protonscr

Vega 64 GPU hang when opening dashboard on mesa git

steamvrclosed
ValveSoftware/SteamVR-for-Linux#165 · opened 2018-12-20 by ChristophHaag · updated 2020-11-16 · 5 comments · github
CChristophHaag 2018-12-20 github

This has been an issue for a few weeks so I don't have precise version info. As soon as I open the dashboard, a GPU hang happens and I have to reboot.

It happens on mesa git 96063100812d1de46f4146fde92975c98e78c3ce + llvm 8 git 0d758b2e709. but it does not happen on mesa 18.3.1 with the same llvm 8 build.

In the end I maybe have to bisect mesa.

On mesa 18.3.1 where I don't have the GPU hang, I have another issue, but it's happening maybe 1/5 of the time I start steamvr and it's seemingly random:

Dez 20 19:00:00 c-pc kernel: [drm:gfx_v9_0_priv_reg_irq [amdgpu]] *ERROR* Illegal register access in command stream
Dez 20 19:00:00 c-pc kernel: amdgpu 0000:0a:00.0: GPU reset begin!

Not sure if it's related.

CChristophHaag 2018-12-20 github

Looks like disabling the camera fixes the hangs. The culprit should be the small camera overlay that is attached to the controller.

CChristophHaag 2018-12-20 github

Sorry it looks like it was just accidentally working when I tried several times.

I get GPU hangs even with the camera support completely disabled.

CChristophHaag 2019-04-12 github

I'm relatively sure that this does not happen with up to date drivers anymore.

CChristophHaag 2019-04-17 github

Seems to be rare now, but I just got another one:

Apr 17 21:56:09 c-pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, signaled seq=9420039, emitted seq=9420041
Apr 17 21:56:09 c-pc kernel: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process vrdashboard pid 7059 thread vrdashboar:cs0 pid 7130
CChristophHaag 2020-11-16 github

I have not seen this in a long time.

Nothing extracted yet.