FS#64866 - [linux] GPU resets since linux 5.4 on Ryzen 5 2500U

Attached to Project: Arch Linux
Opened by rainer (raneon) - Monday, 16 December 2019, 21:47 GMT
Last edited by freswa (frederik) - Friday, 21 February 2020, 14:49 GMT
Task Type Bug Report
Category Packages: Core
Status Closed
Assigned To No-one
Architecture x86_64
Severity Medium
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 0
Private No

Details

Since the upgrade to linux 5.4 I get a lot of graphic resets/crashes with a Ryzen 5 2500U notebook (Vega 8 GPU, HP Envy x360). Basically I only have to open on Plasma the system settings or Firefox with Webrender enabled. Then the GPU will crash and try to reset, which partially works but takes 20 seconds. The system will not crash completely, only the GPU will do a reset.

Testing on Plasma Wayland was easier, as the system was able to recover better from the GPU reset, X11 must be killed before doing another test.

Linux 5.3 worked fine. Downgrading to linux-lts solves the GPU reset issue, but then standby is broken, so I cannot use this as a stable workaround.


Additional info:
* Linux 5.4

Steps to reproduce:
- Open Plasma (Wayland)
- Open system settings and scroll down in the menu entry list or open Firefox with Webrender enabled
=> GPU crashes

journalctl -f (Plasma Wayland)
Dez 16 22:25:11 pc kernel: [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for fences timed out!
Dez 16 22:25:11 pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, but soft recovered
Dez 16 22:25:33 pc kernel: [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for fences timed out!
Dez 16 22:25:33 pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, but soft recovered
Dez 16 22:25:55 pc kernel: [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for fences timed out!
Dez 16 22:25:55 pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, but soft recovered
Dez 16 22:26:15 pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, but soft recovered
Dez 16 22:26:38 pc kernel: [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for fences timed out!
Dez 16 22:26:38 pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, but soft recovered


journalctl -f (Plasma)
Dez 16 22:32:23 pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, but soft recovered
Dez 16 22:36:42 pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, but soft recovered
Dez 16 22:37:28 pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, but soft recovered
This task depends upon

Closed by  freswa (frederik)
Friday, 21 February 2020, 14:49 GMT
Reason for closing:  No response
Additional comments about closing:  This seems pretty stalled to me. If it's still an issue. Please fill a re-open request. Thank you :)
Comment by loqs (loqs) - Tuesday, 17 December 2019, 00:06 GMT
Is the issue also present in linux 5.5-rc2 or https://cgit.freedesktop.org/~agd5f/linux/log/?h=drm-next ?
Comment by rainer (raneon) - Saturday, 21 December 2019, 11:27 GMT
I did compile Linux 5.5-rc2 as recommended from drm-next, but unfortunately it shows the same bugs as Linux 5.4.
Comment by loqs (loqs) - Saturday, 21 December 2019, 12:29 GMT
Thank you for testing that.
There is some similarity to https://gitlab.freedesktop.org/drm/amd/issues/934
Or you could open a new bug on https://gitlab.freedesktop.org/drm/amd/issues

Loading...