[Bug 222109] [amdgpu Picasso 1002:15d8] gfxhub UTCL2 SQC permission fault -> ring gfx timeout when DPM forced high (repro on LTS 6.18 too)

[bugzilla-daemon@kernel.org]

https://bugzilla.kernel.org/show_bug.cgi?id=222109

--- Comment #7 from Alex (aleksey16.11.2000@gmail.com) ---
Additional confirmation from reporter (to use all triage info):

- Reproduces on stock Arch LTS kernel 6.18.55-1 as well, same cold-boot + high
+ immediate-GUI pattern. So NOT CachyOS/sched_ext specific (CachyOS 7.2.9 has
lavd warnings, LTS does not, both wedge).
- No RADV involvement: faulting CS are kitty (OpenGL) and kwin_wayland (OpenGL
radeonsi raven ACO). RADV installed but idle in this repro.
- devcoredump: kernel printed `[drm] AMDGPU device coredump file has been
created` twice in faulty boot, but it is RAM-only and expired after reboot
(nothing in /sys now). Reporter will set a udev rule to auto-save
`/sys/class/drm/card*/device/devcoredump/data` on next `high` repro if you need
it - please tell which artifact you prefer (devcoredump vs umr register dump vs
full dmesg with drm.debug=0x1ff).
- Will run memtest86+ to fully exclude shared DDR4 (VRAM is carved from system
RAM) and `amdgpu.vm_update_mode` / `cwsr_enable` tests on request. Currently
stable on `auto`, so can A/B on demand.
- 3.9MB full journal of faulty boot retained locally (attached trimmed
kernel-only 109KB due to size); can provide compressed full log or split on
request.

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are watching the assignee of the bug.