radeon_si + navi10: KSP hangs system: *ERROR* Waiting for fences timed out!
I'm experiencing a reproducible whole system hang in Kerbal Space Program 1.8.1 that's rebootable using Magic SYSRQ.
To reproduce, I take a vessel from Kerbin to Minmus and land on the flats. If it doesn't hang when I land, it does when I plant a flag.
The kernel log is unremarkable until:
Dec 03 14:02:26 alfsgard kernel: [drm:amdgpu_dm_atomic_commit_tail] *ERROR* Waiting for fences timed out!
Dec 03 14:02:26 alfsgard kernel: [drm:amdgpu_job_timedout] *ERROR* ring gfx_0.0.0 timeout, signaled seq=239915, emitted seq=239917
Dec 03 14:02:26 alfsgard kernel: [drm:amdgpu_job_timedout] *ERROR* Process information: process KSP.x86_64 pid 35741 thread KSP.x86_64:cs0 pid 35743
Dec 03 14:02:26 alfsgard kernel: [drm] GPU recovery disabled.
Dec 03 14:03:11 alfsgard kernel: sysrq: Keyboard mode set to system default
Dec 03 14:03:15 alfsgard systemd-journald[761]: Journal stopped
An apitrace is available here: https://www.icloud.com/iclouddrive/0kSlKNQqckwXjlgyGjSN0v9tg#KSP.x86%5F64.trace
System Info:
Gentoo Linux
Ryzen R9 3900X
Radeon RX 5700 XT
Kernel 5.4.1 + fix for https://bugzilla.kernel.org/show_bug.cgi?id=205169
Mesa-20-git
llvm-10-git
gcc-9.2.0
apitrace-git