Re: [LTP] [PATCH v1] tst_cgroup.c: Force tst_cg_scan only scan specific cgroup version if needs_ver exist

[Cyril Hrubis <chrubis@suse.cz>]

Hi!
> > You are supposed to get an error here, at least that is what I thought.
> 
> That depends -- if the controller's tasks are all in its (only) root
> cgroup, it can be re-bind between hierarchies (v1 to v2 or vice versa).
> 
> > I do get error here on vanilla 6.10 but on debian 6.1 the mount succeeds
> > as well. CCing Michal.
> 
> If /sys/fs/cgroup/cgroup.subtree_control contains `cpuset`, it's likely
> used by some of the services, hence not all tasks are in the root cgroup
> and re-binding fails. That's what can differ between systems.
> 
> > Michal I was under an impression that a controller that has been bound
> > to v2 cannot be removed from there and bound to v1 anymore, but it seems
> > that it may happen in some cases.
> 
> It can happen under the condition above.
> (And in a future kernel it may be truly unavailable in v1 with
> !CONFIG_CPUSETS_V1.)

Thanks for the clarifications.

I still do not think that re-binding controllers to v1 is a good idea
though. Firstly v1 is being phased out and we will eventually get rid
of it, so there is no point in investing too much effort into v1 testing
in hybrid environments.

And secondly I fear that we may end up skipping v2 tests because of the
rebinding. What I think may happen is that the cgroup cleanup is
asynchronous which means that there still would be remnants of the v1
cpuset cgroup even after the particular v1 test would exit, which would
cause next test a v2 test to be skipped, because the rebinding would
fail.

-- 
Cyril Hrubis
chrubis@suse.cz

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp