lkml.org 
[lkml]   [2012]   [Feb]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v3 02/21] memcg: make mm_match_cgroup() hirarchical
On Thu, Feb 23, 2012 at 11:46:22PM +0400, Konstantin Khlebnikov wrote:
> Johannes Weiner wrote:
> >On Thu, Feb 23, 2012 at 05:51:46PM +0400, Konstantin Khlebnikov wrote:
> >>@@ -821,6 +821,26 @@ struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
> >> struct mem_cgroup, css);
> >> }
> >>
> >>+/**
> >>+ * mm_match_cgroup - cgroup hierarchy mm membership test
> >>+ * @mm mm_struct to test
> >>+ * @cgroup target cgroup
> >>+ *
> >>+ * Returns true if mm belong this cgroup or any its child in hierarchy
> >>+ */
> >>+int mm_match_cgroup(const struct mm_struct *mm, const struct mem_cgroup *cgroup)
> >>+{
> >>+ struct mem_cgroup *memcg;
> >>+
> >>+ rcu_read_lock();
> >>+ memcg = mem_cgroup_from_task(rcu_dereference((mm)->owner));
> >>+ while (memcg != cgroup&& memcg&& memcg->use_hierarchy)
> >>+ memcg = parent_mem_cgroup(memcg);
> >>+ rcu_read_unlock();
> >>+
> >>+ return cgroup == memcg;
> >>+}
> >
> >Please don't duplicate mem_cgroup_same_or_subtree()'s functionality in
> >a worse way. The hierarchy information is kept in a stack such that
> >ancestry can be detected in linear time, check out css_is_ancestor().
>
> Ok, there will be something like that:
>
> +bool mm_match_cgroup(const struct mm_struct *mm,
> + const struct mem_cgroup *cgroup)
> +{
> + struct mem_cgroup *memcg;
> + bool ret;
> +
> + rcu_read_lock();
> + memcg = mem_cgroup_from_task(rcu_dereference((mm)->owner));
> + ret = memcg && mem_cgroup_same_or_subtree(cgroup, memcg);
> + rcu_read_unlock();
> +
> + return ret;
> +}
> +

It would be unfortunate to nest rcu_read_lock(), but I think this
looks good otherwise.

> >If you don't want to nest rcu_read_lock(), you could push the
> >rcu_read_lock() from css_is_ancestor() into its sole user and provide
> >a __mem_cgroup_is_ancestor() that assumes rcu already read-locked.
> >
> >No?
>
> It is not a problem.
>
> looks like mem_cgroup_same_or_subtree() check something different,
> because it does not check ->use_hierarchy flag on tested cgroup, only on target cgroup.

If a memcg has hierarchy enabled, any memcg that turns out to be its
child is guaranteed to have hierarchy enabled.

> Or just all this hierarchical stuff is out of sync in different parts of code.
> For example memcg_get_hierarchical_limit() start from deepest cgroup and go upper
> while ->use_hierarchy is set.

This one really has to walk up and find the smallest applicable limit,
there is no way around looking at every single level.

But checking for ancestry can and has been optimized.


\
 
 \ /
  Last update: 2012-02-23 23:09    [W:0.106 / U:6.892 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site