Files
palemoon27/xpcom/threads/TimerThread.cpp
T
roytam1 ebda1f73c4 import changes from `dev' branch of rmottola/Arctic-Fox:
- Bug 1208341 - Make NativeObject::goodElementsAllocationAmount fallible. r=Waldo (64eac96aa2)
- minor reformat (95d27e5332)
- Bug 1214049 - Use ReportOutOfMemory in NativeObject::goodElementsAllocationAmount. r=Waldo (6779dd81c6)
- Bug 1206700 - Fix an bug in property assignment, recently exposed by Reflect.set. r=waldo (0a7484ab70)
- Bug 1186810 (part 1) - Replace nsBaseHashtable::EnumerateRead() calls in dom/workers/ with iterators. r=nsm. (05a550c94f)
- Bug 1186810 (part 2) - Replace nsBaseHashtable::EnumerateRead() calls in dom/workers/ with iterators. r=nsm. (f1659abf38)
- Bug 1186810 (part 3) - Replace nsBaseHashtable::EnumerateRead() calls in dom/workers/ with iterators. r=nsm. (cda73950f3)
- Bug 1186810 (part 4) - Replace nsBaseHashtable::EnumerateRead() calls in dom/workers/ with iterators. r=nsm. (555e83bacb)
- Bug 1186810 (part 5) - Replace nsBaseHashtable::EnumerateRead() calls in dom/workers/ with iterators. r=nsm. (c9520dbc5f)
- Bug 1186810 (part 6) - Replace nsBaseHashtable::EnumerateRead() calls in dom/workers/ with iterators. r=khuey. (24fcdfd772)
- Bug 1186810 (part 7) - Replace nsBaseHashtable::EnumerateRead() calls in dom/workers/ with iterators. r=khuey. (657cd575fb)
- Bug 1186810 (part 8) - Replace nsBaseHashtable::Enumerate() calls in dom/workers/ with iterators. r=khuey. (0ff6a5e5ff)
- Bug 1205020 - Disable optimizations in TestTuple.cpp since it breaks the build on VS2013 Windows 10. r=froydnj (1c98d8485c)
- Bug 1199068 - Soothe a unused variable warning in TestSplayTree.cpp. r=nfroyd (d4d4ab56fe)
- Bug 1209751 - Make sync decoding more reliable when nsImageRenderer is used with -moz-element. r=roc (95c99bb907)
- Bug 1209731 - Return DrawResult::NOT_READY from PaintBackgroundWithSC if we couldn't prepare the nsImageRenderer. r=tn (8bd98872cb)
- Bug 1209715 - Add operator overloads for combining DrawResults. r=tn (f3a9806d03)
- Bug 1209725 - Use DrawResult combining operators in nsCSSRendering::PaintBackgroundWithSC. r=tn (420cfde63e)
- Bug 1211363. Calculate min inset blur with max border radius. r=mstange (5e408bc67f)
- Bug 1213545. Carry x,y offsets with inset box shadows. r=mstange (52c20715bc)
- Bug 1209739 - Make nsImageRenderer::PrepareImage() fail unless an image has STATUS_LOAD_COMPLETE. r=tn (c1cc4d79d8)
- Bug 1209765 (Part 1) - Support sync decoding of border-image in nsCSSRendering. r=tn (3a54b7d117)
- Bug 1209765 (Part 2) - Support sync decoding and track draw results in nsDisplayBorder. r=tn (03abe63e67)
- Bug 1209765 (Part 3) - Support sync decoding and track draw results when drawing borders in nsButtonFrameRenderer. r=tn (107b9bb851)
- Bug 1209765 (Part 4) - Support sync decoding and track draw results when drawing borders in nsFieldSetFrame. r=tn (16fd30aba3)
- Bug 1209765 (Part 5) - Support sync decoding and track draw results when drawing borders in nsRangeFrame. r=tn (289de3c17e)
- Bug 1209765 (Part 6) - Support sync decoding and track draw results when drawing borders in nsColumnSetFrame. r=tn (0317d6a914)
- Bug 1143267 - Make nsDisplayAltFeedback use nsGenericImageGeometry. r=tn (8fc6372d48)
- Bug 1176081. Add a pref to control drawing of the image loading place holder icon and border, and disable it. r=seth (792e8ec9b3)
- Bug 1209765 (Part 7) - Support sync decoding and track draw results when drawing borders in nsImageFrame. r=tn (70d9af60a3)
- Bug 1209765 (Part 8) - Support sync decoding and track draw results when drawing borders in MathML code. r=tn (555bff2635)
- Bug 1209765 (Part 9) - Support sync decoding and track draw results when drawing borders in nsTableFrame. r=tn (d242dc5c79)
- Bug 1209765 (Part 10) - Support sync decoding and track draw results when drawing borders in nsGroupBoxFrame. r=tn (a7187cfa5d)
- Bug 1209765 (Part 11) - Support sync decoding and track draw results when drawing borders in nsTreeBodyFrame. r=tn (21bd2b3285)
- Bug 1209765 (Followup) - Add very minor fuzz to underline-button-2.html on OS X 10.10. r=me (04cff3c2ff)
- Bug 1209765 (Followup) - Add very minor fuzz to underline-button-1.html on OS X 10.10. r=me DONTBUILD (187fdb4106)
- Bug 1217069 - Don't attempt to mis-optimize JSON.stringify's filter-list creation for replacer arrays with trailing holes. r=arai (04fb907d42)
- Bug 1156463 - Add a preference option to disable the scale transform applied to MathML operators. r=karl (415a041117)
- Bug 1174143 - Remove private tables for Standard Symbols L and MathJax fonts. r=karlt (1f58d14229)
- Bug 1222297 - Use channel->Open2() in netwerk/base/nsNetUtil.cpp (r=sicking) (cbd12ebb0a)
- Bug 1139709. Remove unused parameters from nsMathMLChar::GetMaxWidth. r=mats (6ea90a0a02)
- Bug 1160635 patch 1 - Add reftests. r=dholbert (b58f8fd7fa)
- Bug 1160635 patch 2 - Stop making image resizing optimization when image has percent width or height. r=dholbert (9ae3621baf)
- Bug 1156097 - Make nsImageFrame::mImageMap an nsRefPtr; r=roc (b49157e424)
- Bug 1176156. Make sure images are asked to decode if we are drawing the alt feedback instead of the image. r=seth (8d7a09622e)
- Bug 1207355 (Part 2) - Stop requesting decodes in nsImageLoadingContent. r=tn (c781285098)
- Bug 1207355 (Part 3) - Request decodes intelligently in nsImageFrame. r=tn (c61b7591fb)
- Bug 1156028 - Teach clang-analyzer about XPCOM assertion macros; r=froydnj (4d8b7ecf03)
- Bug 1207355 (Part 4) - Request decodes intelligently in nsBulletFrame. r=tn (2d7ced9c0f)
- Bug 1207355 (Part 5) - Request decodes intelligently in MultipartImage. r=tn (c0cb580838)
- Bug 1207355 (Part 6) - Only respect StartDecoding() in imgRequest/imgRequestProxy. r=tn (0e108cda4c)
- Bug 1207355 (Part 7) - Only trigger intrinsic size decode in FinalizeDecoder() if StartDecoding() was called. r=tn (a71fc0d1b7)
- Bug 1207355 (Part 8) - Remove imgIContainer::RequestDecode() and imgIRequest::RequestDecode(). r=tn (fbce1ccbe1)
- Bug 1215596 - outline: auto; will render as outline: solid; when disabled. r=roc (b8ea369628)
- Bug 1180966. Remove code that made no sense. r=mstange (eac62913ef)
- Bug 1217905. Don't transform dest rects if destination context has a rotation with inset box shadows. r=mstange (53cf4d6f86)
- Bug 1189010. In ScrollFrameHelper::BuildDisplayList decide if we are building a scrollable layer before we call AppendScrollPartsTo because it needs this information. r=mstange (a008c7e34d)
- Bug 1204535 - Flatten conditional structure one level. r=tn (152ff398a0)
- Bug 1204535 - Remove no-op code. r=tn (95d17e64c3)
- Bug 1204535 - Collapse redundant conditional block. r=tn (d0e90b8d80)
- Bug 1204535 - Rename usingDisplayport to usingDisplayPort for consistency. r=tn (516757a933)
- Bug 1204535 - Relocate comment to go with the line of code it's commenting. r=tn (70ae0c5e71)
- Bug 1204535 - Hoist out the code to initially compute usingDisplayPort. r=tn (4a5336a480)
- Bug 1210578. Part 1. Rename mShouldBuildScrollableLayer to mWillBuildScrollableLayer to better capture the meaning of the variable. r=mstange (d13bb100f4)
- Bug 1210578. Part 2. Rename shouldBuildLayer to couldBuildLayer to better reflect it's actual meaning. r=mstange (ae21c687c2)
- Bug 1151663 - Only call GetOrMaybeCreateDisplayPort() in nsLayoutUtils::PaintFrame() if we are using containers for root scroll frames. r=tn (2bef3f9457)
- Bug 1195431 - Small refactoring to use conditionals consistently. r=tn (897eb0f42a)
- Bug 1167069 - Initialize nsPluginFrame::mInstanceOwner to null. r=jmathies (6ce1fcb8c2)
- Bug 1137944 - Add support for hiding plugin windows during async dom scroll operations managed in the content process. r=roc (730a0ed004)
- Bug 1210578. Part 3. Create DecideScrollableLayer that encapsulates all logic to create display ports and build scrollable layers. r=mstange (994ac6ec5e)
- Bug 376679 part.1 Change the result of EventStateManager::ComputeScrollTarget() from nsIScrollableFrame* to nsIFrame* r=smaug (914878edbd)
- Bug 376679 part.2 EventStateManager should treat plugin frame as scrollable frame if the plugin wants to handle wheel events as default action r=smaug (58c146b7ba)
- Bug 376679 part.3 Compute default action target frame for wheel event before deciding the action because plugin should decide what is the default action when the target is a plugin frame r=smaug (334cbfcf53)
- Bug 376679 part.4 Manage wheel transaction at sending a wheel event to target plugin r=smaug (d76062a429)
- Bug 376679 part.5 nsPluginInstanceOwner::ProcessEvent() should create WM_MOUSE*WHEEL message from WidgetWheelEvent data r=smaug+jimm (56ee6bdf90)
- Bug 376679 part.6 nsPluginInstanceOwner::ProcessEvent() should refer both deltaMode and system scroll amount settings when it generates WM_MOSUE*WHEEL messages r=jimm (927cb10a3b)
- Bug 1145546 - Add missing nsBidiPresUtils.h include to nsGfxScrollFrame.cpp to fix non-unified bustage. r=jfkthame (af4043c024)
- Bug 1137944 - Add plugin window handling for apz related notifications received by the content process. r=mstange (798e9676fd)
- Bug 1208819 - Fix irregexp overrecursion check to ignore interrupts. r=bhackett (c845e65a2e)
- Bug 1218640 - IonMonkey: MIPS64: Add support into irregexp. r=arai (613ddd95ef)
- Bug 1211964 - Fix a couple of OOM handling issues in regexp code r=terrence (bb269eef3e)
- Bug 1200734 - Use stable hashing for LiveScopesMap; r=shu (1d717c8f53)
- Bug 1223853 - Use stable hashing for ObjectValueMap; r=jonco (c91f3b1b21)
- Bug 1223863 - Use stable hashing for WeakGlobalObjectSet; r=jonco (eb29e841f1)
- Bug 1223918 - Use stable hashing for WatchPointMap; r=jonco (7a69630c83)
- Bug 1212011 - Replace more uses of CrashAtUnhandlableOOM() with AutoEnterOOMUnsafeRegion r=terrence (0079439d90)
- Bug 1224347 - Use stable hashing for DebuggerWeakMaps; r=jonco (d2a627e077)
- Bug 1219418 - Always access minEmptyChunkCount under the GC lock; r=jonco (f67f74ca57)
- Bug 1202923 - Encapsulate GC begin and end notification callbacks in an RAII guard; r=jonco (46f63fd090)
- Bug 1212624 - Tests for LinkedList, r=Waldo (3d48b12918)
- Bug 1215752 - Weakmap refactoring, r=terrence (90cbf4d294)
- Bug 1216744 - Move weakKeys to the zone, r=terrence (5fdb1d57bf)
- Bug 1215336 - Measure unmark gray time, r=terrence (278e0ee562)
- Bug 1211939 - Don't call JSRuntime::onOutOfMemory from helper threads. r=jonco (86c6906421)
- Bug 1218638 - IonMonkey: MIPS64: Add support into jit/shared. r=lth (e4f2904c53)
- Bug 1216277 - Don't assert properties of the buffer location if the assembler is out of memory r=jandem (65556e83ce)
- Bug 1211962 - addNativeToBytecodeEntry: Check that the masm.currentOffset() returns a valid result. r=djvj (3c5723c65a)
- Bug 1090957 - IonMonkey: Add temp registers for mips atomics. r=lth (00f23fe2e9)
- Bug 1218967 - Do not stop executing autospider.sh on test failure. r=sfink (75f912efbb)
- Bug 1213881 - Run jsapi-tests in the arm64-sim autospider. r=sfink (2e731f14e7)
- Bug 1216607 - Fix assertion failure ARM assembler on OOM r=jandem (0bc2dcb5db)
- Bug 1209515 part 1 - IonBuilder: Attach hit counts on the MIRGraph. r=bhackett (f5448d9596)
- Bug 1209515 part 2 - Ensure that MPhi removal considers removed uses. r=bhackett (95de413c50)
- Bug 1209515 part 3 - IonMonkey: Add branch pruning based on code coverage counters. r=bhackett (fb71022658)
- Bug 1209515 part 0 - Move callerResumePoint field next to other resume points of basic blocks. r=bbouvier (394b52860d)
- Bug 1209515 part 4 - Increment code coverage counters on bailouts. r=bhackett (22edac5c24)
- Bug 1209515 part 5 - Ensure that we recompile when we have frequent bailouts caused by the first execution. r=jandem (bca0fa4118)
- Bug 1209515 part 7 - IonBuilder: pushLoop should not infer the continuepc from the entry block. r=jandem (025d10fd57)
- Bug 1209515 part 8 - IonBuilder: newOsrPreheader should not use the hit-count of the loop. r=jandem (5db05ef17f)
- Bug 1209515 part 9 - IonBuilder: Loop header should use the hit count of the loop, instead of before the loop. r=jandem (b259b490bf)
- Bug 1209515 part 10 - Prevent PruneUnusedBranches from being greedy while removing branches which have only be visited a few times. r=bhackett (4a1a7b34b6)
- Bug 1209515 part 11 - Increase JitSpewer mask size to avoid overflow. r=bbouvier (28546de94e)
- Bug 1219401 - Don't call script->hasIonScript() off-thread, fixes a TSan race. r=h4writer (8f31724484)
- Bug 1172076 - Assert that SnapshotIterator::initInstructionResults always match the recover instructions data. r=jandem (6ed0eebc65)
- Bug 1209260 - Don't use sincos on iOS r=npb (eaa3cacac7)
- Bug 1222499 - "JitSpew/IonSpew: warmupCounter should use PRIu32 instead of PRIuSIZE". r=nbp (50b80b985a)
- Bug 1216182: Rename MODE_MASK to MODE_BITS_MASK, r=hv1989 (f28e461d94)
- Bug 1137944 - Add a pref to control plugin window hiding. r=roc (691dc99b94)
- Bug 1214878 - When async scrolling with e10s hide windowed plugins in sub documents. r=roc (79c0eac459)
- Bug 1215434 part.1 If scroll target is a plugin frame, EventStateManager::PostHandleEvent() should send the wheel event to the plugin frame even if APZC already handled it r=smaug (f2e8e33d7f)
- Bug 1215434 part.2 Add plugin frame rect to dispatch-to-content region if it wants to handle wheel event as default action r=mstange (dbecb1e256)
- Fix inactive scroll frames sometimes creating very large, incorrect hit regions. (bug 1190112, r=tn) (2056d8d63e)
- Fix scrollbars on subframes not directing APZ events to the correct scrollframe. (bug 1213324, r=tn) (0594923ab2)
- Bug 1182017. Call Send__delete__ immediately rather than through an event. r=nical (90714b065b)
- Bug 1212656 - ARM64: Fix automation build failures. (r=efaust) (5695518dd1)
- Bug 1137944 - Simplify the api associated with hiding e10s plugin widgets. r=roc (dfbfda4d31)
- Bug 1137944 - Fire before and after composite events. r=matt.woodrow (0048aeb8cb)
- Bug 1201660 - When layer tree updates get dropped, hide plugins associated with the tree. Fixes a bug where we were hiding all plugins associated with the window. r=matt.woodrow (9352150c73)
- Bug 1137944 - Backout compositor bits (cset b4711fa5f734) which caused a telemetry metric regression. r=backout (241659d6a8)
- Bug 1127270: Only acquire a hold on the compositor thread once the channel for a parent actor has been succesfully connected. r=nical (1d1b077d68)
- Bug 1207270 - Ensure that the TaskThrottler objects are created on the main thread. r=botond (b43af025f1)
- Bug 1194876 - Add test. r=botond (440c72ec81)
- Bug 1152051 - Gtest. r=Cwiiis,kats (011f066f37)
- Bug 1144112 - Update the FlingStop tests to check if events are flowing properly after the fast-motion blocks are done. r=botond (25f71629d7)
- Bug 1092128 - Ensure that when touch-action is enabled the gtests always provide default touch behaviours where expected. r=botond (3f0d1229f3)
- Bug 1144112 - Only set the default allowed touch behaviour if the event is not consumed, to have the tests better reflect reality. r=botond (1548bd14bb)
- Bug 1163845 - Unify handling of time in APZ gtests. r=kats (b73a16ef5e)
- Bug 1159398 - Set the controller thread when running APZ gtests. r=dvander (96b551036e)
- Bug 1159985 - Gtest. r=kats (23d7d86492)
- Bug 1159985 - In APZ gtests, allow panning in both directions. r=kats (14f4b1156a)
- Bug 1169695 - Make the test timestamp privately held by the MockContentControllerDelayed. r=botond (ffe04249c1)
- Bug 1198900 - Add a gtest to ensure we don't crash when prevent-defaulting a wheel event. r=botond, a=topcrash (c9e5993f18)
- Bug 1169690 - Gtests. r=kats (7e825b34ac)
- Bug 1207270 - Update tests to ensure that TaskThrottlers are created as appropriate. r=botond (7bb6f1f406)
- Bug 1169695 - Ensure timed tasks are dispatched appropriately when advancing the time. r=botond (af1fbe0a85)
- Bug 1204837, Part 1: Make ClosingService thread Nuwa-friedly. r=khuey (eb790b5215)
- Bug 1204837, Part 2: Cloned ImageBridgeParent and CompositorParent should hold a reference to the compositor thread. r=nical (cd742df2e2)
- Bug 1205016: double ClosingService thread stack size to avoid stack overflow r=glandium (cb9acc828f)
- Bug 1163570 - Ignore pending plugin updates and hide plugin windows when a remote layer tree associated with a tab is no longer in use. r=roc (ae852ad442)
- Bug 1196539 - Apply plugin config using AutoResolveRefLayers during composition. r=matt.woodrow (12dbdb6222)
- Bug 1137944 - Cache compositor plugin update data to avoid sending updates to the main thread when the data doesn't change. r=roc (63c503d2f6)
- Bug 1137944 - On Windows defer composition until we receive confirmation plugin window metrics have been updated. r=roc (39d5786519)
- Bug 1137944 - Move plugin window metrics updates to a point during comososition where we know the remote layer tree is hooked up to the chrome layer tree. r=matt.woodrow (a32f77ffb1)
- Bug 1213671 - CompositorParent::UpdatePluginWindowState should always return an accurate result indicating when an ipc request for plugin geometry updates was sent. r=roc (311e65a8ce)
- Bug 1214267 - Send the side to which fixed-pos items are attached so that we can move them in the compositor more correctly when the viewport size is changing. r=mattwoodrow (fa2de207cd)
- Bug 1209964 - In AlignFixedAndStickyLayers(), properly handle the case where a fixed or sticky layer is its own subtree root and has a local transform. r=kats (1a6922699a)
- Bug 1213762 - Clear AsyncCompositionManager::mIsFirstPaint on non-Android platforms, too. r=mstange (c4cf2b722a)
- Bug 1223436 - Fix AsyncCompositionManager so that mLayersUpdated is only set to false after SyncFrameMetrics has been called. r=kats (f3fa84325b)
- bits of 1073545 (3bc7d76081)
- Bug 1213795, Part 1: Don't fire timers in the Nuwa process to fix test case test_NuwaProcessDeadlock.html. r=khuey (5ebca946e5)
- Bug 1213795, Part 2: Fix protocol cloning of actor CrossProcessCompositorParent. r=khuey (92969e2b06)
- spacing (f4857ed898)
- Bug 1213758 - Aquire the inderect layer trees data lock during plugin window update calculations. r=matt.woodrow (ae70d955ec)
- Bug 1208502 - Use more smart pointers in XULDocument. r=smaug (20a92d72e8)
- Bug 1193086 - Remove some null checks of infallible new code in XUL. r=baku (2f463c5a3d)
- Bug 1217190 - Make PrintingParent return an nsresult when attempting to show print progress. r=bobowen (addf31f3a7)
- Bug 1217190 - Make sure to initialize outparam in PrintingParent::RecvShowProgress. r=bobowen (f6c11a0181)
- Bug 1152921 - Don't send an uninitialized PrintData over IPC when cancelling print dialog (or failing ShowPrintDialog for other reasons). r=mconley (4ea146ccf1)
- Bug 1178233 - [non-e10s] The update process doesn't work within about:serviceworkers in non-e10s mode. r=baku (7a08441bec)
- Bug 929846 - Considering key size when calculating quota usage. r=jdm (a97ddc7906)
- Bug 1194049 - Part 3: support device update; r=fabrice (6ba600a03d)
- Bug 1194049 - Part 1: add thread assertion; r=schien (68c4658ae0)
- Bug 1194049 - Part 2: add discovery timeout support; r=schien (14f5f537d5)
- Bug 1194049 - Part 4: clear discovered devices when re-discover; r=schien (e2ecdb573f)
- Bug 1136565 - report session request from device provider. r=xeonchen,junior,seanlin. (19daa67751)
- Bug 1216398 - support non-discoverable controlling device. r=junior,xeonchen. (7383401dbc)
- Bug 1200132 - Use adderss to create socketTransport, r=schien, junior (263d253562)
- Bug 1200132 - Part1: Get address info after service resolved, r=mcmanus (9bcc344d20)
- Bug 1200132 - Part2: Get IP address by calling InetAddress.getHostAddress(), r=mcmanus, rnewman (502c243314)
- Bug 1173334: Open nfcd connections with socket type SOCK_SEQPACKET, allstars.chh (03a11dc077)
- Bug 1173802: Move NFC socket and messaging code into |NfcConsumer|, r=allstars.chh (f92946b40e)
- Bug 1173802: Run |NfcConsumer| on NFC thread, r=allstars.chh (87989483a7)
- Bug 1173802: Clean up |NfcEventDispatcher|, r=allstars.chh (87f58b7310)
- Bug 1173802: Replace |assertIsNfcServiceThread| with thread-safe checks, r=allstars.chh (1a6a76ae05)
- Bug 1173802: Store global NFC service in ref-counted variable, r=allstars.chh (a33c025976)
- Bug 1188639 - Fix typo in BrowserElement.webidl and add tests for it. r=yoshi,bz (7b657a335d)
- Bug 1175387 - close file descriptor after detached from STS. r=mcmanus (377943bad5)
- Bug 1190069 - Fix uninitialized variable in MDNSResponderOperator.cpp. r=xeonchen (ebf47b613e)
- Bug 1217807 - Part 1: use ServiceWatcher to extend life cycle of mDNS operators; r=schien (70201259d2)
- missing namespace (c6bccda1e5)
- Bug 1209843 - Stop checking for UNKNOWN_APP_ID in all places except those where AppId() is explicitly queried. r=sicking (e364223906)
- Bug 1168337: Kill the processes on startup of the Nuwa test case fix the abort in the preallocated process. r=khuey (501a30cb9b)
- Bug 1202424 - Free shared memory when IPC system shuts down, not after. r=mrbkap (f718cbf801)
- Bug 1209987 - webrtc.org Engine creation and destruction should happen on the WebRTC threads. r=jesup (4bbab795cf)
- Bug 1214675 - Factor out cleanup functions to avoid deadlock/dispatch-without-IPC. r=jesup (bd466c85e4)
- Bug 1210560 - Part 2: Convert some simple users to use PushGroupForBlendBack. r=jrmuizel (158b532f9f)
- Bug 1174785 - Part 0: Use mozilla/StaticPtr.h in signaling. r=jesup (c98ec23ee0)
- Bug 1174785 - Part 1: Add LogModule, LogModuleManager, and LazyLogModule. r=froydnj (e797bca424)
- Bug 1195867 - Hoist StateWatching and StateMirroring into XPCOM. r=froydnj (a62fbbf749)
- Bug 1194631 - Make firefox build with --disable-logging. r=jrmuizel (ba8238dc42)
- Bug 1174785 - Part 2: Convert xpcom over to LogModule. r=froydnj (ccb119d496)
- Bug 1205237 - Part 2: add PresentationNetworkHelper; r=schien (be47702ffe)
- Bug 1205237 - Part 3: get self Wi-Fi IP address; r=seanlin (f31163e895)
- Bug 1217683 - Add sendIceCandidate and implement |close(aReason)| in TCPControlChannel. r=fabrice (0004ea93f1)
- Bug 1217807 - Part 2: Handle network online/offline event in TCPPresentationServer; r=junior (48c9236adb)
- Bug 1217807 - disable MDNS discovery. r=xeonchen. (ee3cb0ddb6)
2022-11-23 13:32:46 +08:00

784 lines
20 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 "nsTimerImpl.h"
#include "TimerThread.h"
#include "nsThreadUtils.h"
#include "plarena.h"
#include "pratom.h"
#include "nsIObserverService.h"
#include "nsIServiceManager.h"
#include "mozilla/Services.h"
#include "mozilla/ChaosMode.h"
#include "mozilla/ArrayUtils.h"
#include "mozilla/BinarySearch.h"
#include <math.h>
using namespace mozilla;
NS_IMPL_ISUPPORTS(TimerThread, nsIRunnable, nsIObserver)
TimerThread::TimerThread() :
mInitInProgress(false),
mInitialized(false),
mMonitor("TimerThread.mMonitor"),
mShutdown(false),
mWaiting(false),
mNotified(false),
mSleeping(false)
{
}
TimerThread::~TimerThread()
{
mThread = nullptr;
NS_ASSERTION(mTimers.IsEmpty(), "Timers remain in TimerThread::~TimerThread");
}
nsresult
TimerThread::InitLocks()
{
return NS_OK;
}
namespace {
class TimerObserverRunnable : public nsRunnable
{
public:
explicit TimerObserverRunnable(nsIObserver* aObserver)
: mObserver(aObserver)
{
}
NS_DECL_NSIRUNNABLE
private:
nsCOMPtr<nsIObserver> mObserver;
};
NS_IMETHODIMP
TimerObserverRunnable::Run()
{
nsCOMPtr<nsIObserverService> observerService =
mozilla::services::GetObserverService();
if (observerService) {
observerService->AddObserver(mObserver, "sleep_notification", false);
observerService->AddObserver(mObserver, "wake_notification", false);
observerService->AddObserver(mObserver, "suspend_process_notification", false);
observerService->AddObserver(mObserver, "resume_process_notification", false);
}
return NS_OK;
}
} // namespace
namespace {
// TimerEventAllocator is a thread-safe allocator used only for nsTimerEvents.
// It's needed to avoid contention over the default allocator lock when
// firing timer events (see bug 733277). The thread-safety is required because
// nsTimerEvent objects are allocated on the timer thread, and freed on another
// thread. Because TimerEventAllocator has its own lock, contention over that
// lock is limited to the allocation and deallocation of nsTimerEvent objects.
//
// Because this allocator is layered over PLArenaPool, it never shrinks -- even
// "freed" nsTimerEvents aren't truly freed, they're just put onto a free-list
// for later recycling. So the amount of memory consumed will always be equal
// to the high-water mark consumption. But nsTimerEvents are small and it's
// unusual to have more than a few hundred of them, so this shouldn't be a
// problem in practice.
class TimerEventAllocator
{
private:
struct FreeEntry
{
FreeEntry* mNext;
};
PLArenaPool mPool;
FreeEntry* mFirstFree;
mozilla::Monitor mMonitor;
public:
TimerEventAllocator()
: mFirstFree(nullptr)
, mMonitor("TimerEventAllocator")
{
PL_InitArenaPool(&mPool, "TimerEventPool", 4096, /* align = */ 0);
}
~TimerEventAllocator()
{
PL_FinishArenaPool(&mPool);
}
void* Alloc(size_t aSize);
void Free(void* aPtr);
};
} // namespace
class nsTimerEvent : public nsRunnable
{
public:
NS_IMETHOD Run();
nsTimerEvent()
: mTimer()
, mGeneration(0)
{
MOZ_COUNT_CTOR(nsTimerEvent);
// Note: We override operator new for this class, and the override is
// fallible!
sAllocatorUsers++;
}
TimeStamp mInitTime;
static void Init();
static void Shutdown();
static void DeleteAllocatorIfNeeded();
static void* operator new(size_t aSize) CPP_THROW_NEW
{
return sAllocator->Alloc(aSize);
}
void operator delete(void* aPtr)
{
sAllocator->Free(aPtr);
DeleteAllocatorIfNeeded();
}
already_AddRefed<nsTimerImpl> ForgetTimer()
{
return mTimer.forget();
}
void SetTimer(already_AddRefed<nsTimerImpl> aTimer)
{
mTimer = aTimer;
mGeneration = mTimer->GetGeneration();
}
private:
~nsTimerEvent()
{
MOZ_COUNT_DTOR(nsTimerEvent);
MOZ_ASSERT(!sCanDeleteAllocator || sAllocatorUsers > 0,
"This will result in us attempting to deallocate the nsTimerEvent allocator twice");
sAllocatorUsers--;
}
RefPtr<nsTimerImpl> mTimer;
int32_t mGeneration;
static TimerEventAllocator* sAllocator;
static Atomic<int32_t> sAllocatorUsers;
static bool sCanDeleteAllocator;
};
TimerEventAllocator* nsTimerEvent::sAllocator = nullptr;
Atomic<int32_t> nsTimerEvent::sAllocatorUsers;
bool nsTimerEvent::sCanDeleteAllocator = false;
namespace {
void*
TimerEventAllocator::Alloc(size_t aSize)
{
MOZ_ASSERT(aSize == sizeof(nsTimerEvent));
mozilla::MonitorAutoLock lock(mMonitor);
void* p;
if (mFirstFree) {
p = mFirstFree;
mFirstFree = mFirstFree->mNext;
} else {
PL_ARENA_ALLOCATE(p, &mPool, aSize);
if (!p) {
return nullptr;
}
}
return p;
}
void
TimerEventAllocator::Free(void* aPtr)
{
mozilla::MonitorAutoLock lock(mMonitor);
FreeEntry* entry = reinterpret_cast<FreeEntry*>(aPtr);
entry->mNext = mFirstFree;
mFirstFree = entry;
}
} // namespace
void
nsTimerEvent::Init()
{
sAllocator = new TimerEventAllocator();
}
void
nsTimerEvent::Shutdown()
{
sCanDeleteAllocator = true;
DeleteAllocatorIfNeeded();
}
void
nsTimerEvent::DeleteAllocatorIfNeeded()
{
if (sCanDeleteAllocator && sAllocatorUsers == 0) {
delete sAllocator;
sAllocator = nullptr;
}
}
NS_IMETHODIMP
nsTimerEvent::Run()
{
if (mGeneration != mTimer->GetGeneration()) {
return NS_OK;
}
if (MOZ_LOG_TEST(GetTimerLog(), LogLevel::Debug)) {
TimeStamp now = TimeStamp::Now();
MOZ_LOG(GetTimerLog(), LogLevel::Debug,
("[this=%p] time between PostTimerEvent() and Fire(): %fms\n",
this, (now - mInitTime).ToMilliseconds()));
}
mTimer->Fire();
// Since nsTimerImpl is not thread-safe, we should release |mTimer|
// here in the target thread to avoid race condition. Otherwise,
// ~nsTimerEvent() which calls nsTimerImpl::Release() could run in the
// timer thread and result in race condition.
mTimer = nullptr;
return NS_OK;
}
nsresult
TimerThread::Init()
{
MOZ_LOG(GetTimerLog(), LogLevel::Debug,
("TimerThread::Init [%d]\n", mInitialized));
if (mInitialized) {
if (!mThread) {
return NS_ERROR_FAILURE;
}
return NS_OK;
}
nsTimerEvent::Init();
if (mInitInProgress.exchange(true) == false) {
// We hold on to mThread to keep the thread alive.
nsresult rv = NS_NewThread(getter_AddRefs(mThread), this);
if (NS_FAILED(rv)) {
mThread = nullptr;
} else {
RefPtr<TimerObserverRunnable> r = new TimerObserverRunnable(this);
if (NS_IsMainThread()) {
r->Run();
} else {
NS_DispatchToMainThread(r);
}
}
{
MonitorAutoLock lock(mMonitor);
mInitialized = true;
mMonitor.NotifyAll();
}
} else {
MonitorAutoLock lock(mMonitor);
while (!mInitialized) {
mMonitor.Wait();
}
}
if (!mThread) {
return NS_ERROR_FAILURE;
}
return NS_OK;
}
nsresult
TimerThread::Shutdown()
{
MOZ_LOG(GetTimerLog(), LogLevel::Debug, ("TimerThread::Shutdown begin\n"));
if (!mThread) {
return NS_ERROR_NOT_INITIALIZED;
}
nsTArray<nsTimerImpl*> timers;
{
// lock scope
MonitorAutoLock lock(mMonitor);
mShutdown = true;
// notify the cond var so that Run() can return
if (mWaiting) {
mNotified = true;
mMonitor.Notify();
}
// Need to copy content of mTimers array to a local array
// because call to timers' ReleaseCallback() (and release its self)
// must not be done under the lock. Destructor of a callback
// might potentially call some code reentering the same lock
// that leads to unexpected behavior or deadlock.
// See bug 422472.
timers.AppendElements(mTimers);
mTimers.Clear();
}
uint32_t timersCount = timers.Length();
for (uint32_t i = 0; i < timersCount; i++) {
nsTimerImpl* timer = timers[i];
timer->ReleaseCallback();
ReleaseTimerInternal(timer);
}
mThread->Shutdown(); // wait for the thread to die
nsTimerEvent::Shutdown();
MOZ_LOG(GetTimerLog(), LogLevel::Debug, ("TimerThread::Shutdown end\n"));
return NS_OK;
}
#ifdef MOZ_NUWA_PROCESS
#include "ipc/Nuwa.h"
#endif
namespace {
struct MicrosecondsToInterval
{
PRIntervalTime operator[](size_t aMs) const {
return PR_MicrosecondsToInterval(aMs);
}
};
struct IntervalComparator
{
int operator()(PRIntervalTime aInterval) const {
return (0 < aInterval) ? -1 : 1;
}
};
} // namespace
NS_IMETHODIMP
TimerThread::Run()
{
PR_SetCurrentThreadName("Timer");
#ifdef MOZ_NUWA_PROCESS
if (IsNuwaProcess()) {
NuwaMarkCurrentThread(nullptr, nullptr);
}
#endif
MonitorAutoLock lock(mMonitor);
// We need to know how many microseconds give a positive PRIntervalTime. This
// is platform-dependent and we calculate it at runtime, finding a value |v|
// such that |PR_MicrosecondsToInterval(v) > 0| and then binary-searching in
// the range [0, v) to find the ms-to-interval scale.
uint32_t usForPosInterval = 1;
while (PR_MicrosecondsToInterval(usForPosInterval) == 0) {
usForPosInterval <<= 1;
}
size_t usIntervalResolution;
BinarySearchIf(MicrosecondsToInterval(), 0, usForPosInterval, IntervalComparator(), &usIntervalResolution);
MOZ_ASSERT(PR_MicrosecondsToInterval(usIntervalResolution - 1) == 0);
MOZ_ASSERT(PR_MicrosecondsToInterval(usIntervalResolution) == 1);
// Half of the amount of microseconds needed to get positive PRIntervalTime.
// We use this to decide how to round our wait times later
int32_t halfMicrosecondsIntervalResolution = usIntervalResolution / 2;
bool forceRunNextTimer = false;
while (!mShutdown) {
// Have to use PRIntervalTime here, since PR_WaitCondVar takes it
PRIntervalTime waitFor;
bool forceRunThisTimer = forceRunNextTimer;
forceRunNextTimer = false;
if (mSleeping
#ifdef MOZ_NUWA_PROCESS
|| IsNuwaProcess() // Don't fire timers or deadlock will result.
#endif
) {
// Sleep for 0.1 seconds while not firing timers.
uint32_t milliseconds = 100;
if (ChaosMode::isActive(ChaosFeature::TimerScheduling)) {
milliseconds = ChaosMode::randomUint32LessThan(200);
}
waitFor = PR_MillisecondsToInterval(milliseconds);
} else {
waitFor = PR_INTERVAL_NO_TIMEOUT;
TimeStamp now = TimeStamp::Now();
nsTimerImpl* timer = nullptr;
if (!mTimers.IsEmpty()) {
timer = mTimers[0];
if (now >= timer->mTimeout || forceRunThisTimer) {
next:
// NB: AddRef before the Release under RemoveTimerInternal to avoid
// mRefCnt passing through zero, in case all other refs than the one
// from mTimers have gone away (the last non-mTimers[i]-ref's Release
// must be racing with us, blocked in gThread->RemoveTimer waiting
// for TimerThread::mMonitor, under nsTimerImpl::Release.
RefPtr<nsTimerImpl> timerRef(timer);
RemoveTimerInternal(timer);
timer = nullptr;
MOZ_LOG(GetTimerLog(), LogLevel::Debug,
("Timer thread woke up %fms from when it was supposed to\n",
fabs((now - timerRef->mTimeout).ToMilliseconds())));
// We are going to let the call to PostTimerEvent here handle the
// release of the timer so that we don't end up releasing the timer
// on the TimerThread instead of on the thread it targets.
timerRef = PostTimerEvent(timerRef.forget());
if (timerRef) {
// We got our reference back due to an error.
// Unhook the nsRefPtr, and release manually so we can get the
// refcount.
nsrefcnt rc = timerRef.forget().take()->Release();
(void)rc;
// The nsITimer interface requires that its users keep a reference
// to the timers they use while those timers are initialized but
// have not yet fired. If this ever happens, it is a bug in the
// code that created and used the timer.
//
// Further, note that this should never happen even with a
// misbehaving user, because nsTimerImpl::Release checks for a
// refcount of 1 with an armed timer (a timer whose only reference
// is from the timer thread) and when it hits this will remove the
// timer from the timer thread and thus destroy the last reference,
// preventing this situation from occurring.
MOZ_ASSERT(rc != 0, "destroyed timer off its target thread!");
}
if (mShutdown) {
break;
}
// Update now, as PostTimerEvent plus the locking may have taken a
// tick or two, and we may goto next below.
now = TimeStamp::Now();
}
}
if (!mTimers.IsEmpty()) {
timer = mTimers[0];
TimeStamp timeout = timer->mTimeout;
// Don't wait at all (even for PR_INTERVAL_NO_WAIT) if the next timer
// is due now or overdue.
//
// Note that we can only sleep for integer values of a certain
// resolution. We use halfMicrosecondsIntervalResolution, calculated
// before, to do the optimal rounding (i.e., of how to decide what
// interval is so small we should not wait at all).
double microseconds = (timeout - now).ToMilliseconds() * 1000;
if (ChaosMode::isActive(ChaosFeature::TimerScheduling)) {
// The mean value of sFractions must be 1 to ensure that
// the average of a long sequence of timeouts converges to the
// actual sum of their times.
static const float sFractions[] = {
0.0f, 0.25f, 0.5f, 0.75f, 1.0f, 1.75f, 2.75f
};
microseconds *=
sFractions[ChaosMode::randomUint32LessThan(ArrayLength(sFractions))];
forceRunNextTimer = true;
}
if (microseconds < halfMicrosecondsIntervalResolution) {
forceRunNextTimer = false;
goto next; // round down; execute event now
}
waitFor = PR_MicrosecondsToInterval(
static_cast<uint32_t>(microseconds)); // Floor is accurate enough.
if (waitFor == 0) {
waitFor = 1; // round up, wait the minimum time we can wait
}
}
if (MOZ_LOG_TEST(GetTimerLog(), LogLevel::Debug)) {
if (waitFor == PR_INTERVAL_NO_TIMEOUT)
MOZ_LOG(GetTimerLog(), LogLevel::Debug,
("waiting for PR_INTERVAL_NO_TIMEOUT\n"));
else
MOZ_LOG(GetTimerLog(), LogLevel::Debug,
("waiting for %u\n", PR_IntervalToMilliseconds(waitFor)));
}
}
mWaiting = true;
mNotified = false;
mMonitor.Wait(waitFor);
if (mNotified) {
forceRunNextTimer = false;
}
mWaiting = false;
}
return NS_OK;
}
nsresult
TimerThread::AddTimer(nsTimerImpl* aTimer)
{
MonitorAutoLock lock(mMonitor);
// Add the timer to our list.
int32_t i = AddTimerInternal(aTimer);
if (i < 0) {
return NS_ERROR_OUT_OF_MEMORY;
}
// Awaken the timer thread.
if (mWaiting && i == 0) {
mNotified = true;
mMonitor.Notify();
}
return NS_OK;
}
nsresult
TimerThread::TimerDelayChanged(nsTimerImpl* aTimer)
{
MonitorAutoLock lock(mMonitor);
// Our caller has a strong ref to aTimer, so it can't go away here under
// ReleaseTimerInternal.
RemoveTimerInternal(aTimer);
int32_t i = AddTimerInternal(aTimer);
if (i < 0) {
return NS_ERROR_OUT_OF_MEMORY;
}
// Awaken the timer thread.
if (mWaiting && i == 0) {
mNotified = true;
mMonitor.Notify();
}
return NS_OK;
}
nsresult
TimerThread::RemoveTimer(nsTimerImpl* aTimer)
{
MonitorAutoLock lock(mMonitor);
// Remove the timer from our array. Tell callers that aTimer was not found
// by returning NS_ERROR_NOT_AVAILABLE. Unlike the TimerDelayChanged case
// immediately above, our caller may be passing a (now-)weak ref in via the
// aTimer param, specifically when nsTimerImpl::Release loses a race with
// TimerThread::Run, must wait for the mMonitor auto-lock here, and during the
// wait Run drops the only remaining ref to aTimer via RemoveTimerInternal.
if (!RemoveTimerInternal(aTimer)) {
return NS_ERROR_NOT_AVAILABLE;
}
// Awaken the timer thread.
if (mWaiting) {
mNotified = true;
mMonitor.Notify();
}
return NS_OK;
}
// This function must be called from within a lock
int32_t
TimerThread::AddTimerInternal(nsTimerImpl* aTimer)
{
mMonitor.AssertCurrentThreadOwns();
if (mShutdown) {
return -1;
}
TimeStamp now = TimeStamp::Now();
TimerAdditionComparator c(now, aTimer);
nsTimerImpl** insertSlot = mTimers.InsertElementSorted(aTimer, c);
if (!insertSlot) {
return -1;
}
aTimer->mArmed = true;
NS_ADDREF(aTimer);
#ifdef MOZ_TASK_TRACER
// Caller of AddTimer is the parent task of its timer event, so we store the
// TraceInfo here for later used.
aTimer->GetTLSTraceInfo();
#endif
return insertSlot - mTimers.Elements();
}
bool
TimerThread::RemoveTimerInternal(nsTimerImpl* aTimer)
{
mMonitor.AssertCurrentThreadOwns();
if (!mTimers.RemoveElement(aTimer)) {
return false;
}
ReleaseTimerInternal(aTimer);
return true;
}
void
TimerThread::ReleaseTimerInternal(nsTimerImpl* aTimer)
{
if (!mShutdown) {
// copied to a local array before releasing in shutdown
mMonitor.AssertCurrentThreadOwns();
}
// Order is crucial here -- see nsTimerImpl::Release.
aTimer->mArmed = false;
NS_RELEASE(aTimer);
}
already_AddRefed<nsTimerImpl>
TimerThread::PostTimerEvent(already_AddRefed<nsTimerImpl> aTimerRef)
{
mMonitor.AssertCurrentThreadOwns();
RefPtr<nsTimerImpl> timer(aTimerRef);
if (!timer->mEventTarget) {
NS_ERROR("Attempt to post timer event to NULL event target");
return timer.forget();
}
// XXX we may want to reuse this nsTimerEvent in the case of repeating timers.
// Since we already addref'd 'timer', we don't need to addref here.
// We will release either in ~nsTimerEvent(), or pass the reference back to
// the caller. We need to copy the generation number from this timer into the
// event, so we can avoid firing a timer that was re-initialized after being
// canceled.
RefPtr<nsTimerEvent> event = new nsTimerEvent;
if (!event) {
return timer.forget();
}
if (MOZ_LOG_TEST(GetTimerLog(), LogLevel::Debug)) {
event->mInitTime = TimeStamp::Now();
}
// If this is a repeating precise timer, we need to calculate the time for
// the next timer to fire before we make the callback. But don't re-arm.
if (timer->IsRepeatingPrecisely()) {
timer->SetDelayInternal(timer->mDelay);
}
#ifdef MOZ_TASK_TRACER
// During the dispatch of TimerEvent, we overwrite the current TraceInfo
// partially with the info saved in timer earlier, and restore it back by
// AutoSaveCurTraceInfo.
AutoSaveCurTraceInfo saveCurTraceInfo;
(timer->GetTracedTask()).SetTLSTraceInfo();
#endif
nsIEventTarget* target = timer->mEventTarget;
event->SetTimer(timer.forget());
nsresult rv;
{
// We release mMonitor around the Dispatch because if this timer is targeted
// at the TimerThread we'll deadlock.
MonitorAutoUnlock unlock(mMonitor);
rv = target->Dispatch(event, NS_DISPATCH_NORMAL);
}
if (NS_FAILED(rv)) {
timer = event->ForgetTimer();
RemoveTimerInternal(timer);
return timer.forget();
}
return nullptr;
}
void
TimerThread::DoBeforeSleep()
{
// Mainthread
MonitorAutoLock lock(mMonitor);
mSleeping = true;
}
// Note: wake may be notified without preceding sleep notification
void
TimerThread::DoAfterSleep()
{
// Mainthread
MonitorAutoLock lock(mMonitor);
mSleeping = false;
// Wake up the timer thread to re-process the array to ensure the sleep delay is correct,
// and fire any expired timers (perhaps quite a few)
mNotified = true;
mMonitor.Notify();
}
NS_IMETHODIMP
TimerThread::Observe(nsISupports* /* aSubject */, const char* aTopic,
const char16_t* /* aData */)
{
if (strcmp(aTopic, "sleep_notification") == 0 ||
strcmp(aTopic, "suspend_process_notification") == 0) {
DoBeforeSleep();
} else if (strcmp(aTopic, "wake_notification") == 0 ||
strcmp(aTopic, "resume_process_notification") == 0) {
DoAfterSleep();
}
return NS_OK;
}