Bug 78922

Created attachment 99352 [details]
dmesg output

Created attachment 99353 [details]
glxinfo output

LIBGL_DEBUG=verbose glxinfo outputs the following : 

libGL: screen 0 does not appear to be DRI3 capable
libGL: pci id for fd 4: 8086:0152, driver i965
libGL: OpenDriver: trying /usr/lib/x86_64-linux-gnu/dri/tls/i965_dri.so
libGL: OpenDriver: trying /usr/lib/x86_64-linux-gnu/dri/i965_dri.so
libGL: Can't open configuration file /home/andrew/.drirc: No such file or directory.
libGL: Can't open configuration file /home/andrew/.drirc: No such file or directory.
libGL: Can't open configuration file /home/andrew/.drirc: No such file or directory.

Created attachment 99354 [details]
xorg log

First please update mesa to eliminate one known IVB display corruption bug.

Mesa updated to 10.3.0-devel from oibaf's PPA. No change. glxgears shows severe rendering artefacts. (see attachment)

Created attachment 99357 [details]
glxgears rendering artifacts

How about going the other way, and looking for a version of packages that used to work? Then find which package seems to cause the regression. The choice is kernel / ddx / mesa. So far we know that mesa has a few issues, but they may be unrelated.

Added the following to xorg.conf

Option "AccelMethod" "uxa"

This has removed the speckling artifacts. I am still having issues with glxgears not rendering correctly, but this is definitely an improvement.

Sadly, that still doesn't rule out a root cause in mesa.

I've tried the latest Mesa Release Candidate (10.2 rc3) and 9.2.5 - no change. As I said, changing the acceleration mode fixes the speckling artifacts but OpenGL applications have severe vertical tearing artifacts when rendering in hardware.

Unsure where to go from here - I haven't seen anyone else report the problem so is this likely a hardware issue? Windows drivers exhibit no problems but I'm not sure if that means anything.

Cheers,
Andrew.

The speckle looks to be a blend failure. It looks to be in the compositor as the pattern seems to have some of the overdrawn content, which is why I suspect that mesa is involved. At first I thought the speckle would be a cache invalidate issue, but it doesn't actually seem to be wide enough for a cacheline issue - so it may be a bad EU, sampler or killpixel instruction.

When you tried 9.2.6, what changed? Presumably the glxgears returned to normal?

9.2.5 still rendered with artifacts. I tried with 9.0.3 and it renders fine. Would it be helpful if I find the exact version where it breaks?

Cheers,
Andrew.

Ok, it definitely broke between 9.0.3 and 9.1, and seen as that's when OpenGL 3 support was introduced I suspect that's probably important.

Let me know if there's anything else I can do.

Cheers,
Andrew.

I've lost track. Are you referring to the speckle or the glxgears corruption that disappears with 9.0?

Ok it looks like the two issues are distinct.

 - Speckling disappears by changing the kernel module's acceleration method to UXA, but is unaffected by any of the Mesa version changes

 - GL rendering artifacts clear up on reverting to Mesa 9.0.3 ( and reappear after 9.1 ) but are unaffected by the kernel module's acceleration method.

Time to do a new round of testing: kernel for any w/a, ddx and mesa for any fixes.

-- 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-intel/issues/173.