Arch Linux

AerialX

Member

Registered: 2009-11-23

Posts: 5

Intel GPU crash on Sager NP5160

Hi,

I just recently bought a Sager NP5160, which works great for the most part. The Sandy Bridge / NVIDIA Optimus graphics seems to be a bit of an issue, however. I've disabled the NVIDIA card with acpi_call (I don't need it), so am using the integrated card exclusively. Problem is, I occasionally get random GPU lockups after a period of use (anywhere from 30 minutes to 24 hours). Sometimes they completely kill the system, others the mouse cursor continues to move (though everything else is unresponsive) and the network is still up for me to SSH in. I can consistently reproduce the issue by running an OpenGL-intensive program for about 30 seconds.

As for info on the crash, the whole X screen locks up. If it doesn't hard-lock everything, the sound generally keeps going for about 30 more seconds. the mouse continues to respond while the keyboard does not (can't switch to a TTY), and the network stays up. When that happens I can grab logs and shut down, not much more than that (can't seem to recover the system to a working state). As for the info dump, dmesg reports:

[  120.601980] [drm:i915_hangcheck_elapsed] *ERROR* Hangcheck timer elapsed... GPU hung
[  120.601999] [drm] capturing error event; look for more information in /debug/dri/0/i915_error_state
[  120.627446] [drm:i915_wait_request] *ERROR* i915_wait_request returns -11 (awaiting 29794 at 29790, next 29795)

X just kinda goes:

[   120.435] (EE) intel(0): Detected a hung GPU, disabling acceleration.
[   120.435] (EE) intel(0): When reporting this, please include i915_error_state from debugfs and the full dmesg.

For those interested, an info dump:
i915_error_state
dmesg
lscpi

Does anyone have any suggestions? Are there other packages I should try? Has anyone seen similar issues? Should I submit this as a bug report (to who)? The crashes have been happening with packages both from the main repos as well as testing. I'll probably try compiling the whole shebang (intel-dri, xf86-video-intel, mesa, libgl, etc.) from the AUR -git packages, but can't see that being a whole lot better than what's in testing right now.

Thanks,
- Aaron

Also, random issue with it that's been bothering me, just going to throw this out there:

$ time cat /sys/class/power_supply/AC/online 
1

real    0m0.806s
user    0m0.000s
sys    0m0.003s

EDIT UPDATE: Well, -git packages didn't help. Except that mesa-git is slightly broken in the AUR, and didn't install all the libs the DRI driver needed... It caused X to use software rendering - which surprisingly performs so well I didn't even notice OpenGL was software-rendering until I wondered why my CPU usage was so high (and it wasn't vsyncing).

Last edited by AerialX (2011-06-12 17:45:43)