linux i686

debian packages:
mesa_8.0.4-1
linux-image-3.2.0-3-686-pae_3.2.23-1 (also crashes but does not print trace)
linux-image-3.5-trunk-686-pae_3.5-1~experimental.1
xserver-xorg_1:7.7+1

git:
xf86-video-intel 5ff749727d3590368806508ac0d0fa8efd1d1d51

Attaching dmesg and i915_error_state

Created attachment 65435 [details]
dmesg from normal boot

Created attachment 65436 [details]
i915_error_state from normal boot

One oddity is that the fbcon is in the CPU read domain, but otherwise it would look like we tried to rewrite the GTT entries for the currently bound fb even whilst we had pending render operations. Impossible!

Perhaps a relevant comment is that I have to boot with: video=SVIDEO-1:d or my laptop panel is off until KMS loads?

So for example, if I boot the standard Debian, I will get a black (but powered) panel. But if I install plymouth or add i915 to initramfs I get the video output through my laptop panel. Otherwise I have to add that option to kernel line.

This is usually indicative of some sort of BIOS / vesafb takeover race. That it has begun to show up recently offers hope that you could bisect it and find the cause. :)

(In reply to comment #6)
> This is usually indicative of some sort of BIOS / vesafb takeover race. That it
> has begun to show up recently offers hope that you could bisect it and find the
> cause. :)

Well I do not know if recently, I remember it from a long time ago... If you suspect of something I can try specific commits or date ranges.

Kernel bisecting is new grounds for me, also. Advice appreciated :)

*** Bug 54677 has been marked as a duplicate of this bug. ***

Do you see the error even if you do not boot into X?

Working on the theory that this is also related to the cpu-relocs issue, does using 3.7 help?

Ok, this isn't the cpu-relocs one, but I do believe that with the extra GTT flush and more careful waiting upon pending operations before modeset that this should be fixed in 3.7.

(In reply to comment #9)
> Do you see the error even if you do not boot into X?

Hi Cris, sorry for being away.

After your last comments, do you still need me to test this case, would it be useful to you? Or should I just wait for a testable 3.7 release?

Diego

I'm pretty certain we have these resolved now and in fact this category of PGTBL_ER is mostly harmless so whilst I would like for you to test and confirm everything works, there is no rush and we can wait for 3.7 to be available.