Re: [patch v2 3/6] memcg: rework mem_cgroup_iter to use cgroupiterators

[Michal Hocko]

On Fri 07-12-12 09:12:25, Ying Han wrote:
> On Fri, Dec 7, 2012 at 12:58 AM, Michal Hocko <mhocko@xxxxxxx> wrote:
> > On Thu 06-12-12 19:43:52, Ying Han wrote:
> > [...]
> >> Forgot to mention, I was testing 3.7-rc6 with the two cgroup changes :
> >
> > Could you give a try to -mm tree as well. There are some changes for
> > memcgs removal in that tree which are not in Linus's tree.
> 
> I will give a try, which patchset you have in mind so i can double check?

Have a look at ba5e0e6be1c76fd37508b2825372b28a90a5b729 in my tree.
 
> I didn't find the place where the css pins the memcg, which either
> something i missed or not in place in my tree.

Yeah, it is carefully hidden ;).
css pins cgroup (last css_put will call dput on the cgroup dentry - via
css_dput_fn) and the last reference to memcg is dropped from ->destroy
callback (mem_cgroup_destroy) called from cgroup_diput.

> I twisted the patch a bit to make it closer to your v2 version,
> but instead keep the mem_cgroup_put() as well as using
> css_tryget(). Again, my test is happy with it:

This is really strange, there must be something weird with ref counting
going on.
Anyway, thanks for your testing! I will try to enahance my testing some
more next week.

> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> index f2eeee6..acec05a 100644
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -1004,6 +1004,7 @@ struct mem_cgroup *mem_cgroup_iter(struct
> mem_cgroup *root,
>                         if (prev && reclaim->generation != iter->generation) {
>                                 if (last_visited) {
>                                         css_put(&last_visited->css);
> +                                       mem_cgroup_put(last_visited);
>                                         iter->last_visited = NULL;
>                                 }
>                                 spin_unlock(&iter->iter_lock);
> @@ -1041,15 +1042,22 @@ struct mem_cgroup *mem_cgroup_iter(struct
> mem_cgroup *root,
>                 if (reclaim) {
>                         struct mem_cgroup *curr = memcg;
> 
> -                       if (last_visited)
> +                       if (last_visited) {
>                                 css_put(&last_visited->css);
> +                               mem_cgroup_put(last_visited);
> +                       }
> 
>                         if (css && !memcg)
>                                 curr = container_of(css, struct
> mem_cgroup, css);
> 
>                         /* make sure that the cached memcg is not removed */
> -                       if (curr)
> -                               css_get(&curr->css);
> +                       if (curr) {
> +                               mem_cgroup_get(curr);
> +                               if (!css_tryget(&curr->css)) {
> +                                       mem_cgroup_put(curr);
> +                                       curr = NULL;
> +                               }
> +                       }
>                         iter->last_visited = curr;
> 
>                         if (!css)
> 
> 
> --Ying
> 
> > --
> > Michal Hocko
> > SUSE Labs
> --
> 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/

-- 
Michal Hocko
SUSE Labs
--
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/