Re: [PATCHSET cgroup/for-3.8] cpuset: decouple cpuset locking fromcgroup core

[Glauber Costa]

On 11/30/2012 01:49 PM, Michal Hocko wrote:
> On Fri 30-11-12 13:42:28, Glauber Costa wrote:
> [...]
>> Speaking of it: Tejun's tree still lacks the kmem bits. How hard would
>> it be for you to merge his branch into a temporary branch of your tree?
> 
> review-cpuset-locking is based on a post merge window merges so I cannot
> merge it as is. I could cherry-pick the series after it is settled. I
> have no idea how much conflicts this would bring, though.
> 
Ok.

I believe the task problem only exist for us for kmem. So I could come
up with a patchset that only deals with child cgroup creation, and
ignore attach for now. So long as we have a mechanism that will work for
it, and don't get lost and forget to patch it when the trees are merged.

Now, what I am actually seeing with cgroup creation, is that the
children will copy a lot of the values from the parent, like swappiness,
hierarchy, etc. Once the child copies it, we should no longer be able to
change those values in the parent: otherwise we'll get funny things like
parent.use_hierarchy = 1, child.use_hierarchy = 0.

One option is to take a global lock in memcg_alloc_css(), and keep it
locked until we did all the cgroup bookkeeping, and then unlock it in
css_online. But I am guessing Tejun won't like it very much.

What do you think about a children counter? If we are going to do things
similar to the attach_in_progress of cpuset, we might very well turn it
into a direct counter so we don't have to iterate at all.

The code would look like: (simplified example for use_hierarchy)

memcg_lock();
if (memcg->nr_children != 0)
    return -EINVAL;
else
    memcg->use_hierarchy = val
memcg_unlock()


--
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/