Created attachment 87953 [details] Xorg.0.log.old Distribution: Arch Linux ddx driver: xf86-video-ati 1.7.2.0 xorg-server 1.14.3 glamor-egl 0.5.1 kernel: 3.11.6 / 3.12.0-rc6 When I press the logout button in Xfce! entire DE becomes extremly unresponsive for some 60sec and then I get the logout dialog. If I do a VT switch - logout dialog becomes available instantly. After switching from glamor to EXA - everything works as expected, but I do get a delay after logging in via lightdm, untill DE appears. If I switch the VT to tty2 - tty1 gets activated automatically with DE loaded. After compiling xf86-video-ati from git I cant even get the lightdm or X to start, last message is [ 6.760] (==) RADEON(0): Silken mouse enabled right before randr should get initialized (but this probably belongs to another bug report and would benefit from bisecting. Radeon kernel driver is built-in along PITCAIRN and TAHITI_uvd blobs. There is no separate Curacao XT firmware available yet. I'm attaching the Xorg.0.log file.
Please attach your dmesg output.
Created attachment 88008 [details] dmesg_3.11.6 Here is dmesg output with 3.11.6. This one contains lockdep regarding logitech unified reciever driver, I thought this could be related but 3.12.0-rc6 doesn't show the same warning. Even when I plug a different keyboard there is no change - bug is still present.
Created attachment 88009 [details] dmesg_3.12.0-rc6
The way the radeon driver uses glamor currently results in pathological behaviour sometimes when hittin software rendering fallback paths. See also bug 68524.
Created attachment 88016 [details] xorg.log after logout/login Well, that's another way to trigger this bug. :) At first I got those results posted on the other bug report when I first started gtkperf. After that I disabled xfwm internal compositor, ran gtkperf again and bam - [mi] EQ overflowing backtrace in Xorg.0.log popped up. I've enabled the compositor, logged out/in again and ran gtkperf again with compositor turned on and same thing happened again. In Lines test - a line (singular) was drawed every 3sec. Maybe this isn't exclusive to glamor, while using EXA some timout occured when logging in and VT switch helped again (black screen with pointer in the middle) as described in the first post, but without the [mi] backtrace. For now I'll try to keep testing the latest git versions of glamor and xf86-video-ati. If there is anything else I could help with, apply/test an experimental patch ie...
Hm, little update. Git snapshots of xf86-video-ati and glamor didn't help, however forcing ColorTiling and ColorTiling2D on did - I can't reproduce the issue any more. I have a certain lag when I login, and when I press logout button (around 3-5 sec) but there are no warnings in logs. gtkperf with lines remained the same, while overall test finished under 5min. 3D performance is stunning, UrbanTerror 125fps constantly. When I switch to EXA, gtkperf runs under 2secs, but 3D performance is awfull, UrbanTerror is unplayable with ~20fps.
(In reply to comment #6) > Hm, little update. Git snapshots of xf86-video-ati and glamor didn't help, > however forcing ColorTiling and ColorTiling2D on did - I can't reproduce the > issue any more. I have a certain lag when I login, and when I press logout > button (around 3-5 sec) but there are no warnings in logs. FWIW, whether or not the EQ (event queue) overflowing warnings appear also depends on how many input events you generate, i.e. how much you move the mouse. It sounds like enabling 2D tiling makes things just fast enough for you to avoid the warnings normally. > When I switch to EXA, gtkperf runs under 2secs, but 3D performance is > awfull, UrbanTerror is unplayable with ~20fps. There is no EXA support for Southern Islands and newer cards. You're probably getting pure software rendering if you try to use EXA.
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/xorg/driver/xf86-video-ati/issues/81.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.