vkd3d-proton icon indicating copy to clipboard operation
vkd3d-proton copied to clipboard

AMD GPU hangs randomly with Cyberpunk 2077 Phantom Liberty DLC

Open JulienDlq opened this issue 1 year ago • 11 comments

The AMD GPU hangs randomly since I installed Phantom Liberty DLC, everything was fine before installing it.

Software information

Cyberpunk 2077 Phantom Liberty v2.1, FSR 2.1, RayTracing (no PathTracing).

System information

  • GPU: AMD Radeon RX 7800 XT (RADV NAVI32)
  • Driver: 1.3.267
  • Wine version: wine-ge-8-25-x86_64
  • VKD3D-Proton version: 2.11.1

Log files

dmesg:

[mer. 27 déc. 07:08:51 2023] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring comp_1.1.0 timeout, signaled seq=46708, emitted seq=46709                                                                                                                                           
[mer. 27 déc. 07:08:51 2023] [drm:amdgpu_job_timedout [amdgpu]] *ERROR* Process information: process GameThread pid 5900 thread redDispatcher1 pid 5927                                                                                                                          
[mer. 27 déc. 07:08:51 2023] amdgpu 0000:03:00.0: amdgpu: GPU reset begin!                                                                                                                                                                                                       
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu: IP block:gfx_v11_0 is hung!                                                                                                                                                                                            
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu: [gfxhub] page fault (src_id:0 ring:173 vmid:0 pasid:0, for process  pid 0 thread  pid 0)                                                                                                                               
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu:   in page starting at address 0x0000000000000000 from client 10                                                                                                                                                        
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu: GCVM_L2_PROTECTION_FAULT_STATUS:0x00040B5A                                                                                                                                                                             
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu:        Faulty UTCL2 client ID: CPC (0x5)                                                                                                                                                                               
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu:        MORE_FAULTS: 0x0                                                                                                                                                                                                
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu:        WALKER_ERROR: 0x5                                                                                                                                                                                               
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu:        PERMISSION_FAULTS: 0x5                                                                                                                                                                                          
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu:        MAPPING_ERROR: 0x1                                                                                                                                                                                              
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu:        RW: 0x1                                                                                                                                                                                                         
[mer. 27 déc. 07:08:53 2023] [drm] kiq ring mec 3 pipe 1 q 0                                                                                                                                                                                                                     
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu: recover vram bo from shadow start                                                                                                                                                                                      
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu: recover vram bo from shadow done                                                                                                                                                                                       
[mer. 27 déc. 07:08:53 2023] [drm] Skip scheduling IBs!                                                                                                                                                                                                                          
[mer. 27 déc. 07:08:53 2023] [drm] Skip scheduling IBs!                                                                                                                                                                                                                          
[mer. 27 déc. 07:08:53 2023] amdgpu 0000:03:00.0: amdgpu: GPU reset(1) succeeded!                                                                                                                                                                                                ```

JulienDlq avatar Dec 27 '23 06:12 JulienDlq

game.txt

JulienDlq avatar Dec 27 '23 06:12 JulienDlq

Is there any way to reproduce this? How long does it take to happen?

K0bin avatar Dec 27 '23 16:12 K0bin

Unfortunately I was not able to reproduce it again for the moment. Last times, it happened after 10 minutes of gameplay, or 1 h, or never… I’ll post an update if it happens again.

JulienDlq avatar Dec 27 '23 16:12 JulienDlq

And does it also happen in other games? I'm seeing a bunch of random gpu hang reports from rdna3 users in various games recently.

mbriar avatar Dec 27 '23 16:12 mbriar

For the other games I currently play (Diablo IV, Control, Metal Gear Solid 5: Phantom Pain), I have no issue at all.

Also, for information, before installing Cyberpunk DLC, I had hangs with v2.11 of vkd3d, but updating to v2.11.1 fixed the issue. And now with Cyberpunk DLC, it happens again. (But as I said, cannot reproduce it for the moment).

JulienDlq avatar Dec 27 '23 16:12 JulienDlq

What is your mesa driver version. Assuming you are using mesa

Blisto91 avatar Dec 27 '23 17:12 Blisto91

I use mesa v23.3.1.

JulienDlq avatar Dec 27 '23 18:12 JulienDlq

Cyberpunk 2.1 doesn't launch at all with vkd3d-proton 2.11.1 and later. That's why i had to revert back to vkd3d-proton 2.11 version. As for the hangs try to launch with this environment variable and see if that helps: VKD3D_CONFIG=force_host_cached

fakhraldin avatar Jan 12 '24 21:01 fakhraldin

@fakhraldin Can you make a new issue with some logs for the does not launch part?

Blisto91 avatar Jan 12 '24 21:01 Blisto91

I deleted CP2077's and vkd3d's cache, set up a new winepreset. It runs per vkd3d v2.11.1 without a crash in my half an hour test or so. The no launching bug disappeared and i can't bisect anymore due to many trial and error settings i did since then. However i still get a main error filling up the log: 9121.067:00cc:01d8:err:vkd3d-proton:vkd3d_waiting_event_signal: Failed to signal event, hr #80004005. this goes on untill i terminate the app. then it stopped here: 9283.702:00cc:01d8:err:vkd3d-proton:vkd3d_waiting_event_signal: Failed to signal event, hr #80004005. I don't know if this plays a role in the crash or not.

fakhraldin avatar Jan 15 '24 14:01 fakhraldin

I found the reason. I used CyberFSR on my wineprefix including a register overwrite. I did this to mod in a newer version of FSR, by spoofing it as dlss, which caused the trouble. I removed the mod from the install folder, set up a new wineprefix and it works.

fakhraldin avatar Jul 18 '24 18:07 fakhraldin