[PATCH 0/1][SRU][Plucky/Oracular] HW accelerated video playback causes VCN timeout on VCN 4.0.5 (AMD Strix)

[You-Sheng Yang]

BugLink: https://bugs.launchpad.net/bugs/2112582

[ Impact ]

amdgpu has functionality that will reset the GPU when a job in a ring
buffer times out to try to recover. When this happens, the application
submitting the job will crash and there will be a flicker for ~2s while
the GPU recovers.

Proposed fix in v6.15-rc7 as commit 07c9db090b86 ("drm/amdgpu: read back
register after written for VCN v4.0.5").

[ Test Plan ]

This was originally reported in
https://gitlab.freedesktop.org/mesa/mesa/-/issues/12528. The reproduce
steps are:
1. enabled HW accelerated video decode in Firefox 134 using about:config
2. run video playback streams as stress test for hours.

[ Where problems could occur ]

The fix reads back registers to ensure a previous write has completed. No
known problem as this moment.

[ Other Info ]

While this affects VCN 4.0.5 (AMD Strix Point) and the fix has been
committed to v6.15-rc7, nominate for 6.11+ - 6.14 kernels, e.g.
linux/oracular, linux-oem-6.11/noble, linux/plucky, linux-oem-6.14/noble.

David (Ming Qiang) Wu (1):
  drm/amdgpu: read back register after written for VCN v4.0.5

 drivers/gpu/drm/amd/amdgpu/vcn_v4_0_5.c | 8 ++++++++
 1 file changed, 8 insertions(+)

-- 
2.48.1