linux-kernel - Re: [patch 1/3] mm: vmscan: fix numa reclaim balance problem in kswapd

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 31 Jul 2013 08:43:50 -0400
From:	Johannes Weiner <hannes@...xchg.org>
To:	Andrew Morton <akpm@...ux-foundation.org>
Cc:	Mel Gorman <mgorman@...e.de>, Rik van Riel <riel@...hat.com>,
	Andrea Arcangeli <aarcange@...hat.com>, linux-mm@...ck.org,
	linux-kernel@...r.kernel.org
Subject: Re: [patch 1/3] mm: vmscan: fix numa reclaim balance problem in
 kswapd

On Tue, Jul 30, 2013 at 01:45:39PM -0400, Johannes Weiner wrote:
> On Fri, Jul 26, 2013 at 03:53:19PM -0700, Andrew Morton wrote:
> > On Fri, 19 Jul 2013 16:55:23 -0400 Johannes Weiner <hannes@...xchg.org> wrote:
> > 
> > > When the page allocator fails to get a page from all zones in its
> > > given zonelist, it wakes up the per-node kswapds for all zones that
> > > are at their low watermark.
> > > 
> > > However, with a system under load and the free page counters being
> > > per-cpu approximations, the observed counter value in a zone can
> > > fluctuate enough that the allocation fails but the kswapd wakeup is
> > > also skipped while the zone is still really close to the low
> > > watermark.
> > > 
> > > When one node misses a wakeup like this, it won't be aged before all
> > > the other node's zones are down to their low watermarks again.  And
> > > skipping a full aging cycle is an obvious fairness problem.
> > > 
> > > Kswapd runs until the high watermarks are restored, so it should also
> > > be woken when the high watermarks are not met.  This ages nodes more
> > > equally and creates a safety margin for the page counter fluctuation.
> > 
> > Well yes, but what guarantee is there that the per-cpu counter error
> > problem is reliably fixed?  AFAICT this patch "fixes" it because the
> > gap between the low and high watermarks happens to be larger than the
> > per-cpu counter fluctuation, yes?  If so, there are surely all sorts of
> > situations where it will break again.
> > 
> > To fix this reliably, we should be looking at constraining counter
> > batch sizes or performing a counter summation to get the more accurate
> > estimate?
> 
> Thinking about this some more, the per-cpu fluctuation appears to be a
> red herring in this case.  Kswapd wakeup uses the safe version of the
> watermark checking code.  A percpu inaccuracy would result in the
> allocator waking kswapd a little too early or too late, but once it
> decides to wake it, we are getting an accurate picture on all zones.
> The safe watermark checks are designed to catch percpu inaccuracies
> between the low and the min watermark, the distance between low and
> high is twice as big, so we should be good.
> 
> The fluctuation that makes individual zones miss full aging cycles
> comes from true free page variation under load when the NOT OK
> threshold for the allocator is the same value as the OK threshold for
> skipping kswapd wakeups.

Andrew, could you please update the patch to this new changelog with
the percpu counter drivel removed?  Thanks!

---
When the page allocator fails to get a page from all zones in its
given zonelist, it wakes up the per-node kswapds for all zones that
are at their low watermark.

However, with a system under load the free pages in a zone can
fluctuate enough that the allocation fails but the kswapd wakeup is
also skipped while the zone is still really close to the low
watermark.

When one node misses a wakeup like this, it won't be aged before all
the other node's zones are down to their low watermarks again.  And
skipping a full aging cycle is an obvious fairness problem.

Kswapd runs until the high watermarks are restored, so it should also
be woken when the high watermarks are not met.  This ages nodes more
equally and creates a safety margin for the page counter fluctuation.

By using zone_balanced(), it will now check, in addition to the
watermark, if compaction requires more order-0 pages to create a
higher order page.

Signed-off-by: Johannes Weiner <hannes@...xchg.org>
---
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/