Systematic crash with Yuzu

[DistantThunder]

System info:

• Kernel: 5.13.4-zen1-1-zen
• vkBasalt: 0.3.2.4-1
• Mesa: 21.1.5-1
• Yuzu dev build: r19229.7f13104c1-1

Vulkan renderer.

Each time I tried to run VKBASALT with Yuzu I got an amdgpu crash though the rest of the kernel remained functional. It still crashes the entire GUI session though.

[drm:gfx_v10_0_priv_reg_irq [amdgpu]] ERROR Illegal register access in command stream
[drm:amdgpu_job_timedout [amdgpu]] ERROR ring gfx_0.0.0 timeout, signaled seq=1526461, emitted seq=1526463
[drm:amdgpu_job_timedout [amdgpu]] ERROR Process information: process yuzu pid 22900 thread yuzu:cs0 pid 22938
amdgpu 0000:28:00.0: amdgpu: GPU reset begin!
amdgpu 0000:28:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] ERROR ring kiq_2.1.0 test failed (-110)
[drm:gfx_v10_0_hw_fini [amdgpu]] ERROR KGQ disable failed
amdgpu 0000:28:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] ERROR ring kiq_2.1.0 test failed (-110)
[drm:gfx_v10_0_hw_fini [amdgpu]] ERROR KCQ disable failed
[drm:gfx_v10_0_hw_fini [amdgpu]] ERROR failed to halt cp gfx
[drm] free PSP TMR buffer
amdgpu 0000:28:00.0: amdgpu: MODE1 reset
amdgpu 0000:28:00.0: amdgpu: GPU mode1 reset
amdgpu 0000:28:00.0: amdgpu: GPU smu mode1 reset
amdgpu 0000:28:00.0: amdgpu: GPU reset succeeded, trying to resume
[drm] PCIE GART of 512M enabled (table at 0x0000008000300000).
[drm] VRAM is lost due to GPU reset!
[drm] PSP is resuming...
[drm] psp command (0x22) failed and response status is (0x80000310)

Does someone reproduce? can it be solved?

[v-fox]

The way GPU drivers are developed, the hardware is not protected from errors on the userspace level, meaning that kernel driver lets through bad code without error handling from Mesa (which may or may not come directly from an app) all the time. I had particularly bad cases with memory leak with mpv's vulkan backend and hard crash like this when running Dishonored 2 under wine/dxvk. Recently got one with cemu, which is similar to yuzu, even without vkBasalt.

In my case of the crash, there was some unknown bug in Mesa-21.x that was fixed (i think, for now) in 22.x update, leak was also fixed at some point. It also very crashy if 64-bit PCIe memory mapping is enabled in BIOS/EFI (so-called "SAM"). amdgpu parameters and Mesa debug/drirc variables also may cause that. So check if it got better in kernel-6.x and Mesa-22.x. Also try flipping mes and mes_kiq amdgpu options.