Compatibility Report

Tried kisak-mesa, the crash occurs in the same place on the mountain path.

Updated driver info:
Driver: AMD AMD Radeon RX 480 Graphics (polaris10, LLVM 15.0.7, DRM 3.42, 5.15.0-76-generic)
Driver Version: 4.6 (Compatibility Profile) Mesa 23.1.3 - kisak-mesa PPA
OpenGL Version: 4.6

Proton log using the updated driver:
steam-1668510.zip

I did not test with the oibaf PPA, because I have a negative experience with this PPA in the past.

Some additional details from further testing:

• The bug DOES occur in 7.0-6. Sorry for the confusion. (I was probably fooled by the flakiness mentioned below.)
• The bug in this location seems to depend on following a specific path. The most reliable way I've seen is this: After passing the Humming Gator, hug the south of the mountain path. If I don't see it at first, I can run back and forth while obscured by the trees, and the crash will occur eventually.

I need to do some additional verification after removing the updated Mesa drivers, just for consistency.

Did some further testing after downgrading to the 22.04 provided mesa drivers.

• The issue occurs in 7.0-6, 8.0-2, and Experimental.
• The crash is less likely to occur in 720p windowed mode, but I can still trigger a crash by running back and forth against the south of the path.

I continued playing the game with the environment variable PROTON_USE_WINED3D=1 and didn't observe any crashes.

@RobertTheSable Do you still experience this issue with dxvk and latest radv drivers?

I can still reproduce this on RX 470 and Mesa main 97efa575, so it's still there. I tested with RADV_DEBUG=hang, but that works around the issue, so I couldn't get a UMR dump. RADV_DEBUG=syncshaders also works around it.

[drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, signaled seq=176237, emitted seq=176239
[drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process zero.exe pid 115957 thread dxvk-submit pid 116082
amdgpu 0000:01:00.0: amdgpu: GPU reset begin!
amdgpu: cp is busy, skip halt cp
amdgpu: rlc is busy, skip halt rlc
amdgpu 0000:01:00.0: amdgpu: BACO reset
amdgpu 0000:01:00.0: amdgpu: GPU reset succeeded, trying to resume
[drm] PCIE GART of 256M enabled (table at 0x000000F4FFF80000).
[drm] VRAM is lost due to GPU reset!
amdgpu 0000:01:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] *ERROR* ring comp_1.2.0 test failed (-110)
amdgpu 0000:01:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] *ERROR* ring comp_1.3.0 test failed (-110)
[drm] UVD and UVD ENC initialized successfully.
[drm] VCE initialized successfully.
amdgpu 0000:01:00.0: amdgpu: recover vram bo from shadow start
amdgpu 0000:01:00.0: amdgpu: recover vram bo from shadow done
amdgpu 0000:01:00.0: amdgpu: GPU reset(2) succeeded!
[drm] Skip scheduling IBs!
...
[drm] Skip scheduling IBs!
[drm:amdgpu_cs_ioctl [amdgpu]] *ERROR* Failed to initialize parser -125!

The comp ring error messages shown above seem to be unrelated as they've been appearing regularly in the log before launching this game.

I haven't been able to get any more info out of this. I tried to get an apitrace, but the tracing ran too slowly on Windows to reproduce there, and it would crash shortly after I started tracing via WineD3D on Linux. I also tried to get an UMR dump manually without RADV_DEBUG=hang, after double checking that I could get waves dumped with a test program, but I got no active waves from this hang.

It doesn't seem like RADV_DEBUG=syncshaders affects performance much, so that's a workaround for now, at least.

Looks like the Polaris GPU hang was fixed in Mesa 24.1.0, I can no longer reproduce on RX 470 since commit fb323ae4.