Created attachment 176769 [details] Hang callstack SUMMARY kwin_x11 frequently hangs on shutdown/reboot causing a delay and this message to appear: "User job plasma-kwin_x11.service/stop running (XYs / 1min 29s)" It appears to not terminate after being sent SIGTERM and has to be SIGKILL-ed later. The hang callstack is attached. The thread "vsync event monitor" seems to be stuck deallocating a thread-local GLX resource. The main thread is also waiting for something, probably related to the vsync thread. If kwin doesn't hang, it crashes immediately with SIGSEGV. Unfortunately, I was not able to get the crash callstack as systemd-coredump does not seem to work during shutdown and nothing shows up in coredumpctl. STEPS TO REPRODUCE 1. Shutdown or restart. OBSERVED RESULT Shutdown gets delayed due to kwin hanging. EXPECTED RESULT No delays. SOFTWARE/OS VERSIONS OS: Debian Testing KDE Plasma Version: 6.2.4 KDE Frameworks Version: 6.8.0 Qt Version: 6.7.2 ADDITIONAL INFORMATION Kernel Version: 6.11.10-amd64 (64-bit) Graphics Platform: X11 Processors: 16 × AMD Ryzen 7 5800X 8-Core Processor Memory: 31,3 GiB of RAM Graphics Processor: NVIDIA GeForce RTX 3070/PCIe/SSE2 Graphics Drivers: Proprietary nvidia-driver 565.77 Setting KWIN_X11_FORCE_SOFTWARE_VSYNC=1 fixes both the hang and the crash. I think this prevents the existence of the vsync thread. This issue seems to be similar to https://bugs.kde.org/show_bug.cgi?id=453928 in terms of symptoms but could have a different root cause.
Created attachment 176770 [details] Journal of kwin_x11 when the hang occurred.
Created attachment 176771 [details] Journal of kwin_x11 when the crash occurred.
Created attachment 176776 [details] Crash callstack. > Unfortunately, I was not able to get the crash callstack as systemd-coredump does not seem to work during shutdown and nothing shows up in coredumpctl. This is a known issue: https://github.com/systemd/systemd/issues/28338 Anyway, I was able to work around it by restoring the legacy coredump handling per https://systemd.io/COREDUMP/#restoring-legacy-coredump-handling and managed to acquire the crash callstack.
Adding the x11-only keyword