Hi,

This looks like the same WARN as drm/amd issue #5883, which I can't
comment on, so reporting here. In my case it only shows up when
compositor VRR is enabled, which the issue doesn't mention.

  https://gitlab.freedesktop.org/drm/amd/-/issues/5883

Hardware / software:
  - RX 6600 (Navi 23, DCN 3.0.2), single DP monitor, 1920x1080 @ 240 Hz,
    HDR on
  - CachyOS, kernels 7.2.2 and 7.2.8 (linux-cachyos-bore-lto),
    KDE Plasma 6.7.5 / KWin Wayland

Same WARN, same resume path:

  WARNING: drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm.c:10219 at 
amdgpu_dm_atomic_commit_tail+0x1fcc/0x44f0 [amdgpu]
  Call Trace:
   commit_tail
   drm_atomic_helper_commit
   drm_atomic_commit
   drm_atomic_helper_commit_duplicated_state
   drm_atomic_helper_resume
   dm_resume [amdgpu]
   amdgpu_ip_block_resume [amdgpu]
   amdgpu_device_resume [amdgpu]
   amdgpu_pmops_resume [amdgpu]

(Reported as amdgpu_dm.c:10114 on 7.2.2 -- the same statement.)

Unlike the GitLab report, it is intermittent here (3 WARNs across ~30
resumes in my retained logs, counting "PM: suspend exit"), the monitor
always comes back, and nothing follows it (no vblank refcount,
flip-timeout or put-underflow messages).

VRR correlation:

All three WARNs were on days when KWin's VRR setting was enabled
("Always" for the first two, "Automatic" for the third). On the two
dates in between when I checked the setting it was "Never", and no
resume between those WARNs produced it.

That fits the code. In 7.2.8, line 10219 is

  WARN_ON(drm_crtc_vblank_get(new_state->base.crtc) != 0);

in amdgpu_dm_handle_vrr_transition(), inside the VRR off->on branch.
With the compositor's vrr_enabled = 0 (and freesync_video = 0),
get_freesync_config_for_crtc() leaves the state at VRR_STATE_INACTIVE,
so that branch -- and the WARN -- can't be reached. Presumably it
triggers when the resume commit restores a pre-suspend state that had
VRR active.

If it helps confirm a shared trigger: it would be useful to know
whether VRR/FreeSync is enabled in the GitLab reporter's setup, and
whether disabling it also stops their black screen.

Thanks,
Francis

Reply via email to