I am running a 5070 Ti on an Arch Linux machine, and was using the Linux 6.18-rc7 kernel, using the linux-cachyos-rc PKGBUILD, with some extra patches added from Andrea Righi’s scx tree for dl-server implementation. This should have allowed system saturation with sched-ext schedulers, and has in the past on previous tests. Maybe the patches are broken now.
Anyway, running scx_beerland from scx-scheds-git 1.0.18.r156.g70ab12a7-1, and using scx-tools-git 1:1.0.18.r6.g0bacdf7-3, and someone else using my machine remotely was running the archiso tools to generate an install image, and when mksquashfs saturated my CPU’s 24 threads, KWin_Wayland eventually hung up.
dmesg reported first some page flip hangs, then within 15 seconds, the kernel reported that the beerland scheduler had hung for approximately 6 seconds, so it may or may not have been the scheduler that hung up first.
The scheduler was recovered and replaced with a new instance, and the GPU recovered the same desktop session. Log attached, in case it can be determined to be caused by the GPU drivers:
nvidia-bug-report.log.gz (2.4 MB)