From: Johannes Weiner <hannes@cmpxchg.org>
To: Konstantin Khlebnikov <khlebnikov@openvz.org>
Cc: linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>,
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>,
cgroups@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/2] memcg: fix page_referencies cgroup filter on global reclaim
Date: Tue, 21 Feb 2012 11:46:22 +0100 [thread overview]
Message-ID: <20120221104622.GB1676@cmpxchg.org> (raw)
In-Reply-To: <20120215162830.13902.60256.stgit@zurg>
On Wed, Feb 15, 2012 at 08:28:30PM +0400, Konstantin Khlebnikov wrote:
> Global memory reclaimer should't skip referencies for any pages,
> even if they are shared between different cgroups.
Agreed: if we reclaim from one memcg because of its limit, we want to
reclaim those pages that this group is not using. If it's used by
someone else, it should be evicted and refaulted by the group that
needs it.
If we reclaim globally, all references are "true" because we want to
evict those pages that are not used by any cgroup.
But if we reclaim a hierarchical subgroup, we don't want to evict
pages that are shared among this hierarchy, either, even if the memcg
that has the page charged to it is not using it. Bouncing the page
around the hierarchy is not sensible, because it does not solve the
problem of the parent hitting its limit when the sibling group will
refault it in a blink of an eye. It should only be evicted if the
memcg that's not using it nears its own limit, because only in that
case would reclaiming the page remedy the situation.
> This patch adds scan_control->current_mem_cgroup, which points to currently
> shrinking sub-cgroup in hierarchy, at global reclaim it always NULL.
So to be consistent, I'm wondering if we should pass
sc->target_mem_cgroup - the limit-hitting hierarchy root - to
page_referenced() and then have mm_match_cgroup() do a
mem_cgroup_same_or_subtree() check to see if the vma is in the
hierarchy rooted at sc->target_mem_cgroup.
Global reclaim is handled automatically, because mm_match_cgroup() is
not checked when the passed memcg is NULL, which sc->target_mem_cgroup
is for global reclaim.
--
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/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
next prev parent reply other threads:[~2012-02-21 10:46 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-02-15 16:28 Konstantin Khlebnikov
2012-02-15 16:28 ` [PATCH 2/2] memcg: use vm_swappiness from current memcg Konstantin Khlebnikov
2012-02-16 0:15 ` KAMEZAWA Hiroyuki
2012-02-16 7:02 ` Konstantin Khlebnikov
2012-02-16 0:09 ` [PATCH 1/2] memcg: fix page_referencies cgroup filter on global reclaim KAMEZAWA Hiroyuki
2012-02-21 10:46 ` Johannes Weiner [this message]
2012-02-21 11:38 ` Konstantin Khlebnikov
2012-02-21 13:11 ` Johannes Weiner
2012-02-23 13:57 ` Konstantin Khlebnikov
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=20120221104622.GB1676@cmpxchg.org \
--to=hannes@cmpxchg.org \
--cc=akpm@linux-foundation.org \
--cc=cgroups@vger.kernel.org \
--cc=kamezawa.hiroyu@jp.fujitsu.com \
--cc=khlebnikov@openvz.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
/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