enabling the toplevel.flow category causes crashes

[chiniforooshan]

Steps to reproduce:

• Run ToT chrome (50.0.2626.0 64-bit) on Linux.
• Go to chrome://tracing and record a trace with "toplevel.flow" category enabled.
• Open a new tab and navigate to an address. It will crash in the first 1-2 seconds.

Note: I couldn't reproduce this on stable (48.0.2564.97).

Stack Trace: Received signal 11 SEGV_MAPERR 000000000000

0 0x7f614ff5165b base::debug::(anonymous namespace)::StackDumpSignalHandler()

1 0x7f6147eef330

2 0x7f614ff9920f base::ThreadTaskRunnerHandle::Get()

3 0x7f614ffb11cc base::trace_event::TraceLog::ThreadLocalEventBuffer::ThreadLocalEventBuffer()

4 0x7f614ffb1f3e base::trace_event::TraceLog::InitializeThreadLocalEventBufferIfSupported()

5 0x7f614ffb4db0 base::trace_event::TraceLog::AddTraceEventWithThreadIdAndTimestamp()

6 0x7f614ff52649 base::debug::TaskAnnotator::DidQueueTask()

7 0x7f614ffc39e0 base::internal::IncomingTaskQueue::PostPendingTask()

8 0x7f614ffc3965 base::internal::IncomingTaskQueue::AddToIncomingQueue()

9 0x7f61502ada41 mojo::edk::MessagePipeDispatcher::RequestNontransferableChannel()

10 0x7f61502af417 mojo::edk::MessagePipeDispatcher::WriteMessageImplNoLock()

11 0x7f61502ac1cc mojo::edk::Dispatcher::WriteMessage()

12 0x7f61502a850e mojo::edk::Core::WriteMessage()

13 0x7f614d87d64e MojoWriteMessage

14 0x7f614d42bd13 mojo::common::MessagePumpMojo::SignalControlPipe()

15 0x7f614ffc3b80 base::internal::IncomingTaskQueue::StartScheduling()

16 0x7f614ff68483 base::MessageLoop::BindToCurrentThread()

17 0x7f614ff9d247 base::Thread::ThreadMain()

18 0x7f614ff9997d base::(anonymous namespace)::ThreadFunc()

19 0x7f6147ee7184 start_thread

20 0x7f61450d334d clone

r8: 7fffffffffffffff r9: 0000002f6dc29f60 r10: 0000000000000001 r11: 0000000000000293 r12: 00007f613a6e50f8 r13: 00000ad16dc7ca80 r14: 00000ad16dbe9500 r15: 00000ad16de60008 di: 0000000000000009 si: 0000000000000000 bp: 0000000000000000 bx: 00007f613a6e50f8 dx: 00007f613a6e6aa0 ax: 0000000000000000 cx: 00000ad16de60000 sp: 00007f613a6e50d0 ip: 00007f614ff9920f efl: 0000000000010246 cgf: 0000000000000033 erf: 0000000000000004 trp: 000000000000000e msk: 0000000000000000 cr2: 0000000000000000 [end of stack trace]

[chiniforooshan]

@skyostil

[chiniforooshan]

It looks like migration to ports EDK has fixed this. I rebased my repo to after that patch and the crash is gone.

[chiniforooshan]

Probably, the issue was caused by this EDK change, back in December, that introduces new PostTasks that in turn generate trace events, apparently too early, before the thread has a message loop(?).

Closing this since it's been resolved.