[RADV] The Dead Rising 4 is causing a GPU hang with LLVM backend
The Dead Rising 4 is causing a GPU hang with LLVM backend, with ACO backend the hang did not happens.
when it happened I see in kernel log a lot of messages:
[gfxhub0] no-retry page fault (src_id:0 ring:24 vmid:5 pasid:32778, for process deadrising4.exe pid 7253 thread deadrising:cs0 pid 7844)
in page starting at address 0x000080020b2d9000 from client 27
VM_L2_PROTECTION_FAULT_STATUS:0x00000000
MORE_FAULTS: 0x0
WALKER_ERROR: 0x0
PERMISSION_FAULTS: 0x0
MAPPING_ERROR: 0x0
RW: 0x0
which are ended with message:
[drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for fences timed out!
- Mesa 19.3 RC3 + LLVM 9
- Kernel 5.4 RC8
./umr -O halt_waves -wa
./umr -R gfx[.]
./umr -O many,bits -r *.*.mmGRBM_STATUS*
./umr -O many,bits -r *.*.mmCP_EOP_*
./umr -O many,bits -r *.*.mmCP_PFP_HEADER_DUMP
./umr -O many,bits -r *.*.mmCP_ME_HEADER_DUMP
dmesg
I am also created RenderDoc capture with ACO backend:
wine64-preloader_2019.11.24_18.15_frame1865.rdc
but replaying this capture always invoke OOM killer on my machine
Edited by Mikhail Gavrilov