[Differential] [Commented On] D3132: [platformx/x11] Add a freeze protection against OpenGL

antlarr (Antonio Larrosa Jimenez) noreply at phabricator.kde.org
Sat Oct 22 08:49:55 UTC 2016


antlarr added a comment.


  > How many frames do we need to detect the freeze? I think 30 frames is too much by default. It's adding quite some stress on startup (lots of write to config, lots of additional threads started). Assuming we freeze with tripple buffer when trying to get the next buffer 4 frames should be sufficient to trigger the freeze?
  
  Yesterday I tried with 20 and 10 frames and it detected the freeze correctly, then tried 5 and got a strange behaviour (the desktop wasn't frozen but the menus were shown displaced and there was no window decoration). I decided I would remove an option I added to the nouveau driver to test if it helped and rebooted the machine to be sure I tested with a clean system. But then I lost the connection to the machine and it seems to have a problem booting up so I can't do more tests until Monday when someone goes to the SUSE offices where the machine is and check what the problem is.
  
  In the meantime, I will improve a bit the PreFrame and PostFrame code so it doesn't add so much overhead (reusing the thread and timer, and only write the config option in the lambda function, which wouldn't work to detect a crash, but it should work to detect a freeze, so I'll separate the code of PreFrame/PostFrame from the code in PreInit/PostInit).
  
  I'll update this phab later today with those optimizations.

REPOSITORY
  rKWIN KWin

REVISION DETAIL
  https://phabricator.kde.org/D3132

EMAIL PREFERENCES
  https://phabricator.kde.org/settings/panel/emailpreferences/

To: antlarr, #kwin, #plasma, davidedmundson
Cc: graesslin, kwin, plasma-devel, davidedmundson, lesliezhai, ali-mohamed, hardening, jensreuterberg, abetts, sebas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/plasma-devel/attachments/20161022/bf511aad/attachment.html>


More information about the Plasma-devel mailing list