Re: [PATCH v6 2/2] drm/amd/display: add DMU timeout recovery support

[LetsDank <ichi2209@icloud.com>]

Hi,

An important follow-up arrived shortly after my previous message.

The same separate gfxhub failure has now reproduced on the Fedora
7.1.3-200 kernel with amdgpu.dcdebugmask=0x810, after about 30.5 hours of
uptime. The sequence was:

  [gfxhub] page fault, process Xorg, ring 24, vmid 5
  GCVM_L2_PROTECTION_FAULT_STATUS: 0x00501031
  Faulty UTCL2 client ID: TCP
  PERMISSION_FAULTS: 0x3
  ring gfx_0.0.0 timeout
  Ring gfx_0.0.0 reset succeeded

Mesa then aborted Xorg in amdgpu_ctx_set_sw_reset_status(), ending the
X11 session and returning to SDDM. As in the v7.2-rc3 incident, there
were no DMUB, DMCUB, flip_done, or commit-wait timeout messages.

I captured a second complete 5.4 MB AMDGPU devcoredump. The fault status,
client, Xorg context, Mesa abort path, and ring-reset sequence match the
v7.2-rc3 incident.

This corrects any impression that the machine is stable against all GPU
faults with dcdebugmask=0x810. The workaround still appears to avoid the
original DMCUB/page-flip timeout signature, but it does not prevent this
separate gfxhub VM fault. The reproduction on both kernels also makes it
unlikely that the gfxhub fault was introduced by this patch series.

Both devcoredumps and the matching kernel logs are available if useful.

Regards,
LetsDank