Re: [PATCH 2/2] cgroups: convert ss->attach to use whole threadgroup flex_array (cpuset, memcontrol)

[Frederic Weisbecker <fweisbec@gmail.com>]

On Fri, Oct 14, 2011 at 09:54:57AM -0400, Ben Blum wrote:
> On Fri, Oct 14, 2011 at 09:53:23AM -0400, Ben Blum wrote:
> > On Fri, Oct 14, 2011 at 02:21:30PM +0200, Frederic Weisbecker wrote:
> > > On Thu, Oct 13, 2011 at 08:36:01PM -0400, Ben Blum wrote:
> > > > Convert ss->attach to take a flex_array of tasks instead of just the leader.
> > > > 
> > > > From: Ben Blum <bblum@andrew.cmu.edu>
> > > > 
> > > > This lets subsystems with whole-threadgroup attach calls (i.e., cpuset and
> > > > memcontrol) to accurately find the group's mm even when a non-leader does exec
> > > > and leaves the leader with a NULL mm pointer.
> > > > 
> > > > Also converts cpuset and memcontrol to take the flex_array and iterate down it
> > > > until an mm is found, instead of just attempting to use the leader's mm.
> > > > 
> > > > Signed-off-by: Ben Blum <bblum@andrew.cmu.edu>
> > > 
> > > I think there are patches from Tejun that handle that did that already?
> > > 
> > > https://lkml.org/lkml/2011/8/23/418
> > > 
> > 
> > Yeah, I'm not hoping to preempt them or anything; I'm just presenting
> > this alternate approach, since (1) this way addresses the ss->attach()
> > problem directly without needing to add locking anywhere else, while the
> > focus of the discussion around Tejun's patches seems to have moved to
> > more global concerns, and (2) these patches and his patches ought to be
> > compatible with each other.
> 
> though, I did wonder if, given this approach, I shouldn't just get rid
> of attach_task and can_attach_task and use the flex_array for all of
> the calls. would that be nicer?

That's what Tejun's patches do.