Hello. I came across a cgroup_enable_task_cg_lists() call from the cpuset controller... On Fri, Jul 19, 2019 at 03:59:55PM +0200, Juri Lelli <juri.lelli@xxxxxxxxxx> wrote: > --- a/kernel/cgroup/cpuset.c > +++ b/kernel/cgroup/cpuset.c > [...] > +static void rebuild_root_domains(void) > +{ > + struct cpuset *cs = NULL; > + struct cgroup_subsys_state *pos_css; > + > + lockdep_assert_held(&cpuset_mutex); > + lockdep_assert_cpus_held(); > + lockdep_assert_held(&sched_domains_mutex); > + > + cgroup_enable_task_cg_lists(); ...and I wonder why is it necessary to call at this place? (IIUC, before cpuset hierarchy is anywhere mounted it's mere top_cpuset, i.e. processing the top_cpuset alone is enough. And if anyone wants to create any non-root cpusets, they have to mount the hierachy first, i.e. no need to call cgroup_enable_task_cg_lists() manually. Also if I'm not overlooking anything, the race between hotplug and mount (more precisely new cpuset creation) should be synchronized by cpuset_mutex.) Thanks, Michal
Attachment:
signature.asc
Description: Digital signature