Re: [patch] cgroups: disallow attaching kthreadd

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Wed, 2012-04-04 at 23:07 -0700, David Rientjes wrote:
> On Thu, 5 Apr 2012, Mike Galbraith wrote:
> 
> > I submitted that, and it didn't fly.  I like the generic exclusion
> > better, so I submit that for consideration.
> > 
> 
> [+akpm]
> 
> The last time we went through this, it was left after Andrew had fixed it 
> up when the cpusets version was merged in -mm without any disagreement 
> from Peter who was cc'd and that version was acked both by myself and Paul 
> Menage at https://lkml.org/lkml/2011/12/14/402.  Andrew dropped it and 
> asked for a repost since there was some on-going scheduler work going on 
> in linux-next that caused that version not to apply.  No follow-up was 
> ever offered.
> 
> Why have we now gone in a completely different direction again?

I really absolutely hate the cpuset only feature. The whole point of
kthreadd's existence is to provide a clean environment to spawn kthreads
from. Adding kthreadd to controllers violates this premise.

New and fresh kthreads should always get a bare minimum state, this
includes not being part of any particular cgroup.