Created attachment 14809 [details]
Output of lspci -vv
Many openGL applications on a T61 running Ubuntu Gutsy crash X. Specific example is /usr/lib/xscreensaver/skyrocket, which will crash the server and cause a restart in window mode or lock the server up completely in full screen mode.
Created attachment 14810 [details]
Created attachment 14811 [details]
Created attachment 14812 [details]
Created attachment 14813 [details]
Created attachment 14814 [details]
Output of ps axww at time of crash
Created attachment 14815 [details]
dmesg results with /sys/module/drm/parameters/debug set to 1
Created attachment 14816 [details]
You're using a very old 3d driver, and we've changed just about everything about the code surrounding the assert you're getting a failure on. It seems to work with Mesa and X Server from master, so could you re-try with those to confirm that the bug is fixed for you?
(It may even be fixed in the 7.0.3 release candidates, as it's working for me in direct rendering with the mesa stable branch. I can't test indirect mode easily with old Mesa to confirm)
John, are you able to test the git master tip?
I've reproduced it on 7.0.3-rc2 backported from Ubuntu Hardy. I'll get DRM debug output and a backtrace shortly. I'm working on getting git HEAD built and tests.
Created attachment 15041 [details]
dmesg output from 7.0.3-rc2 with drm debug
Created attachment 15042 [details]
X server backtrace from 7.0.3-rc2
Created attachment 15043 [details]
Xorg log from 7.0.3-rc2
HEAD appears to require kernel > 2.6.22, which is a bit of a setback.
Never mind. Mis-googled. Looks like it's an out-of-date libdrm resulting in a missing radeon define.
libdrm-head plus mesa-head on Gutsy appears to result in a horribly broken mess openGL-wise, so I'll try with xserver-head as well. Next time I'll actually read the build instructions before hitting 'make'.
(In reply to comment #16)
> libdrm-head plus mesa-head on Gutsy appears to result in a horribly broken mess
> openGL-wise, so I'll try with xserver-head as well. Next time I'll actually
> read the build instructions before hitting 'make'.
John, not quite understand this.. is the issue gone with git tip build or what?
So I think I'm confused now. I have "the latest" libdrm from git, "the latest" mesa from git, and "the latest" xserver from git. I've built them all (to the best of my ability/knowledge), but the result does not appear to be stable enough to provide any insight as to whether this issue still exists. I can run 'skyrocket' for and arbitrary length of time, but I can't get the Gutsy version of compiz to even start up, which makes me believe I'm not reproducing the original failure.
Is there a specific set of branches/tags/whatever I should be testing against ? I guess I'll try pulling the latest again and rebuilding.
master of server, drm, and mesa, plus any dependencies you required to get them built. So it sounds like you're there, and that you've confirmed the bug is fixed. compiz does have problems on mesa master, see #14441. If all you're seeing is #14441, then we can probably close this bug.
I still can't seem to get it to happen, so I'll move it to fixed. I'm going to ask about the chances of a backport of the fix to 7.0.x, but I'm guessing the chances are practically zero.