Arch Linux

Amphitryon

Member

Registered: 2013-09-20

Posts: 39

5.14, alsa, pulseaudio and unkillable processes

When running Linux 5.14 I am getting issues with the pulseaudio process which seems to be unkillable.  In the system journal I got the following with linux-5.14.3.arch1-1:

Sep 14 17:00:46 ecrins systemd[1]: user@120.service: State 'stop-sigterm' timed out. Killing.
Sep 14 17:00:46 ecrins systemd[1]: user@120.service: Killing process 571 (systemd) with signal SIGKILL.
Sep 14 17:00:46 ecrins systemd[1]: user@120.service: Killing process 648 (pulseaudio) with signal SIGKILL.
Sep 14 17:00:46 ecrins systemd[1]: user@120.service: Killing process 766 (alsa-sink-ALC88) with signal SIGKILL.
Sep 14 17:00:46 ecrins systemd[1]: user@120.service: Killing process 767 (alsa-source-ALC) with signal SIGKILL.
Sep 14 17:00:46 ecrins systemd[1]: user@120.service: Main process exited, code=killed, status=9/KILL
Sep 14 17:00:46 ecrins systemd[1]: user@120.service: Killing process 648 (pulseaudio) with signal SIGKILL.
Sep 14 17:00:46 ecrins systemd[1]: user@120.service: Killing process 766 (alsa-sink-ALC88) with signal SIGKILL.
Sep 14 17:00:46 ecrins systemd[1]: user@120.service: Killing process 767 (alsa-source-ALC) with signal SIGKILL.
Sep 14 17:02:46 ecrins systemd[1]: user@120.service: Processes still around after final SIGKILL. Entering failed mode.
Sep 14 17:02:46 ecrins systemd[1]: user@120.service: Failed with result 'timeout'.
Sep 14 17:02:46 ecrins systemd[1]: user@120.service: Unit process 648 (pulseaudio) remains running after unit stopped.
Sep 14 17:02:46 ecrins systemd[1]: Stopped User Manager for UID 120.

If I reboot the system this pulseaudio process does not die any time before rebooting, i.e. the shutdown screen contains several rounds oftrying to kill it and waiting for a timeout.  The problem occurs with all the 5.14 kernel versions including .1 and .2 as well as the current .3

Has anyone seen anything similar.  Do you agree with me that it is a kernel bug in that a user mode process should always be killable with SIGKILL, i.e. this seems to be deadlocking inside the ALSA driver somewhere?