1548941 - Stop allowing e10s to be turned off with a pref for user-facing desktop builds

Status: RESOLVED FIXED

(Firefox :: General, defect)

Description

[:Gijs (he/him)]

Right now an about:config pref-flip is all it takes to turn off e10s (and, by implication, sandboxing). It shouldn't be that easy.

Ideally, I'd like to get rid of the pref entirely, but it's used by most of our no-e10s/1proc automation suites to disable e10s. They can be switched to an env var, for which we already have code, but I don't know the suites' python code well and we have approx. no time before soft freeze, and somehow I expect nobody will thank me for messing with most of our test suites in the soft freeze week.

Instead, for 68, it seems to me like we could make nsAppRunner read the pref only if at least one of the following holds:

• we're not MOZ_BUILD_APP_IS_BROWSER (hi, fennec)
• we're running in automation
• we're a non-official build (hi, local developers who want to easily turn off e10s.

The only downside I can see is that a profile you share between nightly and a local build that has the pref set is going to behave differently because of MOZ_OFFICIAL. I don't see a great way around that, but also, given the profile separation work that's happened I don't think that's that important a usecase, and about:support will give the right info. We can announce this has happened in the newsgroup, and those factors together I think provide enough aircover to make this change.

On 69, we can follow this up by fully removing support for the pref except on fennec, transitioning the test suites to the env var.

Comment 1

[:Gijs (he/him)]

Attachment: Bug 1548941 - remove e10s force-enable and force-disable prefs, and on desktop restrict 'normal' e10s pref to automation and unofficial builds, r=bholley,ahal

Comment 2

[:Gijs (he/him)]

Posted https://groups.google.com/d/msg/mozilla.dev.platform/cJMzxi7_PmI/R8Qck-dcAQAJ to make sure people aren't caught unawares.

Comment 3

[:Gijs (he/him)]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=4dea476efe9f917fbdc75e117971d182c3834a9a

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/5d97da3c65b5
remove e10s force-enable and force-disable prefs, and on desktop restrict 'normal' e10s pref to automation and unofficial builds, r=bholley

Comment 5

[Dorel Luca [:dluca]]

Backed out changeset 5d97da3c65b5 (bug 1548941) for Valgrind failure

Log:
https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=244979607&repo=autoland&lineNumber=40399

Push with failures:
https://treeherder.mozilla.org/#/jobs?repo=autoland&revision=5d97da3c65b5376c9bc6df4add1646fc63937693

Backout:
https://hg.mozilla.org/integration/autoland/rev/4574274db3f95575dca7da08e16d6f20242fcb2a

Comment 6

[:Gijs (he/him)]

I have no idea how I broke valgrind. Log output:

[task 2019-05-06T22:39:57.072Z] 22:39:57     INFO -   7:56.59 TEST-UNEXPECTED-FAIL | valgrind-test | Syscall param sendmsg(msg.msg_iov[0]) points to uninitialised byte(s) at ??? / IPC::Channel::ChannelImpl::ProcessOutgoingMessages / IPC::Channel::ChannelImpl::Send / applyImpl
[task 2019-05-06T22:39:57.072Z] 22:39:57     INFO -   7:56.59 ==3274== Syscall param sendmsg(msg.msg_iov[0]) points to uninitialised byte(s)
[task 2019-05-06T22:39:57.072Z] 22:39:57     INFO -   7:56.59 ==3274==    at 0x4E4372D: ??? (syscall-template.S:82)
[task 2019-05-06T22:39:57.072Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF2A2963: IPC::Channel::ChannelImpl::ProcessOutgoingMessages()+835 (ipc/chromium/src/chrome/common/ipc_channel_posix.cc:689)
[task 2019-05-06T22:39:57.072Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF2A3CBA: IPC::Channel::ChannelImpl::Send(IPC::Message*)+250 (ipc/chromium/src/chrome/common/ipc_channel_posix.cc:796)
[task 2019-05-06T22:39:57.072Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF2D7CBA: applyImpl<IPC::Channel, bool (IPC::Channel::*)(IPC::Message *), StorePtrPassByPtr<IPC::Message> , 0> (dist/include/nsThreadUtils.h:1122)
[task 2019-05-06T22:39:57.072Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF2D7CBA: apply<IPC::Channel, bool (IPC::Channel::*)(IPC::Message *)> (dist/include/nsThreadUtils.h:1128)
[task 2019-05-06T22:39:57.072Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF2D7CBA: mozilla::detail::RunnableMethodImpl<IPC::Channel*, bool (IPC::Channel::*)(IPC::Message*), false, (mozilla::RunnableKind)0, IPC::Message*>::Run()+42 (dist/include/nsThreadUtils.h:1174)
[task 2019-05-06T22:39:57.072Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF296E52: RunTask (ipc/chromium/src/base/message_loop.cc:442)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF296E52: DeferOrRunPendingTask (ipc/chromium/src/base/message_loop.cc:450)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF296E52: MessageLoop::DoWork()+834 (ipc/chromium/src/base/message_loop.cc:523)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF297D68: base::MessagePumpLibevent::Run(base::MessagePump::Delegate*)+280 (ipc/chromium/src/base/message_pump_libevent.cc:321)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF296485: RunInternal (ipc/chromium/src/base/message_loop.cc:315)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF296485: RunHandler (ipc/chromium/src/base/message_loop.cc:308)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF296485: MessageLoop::Run()+85 (ipc/chromium/src/base/message_loop.cc:290)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF29EFE8: base::Thread::ThreadMain()+440 (ipc/chromium/src/base/thread.cc:192)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF29BB29: ThreadFunc(void*)+9 (ipc/chromium/src/base/platform_thread_posix.cc:40)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0x4E3BB4F: start_thread+207 (pthread_create.c:304)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0x57BCFBC: clone+108 (clone.S:112)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==  Address 0x19810765 is 1,413 bytes inside a block of size 8,192 alloc'd
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    at 0x4C2B240: malloc+112 (vg_replace_malloc.c:298)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0x11076D: moz_xmalloc+13 (memory/mozalloc/mozalloc.cpp:68)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF29D370: pod_malloc<char> (dist/include/mozilla/mozalloc.h:210)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.59 ==3274==    by 0xF29D370: mozilla::BufferList<InfallibleAllocPolicy>::AllocateSegment(unsigned long, unsigned long)+32 (dist/include/mozilla/BufferList.h:383)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF2980C7: BufferList (dist/include/mozilla/BufferList.h:87)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF2980C7: Pickle::Pickle(unsigned int, unsigned long)+103 (ipc/chromium/src/base/pickle.cc:130)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF2A41CC: IPC::Message::Message(int, unsigned int, unsigned int, IPC::Message::HeaderFlags, bool)+44 (ipc/chromium/src/chrome/common/ipc_message.cc:57)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF429111: Msg_Update (PLayerTransaction.cpp:31)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF429111: mozilla::layers::PLayerTransactionChild::SendUpdate(mozilla::layers::TransactionInfo const&)+81 (PLayerTransactionChild.cpp:52)
[task 2019-05-06T22:39:57.073Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xFB1980C: mozilla::layers::ShadowLayerForwarder::EndTransaction(mozilla::gfx::IntRegionTyped<mozilla::gfx::UnknownUnits> const&, mozilla::layers::BaseTransactionId<mozilla::layers::TransactionIdType>, bool, unsigned int, bool, mozilla::layers::BaseTransactionId<mozilla::VsyncIdType> const&, mozilla::TimeStamp const&, mozilla::TimeStamp const&, mozilla::TimeStamp const&, bool, nsTString<char> const&, bool*, nsTArray<mozilla::layers::CompositionPayload> const&)+6044 (gfx/layers/ipc/ShadowLayers.cpp:710)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xFA9F272: mozilla::layers::ClientLayerManager::ForwardTransaction(bool)+626 (gfx/layers/client/ClientLayerManager.cpp:705)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xFA9EF77: mozilla::layers::ClientLayerManager::EndTransaction(void (*)(mozilla::layers::PaintedLayer*, gfxContext*, mozilla::gfx::IntRegionTyped<mozilla::gfx::UnknownUnits> const&, mozilla::gfx::IntRegionTyped<mozilla::gfx::UnknownUnits> const&, mozilla::layers::DrawRegionClip, mozilla::gfx::IntRegionTyped<mozilla::gfx::UnknownUnits> const&, void*), void*, mozilla::layers::LayerManager::EndTransactionFlags)+183 (gfx/layers/client/ClientLayerManager.cpp:385)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1180F4C8: nsDisplayList::PaintRoot(nsDisplayListBuilder*, gfxContext*, unsigned int)+1704 (layout/painting/nsDisplayList.cpp:2878)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1158ABE6: nsLayoutUtils::PaintFrame(gfxContext*, nsIFrame*, nsRegion const&, unsigned int, nsDisplayListBuilderMode, nsLayoutUtils::PaintFrameFlags)+9142 (layout/base/nsLayoutUtils.cpp:3994)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1153A011: mozilla::PresShell::Paint(nsView*, nsRegion const&, mozilla::PaintFlags)+1265 (layout/base/PresShell.cpp:6074)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1133EC13: nsViewManager::ProcessPendingUpdatesPaint(nsIWidget*)+435 (view/nsViewManager.cpp:462)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1133E8D4: nsViewManager::ProcessPendingUpdatesForView(nsView*, bool)+532 (view/nsViewManager.cpp:397)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1133D3FD: ProcessPendingUpdates (view/nsViewManager.cpp:1020)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1133D3FD: nsViewManager::WillPaintWindow(nsIWidget*)+221 (view/nsViewManager.cpp:667)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1133D2DA: nsView::WillPaintWindow(nsIWidget*)+26 (view/nsView.cpp:990)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1135DB67: mozilla::widget::PuppetWidget::Paint()+439 (widget/PuppetWidget.cpp:996)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x110C796E: mozilla::dom::BrowserChild::RecvRenderLayers(bool const&, bool const&, mozilla::layers::LayersObserverEpoch const&)+814 (dom/ipc/BrowserChild.cpp:2496)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF58F165: mozilla::dom::PBrowserChild::OnMessageReceived(IPC::Message const&)+2821 (PBrowserChild.cpp:5311)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF362B7F: mozilla::dom::PContentChild::OnMessageReceived(IPC::Message const&)+927 (PContentChild.cpp:6510)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF2CEEC1: mozilla::ipc::MessageChannel::DispatchAsyncMessage(IPC::Message const&)+97 (ipc/glue/MessageChannel.cpp:2151)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF2CE0C1: mozilla::ipc::MessageChannel::DispatchMessage(IPC::Message&&)+289 (ipc/glue/MessageChannel.cpp:2078)
[task 2019-05-06T22:39:57.074Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF2CE7FE: mozilla::ipc::MessageChannel::RunMessage(mozilla::ipc::MessageChannel::MessageTask&)+446 (ipc/glue/MessageChannel.cpp:1937)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF2CEA7D: mozilla::ipc::MessageChannel::MessageTask::Run()+77 (ipc/glue/MessageChannel.cpp:1968)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEDF485D: mozilla::SchedulerGroup::Runnable::Run()+45 (xpcom/threads/SchedulerGroup.cpp:295)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEE03DCC: nsThread::ProcessNextEvent(bool, bool*)+1596 (xpcom/threads/nsThread.cpp:1180)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEE05E17: NS_ProcessNextEvent(nsIThread*, bool)+71 (xpcom/threads/nsThreadUtils.cpp:486)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF2D1099: mozilla::ipc::MessagePump::Run(base::MessagePump::Delegate*)+217 (ipc/glue/MessagePump.cpp:88)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: RunInternal (ipc/chromium/src/base/message_loop.cc:315)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: RunHandler (ipc/chromium/src/base/message_loop.cc:308)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: MessageLoop::Run()+85 (ipc/chromium/src/base/message_loop.cc:290)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x11369F88: nsBaseAppShell::Run()+40 (widget/nsBaseAppShell.cpp:137)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1252D173: XRE_RunAppShell()+67 (toolkit/xre/nsEmbedFunctions.cpp:919)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: RunInternal (ipc/chromium/src/base/message_loop.cc:315)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: RunHandler (ipc/chromium/src/base/message_loop.cc:308)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: MessageLoop::Run()+85 (ipc/chromium/src/base/message_loop.cc:290)
[task 2019-05-06T22:39:57.075Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1252CFF7: XRE_InitChildProcess(int, char**, XREChildData const*)+1031 (toolkit/xre/nsEmbedFunctions.cpp:757)
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x10FBB7: content_process_main (browser/app/../../ipc/contentproc/plugin-container.cpp:56)
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x10FBB7: main+423 (browser/app/nsBrowserApp.cpp:263)
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60 ==3274==
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60 {
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    <insert_a_suppression_name_here>
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    Memcheck:Param
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    sendmsg(msg.msg_iov[0])
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    obj:/lib/x86_64-linux-gnu/libpthread-2.13.so
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    fun:_ZN3IPC7Channel11ChannelImpl23ProcessOutgoingMessagesEv
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    fun:_ZN3IPC7Channel11ChannelImpl4SendEPNS_7MessageE
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    fun:applyImpl<IPC::Channel, bool (IPC::Channel::*)(IPC::Message *), StorePtrPassByPtr<IPC::Message> , 0>
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    fun:apply<IPC::Channel, bool (IPC::Channel::*)(IPC::Message *)>
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    fun:_ZN7mozilla6detail18RunnableMethodImplIPN3IPC7ChannelEMS3_FbPNS2_7MessageEELb0ELNS_12RunnableKindE0EJS6_EE3RunEv
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    fun:RunTask
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    fun:DeferOrRunPendingTask
[task 2019-05-06T22:39:57.076Z] 22:39:57     INFO -   7:56.60    fun:_ZN11MessageLoop6DoWorkEv
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60    fun:_ZN4base19MessagePumpLibevent3RunEPNS_11MessagePump8DelegateE
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60    fun:RunInternal
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60    fun:RunHandler
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60    fun:_ZN11MessageLoop3RunEv
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60    fun:_ZN4base6Thread10ThreadMainEv
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60    fun:_ZL10ThreadFuncPv
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60    fun:start_thread
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60    fun:clone
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 }
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 ==3274==
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 ==3274== 1 errors in context 2 of 302:
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 ==3274== Thread 1 Privileged Conte:
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 TEST-UNEXPECTED-FAIL | valgrind-test | Conditional jump or move depends on uninitialised value(s) at std::sync::once::Once::call_once:: / std::sync::once::Once::call_inner / Servo_StyleSheet_FromUTF8BytesAsync / mozilla::StyleSheet::ParseSheet
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 ==3274== Conditional jump or move depends on uninitialised value(s)
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 ==3274==    at 0x136AFA16: std::sync::once::Once::call_once::{{closure}}+662 (ptr.rs:193)
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x13743B69: std::sync::once::Once::call_inner+265 (once.rs:387)
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x133D7137: Servo_StyleSheet_FromUTF8BytesAsync+423 (once.rs:220)
[task 2019-05-06T22:39:57.077Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x114DE73A: mozilla::StyleSheet::ParseSheet(mozilla::css::Loader*, nsTSubstring<char> const&, mozilla::css::SheetLoadData*)+346 (layout/style/StyleSheet.cpp:961)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x114CB83A: mozilla::css::Loader::ParseSheet(nsTSubstring<char> const&, mozilla::css::SheetLoadData*, mozilla::css::Loader::AllowAsyncParse)+266 (layout/style/Loader.cpp:1594)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x114DA2F0: mozilla::css::StreamLoader::OnStopRequest(nsIRequest*, nsresult)+416 (layout/style/StreamLoader.cpp:109)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEE89CB3: nsBaseChannel::OnStopRequest(nsIRequest*, nsresult)+115 (netwerk/base/nsBaseChannel.cpp:812)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEE89D7C: non-virtual thunk to nsBaseChannel::OnStopRequest(nsIRequest*, nsresult)+12 (netwerk/base/nsBaseChannel.cpp:0)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEEA14AC: nsInputStreamPump::OnStateStop()+300 (netwerk/base/nsInputStreamPump.cpp:655)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEEA0D26: nsInputStreamPump::OnInputStreamReady(nsIAsyncInputStream*)+326 (netwerk/base/nsInputStreamPump.cpp:403)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEDD9170: nsInputStreamReadyEvent::Run()+32 (xpcom/io/nsStreamUtils.cpp:91)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEDF485D: mozilla::SchedulerGroup::Runnable::Run()+45 (xpcom/threads/SchedulerGroup.cpp:295)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEE03DCC: nsThread::ProcessNextEvent(bool, bool*)+1596 (xpcom/threads/nsThread.cpp:1180)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xEE05E17: NS_ProcessNextEvent(nsIThread*, bool)+71 (xpcom/threads/nsThreadUtils.cpp:486)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF2D1099: mozilla::ipc::MessagePump::Run(base::MessagePump::Delegate*)+217 (ipc/glue/MessagePump.cpp:88)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: RunInternal (ipc/chromium/src/base/message_loop.cc:315)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: RunHandler (ipc/chromium/src/base/message_loop.cc:308)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: MessageLoop::Run()+85 (ipc/chromium/src/base/message_loop.cc:290)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x11369F88: nsBaseAppShell::Run()+40 (widget/nsBaseAppShell.cpp:137)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1252D173: XRE_RunAppShell()+67 (toolkit/xre/nsEmbedFunctions.cpp:919)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: RunInternal (ipc/chromium/src/base/message_loop.cc:315)
[task 2019-05-06T22:39:57.078Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: RunHandler (ipc/chromium/src/base/message_loop.cc:308)
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0xF296485: MessageLoop::Run()+85 (ipc/chromium/src/base/message_loop.cc:290)
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x1252CFF7: XRE_InitChildProcess(int, char**, XREChildData const*)+1031 (toolkit/xre/nsEmbedFunctions.cpp:757)
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x10FBB7: content_process_main (browser/app/../../ipc/contentproc/plugin-container.cpp:56)
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60 ==3274==    by 0x10FBB7: main+423 (browser/app/nsBrowserApp.cpp:263)
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60 ==3274==  Uninitialised value was created by a stack allocation
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60 ==3274==    at 0x136AF78A: std::sync::once::Once::call_once::{{closure}}+10 (once.rs:220)
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60 ==3274==
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60 {
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    <insert_a_suppression_name_here>
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    Memcheck:Cond
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN3std4sync4once4Once9call_once28_$u7b$$u7b$closure$u7d$$u7d$17ha9f40b9218583a55E
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN3std4sync4once4Once10call_inner17hb33e6c6eb70a145cE
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:Servo_StyleSheet_FromUTF8BytesAsync
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN7mozilla10StyleSheet10ParseSheetEPNS_3css6LoaderERK12nsTSubstringIcEPNS1_13SheetLoadDataE
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN7mozilla3css6Loader10ParseSheetERK12nsTSubstringIcEPNS0_13SheetLoadDataENS1_15AllowAsyncParseE
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN7mozilla3css12StreamLoader13OnStopRequestEP10nsIRequest8nsresult
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN13nsBaseChannel13OnStopRequestEP10nsIRequest8nsresult
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZThn128_N13nsBaseChannel13OnStopRequestEP10nsIRequest8nsresult
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN17nsInputStreamPump11OnStateStopEv
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN17nsInputStreamPump18OnInputStreamReadyEP19nsIAsyncInputStream
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN23nsInputStreamReadyEvent3RunEv
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN7mozilla14SchedulerGroup8Runnable3RunEv
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_ZN8nsThread16ProcessNextEventEbPb
[task 2019-05-06T22:39:57.079Z] 22:39:57     INFO -   7:56.60    fun:_Z19NS_ProcessNextEventP9nsIThreadb
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:_ZN7mozilla3ipc11MessagePump3RunEPN4base11MessagePump8DelegateE
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:RunInternal
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:RunHandler
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:_ZN11MessageLoop3RunEv
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:_ZN14nsBaseAppShell3RunEv
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:_Z15XRE_RunAppShellv
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:RunInternal
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:RunHandler
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:_ZN11MessageLoop3RunEv
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:_Z20XRE_InitChildProcessiPPcPK12XREChildData
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:content_process_main
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60    fun:main
[task 2019-05-06T22:39:57.080Z] 22:39:57     INFO -   7:56.60 }

AFAICT it's complaining about an uninitialized something, and the something is in stylo or other rust code. So I don't understand how I broke that, as I didn't touch any of it... if valgrind is running with e10s disabled, perhaps that's it, but if so I can find no evidence of such a configuration...

Comment 7

[Bobby Holley (:bholley)]

I seem to remember there being some weirdness / flakiness with valgrind suppressions. If the callstacks changed somehow that could certainly trigger it (because we whitelist stacks in various places), but this patch would have to change that somehow.

Comment 8

[:Gijs (he/him)]

OK, I've figured this out, the issue is that the valgrind test right now runs without e10s. It uses the unittest-required profile preferences through:

https://searchfox.org/mozilla-central/rev/e7d9a8749303b39dadcc0e18ea0d60a570a68145/build/valgrind/mach_commands.py#66-76

which reads profiles.json:

https://searchfox.org/mozilla-central/source/testing/profiles/profiles.json

which has:

    "valgrind": ["base", "common", "unittest-required", "unittest-features"],

and unittest-required has:

https://searchfox.org/mozilla-central/rev/99a2a5a955960b0e58ceade1db1f7652d9db4ba1/testing/profiles/unittest-required/user.js#56

user_pref("browser.tabs.remote.autostart", false);

OK, so why doesn't it still run in non-e10s after my changes? Well, my patch expects that xpc::IsInAutomation returns the right thing here. It doesn't.

IsInAutomation cares about 2 things - the should-be-infamous "let viruses take over my computer" pref (which is set in the "common" user.js file so that's fine) and AreNonLocalConnectionsDisabled(). That looks like this:

  if (disabledForTest == -1) {
    char* s = getenv("MOZ_DISABLE_NONLOCAL_CONNECTIONS");
    if (s) {
      disabledForTest = *s != '0';
    } else {
      disabledForTest = 0;
    }
  }

Looking for MOZ_DISABLE_NONLOCAL_CONNECTIONS shows that a bunch of places are setting this themselves (talos, raptor, wpt, xpcshell, "mozrunner" (used for a variety of tests, I believe), junit, and various C++ unittests). This in addition to build/automation.py.in, which naively I would have expected has some kind of shared routine for this (which it does, it just doesn't seem to be used everywhere).

Anyway, suffice it to say, it's not set for valgrind. Fixing that fixes the failures:

https://treeherder.mozilla.org/#/jobs?repo=try&revision=2a63d5538d003a26207f5ee10ba7cbc7dc572200

Of course, longer-term, we should switch valgrind to use e10s, and we should probably look at how these env vars are set. I'll file follow-ups for that.

Comment 9

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/4e2250bbaed3
remove e10s force-enable and force-disable prefs, and on desktop restrict 'normal' e10s pref to automation and unofficial builds, r=bholley,ahal

Comment 10

[Andreea Pavel [:apavel]]

Backed out for failing test_gpuprocess.js

Push with failures: https://treeherder.mozilla.org/#/jobs?repo=autoland&resultStatus=testfailed%2Cbusted%2Cexception&revision=4e2250bbaed37b7b0a9a2c4b66c0c5455a1551a6&selectedJob=245390286

Failure log: https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=245390286&repo=autoland&lineNumber=13468
Also broke gtest: https://treeherder.mozilla.org/#/jobs?repo=autoland&resultStatus=testfailed%2Cbusted%2Cexception&revision=4e2250bbaed37b7b0a9a2c4b66c0c5455a1551a6&selectedJob=245399312

https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=245399312&repo=autoland&lineNumber=859

Backout: https://hg.mozilla.org/integration/autoland/rev/0313ad8713b5d1ad045af938ae11bef5f7f5f7ca

Comment 11

[:Gijs (he/him)]

This would all be funny if it wasn't so tragic (and a little annoying, at this point).

OK... gtest is easy, it's basically the same as comment #8, it needs MOZ_DISABLE_NONLOCAL_CONNECTIONS setting (and then, again, perhaps a conversation about whether that test should run with e10s turned on, given that it's hitting assertions if not, at a naive level that seems Not Good - but not my problem right now).

xpcshell is more fun. https://searchfox.org/mozilla-central/source/testing/profiles/profiles.json indicates every test framework loads https://searchfox.org/mozilla-central/source/testing/profiles/common/user.js#49 . Every test framework? No, one test framework is bravely holding out... etc. etc.

Right, so xpcshell doesn't set the let_viruses_take_over pref, so isInAutomation returns false, and as a result we ignore xpcshell's attempts to turn off e10s with the usual pref, and then the win7 GPU process infrastructure gets very upset on debug for... reasons (that, to be frank, I haven't fully figured out yet, but ISTM we shouldn't be changing the larger behaviour here so I don't think it's super relevant atm).

I'm not 100% sure what to do about this. Some options:

1. set this pref for all of xpcshell and turn it off for tests that need xpconnect to behave the "normal" way.
2. set this pref for the test_gpuprocess.js (if possible; we can't set prefs in manifests for xpcshell, and I expect setting it at runtime from the test might be too late for it to have the desired effect)
3. fix whatever is making the test fail on win7 debug only when run on infra with e10s on (perhaps turning it off on win7 debug temporarily to move that work to a follow-up bug).
4. switch the main patch here to use something other than xpc::IsInAutomation() to determine whether we're in automation (perhaps only the AreNonLocalConnectionsDisabled() bit?)

Bobby, seeing as you know about almost all the components here (gfx, xpconnect, this patch / e10s longer-term plans), do you have a preference? I'm leaning towards either (3) or (4) but could be convinced to go for any of these...

Comment 12

[:Gijs (he/him)]

Hm, it also seems that gtest is still unhappy after setting MOZ_DISABLE_NONLOCAL_CONNECTIONS, so I'll have to debug that some more... I expect we don't set the pref there either, as I imagine there's no profile.

Comment 13

[:Gijs (he/him)]

OK, based on the fact that there's no prefs in gtest (beyond app defaults), gonna just go for the AreNonLocalConnectionsDisabled() check, even if it's a bit ugly.

Comment 14

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/d62e8172a66f
remove e10s force-enable and force-disable prefs, and on desktop restrict 'normal' e10s pref to automation and unofficial builds, r=bholley,ahal

Comment 15

[Narcis Beleuzu [:NarcisB]]

https://hg.mozilla.org/mozilla-central/rev/d62e8172a66f

Comment 16

[KOLANICH]

Right now an about:config pref-flip is all it takes to turn off e10s (and, by implication, sandboxing). It shouldn't be that easy.

Of course it should. IMHO your e10s quality is damn bad. To avoid serious performance problems, like lags when switching tabs (on desktop it used to be drawing a spinning wheel for some seconds, making the browser completely unusable when e10s is enabled) I always disable e10s on all my devices.

Ideally, I'd like to get rid of the pref entirely

Ideally I want Mozilla to rewrite everything in rust and stop trying to mitigate issues of processes crashing or being exploited by using multiprocessing. These hypothetical security advantages don't worth anything when the browser is completely unusable. If the browser is unusable I have to either to make it usable or just switch to another browser. Unfortunately there is only one "another browser" (there are no other browsers except FF and Chromium and their forks (Anything Goanna-based is not counted), everything else marketed as browseers in fact are subbrowsers - they have poor support of modern standards and poor privacy and security) called Chromium and it is pig shit in the sense of privacy, security, customizability and addons API.

Comment 17

[[e-mail]]

I can't believe you've forced e10s with bugs like this still open after 5 years https://bugzilla.mozilla.org/show_bug.cgi?id=1118086

I'm am now having to install Chrome for all my staff as Firefox is unusable.

Comment 18

[mirh]

So.. I don't want to go all over the place like the other 2 gentlemen in here, I already understood your legit points in bug 1399962 and bug 1520227
And I'm sure having no easy way to revert e10s is for the better (there's no reason every bug couldn't be fixed then)

But.. uh, when did about:config become not difficult enough? You already have a big scary "here to be dragons" warning.
And are you you really going to kill single-process entirely?
I see that speculation attacks, possible future speculation attacks, are a big risk for the average user indeed.

But there are very reasonable situations, where even if these weren't just random hypotheses but certainties, you just wouldn't care.
I for one have a laboratory computer with 0 sensitive data, and the worst an evil ad could know is my glibc version. Probably same story even for an information kiosk scenario, perhaps on a lame rpi.
(both of them having an abysmal amount of RAM, I mean, if the only as usual intrinsic limitation of multiprocess wasn't clear)

I understand you are doing magic with MemShrink, and that no automated testing will come on your side anymore (funnily, a destiny linux32 builds will sooner or later follow), but I don't think that important 1.3x memory usage could be ever be definitively fixed.

Comment 19

[[e-mail]]

Hiding my comment for advocacy is juvenile.

Comment 20

[:Gijs (he/him)]

Bugzilla is a bugtracker, not a discussion forum, and this isn't the right place to try to rehash this decision, so I'm going to restrict comments.

There was a newsgroup thread (see mozilla-dev-platform, subject "Lack of browser mochitests in non-e10s configuration and support for turning off e10s on desktop going forward", April of this year) prior to this change, and broad consensus that no, we do not intend to support single-process mode for desktop going forward, except for some local debugging needs. The security and performance impact to users is such that it would be counter to our values to do so. That's less about speculation attacks and more about sandboxing web content to ensure there are several layers of security between malicious web content and your machine. It wasn't always necessary for users to use about:config to flip the pref in order for that to happen, hence the desire to restrict the use of the pref so people aren't grandfathered into an insecure configuration without realizing.

Yes, memory usage is slightly higher than in single process mode in some cases. We're constantly working to bring that overhead down even further than we already have, and there are responsiveness benefits in turning on multiprocess as well, so it's not some kind of complete performance loss. If there are specific bugs when using multiple processes that you feel aren't getting the right attention then please ping the triage owner of the relevant component in that bug, not here. On the whole, we're confident about the trade-offs here and I don't see us revisiting this decision.