GPUOpen-Drivers / AMDVLK

AMD Open Source Driver For Vulkan
MIT License
1.72k stars 161 forks source link

AMDVLK crashes kernel driver in Strange Brigade if the game has been started with RADV before #228

Closed aufkrawall closed 3 months ago

aufkrawall commented 3 years ago

-Start Strange Brigade (Vulkan) with RADV, then close it and start it again with AMDVLK: amdgpu kernel driver crashes and resets upon reaching the game's main menu. -Start it first with AMDVLK and then with RADV: amdgpu kernel driver doesn't crash. -Start it two times in a row with AMDVLK: It doesn't crash. -Start it two times in a row with RADV: It doesn't crash.

So, it seems RADV puts something in a state that AMDVLK doesn't like. Ping @pendingchaos . :)

Errors in dmesg:

[ 2278.493879] [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for fences timed out!
[ 2283.622891] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx_0.0.0 timeout, signaled seq=227004, emitted seq=227006
[ 2283.622980] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process StrangeBrigade_ pid 151480 thread StrangeBrigade_ pid 151480
[ 2283.623056] amdgpu 0000:03:00.0: amdgpu: GPU reset begin!
[ 2283.962222] amdgpu 0000:03:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] *ERROR* ring kiq_2.1.0 test failed (-110)
[ 2283.962288] [drm:gfx_v10_0_hw_fini [amdgpu]] *ERROR* KGQ disable failed
[ 2284.184339] amdgpu 0000:03:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] *ERROR* ring kiq_2.1.0 test failed (-110)
[ 2284.184400] [drm:gfx_v10_0_hw_fini [amdgpu]] *ERROR* KCQ disable failed
[ 2284.406005] [drm:gfx_v10_0_hw_fini [amdgpu]] *ERROR* failed to halt cp gfx
[ 2284.425024] [drm] free PSP TMR buffer
[ 2284.473807] amdgpu 0000:03:00.0: amdgpu: MODE1 reset
[ 2284.473875] amdgpu 0000:03:00.0: amdgpu: GPU smu mode1 reset
[ 2284.981885] amdgpu 0000:03:00.0: amdgpu: GPU reset succeeded, trying to resume

dmseg.log

Happens with both amdvlk-open 2021.Q2.2 and amdvlk-pro 21.10. RX 6800, Linux 5.12.4, Proton Experimental.

Oschowa commented 3 years ago

I tried this on Polaris and couldn't reproduce it there.

iWeaker commented 3 years ago

I am trying to reproduce it and I have not been able to do it.

aufkrawall commented 3 years ago

With an RDNA2 GPU?

pingubot commented 3 years ago

I had very strange gpu hangs in many games with amdvlk and amd-pro (Strange Brigade and dx12 games via vkd3d).They are all gone now when i do not use radv before. I have a 6900xt.

aufkrawall commented 3 years ago

Yeah, I think it can happen with all sort of applications. Hopefully this reproducible case covers them all.

pingubot commented 3 years ago

Yeah, I think it can happen with all sort of applications. Hopefully this reproducible case covers them all.

Yep. And many thx for figuring out the cause of those hangs !

pingubot commented 3 years ago

That still seems to be an issue with the recent release q26.

NTMan commented 3 years ago

I am able reproduce it bug with 6900XT and several games such as Zombie Army 4, Resident Evil Village, Forza Horizon 4, Avengers. https://gitlab.freedesktop.org/mesa/mesa/-/issues/5055