890362 - Delayed tasks posted to the mozilla UI message loop stop working after sleep/wake and process suspend/resume

Status: RESOLVED FIXED

(Core :: IPC, defect)

Description

[Brian R. Bondy [:bbondy]]

Attachment: Patch v1.

We've noticed that events sometimes get lost in Windows 8 especially around touch when there is a computer sleep/wake or process_suspend/process_resume.
We've noticed in Win8 more than we've ever noticed before because the process gets suspended by windows whenever you switch away from it.

In particular if you use:
> MessageLoop::current()->PostDelayedTask(FROM_HERE, mTask, mDelayMs);
And then before mDelayMS is reached your computer sleeps, or the process suspends, you can reproduce mTask never being executed.

After debugging a little I found that inside TimerThread::DoAfterSleep()
 the value for delay (obtained from timer->GetDelay(&delay)) is a very, very large value.
That comes from a uint, so naturally it is an underflow.

After debugging more I found the first occurrence of that underflow to be passed into nsTimerImpl::InitCommon.
And that comes from a call into MessagePump::ScheduleDelayedWork with a parameter of a value that is slightly less than base::TimeTicks::Now().
Inside we calculate delay which contains a uint: delay = aDelayedTime - base::TimeTicks::Now() which underflows.

The value for the next delayed task time (which is passed into MessagePump::ScheduleDelayedWork) is obtained from: MessageLoop::DoDelayedWork but on sleep/wake and process_suspend/process_resume TimeTicks::Now is later than that value.

The fix here is just to prevent an underflow into the uint 'delay' in such a case.

The reproducible problem is fixed with this fix.

Comment 1

[Benjamin Smedberg]

Please add a comment that you're relying on the TimeDelta constructor initializing to 0, since it looks like you're using it uninitialized.

Comment 2

[Brian R. Bondy [:bbondy]]

will do, thanks for the review.

Comment 3

[Brian R. Bondy [:bbondy]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/3f64ceb4d042

Comment 4

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/3f64ceb4d042