Created attachment 54820 [details] Screenshot demonstrating glyph corruption After upgrading xf86-video-intel to latest GIT (had to wait for libdrm 2.6.29 to hit Debian testing first), I noticed persistent font/glyph corruption when using Firefox. GIT bisect points out b424b10e771b1d3d041efdd2b77f576251364744 ("sna: use tight pitches for a8") as the culprit. This is with SNA enabled, of course. -- Window manager: WindowMaker 0.92 (obviously not using composition or such) -- chipset: 00:02.0 VGA compatible controller: Intel Corporation 82G33/G31 Express Integrated Graphics Controller (rev 10) -- system architecture: i686 -- xf86-video-intel: present since mentioned changeset, also in latest GIT f7593a995a829978ce81397dde5c9ea6f9bb7681 -- xserver/mesa/etc: latest in Debian Testing: X.Org X Server 1.11.2.902 (1.11.3 RC 2) -- libpixman: 0.24.0 -- libdrm version: 2.4.29-1 -- kernel version: 3.1.5 -- Linux distribution: Debian Testing aka Wheezy -- Machine or mobo model: Asus P5KPL-CM -- Display connector: VGA
Looks more like cacheline corruption, if I can you to try http://cgit.freedesktop.org/~danvet/drm/log/?h=my-next that would narrow down one difference between our systems.
commit 2934e778f01cdf1307732b248b11a31c0e79e866 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Mon Dec 26 11:30:39 2011 +0000 sna: Increase the minimum alignment constraint for g33 Still not getting the single-stream mode that I am aiming for! Reported-by: Matti Hamalainen <ccr@tnsp.org> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=44150 Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> If I was watching close enough I would have spotted it, bizarrely since I bisected an issue with very similar corruption caused by the kernel whilst using that dx!
Ahh. Would it still be beneficial for me to try the libdrm from that GIT repo, or ... ?
Created attachment 54823 [details] More severe corruption Upgrading to the latest GIT with the mentioned patch resulted in more severe visual problems (see the screenshot) In any case, happy holidays. :)
Sure you didn't also get a gpu hang? Anyway that looks very much like it could be result of the finish-gpu kernel bug...
libdrm is more or less irrelevant with sna (you can even --disable-uxa to workaround the librm_intel requirement). It would be worth testing Daniel's my-next kernel branch as that does contain the fixes for a couple of gen3 issues.
Weirdly enough it seems that the horrible corruption problem disappeared after rebooting. Perhaps the GPU was being confuzzled and not everything being reset during X restart(s)? I'll keep my eyes on the situation, however.
I've just push a revert of the patch to fix this particular bug as the corruption I was seeing on my machine was related to another bug. Can you retest master and report if you start seeing the corruption again?
Yep, font/glyph corruption is back. Also, gdm now exhibits some kind of corruption on its login screen. Not sure how and if I can get a screenshot of that, though. I am, however, still using X.Org X Server 1.11.2.902 (1.11.3 RC 2) and libdrm 2.4.29-1, as newer isn't available for Debian testing (and I don't wish to bypass package management and build Xorg from tarballs/GIT at this point.)
I made it worse, bad enough that it started to affect me... commit b4ae6dbaedd774f250b610b1e7323897c4abe23a Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Tue Jan 10 21:25:28 2012 +0000 sna: Align the small upload buffers to 2 texels, and the pitch to dwords References: https://bugs.freedesktop.org/show_bug.cgi?id=44150 Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk That clears up the recent bout of corruption, does it fix yours as well? Presuming I've fixed all the other intervening corruption. ;)
Unfortunately I am not entirely sure. At least on a quick glance, I could not see it. However, upgrading to latest xf86-video-intel GIT ( a93c93be76f6d5d2b481971349aabd15f282c3e8), and using danvet/my-next kernel & Xorg 1.11.99.901 (1.12.0 RC 1) I started getting GPU hangs and after each hang XV would stop working etc. Only system reset seemed to normalize things. I can provide dmesg, debugfs, etc information on IRC (or here, if required).
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.