Bug 21539

Is the error 100% reproducible (same location)?  Please attach Xorg.0.log (with ModeDebug), dmesg, and dmidecode output.

Created attachment 25453 [details]
dmesg output

I thought these tests were low level tests that could be run outside of X. X still doesn't start because of bug 21539. I don't know if the problems are related so I raised the test regression as a separate bug.

Created attachment 25454 [details]
dmidecode output as requested

Created attachment 25455 [details]
Xorg.log (ModeDebug=true)

This shows the same behaviour as 21480 although run with startx instead of via xdm. So this just continues until it crashes in libdrm due to a bad linked list.

Yes this bug is 100% reproduce-able. However could you clarify if it meant to be run with X running or just at the framebuffer console screen before starting X.

either way is fine.

Oops, ModeDebug no longer does anything as of KMS, so intel_reg_dumper output is needed (from the 2D driver)

Assuming that you added the swizzle printout to the tools and actually printed the swizzle, given your memory configuration it looks like something's really going wrong and I'll need to see intel_reg_dumper output to see more (hopefully that covers everything I need :) ).

This would be unrelated to any bug other than "glReadPixels returns garbage", though.

(In reply to comment #6)

> Assuming that you added the swizzle printout to the tools and actually printed
> the swizzle, given your memory configuration it looks like something's really
> going wrong and I'll need to see intel_reg_dumper output to see more (hopefully
> that covers everything I need :) ).

Yes, it's just the swizzle value.
 
> This would be unrelated to any bug other than "glReadPixels returns garbage",
> though.

Well the other bug mentioned is probably a confluence of a crash due to failure to allocate memory and some sort of pixel corruption before.

Anyway I'm attaching the reg traces now. 

Created attachment 25593 [details]
Register Dump before running pread_test (bz2)

Run from console before any tests run. I had to compress it to allow bugzilla to upload the file.

Created attachment 25594 [details]
Intel GPU Dump After pread test (bz2)

Again compressed to get around bugzilla limits.

Comment on attachment 25594 [details]
Intel GPU Dump After pread test (bz2)

Sorry, I need intel_reg_dumper output, not intel_gpu_dump.  (Some day we'll get intel_reg_dumper output into intel_gpu_dump, but that hasn't happened yet).

(In reply to comment #10)
> (From update of attachment 25594 [details])
> Sorry, I need intel_reg_dumper output, not intel_gpu_dump.  (Some day we'll get
> intel_reg_dumper output into intel_gpu_dump, but that hasn't happened yet).
> 

OK I found the intel_reg_dumper and am uploading now

Created attachment 25815 [details]
intel reg dumper output as requested

Tested on 2.6.30-rc6-intel-drm-00006-g619ac3b still failing.

(In reply to comment #13)
> Tested on 2.6.30-rc6-intel-drm-00006-g619ac3b still failing.
> 

Tested on .6.31-rc2-intel-drm-01085-g2a2430f still failing

Is it still accurate with latest Intel drivers?

This is a i945g, drm-intel-testing has a patch to fix swizzle testing on that. Please test the tree available at:

http://cgit.freedesktop.org/~danvet/drm-intel/

I presume that this patch fixes this for you, please reopen the bug if it's not the case.

*** This bug has been marked as a duplicate of bug 42625 ***

We reproduced the issue with 2.6.33 kernel, and now the latest kernel on -testing branch could fix it.