Created attachment 139302 [details]
gdb backtrace of hanging plasmashell thread
Starting with bisected commit 44d5f2eb8a2f92571698adec39ac569b71da5a1b - "xf86-video-modesetting: Support new vblank kernel API [v2]" Clients seem to miss events, here plasmashell . This is easily reproduced by opening and closing the menu until plasmashell hangs. It seems to me that the server/driver does not properly report all vblank events.
Working around by calling xcb_poll_for_special_event() instead of xcb_wait_for_special_event() in the client call path "resolves" the issue in a dirty way on the client side.
Just to collect a bit more info:
HW: VGA compatible controller: Intel Corporation Device 591b (rev 04)
tested kernel: 4.16.x & 4.17rc
If you need more info, please let me know!
Created attachment 139311 [details] [review]
Disable new code
Does this patch on top of current xserver Git master still prevent the problem?
(In reply to Michel Dänzer from comment #1)
> Created attachment 139311 [details] [review] [review]
> Disable new code
> Does this patch on top of current xserver Git master still prevent the
Yes the old code/interfaces does still work fine on top of latest master.
This might actually be a duplicate of bug #106351 which i haven't checked :/. Yet we already made a bit of progress on this bugreport with bisecting and testing old & new code, leaving this one open for now seems reasonable...
*** This bug has been marked as a duplicate of bug 106351 ***