If you pass --nodeps to start or restart the cgroup is not created. This is a problem especially for udev. UDev expects to run in its own cgroup and may kill random processes from its cgroup in the wake of a trigger event. So if the cgroup is the root one then typically sshd dies (not cool for physical remote machines where console access is difficult). Reproducible: Always Steps to Reproduce: 1. /etc/init.d/udev -D restart 2. ls -ld /sys/fs/cgroup/unified/udev 3. /etc/init.d/udev restart 4. ls -ld /sys/fs/cgroup/unified/udev Actual Results: cgroup is not created Expected Results: cgroup should be created also with --nodeps