331012 - topcrash on exit [@ TimerThread::UpdateFilter]

Status: VERIFIED FIXED

(Core :: XPCOM, defect)

Description

[Jesse Ruderman]

Talkback says 12.4% of trunk crashes are at [@ TimerThread::UpdateFilter].  Most comments for this crash indicate that it happened while exiting Firefox.  This appears to be happening on all three major platforms (Win, Mac, Linux).

Example stack:

TimerThread::UpdateFilter()  [xpcom/threads/TimerThread.cpp, line 197]
TimerThread::UpdateFilter()  [xpcom/threads/TimerThread.cpp, line 183]
handleTimerEvent()  [xpcom/threads/nsTimerImpl.cpp, line 470]
PL_HandleEvent()  [xpcom/threads/plevent.c, line 689]
PL_ProcessPendingEvents()  [xpcom/threads/plevent.c, line 623]
nsEventQueueImpl::ProcessPendingEvents()  [xpcom/threads/nsEventQueue.cpp, line 423]
NS_ShutdownXPCOM_P()  [xpcom/build/nsXPComInit.cpp, line 830]
ScopedXPCOMStartup::~ScopedXPCOMStartup()  [toolkit/xre/nsAppRunner.cpp, line 557]
XRE_main()  [toolkit/xre/nsAppRunner.cpp, line 2403]
_start()
start()

Comment 1

[Benjamin Smedberg]

Ugh. I changed the shutdown sequence recently so that the timer thread is joined earlier in the shutdown sequence: I wonder if we're processing timer events after the timerthread expects. The line number in nsXPComInit.cpp is bogus, and even the top of the stack is suspect, but I suspect it's crashing here:

http://lxr.mozilla.org/mozilla/source/xpcom/threads/nsTimerImpl.cpp#394

Called through here:

http://lxr.mozilla.org/mozilla/source/xpcom/build/nsXPComInit.cpp#707 or #715

We should probably just null-check gThread there.

I don't suppose there's a way to reproduce this reliably? Since it's timing-related probably not.

Comment 2

[Darin Fisher]

hmm... yeah, nsTimerImpl should protect itself against a shutdown timer thread.

Comment 3

[Benjamin Smedberg]

Attachment: Null-check gThread, rev. 1

Comment 4

[Benjamin Smedberg]

Patch landed on trunk. Jesse, could you please verify that this fixes the topcrash?

Comment 5

[Jesse Ruderman]

Sure, I can look at Talkback again in a few days to see whether the topcrash went away.

Comment 6

[Jesse Ruderman]

I can verify that this is fixed based on Talkback data -- there are no crashes [@ TimerThread::UpdateFilter] recorded for March 22 and later builds.

Comment 7

[Benjamin Smedberg]

Thanks!