protonscr

[D3D11,dxvk] Star Trek Online causes ring gfx_0.0.0 timeout on RDNA4 with MSAA enabled since commit 84f2939

dxvkclosed amd radvcan't reproduce
doitsujin/dxvk#4756 · opened 2025-03-09 by Gnatzelle · updated 2025-03-11 · 10 comments · github
GGnatzelle 2025-03-09 github

As discussed here: https://github.com/doitsujin/dxvk/issues/4750

And as the Title says: Star Trek Online causes ring gfx_0.0.0 timeout on RDNA4 with MSAA enabled since commit https://github.com/doitsujin/dxvk/commit/84f2939b55cd4ce93ba76b08ec049eb5b0250003

Last working commit: https://github.com/doitsujin/dxvk/commit/7503a66789ed80fb0d91ade2f550c3bb76e6214b

How to reproduce (at least on my System with an RDNA4 GPU)
-Boot the Game in Safe-Mode
-Raise the Graphics-Quality up to maximum Details
-Switch AA from None > MSAA
-GPU Hang...

Arch Linux - 6.13.6-zen1
Mesa 25.1.0-devel (git-76883e0b3c)
Linux-Firmware-git
wine-10.2-staging-tkg-amd64
AMD RX 9070 XT

I had no luck with an Apitrace when dxvk was disabled like mentioned in the Wiki. So this is a trace with dxvk enabled, otherwise the Game won't launch at all. And it's a rather small file with only 32 MiB, so i don't know if it really helps...
https://drive.google.com/file/d/1o-ebziFd0zGVjuwQUM5cbQdTy7myxY0K/view?usp=sharing

FWIW the Journal when the crash happend:

Mär 09 22:50:11 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: Dumping IP State
Mär 09 22:50:11 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: Dumping IP State Completed
Mär 09 22:50:11 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: ring gfx_0.0.0 timeout, signaled seq=169961, emitted seq=169963
Mär 09 22:50:11 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: Process information: process Main Thread pid 3138 thread dxvk-submit pid 3254
Mär 09 22:50:11 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: Starting gfx_0.0.0 ring reset
Mär 09 22:50:14 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: MES(1) failed to respond to msg=RESET
Mär 09 22:50:14 gnatz-arch kernel: [drm:amdgpu_mes_reset_legacy_queue [amdgpu]] *ERROR* failed to reset legacy queue
Mär 09 22:50:14 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: reset via MES failed -110
Mär 09 22:50:14 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: Ring gfx_0.0.0 reset failure
Mär 09 22:50:14 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: GPU reset begin!
Mär 09 22:50:17 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: MES(1) failed to respond to msg=REMOVE_QUEUE
Mär 09 22:50:17 gnatz-arch kernel: [drm:amdgpu_mes_unmap_legacy_queue [amdgpu]] *ERROR* failed to unmap legacy queue
Mär 09 22:50:20 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: MES(1) failed to respond to msg=REMOVE_QUEUE
Mär 09 22:50:20 gnatz-arch kernel: [drm:amdgpu_mes_unmap_legacy_queue [amdgpu]] *ERROR* failed to unmap legacy queue
Mär 09 22:50:22 gnatz-arch kernel: amdgpu 0000:09:00.0: amdgpu: MES(1) failed to respond to msg=REMOVE_QUEUE
Mär 09 22:50:22 gnatz-arch kernel: [drm:amdgpu_mes_unmap_legacy_queue [amdgpu]] *ERROR* failed to unmap legacy queue
Ddoitsujin maintainer 2025-03-09 github

I made an apitrace when testing the game earlier, does replaying it also hang your system?

GGnatzelle 2025-03-09 github

Problem is fixed with this branch: https://github.com/doitsujin/dxvk/tree/desktop-no-secondary-cmdbufs

Ddoitsujin maintainer 2025-03-09 github

Can you still try the trace I posted with the non-working master branch? I'd like to know if the problem is RDNA4-specific.

GGnatzelle 2025-03-09 github

Ah okay, was about to ask ^^"
Any specific commit? Or the first when it broke?

EDIT: Sorry if i'm a little slow, but this is kinda new for me ^^" master of course, you only reverted this one change... trying now

GGnatzelle 2025-03-09 github

@doitsujin Replaying the Apitrace File also triggerd the crash with dxvk-git master-branch when dxvk is enabled. Runs fine though if dxvk is disabled.

Ddoitsujin maintainer 2025-03-09 github

Thanks, so this does seem to be specific to RDNA4.

I'll probably revert the change to not hang systems left and right, but this should probably be looked at from the Mesa sice since I'm really not finding any obivous DXVK bug here no matter how hard I look at a capture.

Ddoitsujin maintainer 2025-03-09 github

Merged the workaround for now.

GGnatzelle 2025-03-09 github

I mean, RDNA4 is brand new and i thought that there would be some problems here and there with those changes in the µArch.

But many thanks again for such a quick fix :) Even if it makes things more complicated again -.-"

OT: Noticed some odd behaviour in Shin Megami Tensei III Nocturne HD as well without the fix...it's now functioning normal as well. Will file a bug Report at Mesa for this and STO...

Ddoitsujin maintainer 2025-03-09 github

What was wrong with SMT3 specfically?

GGnatzelle 2025-03-09 github

Ignore my "odd behaviour" saying, i was using a wrong Proton Version at that time, cause no Ingame Videos worked and it looked funny but would load normally.

But i checked SMT3 again...same Error as with STO before the fix. GPU Hang when the Title Screen appears and the Menu Items float in when AA is enabled (Only on or off in the Options, 2xMSAA iirc)
Running normally now with the fix.

Upstream links