[RADV/ACO] Horizon Zero Dawn cause a GPU hung ([gfxhub0] retry page fault (src_id:0 ring:0 vmid:6 pasid:32787, for process HorizonZeroDawn pid 40832 thread HorizonZer:cs0 pid 40834))
Demonstration: https://youtu.be/ozP_LeQUN5c
Yet another demonstration: https://youtu.be/i9u9rOYe-dY
Result of DXVK_HUD=full RADV_TRACE_FILE=$HOME/horizon_zero_dawn.trace RADV_DEBUG=zerovram,syncshaders steam &> horizon_zero_dawn.hang_report
:
horizon_zero_dawn.trace
horizon_zero_dawn.hang_report
dmesg
yet another hang_report with umr logs:
horizon_zero_dawn.trace
horizon_zero_dawn.hang_report
umr logs all in one archive
- GPU: Radeon 7
- Kernel: 5.8
- Mesa: 20.3 (git 57fba85d)
- LLVM: 11
With RADV_DEBUG=llvm
issue not reproduced, so this is ACO specific bug.
For reproduction:
- Reboot the computer (it strange but the issue stop reproduction after I could successfully watch intro video with LLVM backend)
- Remove proton prefix directory and cache:
$ rm -R "$HOME/.steam/root/steamapps/compatdata/1151640"
$ rm "$HOME/.steam/steam/steamapps/common/Horizon Zero Dawn/LocalCacheDX12/PSOCache.bin"
- Launch the game with ACO backend and watch the intro video.
Every time hung happens in a random place.