Bug 359467 - KWin might cause recursive QXcbEventReader::lock()'s
Summary: KWin might cause recursive QXcbEventReader::lock()'s
Status: RESOLVED NOT A BUG
Alias: None
Product: kwin
Classification: Plasma
Component: core (show other bugs)
Version: git master
Platform: Other Linux
: NOR normal
Target Milestone: ---
Assignee: KWin default assignee
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2016-02-16 15:07 UTC by Thomas Lübking
Modified: 2024-02-19 16:29 UTC (History)
2 users (show)

See Also:
Latest Commit:
Version Fixed In:
Sentry Crash Report:


Attachments
Backtrace by Jürgen Scholz, originating from unrelated bug #358245 (39.88 KB, text/plain)
2016-02-16 15:10 UTC, Thomas Lübking
Details

Note You need to log in before you can comment on or make changes to this bug.
Description Thomas Lübking 2016-02-16 15:07:48 UTC
QXcbEventReader uses a non-recursive mutex (one might call this begging for trouble, but well...) and KWin calls QX11Info::getTimestamp() when activating a window after the active window closed (see backtrace)

This sucks a bit, we either need to defer activation to the end of the eventloop (bad idea, since we're gonna loose the formerly active client until then) or avoid fetching the X11 timestamp while we're in the eventloop.
Latter will inevitably collide w/ any solution reg. bug #348569 *AND* it's pretty hard to assure we're not causing other access to the eventreader from the eventfilter.

=> Maybe ask Qt to use a recursive mutex in 5.6?
Other ideas?

Reproducible: Always
Comment 1 Thomas Lübking 2016-02-16 15:10:10 UTC
Created attachment 97250 [details]
Backtrace by Jürgen Scholz, originating from unrelated bug #358245

The backtrace that made me aware of the problem, it's not related to the bug it was originally filed on.
Comment 2 Martin Flöser 2016-02-17 07:20:11 UTC
(In reply to Thomas Lübking from comment #0)
> => Maybe ask Qt to use a recursive mutex in 5.6?

Sounds like the best idea. The API Qt offers shouldn't dead-lock in a correct usage.
Comment 3 Thomas Lübking 2016-02-17 23:15:32 UTC
They might argue that our "correct usage" is a native event filter ;-)

Maybe we'll have to -steal- "borrow" QXcbConnection::getTimestamp()
Comment 4 Martin Flöser 2016-02-18 07:23:22 UTC
(In reply to Thomas Lübking from comment #3)
> They might argue that our "correct usage" is a native event filter ;-)
> 
> Maybe we'll have to -steal- "borrow" QXcbConnection::getTimestamp()

No need to steal, I wrote the code (9c45260942c0ee32267e651d981bfa33a6acb64d) ;-) And the reason for that was that it's impossible to implement it in a native event filter (at least back then).
Comment 5 Justin Zobel 2021-03-10 00:32:38 UTC
Thank you for the bug report.

As this report hasn't seen any changes in 5 years or more, we ask if you can please confirm that the issue still persists.

If this bug is no longer persisting or relevant please change the status to resolved.
Comment 6 Thomas Lübking 2021-03-10 07:12:41 UTC
No idea whether that's still reelvant, this "bug" discusses working around a threading issue in Qt's X11 code that existed in 2016
Comment 7 David Edmundson 2024-02-19 16:29:55 UTC
Lets assume not given we're talking about a hypothetical many years later