From mboxrd@z Thu Jan 1 00:00:00 1970 From: Glauber Costa Subject: Re: [PATCH v2 06/11] memcg: kmem controller infrastructure Date: Wed, 15 Aug 2012 13:18:14 +0400 Message-ID: <502B6956.5030508@parallels.com> References: <1344517279-30646-1-git-send-email-glommer@parallels.com> <1344517279-30646-7-git-send-email-glommer@parallels.com> <50254475.4000201@jp.fujitsu.com> <5028BA9E.7000302@parallels.com> Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: Sender: cgroups-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org List-ID: Content-Type: text/plain; charset="us-ascii" To: Greg Thelen Cc: Kamezawa Hiroyuki , linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-mm-Bw31MaZKKs3YtjvyW6yDsg@public.gmane.org, cgroups-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, devel-GEFAQzZX7r8dnm+yROfE0A@public.gmane.org, Michal Hocko , Johannes Weiner , Andrew Morton , Christoph Lameter , David Rientjes , Pekka Enberg , Pekka Enberg On 08/14/2012 10:58 PM, Greg Thelen wrote: > On Mon, Aug 13 2012, Glauber Costa wrote: > >>>>> + WARN_ON(mem_cgroup_is_root(memcg)); >>>>> + size = (1 << order) << PAGE_SHIFT; >>>>> + memcg_uncharge_kmem(memcg, size); >>>>> + mem_cgroup_put(memcg); >>> Why do we need ref-counting here ? kmem res_counter cannot work as >>> reference ? >> This is of course the pair of the mem_cgroup_get() you commented on >> earlier. If we need one, we need the other. If we don't need one, we >> don't need the other =) >> >> The guarantee we're trying to give here is that the memcg structure will >> stay around while there are dangling charges to kmem, that we decided >> not to move (remember: moving it for the stack is simple, for the slab >> is very complicated and ill-defined, and I believe it is better to treat >> all kmem equally here) > > By keeping memcg structures hanging around until the last referring kmem > page is uncharged do such zombie memcg each consume a css_id and thus > put pressure on the 64k css_id space? I imagine in pathological cases > this would prevent creation of new cgroups until these zombies are > dereferenced. Yes, but although this patch makes it more likely, it doesn't introduce that. If the tasks, for instance, grab a reference to the cgroup dentry in the filesystem (like their CWD, etc), they will also keep the cgroup around. > Is there any way to see how much kmem such zombie memcg are consuming? > I think we could find these with > for_each_mem_cgroup_tree(root_mem_cgroup). Yes, just need an interface for that. But I think it is something that can be addressed orthogonaly to this work, in a separate patch, not as some fundamental limitation. > Basically, I'm wanting to > know where kernel memory has been allocated. For live memcg, an admin > can cat memory.kmem.usage_in_bytes. But for zombie memcg, I'm not sure > how to get this info. It looks like the root_mem_cgroup > memory.kmem.usage_in_bytes is not hierarchically charged. > Not sure what you mean by not being hierarchically charged. It should be, when use_hierarchy = 1. As a matter of fact, I just tested it, and I do see kmem being charged all the way to the root cgroup when hierarchy is used. (we just can't limit it there)