Re: [PATCH v2 0/3] drm/amd/display: Leave self-refresh before an ABM level update
From: Chen, Chen-Yu
Date: Mon Nov 16 2026 - 03:59:04 EST
Thanks for v2.
For the series:
Reviewed-by: Chenyu Chen <chen-yu.chen@xxxxxxx>
On 11/14/2026 3:40 AM, Hugo Sales via B4 Relay wrote:
> [Some people who received this message don't often get email from devnull+hugo.hsal.es@xxxxxxxxxx. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
>
> Writing panel_power_savings while the eDP is in PSR hangs DMUB on a
> Cezanne laptop (DCN 2.1). An ABM level change on its own is a fast
> update, so the new level reaches DMUB with the panel still in
> self-refresh. Chenyu Chen confirmed the same on Renoir
>
> Patch 1 adds helpers that tell whether a PSR or Replay event is already
> set on a stream. Patch 2 is the fix. It sets the PSR and Replay
> hw_programming events around the DC update that carries the new level,
> and clears only the events it set. An event that a full update, a
> modeset or a PSR SU dirty rect change set stays set, for that code to
> clear. Patch 3 stops amdgpu_dm_commit_planes() from sending the same
> level again after the connector update loop sent it, so one level change
> leaves self-refresh once
>
> "drm/amd/display: Don't force a modeset when the ABM level changes" [1]
> sends property changes down this fast path, so it needs this series first
>
> HDR metadata may have the same problem. When the HDR static metadata
> changes while HDR stays on, the commit is a fast update too, and the
> connector loop in amdgpu_dm_atomic_commit_tail() sends the new HDR
> infopacket to DC with the panel still in self-refresh. I have no panel
> with HDR and PSR or Replay to try it on, and the VRR and AVI infopackets
> already reach DC the same way. Does an infopacket update in self-refresh
> need the same pause?
>
> v2:
> - Clear the events right after the DC update, under dc_lock, instead of
> on the next fast commit (Chenyu)
> - Only clear the events this update set, so an event that a full
> update, a modeset or a dirty rect change set in the same commit stays
> set
> - Add amdgpu_dm_psr_event_is_set() and amdgpu_dm_replay_event_is_set(),
> with KUnit tests
> - Send a changed level once per commit, so it leaves self-refresh once
>
> v1: https://lore.kernel.org/r/20261031130002.2335615-1-hugo@xxxxxxx
>
> [1] https://lore.kernel.org/r/20261102-abm-sysfs-commit-v1-2-e5addfd9539a@xxxxxxx
>
> ---
> Hugo Sales (3):
> drm/amd/display: Add helpers to check PSR and Replay events
> drm/amd/display: Leave self-refresh before an ABM level update
> drm/amd/display: Send a changed ABM level once per commit
>
> drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 83 +++++++++++++++--
> .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_psr.c | 24 +++++
> .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_psr.h | 2 +
> .../drm/amd/display/amdgpu_dm/amdgpu_dm_replay.c | 25 +++++
> .../drm/amd/display/amdgpu_dm/amdgpu_dm_replay.h | 3 +
> .../display/amdgpu_dm/tests/amdgpu_dm_psr_test.c | 60 ++++++++++++
> .../amdgpu_dm/tests/amdgpu_dm_replay_test.c | 102 +++++++++++++++------
> 7 files changed, 263 insertions(+), 36 deletions(-)
> ---
> base-commit: ba496bb837882eb1af58f41572fc289e46c77b8c
> change-id: 20261113-amdgpu-psr-abm-update-next-de20bc23b12c
>
> Best regards,
> --
> Hugo Sales <hugo@xxxxxxx>
>
>