All of lore.kernel.org
 help / color / mirror / Atom feed
From: Johannes Weiner <hannes@cmpxchg.org>
To: Tejun Heo <tj@kernel.org>
Cc: akpm@linux-foundation.org, mhocko@kernel.org,
	cgroups@vger.kernel.org, linux-mm@kvack.org,
	vdavydov@parallels.com, kernel-team@fb.com
Subject: Re: [PATCH v3 2/2] memcg: punt high overage reclaim to return-to-userland path
Date: Tue, 15 Sep 2015 09:47:24 +0200	[thread overview]
Message-ID: <20150915074724.GE2858@cmpxchg.org> (raw)
In-Reply-To: <20150913190008.GB25369@htj.duckdns.org>

On Sun, Sep 13, 2015 at 03:00:08PM -0400, Tejun Heo wrote:
> Currently, try_charge() tries to reclaim memory synchronously when the
> high limit is breached; however, if the allocation doesn't have
> __GFP_WAIT, synchronous reclaim is skipped.  If a process performs
> only speculative allocations, it can blow way past the high limit.
> This is actually easily reproducible by simply doing "find /".
> slab/slub allocator tries speculative allocations first, so as long as
> there's memory which can be consumed without blocking, it can keep
> allocating memory regardless of the high limit.
> 
> This patch makes try_charge() always punt the over-high reclaim to the
> return-to-userland path.  If try_charge() detects that high limit is
> breached, it adds the overage to current->memcg_nr_pages_over_high and
> schedules execution of mem_cgroup_handle_over_high() which performs
> synchronous reclaim from the return-to-userland path.

Why can't we simply fail NOWAIT allocations when the high limit is
breached? We do the same for the max limit.

As I see it, NOWAIT allocations are speculative attempts on available
memory. We should be able to just fail them and have somebody that is
allowed to reclaim try again, just like with the max limit.

--
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>

WARNING: multiple messages have this Message-ID (diff)
From: Johannes Weiner <hannes-druUgvl0LCNAfugRpC6u6w@public.gmane.org>
To: Tejun Heo <tj-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
Cc: akpm-de/tnXTf+JLsfHDXvbKv3WD2FQJk+8+b@public.gmane.org,
	mhocko-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org,
	cgroups-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
	linux-mm-Bw31MaZKKs3YtjvyW6yDsg@public.gmane.org,
	vdavydov-bzQdu9zFT3WakBO8gow8eQ@public.gmane.org,
	kernel-team-b10kYP2dOMg@public.gmane.org
Subject: Re: [PATCH v3 2/2] memcg: punt high overage reclaim to return-to-userland path
Date: Tue, 15 Sep 2015 09:47:24 +0200	[thread overview]
Message-ID: <20150915074724.GE2858@cmpxchg.org> (raw)
In-Reply-To: <20150913190008.GB25369-piEFEHQLUPpN0TnZuCh8vA@public.gmane.org>

On Sun, Sep 13, 2015 at 03:00:08PM -0400, Tejun Heo wrote:
> Currently, try_charge() tries to reclaim memory synchronously when the
> high limit is breached; however, if the allocation doesn't have
> __GFP_WAIT, synchronous reclaim is skipped.  If a process performs
> only speculative allocations, it can blow way past the high limit.
> This is actually easily reproducible by simply doing "find /".
> slab/slub allocator tries speculative allocations first, so as long as
> there's memory which can be consumed without blocking, it can keep
> allocating memory regardless of the high limit.
> 
> This patch makes try_charge() always punt the over-high reclaim to the
> return-to-userland path.  If try_charge() detects that high limit is
> breached, it adds the overage to current->memcg_nr_pages_over_high and
> schedules execution of mem_cgroup_handle_over_high() which performs
> synchronous reclaim from the return-to-userland path.

Why can't we simply fail NOWAIT allocations when the high limit is
breached? We do the same for the max limit.

As I see it, NOWAIT allocations are speculative attempts on available
memory. We should be able to just fail them and have somebody that is
allowed to reclaim try again, just like with the max limit.

  reply	other threads:[~2015-09-15  7:47 UTC|newest]

Thread overview: 43+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-09-13 18:59 [PATCH 1/2] memcg: flatten task_struct->memcg_oom Tejun Heo
2015-09-13 18:59 ` Tejun Heo
2015-09-13 19:00 ` [PATCH v3 2/2] memcg: punt high overage reclaim to return-to-userland path Tejun Heo
2015-09-15  7:47   ` Johannes Weiner [this message]
2015-09-15  7:47     ` Johannes Weiner
2015-09-15 15:53     ` Tejun Heo
2015-09-15 16:12       ` Johannes Weiner
2015-09-15 16:22         ` Tejun Heo
2015-09-15 16:22           ` Tejun Heo
2015-09-15 16:33           ` Johannes Weiner
2015-09-15 16:33             ` Johannes Weiner
2015-09-15  7:37 ` [PATCH 1/2] memcg: flatten task_struct->memcg_oom Johannes Weiner
2015-09-15  7:37   ` Johannes Weiner
2015-09-20 14:45 ` Sasha Levin
2015-09-20 14:45   ` Sasha Levin
2015-09-21 20:01   ` Tejun Heo
2015-09-21 20:01     ` Tejun Heo
2015-09-30 18:54     ` Tejun Heo
2015-09-30 18:54       ` Tejun Heo
2015-11-25 14:43     ` Peter Zijlstra
2015-11-25 14:43       ` Peter Zijlstra
2015-11-25 15:02       ` Peter Zijlstra
2015-11-25 15:02         ` Peter Zijlstra
2015-11-25 15:31         ` Andrey Ryabinin
2015-11-25 17:34           ` Dmitry Vyukov
2015-11-25 17:34             ` Dmitry Vyukov
2015-11-25 17:44           ` Peter Zijlstra
2015-12-11 16:25             ` Tejun Heo
2015-12-11 16:25               ` Tejun Heo
2015-12-15 19:22               ` Peter Zijlstra
2015-12-30  9:23                 ` [PATCH v4.4-rc7] sched: isolate task_struct bitfields according to synchronization domains Tejun Heo
2015-12-30 20:10                   ` Linus Torvalds
2015-12-30 20:10                     ` Linus Torvalds
2015-12-30 20:17                     ` Linus Torvalds
2015-12-30 20:41                     ` Tejun Heo
2015-12-30 20:41                       ` Tejun Heo
2015-12-30 20:43                       ` Linus Torvalds
2016-01-01  2:56                     ` [PATCH v4.4-rc7] sched: move sched lock synchronized bitfields in task_struct into ->atomic_flags Tejun Heo
2016-01-01  2:56                       ` Tejun Heo
2016-01-06 13:44                       ` Tejun Heo
2016-01-06 13:44                         ` Tejun Heo
2016-01-06 18:48         ` [tip:sched/core] sched/core: Fix unserialized r-m-w scribbling stuff tip-bot for Peter Zijlstra
2016-01-06 20:17           ` Tejun Heo

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=20150915074724.GE2858@cmpxchg.org \
    --to=hannes@cmpxchg.org \
    --cc=akpm@linux-foundation.org \
    --cc=cgroups@vger.kernel.org \
    --cc=kernel-team@fb.com \
    --cc=linux-mm@kvack.org \
    --cc=mhocko@kernel.org \
    --cc=tj@kernel.org \
    --cc=vdavydov@parallels.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.