1603854 - Intermittent Windows 10 xpcshell crashes in NS_DispatchToMainThread

Status: RESOLVED FIXED

(Core :: Networking, defect, P1)

Description

[Geoff Brown [:gbrown]]

I notice there are a lot of recent xpcshell test failures affecting various tests which involve a crash in NS_DispatchToMainThread.

We seem to be hitting an assertion at

https://searchfox.org/mozilla-central/rev/2f09184ec781a2667feec87499d4b81b32b6c48e/xpcom/threads/nsThreadUtils.cpp#242

  nsCOMPtr<nsIThread> thread;
  nsresult rv = NS_GetMainThread(getter_AddRefs(thread));
  if (NS_WARN_IF(NS_FAILED(rv))) {
    NS_ASSERTION(false,
                 "Failed NS_DispatchToMainThread() in shutdown; leaking");

Comment 1

[Geoff Brown [:gbrown]]

:froydnj - Any interest or insight?

Comment 2

[Geoff Brown [:gbrown]]

It looks like these are all on Windows 10.

https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=281112605&repo=autoland&lineNumber=2098
https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=281096414&repo=autoland&lineNumber=2122
https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=280578062&repo=autoland&lineNumber=2221
https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=281112605&repo=autoland&lineNumber=2098

Comment 3

[Nathan Froyd [:froydnj]]

https://bugzilla.mozilla.org/show_bug.cgi?id=1603840 has the clearest assertion stack, which shows that something is going wrong in nsNotifyAddrListener::NotifyObservers, probably dispatching tasks well after shutdown has completed.

https://bugzilla.mozilla.org/show_bug.cgi?id=1601674 has roughly the same stack.

Other bugs have garbage for stacks, but e.g. https://bugzilla.mozilla.org/show_bug.cgi?id=1592769 has RtlpHpSegPageRangeShrink in the stack, which bug 1603840 does too. I'm going to venture that all of these would be fixed by making nsNotifyAddrListener not dispatch tasks after xpcom-shutdown-threads or thereabouts.

Comment 4

[Geoff Brown [:gbrown]]

Thanks!

Comment 5

[Valentin Gosu [:valentin] (he/him)]

Attachment: Bug 1603854 - nsNotifyAddrListener: Don't call NS_DispatchToMainThread if mShutdown is true r=froydnj

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/2c2dc20f2a21
nsNotifyAddrListener: Don't call NS_DispatchToMainThread if mShutdown is true r=froydnj

Comment 7

[Cristina Coroiu [:ccoroiu]]

https://hg.mozilla.org/mozilla-central/rev/2c2dc20f2a21

Comment 11

[Geoff Brown [:gbrown]]

Thanks Valentin. Your change appears to have stopped the frequent crashes.

However, I noticed one crash that looks very similar and happened well after your fix landed:

https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=281980567&repo=autoland&lineNumber=2069

I haven't seen any others / not sure if it's worth follow-up...just pointing it out.

Comment 12

[Valentin Gosu [:valentin] (he/him)]

(In reply to Geoff Brown [:gbrown] from comment #11)

Thanks Valentin. Your change appears to have stopped the frequent crashes.

However, I noticed one crash that looks very similar and happened well after your fix landed:

https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=281980567&repo=autoland&lineNumber=2069

I haven't seen any others / not sure if it's worth follow-up...just pointing it out.

patched_BaseThreadInitThunk(int, void*, void*) [WindowsDllBlocklist.cpp:23220e6aef9d9290cb4d1b40d435145f979e962c : 582 + 0x

It looks like the blocklist is creating a thread after shutdown?
I think that's a different bug.

Comment 13

[Catalin Sasca, Desktop Test Engineering [:csasca]]

Hi Valentin, is there anything that QA can help with here? And if so, could you provide us with STR? Thanks!

Comment 14

[Valentin Gosu [:valentin] (he/him)]

(In reply to Catalin Sasca, QA [:csasca] from comment #13)

Hi Valentin, is there anything that QA can help with here? And if so, could you provide us with STR? Thanks!

I doubt there's a way to manually reproduce this.