occasional corruption issue with RADV in multiple games, disappears after using amdvlk
Description
Occasionally, newly started games show severe corruption, which disappears after a reboot. The issue doesn't occur very often and no apparent program seems to be accountable. If the issue occurs in one game (e.g. Hitman 2 via Proton DXVK) it also happens in other games when started afterwards (like native Linux Vulkan game Shadow of the Tomb Raider).
Restarting the games doesn't help, but restarting them with amdvlk instead of RADV (be it llvm or aco) makes them render without corruption. Funny thing: Once started with amdvlk, the games also render correctly again when started with radv, without restarting the system.
There seem to be amdgpu errors in dmesg which could be related.
Screenshots/video files
Log files (for system lockups / game freezes / crashes)
Steps to reproduce
It is unclear whether there is a pattern which would make the issue reproducible.
System information
inxi -GSC -xx
System: Host: USER-linux Kernel: 5.11.0-rc6-115-tkg-cfs x86_64 bits: 64 compiler: gcc v: 10.2.1
Desktop: KDE Plasma 5.20.90 tk: Qt 5.15.2 wm: kwin_x11 dm: startx Distro: Arch Linux
CPU: Info: Quad Core model: Intel Core i7-6700K bits: 64 type: MT MCP arch: Skylake-S rev: 3 L2 cache: 8 MiB
flags: avx avx2 lm nx pae sse sse2 sse3 sse4_1 sse4_2 ssse3 vmx bogomips: 63999
Speed: 4015 MHz min/max: 800/4200 MHz Core speeds (MHz): 1: 4015 2: 4119 3: 3989 4: 4060 5: 4160 6: 4018 7: 3493
8: 4169
Graphics: Device-1: Intel HD Graphics 530 vendor: ASRock driver: i915 v: kernel bus ID: 00:02.0 chip ID: 8086:1912
Device-2: Advanced Micro Devices [AMD/ATI] Navi 10 [Radeon RX 5600 OEM/5600 XT / 5700/5700 XT]
vendor: Micro-Star MSI driver: amdgpu v: kernel bus ID: 03:00.0 chip ID: 1002:731f
Display: server: X.org 1.20.10 compositor: kwin_x11 driver: loaded: amdgpu resolution: <missing: xdpyinfo>
OpenGL: renderer: AMD Radeon RX 5700 XT (NAVI10 DRM 3.40.0 5.11.0-rc6-115-tkg-cfs LLVM 11.0.1)
v: 4.6 Mesa 21.1.0-devel (git-e870796113) direct render: Yes
Note: The GPU is undervolted, I can't vouch that this can be ruled out as a cause. However, imho the oddness regarding amdvlk's "healing" effect is peculiar. I'll raise vddgfx and see if the issue ever occurs again. But maybe anyone else has witnessed something similar?