[Intel-gfx] i915 with Dell XPS 9310

wi nk wink at technolu.st
Fri Sep 3 12:37:47 UTC 2021


Hello all,

   I've been following recent kernels on this dell laptop for almost a
year now to work through some issues with the ath11k module.  I've
been experiencing random occasional video artifacts for most of that
time.  These artifacts would cause the i915 module to report some kind
of underrun (sorry I don't have those logs any longer).  At some point
around 5.10 the artifacts turned into full panics that needed a reboot
instead of some waiting and jiggling.  Kalle reported here:
http://lists.infradead.org/pipermail/ath11k/2021-August/001451.html
that there was a commit he could revert to fix it.  I was unable to
reproduce that fix by reverting it.

I'm now running 5.14.0 and it seems to have changed behavior again.
Instead of the machine hard locking (ie: no caps lock even), it seems
to recover after a bit and then I can see this in dmesg:

[226387.152234] DMAR: DRHD: handling fault status reg 3
[226387.152244] DMAR: [DMA Write NO_PASID] Request device
[0x00:0x02.0] fault addr 0xf0afc000 [fault reason 0x07] Next page
table ptr is invalid
[226402.058857] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:0:00000000
[226402.058876] i915 0000:00:02.0: [drm] Resetting rcs0 for stopped
heartbeat on rcs0
[276353.590922] clocksource: timekeeping watchdog on CPU4: hpet
retried 2 times before success

<snip, later>

[345312.963065] DMAR: DRHD: handling fault status reg 3
[345312.963077] DMAR: [DMA Write NO_PASID] Request device
[0x00:0x02.0] fault addr 0xf21ec000 [fault reason 0x07] Next page
table ptr is invalid
[345323.814583] Asynchronous wait on fence
0000:00:02.0:gnome-shell[2707]:1a15a6 timed out
(hint:intel_atomic_commit_ready [i915])
[345327.672581] i915 0000:00:02.0: [drm] GPU HANG: ecode
12:1:85dffffb, in signal-desktop [26051]
[345327.672606] i915 0000:00:02.0: [drm] Resetting rcs0 for stopped
heartbeat on rcs0
[345327.672656] i915 0000:00:02.0: [drm] signal-desktop[26051] context
reset due to GPU hang

I'm not sure what the DMAR messages are about, I included them in case
they're relevant.  How can I debug this further?  I'll gladly enable
whatever is needed :)

Thanks!


More information about the Intel-gfx mailing list