetClassDescription(nsACString& aResult)
1999-04-02 13:20:44 +04:00
{
aResult.SetIsVoid(true);
return NS_OK;
1999-04-02 13:20:44 +04:00
}
NS_IMETHODIMP
nsThreadClassInfo::GetClassID(nsCID** aResult)
1999-04-02 13:20:44 +04:00
{
*aResult = nullptr;
return NS_OK;
1999-04-02 13:20:44 +04:00
}
NS_IMETHODIMP
nsThreadClassInfo::GetFlags(uint32_t* aResult)
1999-04-02 13:20:44 +04:00
{
*aResult = THREADSAFE;
return NS_OK;
1999-04-02 13:20:44 +04:00
}
NS_IMETHODIMP
nsThreadClassInfo::GetClassIDNoAlloc(nsCID* aResult)
1999-04-02 13:20:44 +04:00
{
return NS_ERROR_NOT_AVAILABLE;
}
//-----------------------------------------------------------------------------
NS_IMPL_ADDREF(nsThread)
NS_IMPL_RELEASE(nsThread)
NS_INTERFACE_MAP_BEGIN(nsThread)
NS_INTERFACE_MAP_ENTRY(nsIThread)
NS_INTERFACE_MAP_ENTRY(nsIThreadInternal)
NS_INTERFACE_MAP_ENTRY(nsIEventTarget)
NS_INTERFACE_MAP_ENTRY(nsISerialEventTarget)
NS_INTERFACE_MAP_ENTRY(nsISupportsPriority)
NS_INTERFACE_MAP_ENTRY_AMBIGUOUS(nsISupports, nsIThread)
if (aIID.Equals(NS_GET_IID(nsIClassInfo))) {
static nsThreadClassInfo sThreadClassInfo;
foundInterface = static_cast<nsIClassInfo*>(&sThreadClassInfo);
} else
NS_INTERFACE_MAP_END
NS_IMPL_CI_INTERFACE_GETTER(nsThread, nsIThread, nsIThreadInternal,
nsIEventTarget, nsISupportsPriority)
//-----------------------------------------------------------------------------
class nsThreadStartupEvent final : public Runnable
{
public:
nsThreadStartupEvent()
: Runnable("nsThreadStartupEvent")
, mMon("nsThreadStartupEvent.mMon")
, mInitialized(false)
{
}
// This method does not return until the thread startup object is in the
// completion state.
void Wait()
{
ReentrantMonitorAutoEnter mon(mMon);
while (!mInitialized) {
mon.Wait();
}
}
private:
~nsThreadStartupEvent() = default;
NS_IMETHOD Run() override
{
ReentrantMonitorAutoEnter mon(mMon);
mInitialized = true;
mon.Notify();
1999-04-02 13:20:44 +04:00
return NS_OK;
}
ReentrantMonitor mMon;
bool mInitialized;
};
//-----------------------------------------------------------------------------
struct nsThreadShutdownContext
{
nsThreadShutdownContext(NotNull<nsThread*> aTerminatingThread,
NotNull<nsThread*> aJoiningThread,
bool aAwaitingShutdownAck)
: mTerminatingThread(aTerminatingThread)
, mJoiningThread(aJoiningThread)
, mAwaitingShutdownAck(aAwaitingShutdownAck)
, mIsMainThreadJoining(NS_IsMainThread())
{
MOZ_COUNT_CTOR(nsThreadShutdownContext);
}
~nsThreadShutdownContext()
{
MOZ_COUNT_DTOR(nsThreadShutdownContext);
}
// NB: This will be the last reference.
NotNull<RefPtr<nsThread>> mTerminatingThread;
NotNull<nsThread*> MOZ_UNSAFE_REF("Thread manager is holding reference to joining thread")
mJoiningThread;
bool mAwaitingShutdownAck;
bool mIsMainThreadJoining;
};
// This event is responsible for notifying nsThread::Shutdown that it is time
// to call PR_JoinThread. It implements nsICancelableRunnable so that it can
// run on a DOM Worker thread (where all events must implement
// nsICancelableRunnable.)
class nsThreadShutdownAckEvent : public CancelableRunnable
{
public:
explicit nsThreadShutdownAckEvent(NotNull<nsThreadShutdownContext*> aCtx)
: CancelableRunnable("nsThreadShutdownAckEvent")
, mShutdownContext(aCtx)
{
}
NS_IMETHOD Run() override
{
mShutdownContext->mTerminatingThread->ShutdownComplete(mShutdownContext);
return NS_OK;
}
nsresult Cancel() override
{
return Run();
}
private:
virtual ~nsThreadShutdownAckEvent() { }
NotNull<nsThreadShutdownContext*> mShutdownContext;
};
// This event is responsible for setting mShutdownContext
class nsThreadShutdownEvent : public Runnable
{
public:
nsThreadShutdownEvent(NotNull<nsThread*> aThr,
NotNull<nsThreadShutdownContext*> aCtx)
: Runnable("nsThreadShutdownEvent")
, mThread(aThr)
, mShutdownContext(aCtx)
{
}
NS_IMETHOD Run() override
{
mThread->mShutdownContext = mShutdownContext;
MessageLoop::current()->Quit();
return NS_OK;
}
private:
NotNull<RefPtr<nsThread>> mThread;
NotNull<nsThreadShutdownContext*> mShutdownContext;
};
//-----------------------------------------------------------------------------
static void
SetThreadAffinity(unsigned int cpu)
{
#ifdef HAVE_SCHED_SETAFFINITY
cpu_set_t cpus;
CPU_ZERO(&cpus);
CPU_SET(cpu, &cpus);
sched_setaffinity(0, sizeof(cpus), &cpus);
// Don't assert sched_setaffinity's return value because it intermittently (?)
// fails with EINVAL on Linux x64 try runs.
#elif defined(XP_MACOSX)
// OS X does not provide APIs to pin threads to specific processors, but you
// can tag threads as belonging to the same "affinity set" and the OS will try
// to run them on the same processor. To run threads on different processors,
// tag them as belonging to different affinity sets. Tag 0, the default, means
// "no affinity" so let's pretend each CPU has its own tag `cpu+1`.
thread_affinity_policy_data_t policy;
policy.affinity_tag = cpu + 1;
MOZ_ALWAYS_TRUE(thread_policy_set(mach_thread_self(), THREAD_AFFINITY_POLICY,
&policy.affinity_tag, 1) == KERN_SUCCESS);
#elif defined(XP_WIN)
MOZ_ALWAYS_TRUE(SetThreadIdealProcessor(GetCurrentThread(), cpu) != (DWORD)-1);
#endif
}
static void
SetupCurrentThreadForChaosMode()
{
if (!ChaosMode::isActive(ChaosFeature::ThreadScheduling)) {
return;
}
#ifdef XP_LINUX
// PR_SetThreadPriority doesn't really work since priorities >
// PR_PRIORITY_NORMAL can't be set by non-root users. Instead we'll just use
// setpriority(2) to set random 'nice values'. In regular Linux this is only
// a dynamic adjustment so it still doesn't really do what we want, but tools
// like 'rr' can be more aggressive about honoring these values.
// Some of these calls may fail due to trying to lower the priority
// (e.g. something may have already called setpriority() for this thread).
// This makes it hard to have non-main threads with higher priority than the
// main thread, but that's hard to fix. Tools like rr can choose to honor the
// requested values anyway.
// Use just 4 priorities so there's a reasonable chance of any two threads
// having equal priority.
setpriority(PRIO_PROCESS, 0, ChaosMode::randomUint32LessThan(4));
#else
// We should set the affinity here but NSPR doesn't provide a way to expose it.
uint32_t priority = ChaosMode::randomUint32LessThan(PR_PRIORITY_LAST + 1);
PR_SetThreadPriority(PR_GetCurrentThread(), PRThreadPriority(priority));
#endif
// Force half the threads to CPU 0 so they compete for CPU
if (ChaosMode::randomUint32LessThan(2)) {
SetThreadAffinity(0);
}
}
namespace {
struct ThreadInitData {
nsThread* thread;
const nsACString& name;
};
}
/* static */ mozilla::OffTheBooksMutex&
nsThread::ThreadListMutex()
{
static OffTheBooksMutex sMutex("nsThread::ThreadListMutex");
return sMutex;
}
/* static */ LinkedList<nsThread>&
nsThread::ThreadList()
{
static LinkedList<nsThread> sList;
return sList;
}
/* static */ nsThreadEnumerator
nsThread::Enumerate()
{
return {};
}
/*static*/ void
nsThread::ThreadFunc(void* aArg)
{
using mozilla::ipc::BackgroundChild;
ThreadInitData* initData = static_cast<ThreadInitData*>(aArg);
nsThread* self = initData->thread; // strong reference
self->mThread = PR_GetCurrentThread();
self->mVirtualThread = GetCurrentVirtualThread();
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
self->mEventTarget->SetCurrentThread();
SetupCurrentThreadForChaosMode();
if (!initData->name.IsEmpty()) {
NS_SetCurrentThreadName(initData->name.BeginReading());
}
self->InitCommon();
// Inform the ThreadManager
nsThreadManager::get().RegisterCurrentThread(*self);
mozilla::IOInterposer::RegisterCurrentThread();
// This must come after the call to nsThreadManager::RegisterCurrentThread(),
// because that call is needed to properly set up this thread as an nsThread,
// which profiler_register_thread() requires. See bug 1347007.
if (!initData->name.IsEmpty()) {
PROFILER_REGISTER_THREAD(initData->name.BeginReading());
}
// Wait for and process startup event
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
nsCOMPtr<nsIRunnable> event = self->mEvents->GetEvent(true, nullptr);
MOZ_ASSERT(event);
initData = nullptr; // clear before unblocking nsThread::Init
event->Run(); // unblocks nsThread::Init
event = nullptr;
{
// Scope for MessageLoop.
nsAutoPtr<MessageLoop> loop(
new MessageLoop(MessageLoop::TYPE_MOZILLA_NONMAINTHREAD, self));
// Now, process incoming events...
loop->Run();
BackgroundChild::CloseForCurrentThread();
// NB: The main thread does not shut down here! It shuts down via
// nsThreadManager::Shutdown.
// Do NS_ProcessPendingEvents but with special handling to set
// mEventsAreDoomed atomically with the removal of the last event. The key
// invariant here is that we will never permit PutEvent to succeed if the
// event would be left in the queue after our final call to
// NS_ProcessPendingEvents. We also have to keep processing events as long
// as we have outstanding mRequestedShutdownContexts.
while (true) {
// Check and see if we're waiting on any threads.
self->WaitForAllAsynchronousShutdowns();
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
if (self->mEvents->ShutdownIfNoPendingEvents()) {
break;
}
NS_ProcessPendingEvents(self);
}
}
mozilla::IOInterposer::UnregisterCurrentThread();
// Inform the threadmanager that this thread is going away
nsThreadManager::get().UnregisterCurrentThread(*self);
PROFILER_UNREGISTER_THREAD();
// Dispatch shutdown ACK
NotNull<nsThreadShutdownContext*> context =
WrapNotNull(self->mShutdownContext);
MOZ_ASSERT(context->mTerminatingThread == self);
event = do_QueryObject(new nsThreadShutdownAckEvent(context));
if (context->mIsMainThreadJoining) {
SystemGroup::Dispatch(TaskCategory::Other, event.forget());
} else {
context->mJoiningThread->Dispatch(event, NS_DISPATCH_NORMAL);
}
// Release any observer of the thread here.
self->SetObserver(nullptr);
#ifdef MOZ_TASK_TRACER
FreeTraceInfo();
#endif
NS_RELEASE(self);
1999-04-02 13:20:44 +04:00
}
void
nsThread::InitCommon()
{
mThreadId = uint32_t(PlatformThread::CurrentId());
{
#if defined(XP_LINUX)
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_getattr_np(pthread_self(), &attr);
size_t stackSize;
pthread_attr_getstack(&attr, &mStackBase, &stackSize);
// Glibc prior to 2.27 reports the stack size and base including the guard
// region, so we need to compensate for it to get accurate accounting.
// Also, this behavior difference isn't guarded by a versioned symbol, so we
// actually need to check the runtime glibc version, not the version we were
// compiled against.
static bool sAdjustForGuardSize = ({
#ifdef __GLIBC__
unsigned major, minor;
sscanf(gnu_get_libc_version(), "%u.%u", &major, &minor) < 2 ||
major < 2 || (major == 2 && minor < 27);
#else
false;
#endif
});
if (sAdjustForGuardSize) {
size_t guardSize;
pthread_attr_getguardsize(&attr, &guardSize);
// Note: This assumes that the stack grows down, as is the case on all of
// our tier 1 platforms. On platforms where the stack grows up, the
// mStackBase adjustment is unnecessary, but doesn't cause any harm other
// than under-counting stack memory usage by one page.
mStackBase = reinterpret_cast<char*>(mStackBase) + guardSize;
stackSize -= guardSize;
}
mStackSize = stackSize;
// This is a bit of a hack.
//
// We really do want the NOHUGEPAGE flag on our thread stacks, since we
// don't expect any of them to need anywhere near 2MB of space. But setting
// it here is too late to have an effect, since the first stack page has
// already been faulted in existence, and NSPR doesn't give us a way to set
// it beforehand.
//
// What this does get us, however, is a different set of VM flags on our
// thread stacks compared to normal heap memory. Which makes the Linux
// kernel report them as separate regions, even when they are adjacent to
// heap memory. This allows us to accurately track the actual memory
// consumption of our allocated stacks.
madvise(mStackBase, stackSize, MADV_NOHUGEPAGE);
pthread_attr_destroy(&attr);
#elif defined(XP_WIN)
static const DynamicallyLinkedFunctionPtr<GetCurrentThreadStackLimitsFn>
sGetStackLimits(L"kernel32.dll", "GetCurrentThreadStackLimits");
if (sGetStackLimits) {
ULONG_PTR stackBottom, stackTop;
sGetStackLimits(&stackBottom, &stackTop);
mStackBase = reinterpret_cast<void*>(stackBottom);
mStackSize = stackTop - stackBottom;
}
#endif
}
OffTheBooksMutexAutoLock mal(ThreadListMutex());
ThreadList().insertBack(this);
}
//-----------------------------------------------------------------------------
// Tell the crash reporter to save a memory report if our heuristics determine
// that an OOM failure is likely to occur soon.
// Memory usage will not be checked more than every 30 seconds or saved more
// than every 3 minutes
// If |aShouldSave == kForceReport|, a report will be saved regardless of
// whether the process is low on memory or not. However, it will still not be
// saved if a report was saved less than 3 minutes ago.
bool
nsThread::SaveMemoryReportNearOOM(ShouldSaveMemoryReport aShouldSave)
{
// Keep an eye on memory usage (cheap, ~7ms) somewhat frequently,
// but save memory reports (expensive, ~75ms) less frequently.
const size_t kLowMemoryCheckSeconds = 30;
const size_t kLowMemorySaveSeconds = 3 * 60;
static TimeStamp nextCheck = TimeStamp::NowLoRes()
+ TimeDuration::FromSeconds(kLowMemoryCheckSeconds);
static bool recentlySavedReport = false; // Keeps track of whether a report
// was saved last time we checked
// Are we checking again too soon?
TimeStamp now = TimeStamp::NowLoRes();
if ((aShouldSave == ShouldSaveMemoryReport::kMaybeReport ||
recentlySavedReport) && now < nextCheck) {
return false;
}
bool needMemoryReport = (aShouldSave == ShouldSaveMemoryReport::kForceReport);
#ifdef XP_WIN // XXX implement on other platforms as needed
// If the report is forced there is no need to check whether it is necessary
if (aShouldSave != ShouldSaveMemoryReport::kForceReport) {
const size_t LOWMEM_THRESHOLD_VIRTUAL = 200 * 1024 * 1024;
MEMORYSTATUSEX statex;
statex.dwLength = sizeof(statex);
if (GlobalMemoryStatusEx(&statex)) {
if (statex.ullAvailVirtual < LOWMEM_THRESHOLD_VIRTUAL) {
needMemoryReport = true;
}
}
}
#endif
if (needMemoryReport) {
if (XRE_IsContentProcess()) {
dom::ContentChild* cc = dom::ContentChild::GetSingleton();
if (cc) {
cc->SendNotifyLowMemory();
}
} else {
nsCOMPtr<nsICrashReporter> cr =
do_GetService("@mozilla.org/toolkit/crash-reporter;1");
if (cr) {
cr->SaveMemoryReport();
}
}
recentlySavedReport = true;
nextCheck = now + TimeDuration::FromSeconds(kLowMemorySaveSeconds);
} else {
recentlySavedReport = false;
nextCheck = now + TimeDuration::FromSeconds(kLowMemoryCheckSeconds);
}
return recentlySavedReport;
}
#ifdef MOZ_CANARY
int sCanaryOutputFD = -1;
#endif
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
nsThread::nsThread(NotNull<SynchronizedEventQueue*> aQueue,
MainThreadFlag aMainThread,
uint32_t aStackSize)
: mEvents(aQueue.get())
, mEventTarget(new ThreadEventTarget(mEvents.get(), aMainThread == MAIN_THREAD))
, mShutdownContext(nullptr)
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
, mScriptObserver(nullptr)
, mThread(nullptr)
, mStackSize(aStackSize)
, mNestedEventLoopDepth(0)
, mCurrentEventLoopDepth(-1)
, mShutdownRequired(false)
, mPriority(PRIORITY_NORMAL)
, mIsMainThread(uint8_t(aMainThread))
, mCanInvokeJS(false)
, mCurrentEvent(nullptr)
, mCurrentEventStart(TimeStamp::Now())
, mCurrentPerformanceCounter(nullptr)
{
}
nsThread::~nsThread()
{
NS_ASSERTION(mRequestedShutdownContexts.IsEmpty(),
"shouldn't be waiting on other threads to shutdown");
// We shouldn't need to lock before checking isInList at this point. We're
// destroying the last reference to this object, so there's no way for anyone
// else to remove it in the middle of our check. And the not-in-list state is
// determined the element's next and previous members pointing to itself, so a
// non-atomic update to an adjacent member won't affect the outcome either.
if (isInList()) {
OffTheBooksMutexAutoLock mal(ThreadListMutex());
removeFrom(ThreadList());
}
#ifdef DEBUG
// We deliberately leak these so they can be tracked by the leak checker.
// If you're having nsThreadShutdownContext leaks, you can set:
// XPCOM_MEM_LOG_CLASSES=nsThreadShutdownContext
// during a test run and that will at least tell you what thread is
// requesting shutdown on another, which can be helpful for diagnosing
// the leak.
for (size_t i = 0; i < mRequestedShutdownContexts.Length(); ++i) {
Unused << mRequestedShutdownContexts[i].forget();
}
#endif
}
nsresult
nsThread::Init(const nsACString& aName)
{
// spawn thread and wait until it is fully setup
Bug 1207245 - part 6 - rename nsRefPtr<T> to RefPtr<T>; r=ehsan; a=Tomcat The bulk of this commit was generated with a script, executed at the top level of a typical source code checkout. The only non-machine-generated part was modifying MFBT's moz.build to reflect the new naming. CLOSED TREE makes big refactorings like this a piece of cake. # The main substitution. find . -name '*.cpp' -o -name '*.cc' -o -name '*.h' -o -name '*.mm' -o -name '*.idl'| \ xargs perl -p -i -e ' s/nsRefPtr\.h/RefPtr\.h/g; # handle includes s/nsRefPtr ?</RefPtr</g; # handle declarations and variables ' # Handle a special friend declaration in gfx/layers/AtomicRefCountedWithFinalize.h. perl -p -i -e 's/::nsRefPtr;/::RefPtr;/' gfx/layers/AtomicRefCountedWithFinalize.h # Handle nsRefPtr.h itself, a couple places that define constructors # from nsRefPtr, and code generators specially. We do this here, rather # than indiscriminantly s/nsRefPtr/RefPtr/, because that would rename # things like nsRefPtrHashtable. perl -p -i -e 's/nsRefPtr/RefPtr/g' \ mfbt/nsRefPtr.h \ xpcom/glue/nsCOMPtr.h \ xpcom/base/OwningNonNull.h \ ipc/ipdl/ipdl/lower.py \ ipc/ipdl/ipdl/builtin.py \ dom/bindings/Codegen.py \ python/lldbutils/lldbutils/utils.py # In our indiscriminate substitution above, we renamed # nsRefPtrGetterAddRefs, the class behind getter_AddRefs. Fix that up. find . -name '*.cpp' -o -name '*.h' -o -name '*.idl' | \ xargs perl -p -i -e 's/nsRefPtrGetterAddRefs/RefPtrGetterAddRefs/g' if [ -d .git ]; then git mv mfbt/nsRefPtr.h mfbt/RefPtr.h else hg mv mfbt/nsRefPtr.h mfbt/RefPtr.h fi --HG-- rename : mfbt/nsRefPtr.h => mfbt/RefPtr.h
2015-10-18 08:24:48 +03:00
RefPtr<nsThreadStartupEvent> startup = new nsThreadStartupEvent();
NS_ADDREF_THIS();
mShutdownRequired = true;
ThreadInitData initData = { this, aName };
// ThreadFunc is responsible for setting mThread
if (!PR_CreateThread(PR_USER_THREAD, ThreadFunc, &initData,
PR_PRIORITY_NORMAL, PR_GLOBAL_THREAD,
PR_JOINABLE_THREAD, mStackSize)) {
NS_RELEASE_THIS();
return NS_ERROR_OUT_OF_MEMORY;
}
// ThreadFunc will wait for this event to be run before it tries to access
// mThread. By delaying insertion of this event into the queue, we ensure
// that mThread is set properly.
{
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
mEvents->PutEvent(do_AddRef(startup), EventPriority::Normal); // retain a reference
}
// Wait for thread to call ThreadManager::SetupCurrentThread, which completes
// initialization of ThreadFunc.
startup->Wait();
return NS_OK;
}
nsresult
nsThread::InitCurrentThread()
{
mThread = PR_GetCurrentThread();
mVirtualThread = GetCurrentVirtualThread();
SetupCurrentThreadForChaosMode();
InitCommon();
nsThreadManager::get().RegisterCurrentThread(*this);
return NS_OK;
}
//-----------------------------------------------------------------------------
// nsIEventTarget
NS_IMETHODIMP
nsThread::DispatchFromScript(nsIRunnable* aEvent, uint32_t aFlags)
{
nsCOMPtr<nsIRunnable> event(aEvent);
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
return mEventTarget->Dispatch(event.forget(), aFlags);
}
NS_IMETHODIMP
nsThread::Dispatch(already_AddRefed<nsIRunnable> aEvent, uint32_t aFlags)
{
LOG(("THRD(%p) Dispatch [%p %x]\n", this, /* XXX aEvent */nullptr, aFlags));
return mEventTarget->Dispatch(std::move(aEvent), aFlags);
}
NS_IMETHODIMP
nsThread::DelayedDispatch(already_AddRefed<nsIRunnable> aEvent, uint32_t aDelayMs)
{
return mEventTarget->DelayedDispatch(std::move(aEvent), aDelayMs);
}
NS_IMETHODIMP
nsThread::IsOnCurrentThread(bool* aResult)
{
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
return mEventTarget->IsOnCurrentThread(aResult);
}
NS_IMETHODIMP_(bool)
nsThread::IsOnCurrentThreadInfallible()
{
// Rely on mVirtualThread being correct.
MOZ_CRASH("IsOnCurrentThreadInfallible should never be called on nsIThread");
}
//-----------------------------------------------------------------------------
// nsIThread
NS_IMETHODIMP
nsThread::GetPRThread(PRThread** aResult)
1999-04-02 13:20:44 +04:00
{
*aResult = mThread;
return NS_OK;
1999-04-02 13:20:44 +04:00
}
NS_IMETHODIMP
nsThread::GetCanInvokeJS(bool* aResult)
{
*aResult = mCanInvokeJS;
return NS_OK;
}
NS_IMETHODIMP
nsThread::SetCanInvokeJS(bool aCanInvokeJS)
{
mCanInvokeJS = aCanInvokeJS;
return NS_OK;
}
NS_IMETHODIMP
nsThread::AsyncShutdown()
{
LOG(("THRD(%p) async shutdown\n", this));
// XXX If we make this warn, then we hit that warning at xpcom shutdown while
// shutting down a thread in a thread pool. That happens b/c the thread
// in the thread pool is already shutdown by the thread manager.
if (!mThread) {
return NS_OK;
}
return !!ShutdownInternal(/* aSync = */ false) ? NS_OK : NS_ERROR_UNEXPECTED;
}
nsThreadShutdownContext*
nsThread::ShutdownInternal(bool aSync)
{
MOZ_ASSERT(mThread);
MOZ_ASSERT(mThread != PR_GetCurrentThread());
if (NS_WARN_IF(mThread == PR_GetCurrentThread())) {
return nullptr;
}
// Prevent multiple calls to this method
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
if (!mShutdownRequired.compareExchange(true, false)) {
return nullptr;
}
{
OffTheBooksMutexAutoLock mal(ThreadListMutex());
if (isInList()) {
removeFrom(ThreadList());
}
}
NotNull<nsThread*> currentThread =
WrapNotNull(nsThreadManager::get().GetCurrentThread());
nsAutoPtr<nsThreadShutdownContext>& context =
*currentThread->mRequestedShutdownContexts.AppendElement();
context = new nsThreadShutdownContext(WrapNotNull(this), currentThread, aSync);
// Set mShutdownContext and wake up the thread in case it is waiting for
// events to process.
nsCOMPtr<nsIRunnable> event =
new nsThreadShutdownEvent(WrapNotNull(this), WrapNotNull(context.get()));
// XXXroc What if posting the event fails due to OOM?
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
mEvents->PutEvent(event.forget(), EventPriority::Normal);
// We could still end up with other events being added after the shutdown
// task, but that's okay because we process pending events in ThreadFunc
// after setting mShutdownContext just before exiting.
return context;
}
void
nsThread::ShutdownComplete(NotNull<nsThreadShutdownContext*> aContext)
{
MOZ_ASSERT(mThread);
MOZ_ASSERT(aContext->mTerminatingThread == this);
if (aContext->mAwaitingShutdownAck) {
// We're in a synchronous shutdown, so tell whatever is up the stack that
// we're done and unwind the stack so it can call us again.
aContext->mAwaitingShutdownAck = false;
return;
}
// Now, it should be safe to join without fear of dead-locking.
PR_JoinThread(mThread);
mThread = nullptr;
#ifdef DEBUG
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
nsCOMPtr<nsIThreadObserver> obs = mEvents->GetObserver();
MOZ_ASSERT(!obs, "Should have been cleared at shutdown!");
#endif
// Delete aContext.
MOZ_ALWAYS_TRUE(
aContext->mJoiningThread->mRequestedShutdownContexts.RemoveElement(aContext));
}
void
nsThread::WaitForAllAsynchronousShutdowns()
{
// This is the motivating example for why SpinEventLoop has the template
// parameter we are providing here.
SpinEventLoopUntil<ProcessFailureBehavior::IgnoreAndContinue>([&]() {
return mRequestedShutdownContexts.IsEmpty();
}, this);
}
NS_IMETHODIMP
nsThread::Shutdown()
{
LOG(("THRD(%p) sync shutdown\n", this));
// XXX If we make this warn, then we hit that warning at xpcom shutdown while
// shutting down a thread in a thread pool. That happens b/c the thread
// in the thread pool is already shutdown by the thread manager.
if (!mThread) {
return NS_OK;
}
nsThreadShutdownContext* maybeContext = ShutdownInternal(/* aSync = */ true);
NS_ENSURE_TRUE(maybeContext, NS_ERROR_UNEXPECTED);
NotNull<nsThreadShutdownContext*> context = WrapNotNull(maybeContext);
// Process events on the current thread until we receive a shutdown ACK.
// Allows waiting; ensure no locks are held that would deadlock us!
SpinEventLoopUntil([&, context]() {
return !context->mAwaitingShutdownAck;
}, context->mJoiningThread);
ShutdownComplete(context);
return NS_OK;
}
NS_IMETHODIMP
nsThread::HasPendingEvents(bool* aResult)
{
if (NS_WARN_IF(PR_GetCurrentThread() != mThread)) {
return NS_ERROR_NOT_SAME_THREAD;
}
1999-04-02 13:20:44 +04:00
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
*aResult = mEvents->HasPendingEvent();
return NS_OK;
}
NS_IMETHODIMP
nsThread::IdleDispatch(already_AddRefed<nsIRunnable> aEvent)
{
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
nsCOMPtr<nsIRunnable> event = aEvent;
if (NS_WARN_IF(!event)) {
return NS_ERROR_INVALID_ARG;
}
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
if (!mEvents->PutEvent(event.forget(), EventPriority::Idle)) {
NS_WARNING("An idle event was posted to a thread that will never run it (rejected)");
return NS_ERROR_UNEXPECTED;
}
return NS_OK;
}
#ifdef MOZ_CANARY
void canary_alarm_handler(int signum);
class Canary
{
//XXX ToDo: support nested loops
public:
Canary()
{
if (sCanaryOutputFD > 0 && EventLatencyIsImportant()) {
signal(SIGALRM, canary_alarm_handler);
ualarm(15000, 0);
}
}
~Canary()
{
if (sCanaryOutputFD != 0 && EventLatencyIsImportant()) {
ualarm(0, 0);
}
}
static bool EventLatencyIsImportant()
{
return NS_IsMainThread() && XRE_IsParentProcess();
}
};
void canary_alarm_handler(int signum)
{
void* array[30];
const char msg[29] = "event took too long to run:\n";
// use write to be safe in the signal handler
write(sCanaryOutputFD, msg, sizeof(msg));
backtrace_symbols_fd(array, backtrace(array, 30), sCanaryOutputFD);
}
#endif
#define NOTIFY_EVENT_OBSERVERS(observers_, func_, params_) \
do { \
if (!observers_.IsEmpty()) { \
nsTObserverArray<nsCOMPtr<nsIThreadObserver>>::ForwardIterator \
iter_(observers_); \
nsCOMPtr<nsIThreadObserver> obs_; \
while (iter_.HasMore()) { \
obs_ = iter_.GetNext(); \
obs_ -> func_ params_ ; \
} \
} \
} while(0)
#ifdef MOZ_COLLECTING_RUNNABLE_TELEMETRY
static bool
GetLabeledRunnableName(nsIRunnable* aEvent,
nsACString& aName,
EventPriority aPriority)
{
bool labeled = false;
if (RefPtr<SchedulerGroup::Runnable> groupRunnable = do_QueryObject(aEvent)) {
labeled = true;
MOZ_ALWAYS_TRUE(NS_SUCCEEDED(groupRunnable->GetName(aName)));
} else if (nsCOMPtr<nsINamed> named = do_QueryInterface(aEvent)) {
MOZ_ALWAYS_TRUE(NS_SUCCEEDED(named->GetName(aName)));
} else {
aName.AssignLiteral("non-nsINamed runnable");
}
if (aName.IsEmpty()) {
aName.AssignLiteral("anonymous runnable");
}
if (!labeled && aPriority > EventPriority::Input) {
aName.AppendLiteral("(unlabeled)");
}
return labeled;
}
#endif
mozilla::PerformanceCounter*
nsThread::GetPerformanceCounter(nsIRunnable* aEvent)
{
RefPtr<SchedulerGroup::Runnable> docRunnable = do_QueryObject(aEvent);
if (docRunnable) {
mozilla::dom::DocGroup* docGroup = docRunnable->DocGroup();
if (docGroup) {
return docGroup->GetPerformanceCounter();
}
}
return nullptr;
}
size_t
nsThread::ShallowSizeOfIncludingThis(mozilla::MallocSizeOf aMallocSizeOf) const
{
size_t n = 0;
if (mShutdownContext) {
n += aMallocSizeOf(mShutdownContext);
}
n += mRequestedShutdownContexts.ShallowSizeOfExcludingThis(aMallocSizeOf);
return aMallocSizeOf(this) + aMallocSizeOf(mThread) + n;
}
size_t
nsThread::SizeOfEventQueues(mozilla::MallocSizeOf aMallocSizeOf) const
{
size_t n = 0;
if (mCurrentPerformanceCounter) {
n += aMallocSizeOf(mCurrentPerformanceCounter);
}
if (mEventTarget) {
// The size of mEvents is reported by mEventTarget.
n += mEventTarget->SizeOfIncludingThis(aMallocSizeOf);
}
return n;
}
size_t
nsThread::SizeOfIncludingThis(mozilla::MallocSizeOf aMallocSizeOf) const
{
return ShallowSizeOfIncludingThis(aMallocSizeOf) + SizeOfEventQueues(aMallocSizeOf);
}
NS_IMETHODIMP
nsThread::ProcessNextEvent(bool aMayWait, bool* aResult)
{
LOG(("THRD(%p) ProcessNextEvent [%u %u]\n", this, aMayWait,
mNestedEventLoopDepth));
if (NS_WARN_IF(PR_GetCurrentThread() != mThread)) {
return NS_ERROR_NOT_SAME_THREAD;
}
// When recording or replaying, execute triggers that were activated
// non-deterministically at some point since the last turn of the event loop.
if (recordreplay::IsRecordingOrReplaying()) {
recordreplay::ExecuteTriggers();
// Vsync observers are notified whenever processing events on the main
// thread. Waiting for explicit vsync messages from the UI process can
// result in paints happening at unexpected times when replaying/rewinding.
if (mIsMainThread == MAIN_THREAD) {
recordreplay::child::NotifyVsyncObserver();
}
}
2014-02-01 05:14:00 +04:00
// The toplevel event loop normally blocks waiting for the next event, but
// if we're trying to shut this thread down, we must exit the event loop when
// the event queue is empty.
// This only applys to the toplevel event loop! Nested event loops (e.g.
// during sync dispatch) are waiting for some state change and must be able
// to block even if something has requested shutdown of the thread. Otherwise
// we'll just busywait as we endlessly look for an event, fail to find one,
// and repeat the nested event loop since its state change hasn't happened yet.
bool reallyWait = aMayWait && (mNestedEventLoopDepth > 0 || !ShuttingDown());
Maybe<Scheduler::EventLoopActivation> activation;
if (IsMainThread()) {
DoMainThreadSpecificProcessing(reallyWait);
activation.emplace();
}
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
++mNestedEventLoopDepth;
// We only want to create an AutoNoJSAPI on threads that actually do DOM stuff
// (including workers). Those are exactly the threads that have an
// mScriptObserver.
Maybe<dom::AutoNoJSAPI> noJSAPI;
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
bool callScriptObserver = !!mScriptObserver;
if (callScriptObserver) {
noJSAPI.emplace();
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
mScriptObserver->BeforeProcessTask(reallyWait);
}
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
nsCOMPtr<nsIThreadObserver> obs = mEvents->GetObserverOnThread();
if (obs) {
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
obs->OnProcessNextEvent(this, reallyWait);
}
NOTIFY_EVENT_OBSERVERS(EventQueue()->EventObservers(), OnProcessNextEvent, (this, reallyWait));
#ifdef MOZ_CANARY
Canary canary;
#endif
nsresult rv = NS_OK;
{
// Scope for |event| to make sure that its destructor fires while
// mNestedEventLoopDepth has been incremented, since that destructor can
// also do work.
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
EventPriority priority;
nsCOMPtr<nsIRunnable> event = mEvents->GetEvent(reallyWait, &priority);
if (activation.isSome()) {
activation.ref().SetEvent(event, priority);
}
*aResult = (event.get() != nullptr);
if (event) {
LOG(("THRD(%p) running [%p]\n", this, event.get()));
if (IsMainThread()) {
BackgroundHangMonitor().NotifyActivity();
}
bool schedulerLoggingEnabled = mozilla::StaticPrefs::dom_performance_enable_scheduler_timing();
if (schedulerLoggingEnabled
&& mNestedEventLoopDepth > mCurrentEventLoopDepth
&& mCurrentPerformanceCounter) {
// This is a recursive call, we're saving the time
// spent in the parent event if the runnable is linked to a DocGroup.
mozilla::TimeDuration duration = TimeStamp::Now() - mCurrentEventStart;
mCurrentPerformanceCounter->IncrementExecutionDuration(duration.ToMicroseconds());
}
#ifdef MOZ_COLLECTING_RUNNABLE_TELEMETRY
// If we're on the main thread, we want to record our current runnable's
// name in a static so that BHR can record it.
Array<char, kRunnableNameBufSize> restoreRunnableName;
restoreRunnableName[0] = '\0';
auto clear = MakeScopeExit([&] {
if (IsMainThread()) {
MOZ_ASSERT(NS_IsMainThread());
sMainThreadRunnableName = restoreRunnableName;
}
});
if (IsMainThread()) {
nsAutoCString name;
GetLabeledRunnableName(event, name, priority);
MOZ_ASSERT(NS_IsMainThread());
restoreRunnableName = sMainThreadRunnableName;
// Copy the name into sMainThreadRunnableName's buffer, and append a
// terminating null.
uint32_t length = std::min((uint32_t) kRunnableNameBufSize - 1,
(uint32_t) name.Length());
memcpy(sMainThreadRunnableName.begin(), name.BeginReading(), length);
sMainThreadRunnableName[length] = '\0';
}
#endif
Maybe<AutoTimeDurationHelper> timeDurationHelper;
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
if (priority == EventPriority::Input) {
timeDurationHelper.emplace();
}
// The event starts to run, storing the timestamp.
bool recursiveEvent = false;
RefPtr<mozilla::PerformanceCounter> currentPerformanceCounter;
if (schedulerLoggingEnabled) {
recursiveEvent = mNestedEventLoopDepth > mCurrentEventLoopDepth;
mCurrentEventStart = mozilla::TimeStamp::Now();
mCurrentEvent = event;
mCurrentEventLoopDepth = mNestedEventLoopDepth;
mCurrentPerformanceCounter = GetPerformanceCounter(event);
currentPerformanceCounter = mCurrentPerformanceCounter;
}
event->Run();
// End of execution, we can send the duration for the group
if (schedulerLoggingEnabled) {
if (recursiveEvent) {
// If we're in a recursive call, reset the timer,
// so the parent gets its remaining execution time right.
mCurrentEventStart = mozilla::TimeStamp::Now();
mCurrentPerformanceCounter = currentPerformanceCounter;
} else {
// We're done with this dispatch
if (currentPerformanceCounter) {
mozilla::TimeDuration duration = TimeStamp::Now() - mCurrentEventStart;
currentPerformanceCounter->IncrementExecutionDuration(duration.ToMicroseconds());
}
mCurrentEvent = nullptr;
mCurrentEventLoopDepth = -1;
mCurrentPerformanceCounter = nullptr;
}
}
} else if (aMayWait) {
MOZ_ASSERT(ShuttingDown(),
"This should only happen when shutting down");
rv = NS_ERROR_UNEXPECTED;
}
}
NOTIFY_EVENT_OBSERVERS(EventQueue()->EventObservers(), AfterProcessNextEvent, (this, *aResult));
if (obs) {
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
obs->AfterProcessNextEvent(this, *aResult);
}
if (callScriptObserver) {
if (mScriptObserver) {
mScriptObserver->AfterProcessTask(mNestedEventLoopDepth);
}
noJSAPI.reset();
}
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
--mNestedEventLoopDepth;
return rv;
}
//-----------------------------------------------------------------------------
// nsISupportsPriority
NS_IMETHODIMP
nsThread::GetPriority(int32_t* aPriority)
{
*aPriority = mPriority;
return NS_OK;
}
NS_IMETHODIMP
nsThread::SetPriority(int32_t aPriority)
1999-04-06 10:09:15 +04:00
{
if (NS_WARN_IF(!mThread)) {
return NS_ERROR_NOT_INITIALIZED;
}
// NSPR defines the following four thread priorities:
// PR_PRIORITY_LOW
// PR_PRIORITY_NORMAL
// PR_PRIORITY_HIGH
// PR_PRIORITY_URGENT
// We map the priority values defined on nsISupportsPriority to these values.
mPriority = aPriority;
PRThreadPriority pri;
if (mPriority <= PRIORITY_HIGHEST) {
pri = PR_PRIORITY_URGENT;
} else if (mPriority < PRIORITY_NORMAL) {
pri = PR_PRIORITY_HIGH;
} else if (mPriority > PRIORITY_NORMAL) {
pri = PR_PRIORITY_LOW;
} else {
pri = PR_PRIORITY_NORMAL;
}
// If chaos mode is active, retain the randomly chosen priority
if (!ChaosMode::isActive(ChaosFeature::ThreadScheduling)) {
PR_SetThreadPriority(mThread, pri);
}
return NS_OK;
1999-04-06 10:09:15 +04:00
}
NS_IMETHODIMP
nsThread::AdjustPriority(int32_t aDelta)
{
return SetPriority(mPriority + aDelta);
}
//-----------------------------------------------------------------------------
// nsIThreadInternal
NS_IMETHODIMP
nsThread::GetObserver(nsIThreadObserver** aObs)
{
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
nsCOMPtr<nsIThreadObserver> obs = mEvents->GetObserver();
obs.forget(aObs);
return NS_OK;
}
NS_IMETHODIMP
nsThread::SetObserver(nsIThreadObserver* aObs)
{
if (NS_WARN_IF(PR_GetCurrentThread() != mThread)) {
return NS_ERROR_NOT_SAME_THREAD;
}
Bug 1382922 - Refactor event queue to allow multiple implementations (r=erahm) This patch refactors the nsThread event queue to clean it up and to make it easier to restructure. The fundamental concepts are as follows: Each nsThread will have a pointer to a refcounted SynchronizedEventQueue. A SynchronizedEQ takes care of doing the locking and condition variable work when posting and popping events. For the actual storage of events, it delegates to an AbstractEventQueue data structure. It keeps a UniquePtr to the AbstractEventQueue that it uses for storage. Both SynchronizedEQ and AbstractEventQueue are abstract classes. There is only one concrete implementation of SynchronizedEQ in this patch, which is called ThreadEventQueue. ThreadEventQueue uses locks and condition variables to post and pop events the same way nsThread does. It also encapsulates the functionality that DOM workers need to implement their special event loops (PushEventQueue and PopEventQueue). In later Quantum DOM work, I plan to have another SynchronizedEQ implementation for the main thread, called SchedulerEventQueue. It will have special code for the cooperatively scheduling threads in Quantum DOM. There are two concrete implementations of AbstractEventQueue in this patch: EventQueue and PrioritizedEventQueue. EventQueue replaces the old nsEventQueue. The other AbstractEventQueue implementation is PrioritizedEventQueue, which uses multiple queues for different event priorities. The final major piece here is ThreadEventTarget, which splits some of the code for posting events out of nsThread. Eventually, my plan is for multiple cooperatively scheduled nsThreads to be able to share a ThreadEventTarget. In this patch, though, each nsThread has its own ThreadEventTarget. The class's purpose is just to collect some related code together. One final note: I tried to avoid virtual dispatch overhead as much as possible. Calls to SynchronizedEQ methods do use virtual dispatch, since I plan to use different implementations for different threads with Quantum DOM. But all the calls to EventQueue methods should be non-virtual. Although the methods are declared virtual, all the classes used are final and the concrete classes involved should all be known through templatization. MozReview-Commit-ID: 9Evtr9oIJvx
2017-06-21 05:42:13 +03:00
mEvents->SetObserver(aObs);
return NS_OK;
}
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
uint32_t
nsThread::RecursionDepth() const
{
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
MOZ_ASSERT(PR_GetCurrentThread() == mThread);
return mNestedEventLoopDepth;
}
NS_IMETHODIMP
nsThread::AddObserver(nsIThreadObserver* aObserver)
{
if (NS_WARN_IF(!aObserver)) {
return NS_ERROR_INVALID_ARG;
}
if (NS_WARN_IF(PR_GetCurrentThread() != mThread)) {
return NS_ERROR_NOT_SAME_THREAD;
}
EventQueue()->AddObserver(aObserver);
return NS_OK;
}
NS_IMETHODIMP
nsThread::RemoveObserver(nsIThreadObserver* aObserver)
{
if (NS_WARN_IF(PR_GetCurrentThread() != mThread)) {
return NS_ERROR_NOT_SAME_THREAD;
}
EventQueue()->RemoveObserver(aObserver);
return NS_OK;
}
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
void
nsThread::SetScriptObserver(mozilla::CycleCollectedJSContext* aScriptObserver)
{
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
if (!aScriptObserver) {
mScriptObserver = nullptr;
return;
}
Bug 1179909: Refactor stable state handling. r=smaug This is motivated by three separate but related problems: 1. Our concept of recursion depth is broken for things that run from AfterProcessNextEvent observers (e.g. Promises). We decrement the recursionDepth counter before firing observers, so a Promise callback running at the lowest event loop depth has a recursion depth of 0 (whereas a regular nsIRunnable would be 1). This is a problem because it's impossible to distinguish a Promise running after a sync XHR's onreadystatechange handler from a top-level event (since the former runs with depth 2 - 1 = 1, and the latter runs with just 1). 2. The nsIThreadObserver mechanism that is used by a lot of code to run "after" the current event is a poor fit for anything that runs script. First, the order the observers fire in is the order they were added, not anything fixed by spec. Additionally, running script can cause the event loop to spin, which is a big source of pain here (bholley has some nasty bug caused by this). 3. We run Promises from different points in the code for workers and main thread. The latter runs from XPConnect's nsIThreadObserver callbacks, while the former runs from a hardcoded call to run Promises in the worker event loop. What workers do is particularly problematic because it means we can't get the right recursion depth no matter what we do to nsThread. The solve this, this patch does the following: 1. Consolidate some handling of microtasks and all handling of stable state from appshell and WorkerPrivate into CycleCollectedJSRuntime. 2. Make the recursionDepth counter only available to CycleCollectedJSRuntime (and its consumers) and remove it from the nsIThreadInternal and nsIThreadObserver APIs. 3. Adjust the recursionDepth counter so that microtasks run with the recursionDepth of the task they are associated with. 4. Introduce the concept of metastable state to replace appshell's RunBeforeNextEvent. Metastable state is reached after every microtask or task is completed. This provides the semantics that bent and I want for IndexedDB, where transactions autocommit at the end of a microtask and do not "spill" from one microtask into a subsequent microtask. This differs from appshell's RunBeforeNextEvent in two ways: a) It fires between microtasks, which was the motivation for starting this. b) It no longer ensures that we're at the same event loop depth in the native event queue. bent decided we don't care about this. 5. Reorder stable state to happen after microtasks such as Promises, per HTML. Right now we call the regular thread observers, including appshell, before the main thread observer (XPConnect), so stable state tasks happen before microtasks.
2015-08-11 16:10:46 +03:00
MOZ_ASSERT(!mScriptObserver);
mScriptObserver = aScriptObserver;
}
void
nsThread::DoMainThreadSpecificProcessing(bool aReallyWait)
{
MOZ_ASSERT(IsMainThread());
ipc::CancelCPOWs();
if (aReallyWait) {
BackgroundHangMonitor().NotifyWait();
}
// Fire a memory pressure notification, if one is pending.
if (!ShuttingDown()) {
MemoryPressureState mpPending = NS_GetPendingMemoryPressure();
if (mpPending != MemPressure_None) {
nsCOMPtr<nsIObserverService> os = services::GetObserverService();
if (os) {
if (mpPending == MemPressure_Stopping) {
os->NotifyObservers(nullptr, "memory-pressure-stop", nullptr);
} else {
os->NotifyObservers(nullptr, "memory-pressure",
mpPending == MemPressure_New ? u"low-memory" :
u"low-memory-ongoing");
}
} else {
NS_WARNING("Can't get observer service!");
}
}
}
if (!ShuttingDown()) {
SaveMemoryReportNearOOM(ShouldSaveMemoryReport::kMaybeReport);
}
}
NS_IMETHODIMP
nsThread::GetEventTarget(nsIEventTarget** aEventTarget)
{
nsCOMPtr<nsIEventTarget> target = this;
target.forget(aEventTarget);
return NS_OK;
}
nsIEventTarget*
nsThread::EventTarget()
{
return this;
}
nsISerialEventTarget*
nsThread::SerialEventTarget()
{
return this;
}