[bug][vaapi][h264] The commit 7cbe08a930a132d84b4cf79953b00b074ec7a2a7 on certain video files leads to problems with VAAPI hardware decoding.

[Mikhail Gavrilov]

Hi,

I found a commit that on certain video files leads to problems with
VAAPI hardware decoding.
Reproducing the issue requires mesa to be built with the h264 hardware
encoder enabled and the attached file to be playable in the vlc
player.
Before kernel 5.16 this only led to an artifact in the form of a green
bar at the top of the screen, then starting from 5.17 the GPU began to
freeze.
In 6.0, the problem with GPU freezing is solved, but the kernel itself
freezes when certain actions are performed. And the vlc application
cannot be terminated in any way.

The kernel trace would be like:
[  976.184187] amdgpu 0000:03:00.0: amdgpu: [mmhub] page fault
(src_id:0 ring:40 vmid:1 pasid:32785, for process vlc pid 9905 thread
vlc:cs0 pid 9956)
[  976.184205] amdgpu 0000:03:00.0: amdgpu:   in page starting at
address 0x0000800106b53000 from client 0x12 (VMC)
[  976.184210] amdgpu 0000:03:00.0: amdgpu:
MMVM_L2_PROTECTION_FAULT_STATUS:0x00141651
[  976.184213] amdgpu 0000:03:00.0: amdgpu: Faulty UTCL2 client ID: VCN0 (0xb)
[  976.184216] amdgpu 0000:03:00.0: amdgpu: MORE_FAULTS: 0x1
[  976.184219] amdgpu 0000:03:00.0: amdgpu: WALKER_ERROR: 0x0
[  976.184222] amdgpu 0000:03:00.0: amdgpu: PERMISSION_FAULTS: 0x5
[  976.184225] amdgpu 0000:03:00.0: amdgpu: MAPPING_ERROR: 0x0
[  976.184228] amdgpu 0000:03:00.0: amdgpu: RW: 0x1
[  976.184234] amdgpu 0000:03:00.0: amdgpu: [mmhub] page fault
(src_id:0 ring:40 vmid:1 pasid:32785, for process vlc pid 9905 thread
vlc:cs0 pid 9956)
[  976.184238] amdgpu 0000:03:00.0: amdgpu:   in page starting at
address 0x0000800106b52000 from client 0x12 (VMC)
[  976.184242] amdgpu 0000:03:00.0: amdgpu:
MMVM_L2_PROTECTION_FAULT_STATUS:0x00000000
[  976.184245] amdgpu 0000:03:00.0: amdgpu: Faulty UTCL2 client ID:
unknown (0x0)
[  976.184248] amdgpu 0000:03:00.0: amdgpu: MORE_FAULTS: 0x0
[  976.184251] amdgpu 0000:03:00.0: amdgpu: WALKER_ERROR: 0x0
[  976.184253] amdgpu 0000:03:00.0: amdgpu: PERMISSION_FAULTS: 0x0
[  976.184256] amdgpu 0000:03:00.0: amdgpu: MAPPING_ERROR: 0x0
[  976.184259] amdgpu 0000:03:00.0: amdgpu: RW: 0x0
[  976.184264] amdgpu 0000:03:00.0: amdgpu: [mmhub] page fault
(src_id:0 ring:40 vmid:1 pasid:32785, for process vlc pid 9905 thread
vlc:cs0 pid 9956)
[  976.184268] amdgpu 0000:03:00.0: amdgpu:   in page starting at
address 0x0000800106b53000 from client 0x12 (VMC)
[  976.184271] amdgpu 0000:03:00.0: amdgpu:
MMVM_L2_PROTECTION_FAULT_STATUS:0x00000000
[  976.184273] amdgpu 0000:03:00.0: amdgpu: Faulty UTCL2 client ID:
unknown (0x0)
[  976.184276] amdgpu 0000:03:00.0: amdgpu: MORE_FAULTS: 0x0
[  976.184279] amdgpu 0000:03:00.0: amdgpu: WALKER_ERROR: 0x0
[  976.184281] amdgpu 0000:03:00.0: amdgpu: PERMISSION_FAULTS: 0x0
[  976.184284] amdgpu 0000:03:00.0: amdgpu: MAPPING_ERROR: 0x0
[  976.184286] amdgpu 0000:03:00.0: amdgpu: RW: 0x0


The problematic commit is:
commit 7cbe08a930a132d84b4cf79953b00b074ec7a2a7 (HEAD)
Author: Alex Deucher <alexander.deucher at amd.com>
Date:   Mon Aug 9 11:22:20 2021 -0400

    drm/amdgpu: handle VCN instances when harvesting (v2)

    There may be multiple instances and only one is harvested.

    v2: fix typo in commit message

    Fixes: 83a0b8639185 ("drm/amdgpu: add judgement when add ip blocks (v2)")
    Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/1673
    Reviewed-by: Guchun Chen <guchun.chen at amd.com>
    Reviewed-by: James Zhu <James.Zhu at amd.com>
    Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
    Cc: stable at vger.kernel.org


Thanks!

-- 
Best Regards,
Mike Gavrilov.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test_sample_480_2.mp4
Type: video/mp4
Size: 127816 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20221207/12d3044e/attachment-0001.mp4>