drm i915 hangs on heavy io load

[Norbert Preining]

Hi everyone,

(please Cc)

I am running 3.7-rc2 and got recently hit a few times (under rc1, too)
by hanging drm i915 while doing large io operations.

The efect in the dmesg:
[13193.297751] [drm:i915_hangcheck_hung] *ERROR* Hangcheck timer elapsed... GPU hung
[13193.297758] [drm] capturing error event; look for more information in /debug/dri/0/i915_error_state
[13193.302728] [drm:init_ring_common] *ERROR* failed to set render ring head to zero ctl 00000000 head 85a05e3c tail 00000000 start 00003000
[13193.357584] [drm:init_ring_common] *ERROR* render ring initialization failed ctl 0001f001 head 85a05e3c tail 00000000 start 00003000
[13194.861769] [drm:i915_hangcheck_hung] *ERROR* Hangcheck timer elapsed... GPU hung
[13194.861838] [drm:i915_reset] *ERROR* GPU hanging too fast, declaring wedged!
[13194.861840] [drm:i915_reset] *ERROR* Failed to reset chip.

I captured the i915_error_state and uploaded it here:
	http://www.logic.at/people/preining/drm_i915_error_state.gz

The hangs have been normally initiated on svn up in a very big
repository, or git checkout on a very big repository or so.

Other system is Debian/unstable. The above output and error state is
from after a reboot without any suspends or other tricks inbetween,
uptime 3.5h.

Best wishes and thanks for any suggestions

Norbert
------------------------------------------------------------------------
Norbert Preining            preining@{jaist.ac.jp, logic.at, debian.org}
JAIST, Japan                                 TeX Live & Debian Developer
DSA: 0x09C5B094   fp: 14DF 2E6C 0307 BE6D AD76  A9C0 D2BF 4AA3 09C5 B094
------------------------------------------------------------------------
CORRIEMOILLIE (n.)
The dreadful sinking sensation in a long passageway encounter when
both protagonists immediately realise they have plumped for the
corriedoo (q.v.) much too early as they are still a good thirty yards
apart. They were embarrassed by the pretence of corriecravie (q.v.)
and decided to make use of the corriedoo because they felt silly. This
was a mistake as corrievorrie (q.v.) will make them seem far sillier.
			--- Douglas Adams, The Meaning of Liff
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/