gecko-dev/xpcom/threads/LabeledEventQueue.cpp

286 строки
7.9 KiB
C++

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "LabeledEventQueue.h"
#include "mozilla/dom/TabChild.h"
#include "mozilla/dom/TabGroup.h"
#include "mozilla/Scheduler.h"
#include "mozilla/SchedulerGroup.h"
#include "nsQueryObject.h"
using namespace mozilla::dom;
LinkedList<SchedulerGroup>* LabeledEventQueue::sSchedulerGroups;
size_t LabeledEventQueue::sLabeledEventQueueCount;
SchedulerGroup* LabeledEventQueue::sCurrentSchedulerGroup;
LabeledEventQueue::LabeledEventQueue()
{
// LabeledEventQueue should only be used by one consumer since it uses a
// single static sSchedulerGroups field. It's hard to assert this, though, so
// we assert NS_IsMainThread(), which is a reasonable proxy.
MOZ_ASSERT(NS_IsMainThread());
if (sLabeledEventQueueCount++ == 0) {
sSchedulerGroups = new LinkedList<SchedulerGroup>();
}
}
LabeledEventQueue::~LabeledEventQueue()
{
if (--sLabeledEventQueueCount == 0) {
delete sSchedulerGroups;
sSchedulerGroups = nullptr;
}
}
static SchedulerGroup*
GetSchedulerGroup(nsIRunnable* aEvent)
{
RefPtr<SchedulerGroup::Runnable> groupRunnable = do_QueryObject(aEvent);
if (!groupRunnable) {
// It's not labeled.
return nullptr;
}
return groupRunnable->Group();
}
static bool
IsReadyToRun(nsIRunnable* aEvent, SchedulerGroup* aEventGroup)
{
if (!Scheduler::AnyEventRunning()) {
return true;
}
if (Scheduler::UnlabeledEventRunning()) {
return false;
}
if (aEventGroup) {
return !aEventGroup->IsRunning();
}
nsCOMPtr<nsILabelableRunnable> labelable = do_QueryInterface(aEvent);
if (!labelable) {
return false;
}
return labelable->IsReadyToRun();
}
void
LabeledEventQueue::PutEvent(already_AddRefed<nsIRunnable>&& aEvent,
EventPriority aPriority,
const MutexAutoLock& aProofOfLock)
{
nsCOMPtr<nsIRunnable> event(aEvent);
MOZ_ASSERT(event.get());
SchedulerGroup* group = GetSchedulerGroup(event);
bool isLabeled = !!group;
// Create a new epoch if necessary.
Epoch* epoch;
if (mEpochs.IsEmpty()) {
epoch = &mEpochs.Push(Epoch::First(isLabeled));
} else {
Epoch& lastEpoch = mEpochs.LastElement();
if (lastEpoch.IsLabeled() != isLabeled) {
epoch = &mEpochs.Push(lastEpoch.NextEpoch(isLabeled));
} else {
epoch = &lastEpoch;
}
}
mNumEvents++;
epoch->mNumEvents++;
RunnableEpochQueue* queue = isLabeled ? mLabeled.LookupOrAdd(group) : &mUnlabeled;
queue->Push(QueueEntry(event.forget(), epoch->mEpochNumber));
if (group && group->EnqueueEvent() == SchedulerGroup::NewlyQueued) {
// This group didn't have any events before. Add it to the
// sSchedulerGroups list.
MOZ_ASSERT(!group->isInList());
sSchedulerGroups->insertBack(group);
if (!sCurrentSchedulerGroup) {
sCurrentSchedulerGroup = group;
}
}
}
void
LabeledEventQueue::PopEpoch()
{
Epoch& epoch = mEpochs.FirstElement();
MOZ_ASSERT(epoch.mNumEvents > 0);
if (epoch.mNumEvents == 1) {
mEpochs.Pop();
} else {
epoch.mNumEvents--;
}
mNumEvents--;
}
// Returns the next SchedulerGroup after |aGroup| in sSchedulerGroups. Wraps
// around to the beginning of the list when we hit the end.
/* static */ SchedulerGroup*
LabeledEventQueue::NextSchedulerGroup(SchedulerGroup* aGroup)
{
SchedulerGroup* result = aGroup->getNext();
if (!result) {
result = sSchedulerGroups->getFirst();
}
return result;
}
already_AddRefed<nsIRunnable>
LabeledEventQueue::GetEvent(EventPriority* aPriority,
const MutexAutoLock& aProofOfLock)
{
if (mEpochs.IsEmpty()) {
return nullptr;
}
Epoch epoch = mEpochs.FirstElement();
if (!epoch.IsLabeled()) {
QueueEntry entry = mUnlabeled.FirstElement();
if (!IsReadyToRun(entry.mRunnable, nullptr)) {
return nullptr;
}
PopEpoch();
mUnlabeled.Pop();
MOZ_ASSERT(entry.mEpochNumber == epoch.mEpochNumber);
MOZ_ASSERT(entry.mRunnable.get());
return entry.mRunnable.forget();
}
if (!sCurrentSchedulerGroup) {
return nullptr;
}
// Move active tabs to the front of the queue. The mAvoidActiveTabCount field
// prevents us from preferentially processing events from active tabs twice in
// a row. This scheme is designed to prevent starvation.
if (TabChild::HasActiveTabs() && mAvoidActiveTabCount <= 0) {
for (auto iter = TabChild::GetActiveTabs().ConstIter();
!iter.Done(); iter.Next()) {
SchedulerGroup* group = iter.Get()->GetKey()->TabGroup();
if (!group->isInList() || group == sCurrentSchedulerGroup) {
continue;
}
// For each active tab we move to the front of the queue, we have to
// process two SchedulerGroups (the active tab and another one, presumably
// a background group) before we prioritize active tabs again.
mAvoidActiveTabCount += 2;
// We move |group| right before sCurrentSchedulerGroup and then set
// sCurrentSchedulerGroup to group.
MOZ_ASSERT(group != sCurrentSchedulerGroup);
group->removeFrom(*sSchedulerGroups);
sCurrentSchedulerGroup->setPrevious(group);
sCurrentSchedulerGroup = group;
}
}
// Iterate over each SchedulerGroup once, starting at sCurrentSchedulerGroup.
SchedulerGroup* firstGroup = sCurrentSchedulerGroup;
SchedulerGroup* group = firstGroup;
do {
mAvoidActiveTabCount--;
RunnableEpochQueue* queue = mLabeled.Get(group);
if (!queue) {
// This can happen if |group| is in a different LabeledEventQueue than |this|.
group = NextSchedulerGroup(group);
continue;
}
MOZ_ASSERT(!queue->IsEmpty());
QueueEntry entry = queue->FirstElement();
if (entry.mEpochNumber == epoch.mEpochNumber &&
IsReadyToRun(entry.mRunnable, group)) {
sCurrentSchedulerGroup = NextSchedulerGroup(group);
PopEpoch();
if (group->DequeueEvent() == SchedulerGroup::NoLongerQueued) {
// Now we can take group out of sSchedulerGroups.
if (sCurrentSchedulerGroup == group) {
// Since we changed sCurrentSchedulerGroup above, we'll only get here
// if |group| was the only element in sSchedulerGroups. In that case
// set sCurrentSchedulerGroup to null.
MOZ_ASSERT(group->getNext() == nullptr);
MOZ_ASSERT(group->getPrevious() == nullptr);
sCurrentSchedulerGroup = nullptr;
}
group->removeFrom(*sSchedulerGroups);
}
queue->Pop();
if (queue->IsEmpty()) {
mLabeled.Remove(group);
}
return entry.mRunnable.forget();
}
group = NextSchedulerGroup(group);
} while (group != firstGroup);
return nullptr;
}
bool
LabeledEventQueue::IsEmpty(const MutexAutoLock& aProofOfLock)
{
return mEpochs.IsEmpty();
}
size_t
LabeledEventQueue::Count(const MutexAutoLock& aProofOfLock) const
{
return mNumEvents;
}
bool
LabeledEventQueue::HasReadyEvent(const MutexAutoLock& aProofOfLock)
{
if (mEpochs.IsEmpty()) {
return false;
}
Epoch& frontEpoch = mEpochs.FirstElement();
if (!frontEpoch.IsLabeled()) {
QueueEntry entry = mUnlabeled.FirstElement();
return IsReadyToRun(entry.mRunnable, nullptr);
}
// Go through the labeled queues and look for one whose head is from the
// current epoch and is allowed to run.
uintptr_t currentEpoch = frontEpoch.mEpochNumber;
for (auto iter = mLabeled.Iter(); !iter.Done(); iter.Next()) {
SchedulerGroup* key = iter.Key();
RunnableEpochQueue* queue = iter.Data();
MOZ_ASSERT(!queue->IsEmpty());
QueueEntry entry = queue->FirstElement();
if (entry.mEpochNumber != currentEpoch) {
continue;
}
if (IsReadyToRun(entry.mRunnable, key)) {
return true;
}
}
return false;
}