linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
From: Michal Hocko <mhocko@suse.cz>
To: Ying Han <yinghan@google.com>
Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org,
	KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>,
	Johannes Weiner <hannes@cmpxchg.org>,
	Tejun Heo <htejun@gmail.com>,
	Glauber Costa <glommer@parallels.com>,
	Li Zefan <lizefan@huawei.com>
Subject: Re: [patch v2 3/6] memcg: rework mem_cgroup_iter to use cgroup iterators
Date: Tue, 11 Dec 2012 19:10:37 +0100	[thread overview]
Message-ID: <20121211181037.GH1612@dhcp22.suse.cz> (raw)
In-Reply-To: <20121211161559.GF1612@dhcp22.suse.cz>

On Tue 11-12-12 17:15:59, Michal Hocko wrote:
> On Tue 11-12-12 16:50:25, Michal Hocko wrote:
> > On Sun 09-12-12 08:59:54, Ying Han wrote:
> > > On Mon, Nov 26, 2012 at 10:47 AM, Michal Hocko <mhocko@suse.cz> wrote:
> > [...]
> > > > +               /*
> > > > +                * Even if we found a group we have to make sure it is alive.
> > > > +                * css && !memcg means that the groups should be skipped and
> > > > +                * we should continue the tree walk.
> > > > +                * last_visited css is safe to use because it is protected by
> > > > +                * css_get and the tree walk is rcu safe.
> > > > +                */
> > > > +               if (css == &root->css || (css && css_tryget(css)))
> > > > +                       memcg = mem_cgroup_from_css(css);
> > > >
> > > >                 if (reclaim) {
> > > > -                       iter->position = id;
> > > > +                       struct mem_cgroup *curr = memcg;
> > > > +
> > > > +                       if (last_visited)
> > > > +                               css_put(&last_visited->css);
> > > > +
> > > > +                       if (css && !memcg)
> > > > +                               curr = mem_cgroup_from_css(css);
> > > 
> > > In this case, the css_tryget() failed which implies the css is on the
> > > way to be removed. (refcnt ==0) If so, why it is safe to call
> > > css_get() directly on it below? It seems not preventing the css to be
> > > removed by doing so.
> > 
> > Well, I do not remember exactly but I guess the code is meant to say
> > that we need to store a half-dead memcg because the loop has to be
> > retried. As we are under RCU hood it is just half dead.
> > Now that you brought this up I think this is not safe as well because
> > another thread could have seen the cached value while we tried to retry
> > and his RCU is not protecting the group anymore.
> 
> Hmm, thinking about it some more, it _is_ be safe in the end.
> 
> We are safe because we are under RCU.

And I've just realized that one sentence vanished while I was writing
this.

So either we retry (while(!memcg)) and see the half-dead memcg with a
valid cgroup because we are under rcu so cgroup iterator will find a
next one. Or we race with somebody else on the iterator and that is
described bellow.

> And even if somebody else looked
> at the half-dead memcg from iter->last_visited it cannot disappear
> because the current one will retry without dropping RCU so the grace
> period couldn't have been finished.
> 
> 		CPU0					CPU1
> rcu_read_lock()						rcu_read_lock()
> while(!memcg) {						while(!memcg)
> [...]
> spin_lock(&iter->iter_lock)
> [...]
> if (css == &root->css ||
> 		(css && css_tryget(css)))
> 	memcg = mem_cgroup_from_css(css)
> [...]
> if (css && !memcg)
> 	curr = mem_cgroup_from_css(css)
> if (curr)
> 	css_get(curr);
> spin_unlock(&iter->iter_lock)
> 							spin_lock(&iter->iter_lock)
> 							/* sees the half dead memcg but its cgroup is still valid */ 
> 							[...]
> 							spin_unlock(&iter->iter_lock)
> /* we do retry */
> }
> rcu_read_unlock()
> 
> so the css_get will just helps to prevent from further code obfuscation.
> 
> Makes sense? The code gets much simplified later in the series,
> fortunately.
> -- 
> Michal Hocko
> SUSE Labs
> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

-- 
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

  reply	other threads:[~2012-12-11 18:10 UTC|newest]

Thread overview: 57+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2012-11-26 18:47 rework mem_cgroup iterator Michal Hocko
2012-11-26 18:47 ` [patch v2 1/6] memcg: synchronize per-zone iterator access by a spinlock Michal Hocko
2012-11-26 18:47 ` [patch v2 2/6] memcg: keep prev's css alive for the whole mem_cgroup_iter Michal Hocko
2012-11-28  8:38   ` Kamezawa Hiroyuki
2012-11-26 18:47 ` [patch v2 3/6] memcg: rework mem_cgroup_iter to use cgroup iterators Michal Hocko
2012-11-28  8:47   ` Kamezawa Hiroyuki
2012-11-28  9:17     ` Michal Hocko
2012-11-28  9:23       ` Glauber Costa
2012-11-28  9:33         ` Michal Hocko
2012-11-28  9:35           ` Glauber Costa
2012-11-30  4:07   ` Kamezawa Hiroyuki
2012-12-07  3:39   ` Ying Han
2012-12-07  3:43     ` Ying Han
2012-12-07  8:58       ` Michal Hocko
2012-12-07 17:12         ` Ying Han
2012-12-07 17:27           ` Michal Hocko
2012-12-07 19:16             ` Ying Han
2012-12-07 19:35               ` Michal Hocko
2012-12-07  9:01     ` Michal Hocko
2012-12-09 16:59   ` Ying Han
2012-12-11 15:50     ` Michal Hocko
2012-12-11 16:15       ` Michal Hocko
2012-12-11 18:10         ` Michal Hocko [this message]
2012-12-11 22:43         ` Ying Han
2012-12-12  8:55           ` Michal Hocko
2012-12-12 17:57             ` Ying Han
2012-12-12 18:08               ` Michal Hocko
2012-12-11 22:31       ` Ying Han
2012-12-09 19:39   ` Ying Han
2012-12-11 15:54     ` Michal Hocko
2012-12-11 22:36       ` Ying Han
2012-12-12  9:06         ` Michal Hocko
2012-12-12 18:09           ` Ying Han
2012-12-12 18:34             ` Michal Hocko
2012-12-12 18:42               ` Michal Hocko
2012-12-14  1:06                 ` Ying Han
2012-12-14 10:56                   ` [PATCH] memcg,vmscan: do not break out targeted reclaim without reclaimed pages Michal Hocko
2012-12-12 19:24           ` [patch v2 3/6] memcg: rework mem_cgroup_iter to use cgroup iterators Michal Hocko
2012-12-14  1:14             ` Ying Han
2012-12-14 12:07               ` Michal Hocko
2012-12-14 23:08                 ` Ying Han
2012-12-14 12:37             ` Michal Hocko
2012-11-26 18:47 ` [patch v2 4/6] memcg: simplify mem_cgroup_iter Michal Hocko
2012-11-28  8:52   ` Kamezawa Hiroyuki
2012-11-30  4:09   ` Kamezawa Hiroyuki
2012-12-09 17:01   ` Ying Han
2012-12-11 15:57     ` Michal Hocko
2012-12-11  4:35   ` Ying Han
2012-12-11 16:01     ` Michal Hocko
2012-12-11 22:52       ` Ying Han
2012-11-26 18:47 ` [patch v2 5/6] memcg: further " Michal Hocko
2012-11-30  4:10   ` Kamezawa Hiroyuki
2012-11-30  9:08   ` Glauber Costa
2012-11-30 10:23     ` Michal Hocko
2012-11-26 18:47 ` [patch v2 6/6] cgroup: remove css_get_next Michal Hocko
2012-11-30  4:12   ` Kamezawa Hiroyuki
2012-11-30  8:18     ` Michal Hocko

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20121211181037.GH1612@dhcp22.suse.cz \
    --to=mhocko@suse.cz \
    --cc=glommer@parallels.com \
    --cc=hannes@cmpxchg.org \
    --cc=htejun@gmail.com \
    --cc=kamezawa.hiroyu@jp.fujitsu.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=lizefan@huawei.com \
    --cc=yinghan@google.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).