Created attachment 128484 [details]
GPU keeps freezing and resetting every 10-20 minutes when using an external display. On older kernel versions it would just freeze with mouse pointer still moving, on newer ones (>= 4.8) it freezes and resets.
My machine: Dell XPS 13 (9360)
Distribution: Ubuntu 16.04 (preinstalled by Dell)
Kernel version (tried all of the versions mentioned): 4.4 (vanilla), 4.8, 4.9
Currently I have the following setup:
$ uname -mr
Also, I ran Intel Graphics Update Tool 2.0.3 to update the driver with no success.
Also, I've tried switching off the following flags on by one in /etc/default/grub with no success: i915.enable_rc6=0 intel_idle.max_cstate=1 i915.semaphores=1 i915.enable_dc=0 i915.enable_fbc=0 i915.enable_psr=0
Kernel log contains the following:
[ 4557.659862] [drm] GPU HANG: ecode 9:0:0x85dffffd, in Xorg , reason: Hang on render ring, action: reset
[ 4557.659865] [drm] GPU hangs can indicate a bug anywhere in the entire gfx stack, including userspace.
[ 4557.659866] [drm] Please file a _new_ bug report on bugs.freedesktop.org against DRI -> DRM/Intel
[ 4557.659868] [drm] drm/i915 developers can then reassign to the right component if it's not a kernel issue.
[ 4557.659869] [drm] The gpu crash dump is required to analyze gpu hangs, so please always attach it.
[ 4557.659870] [drm] GPU crash dump saved to /sys/class/drm/card0/error
[ 4557.659990] drm/i915: Resetting chip after gpu hang
[ 4557.660068] [drm] RC6 off
Crash dump is in the attachment.
Hope this is enough. Will supply any additional info if required.
One more thing: it seems that switching to the UXA acceleration method at least postpones the hanging, while slowing things down by quite a bit. I am still testing it, not really sure yet.
One more comment: UXA is slow as hell on higher resolutions. But at least it works.
Sorry Vladimir for neglecting this bug too loooongg. Is the problem still valid with the latest kernel (preferable drm-tip from git://anongit.freedesktop.org/git/drm-tip or 4.10.x from canonical) ?
Yann - please check the crash dump if that tells something about the root cause.
I've just updated to Ubuntu's 4.10.9 (4.10.9-041009-generic x86_64) kernel and SNA mode did not crash for at least a few hours.
Marking this resolved+worksforme based on the previous comment, assuming fixed on 4.10...