Description of the problem: Graphical corruption occasionally appears when using the 2.6.34 vanilla kernel. I never see this corruption when using Ubuntu's 2.6.32-22-generic kernel. Steps to reproduce: 1. Start computer. 2. Use graphical apps? Expected results: Desktop to look normal. Actual results: Strange colour inversion often close to text. How reproducible is the problem? It is highly intermitant. I have yet to find a way to make it happen on demand (but starting a web browser in the background seems helpful). The most recent corruption took around 1.5 hours after starting to occur. The kernel reports it is untainted and no vt switching has been done after the system started but the display was allowed to be turned off. Version information: Ubuntu 10.04 xserver-xorg-video-intel 2:2.9.1-3ubuntu5 xorg 1:7.5+5ubuntu1 kernel 2.6.34
Created attachment 35947 [details] Screenshot of corruption (towards the bottom of the terminal)
I forget to mention this was on an EeePC 900 laptop with an i915 graphics card: 00:00.0 Host bridge [0600]: Intel Corporation Mobile 915GM/PM/GMS/910GML Express Processor to DRAM Controller [8086:2590] (rev 04)
Hmm, stock Ubuntu drivers are quite old, it would be worthwhile to install the xserver-xorg-video-intel from the ppa:xorg-edgers to see whether this bug has been fixed in the many rendering corrections that have not been backported.
I am very interested in knowing whether this is still an issue with 2.12, as the i915 driver was substantially altered, especially the glyph paths.
I did a test with the xorg-edgers packages from a few weeks ago. Playing around for a few minutes didn't show the corruption seen in the screenshot but a new problem where some letters (always the same letter) were corrupted in all locations in a gnome-terminal.
Thanks, that confirms we still have corruption in the glyph cache. I presume tiling related. Out of curiosity more than anything else, does the issue go away with 2.6.35-rc4 which removes the marking of our memory as MOVABLE?
And now? With the magic low-power render fix by Dave Airlie?
Created attachment 38639 [details] Screenshot of (same) corruption with a newer kernel A new kernel (2.6.36-rc3-00036-g3dd7c45 with b7ffdc988523fb57ac1ef454b77d6ecc01dda4d3 "drm: Use a nondestructive mode for output detect when polling" and 300387c0b57d75e5218e2881d6ad2720657a8bcf "drm/i915: Clear the vblank status bit before polling for the next vblank" reverted) with the old Ubuntu 10.04 X userspace described in comment #1 showed the same issue.
When in doubt blame compiz... Can you install the drivers + xorg from ppa:xorg-edgers? If you have compiz running that should fix up a case where the compositing WM could read from the pixmap prior to the DDX actually rendered to it. The corruption in the screenshots is not typical of glyph cache corruption, I was misleading myself. (Glyph cache corruption is the repeated corruption of the same letter, this last screenshot looks more like uninitialised data being used for damage regions.)
OK I've just re-looked at this. I haven't seen the original "inversion style" corruption since 2.6.35 so there's a strong chance that has been cleared up. I was able to reproduce the problem on 2.6.35 by opening a terminal, typing dmesg, starting totem and then holding down F to make it rapidly switch between window and fullscreen mode. Alas, it is very intermittent as to whether this will cause the problem to show. I tried the same thing on 2.6.36 and couldn't get the corruption to appear. Now the second type of corruption (as described in comment 5 and which I will nickname "broken-letters" but is the glyph cache corruption you mentioned) is fairly reproducible with a new Fedora 5 GNOME 3 desktop. It seems to be video related in some shape or form but is definitely a different type of corruption. So, should we close this bug report as it seems to have been quietly solved and open another bug for the glyph corruption?
We can replace one mystery bug with another. :| Sitsofe, if you feel confident that the old style of corruption is no longer reproducible, then lets quietly close it and focus what on the corruption in the glyph cache.
Closing fixed.
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.