Re: [RFC][PATCH 2/7][memcg] cgroup arbitarary ID allocation

[Vivek Goyal]

On Wed, Jul 28, 2010 at 11:35:29AM +0900, KAMEZAWA Hiroyuki wrote:
> On Tue, 27 Jul 2010 22:30:27 -0400
> Vivek Goyal <vgoyal@xxxxxxxxxx> wrote:
> 
> > > Index: mmotm-2.6.35-0719/Documentation/cgroups/cgroups.txt
> > > ===================================================================
> > > --- mmotm-2.6.35-0719.orig/Documentation/cgroups/cgroups.txt
> > > +++ mmotm-2.6.35-0719/Documentation/cgroups/cgroups.txt
> > > @@ -621,6 +621,15 @@ and root cgroup. Currently this will onl
> > >  the default hierarchy (which never has sub-cgroups) and a hierarchy
> > >  that is being created/destroyed (and hence has no sub-cgroups).
> > >  
> > > +void custom_id(struct cgroup_subsys *ss, struct cgroup *cgrp)
> > > +
> > > +Called at assigning a new ID to cgroup subsystem state struct. This
> > > +is called when ss->use_id == true. If this function is not provided,
> > > +a new ID is automatically assigned. If you enable ss->use_id,
> > > +you can use css_lookup()  and css_get_next() to access "css" objects
> > > +via IDs.
> > > +
> > 
> > Couple of lines to explain why a subsystem would like to assign its
> > own ids and not be happy with generic cgroup assigned id be helpful.
> > In this case, I think you are using this id as index into array
> > and want to control the index, hence you seem to be doing it.
> > 
> > But I am not sure again why do you want to control index?
> > 
> 
> Now, the subsystem allocation/id-allocation order is
> 
> 	->create()
> 	alloc_id.
> 
> Otherwise "id" of memory cgroup is just determined by the place in virtual-indexed
> array. 
> As
> 	memcg =	mem_cgroup_base + id
> 
> This "id" is determined at create().
> 
> If "id" is determined regardless of memory cgroup's placement, it's of no use.
> My original design of css_id() allocates id in create() but it was moved to
> generic part. So, this is expected change in my plan.
> 
> We have 2 choices.
> 	id = alloc_id()
> 	create(id)
> or
> 	this patch.
> 
> Both are okay for me. But alloc id before create() may add some ugly rollback.

Ok, so in current design at the time of mem_cgroup instantiation css_id
is not available so you don't know at what index to put the newly
instantiated mem_cgroup object, hence the notion of let subsys decide
the css_id and cgroup can query from subsystem later.

I don't have any preference. Anything simple works..

Vivek
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/