Created attachment 33722 [details]
Bug detailed description:
Glean case readpixPerf will make the system hang, and sometimes it will auto reboot. I find it was caused by the kernel in drm-intel-next branch. Its good commit is 4a42de1a3c2fc04659611338da65e472d04c78b4(built on 2010-01-19), the bad commit is 72ddbd39f39b3cc5f1f661f97e67f2a497c4d57a(built on 2010-01-27). But this branch was rebased. These two commits can't be found now. I will try to find the commits later.
2. ./glean -r readpixPerf -o --quick -v -v -v -t readpixPerf
It only on 934GM do readpixPerf make system hang. And on G45 and GM965 are well.
(In reply to comment #1)
> It only on 934GM do readpixPerf make system hang. And on G45 and GM965 are
A typo. It should be 945GM.
This issuee only exits on our drm-intel-next branch, not exist on 2.6.33. So it not blocks our Q1 release.
It was caused by the commit 060e645ac0181f499478a7be0ccb8a5600edc71a.
Author: Li Peng <email@example.com>
Date: Wed Feb 10 01:54:24 2010 +0800
drm/i915: Fix OGLC performance regression on 945
He Shuang reported an OGLC performance regression introduced in the patch
"enable memory self refresh on 9xx", In that patch, SR on 945 is disabled
everytime when calling intel_mark_busy(), while too much of such operation
will impact performance. Actually disable SR is necessary only when GPU and
Crtc changing from idle to busy. This patch make such optimization.
It fixes upstream bug
Signed-off-by: Li Peng <firstname.lastname@example.org>
Signed-off-by: Eric Anholt <email@example.com>
Peng, can you take a look? This happens only on our 945GM notebook, and fine on our 945GME netbooks. You can remote access if you want.
(In reply to comment #5)
> Peng, can you take a look? This happens only on our 945GM notebook, and fine on
> our 945GME netbooks. You can remote access if you want.
945GME and 945GM should be same. I'll try if I can reproduce on 945GM.
Peng, can you reproduce? If this only happens on this specific 945GM machine, I'm not going to strongly pursue this bug.
decreasing priority, as we are moving the nightly testing from this 945GM to Pineview.
Does this bug still occur?
Jesse, I will look at this bug, assign it to myself.
This problem have gone in current drm-intel-next git tip.
Mark it as verified.
Closing old verified.