From: "Liang, Kan" <kan.liang@intel.com>
To: Linus Torvalds <torvalds@linux-foundation.org>,
        Andi Kleen <ak@linux.intel.com>
CC: Christopher Lameter <cl@linux.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Mel Gorman <mgorman@techsingularity.net>, Mel Gorman <mgorman@suse.de>,
        "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>,
        Tim Chen <tim.c.chen@linux.intel.com>, Ingo Molnar <mingo@elte.hu>,
        Andrew Morton <akpm@linux-foundation.org>,
        Johannes Weiner <hannes@cmpxchg.org>, Jan Kara <jack@suse.cz>,
        linux-mm <linux-mm@kvack.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: RE: [PATCH 1/2] sched/wait: Break up long wake list walk
Thread-Topic: [PATCH 1/2] sched/wait: Break up long wake list walk
Thread-Index: AQHTFWNBYSKZKyu5OE6Y+fM96SxNwqKEIDEAgASaOPD//398AIAAsKCA//+X4wCAAQZZgIAAplUA//+BiwAAFSW90P//ia2AgAASl4CAAAVjAIAEq10A//91k2D//XJ4kIAEpo4AgAANwwCAAAYbgIAAAe8AgAAZmgCAAAREAIAAGKGA//5uVCA=
Date: Wed, 23 Aug 2017 14:51:13 +0000
Message-ID: <37D7C6CF3E00A74B8858931C1DB2F0775378A8BB@SHSMSX103.ccr.corp.intel.com>
References: <20170818185455.qol3st2nynfa47yc@techsingularity.net>
 <CA+55aFwX0yrUPULrDxTWVCg5c6DKh-yCG84NXVxaptXNQ4O_kA@mail.gmail.com>
 <20170821183234.kzennaaw2zt2rbwz@techsingularity.net>
 <37D7C6CF3E00A74B8858931C1DB2F07753788B58@SHSMSX103.ccr.corp.intel.com>
 <37D7C6CF3E00A74B8858931C1DB2F0775378A24A@SHSMSX103.ccr.corp.intel.com>
 <CA+55aFy=4y0fq9nL2WR1x8vwzJrDOdv++r036LXpR=6Jx8jpzg@mail.gmail.com>
 <20170822190828.GO32112@worktop.programming.kicks-ass.net>
 <CA+55aFzPt401xpRzd6Qu-WuDNGneR_m7z25O=0YspNi+cLRb8w@mail.gmail.com>
 <20170822193714.GZ28715@tassilo.jf.intel.com>
 <alpine.DEB.2.20.1708221605220.18344@nuc-kabylake>
 <20170822212408.GC28715@tassilo.jf.intel.com>
 <CA+55aFw_-RmdWF6mPHonnqoJcMEmjhvjzcwp5OU7Uwzk3KPNmw@mail.gmail.com>
In-Reply-To: <CA+55aFw_-RmdWF6mPHonnqoJcMEmjhvjzcwp5OU7Uwzk3KPNmw@mail.gmail.com>
Accept-Language: zh-CN, en-US
Content-Language: en-US
dlp-product: dlpe-windows
dlp-version: 10.0.102.7
dlp-reaction: no-action
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Sender: linux-kernel-owner@vger.kernel.org
Content-Transfer-Encoding: 8bit
Content-Length: 2088
Lines: 51


> Subject: Re: [PATCH 1/2] sched/wait: Break up long wake list walk
> 
> On Tue, Aug 22, 2017 at 2:24 PM, Andi Kleen <ak@linux.intel.com> wrote:
> >
> > I believe in this case it's used by threads, so a reference count
> > limit wouldn't help.
> 
> For the first migration try, yes. But if it's some kind of "try and try again"
> pattern, the second time you try and there are people waiting for the page,
> the page count (not the map count) would be elevanted.
> 
> So it's possible that depending on exactly what the deeper problem is, the
> "this page is very busy, don't migrate" case might be discoverable, and the
> page count might be part of it.
> 
> However, after PeterZ made that comment that page migration should have
> that should_numa_migrate_memory() filter, I am looking at that
> mpol_misplaced() code.
> 
> And honestly, that MPOL_PREFERRED / MPOL_F_LOCAL case really looks like
> complete garbage to me.
> 
> It looks like garbage exactly because it says "always migrate to the current
> node", but that's crazy - if it's a group of threads all running together on the
> same VM, that obviously will just bounce the page around for absolute zero
> good ewason.
> 
> The *other* memory policies look fairly sane. They basically have a fairly
> well-defined preferred node for the policy (although the
> "MPOL_INTERLEAVE" looks wrong for a hugepage).  But
> MPOL_PREFERRED/MPOL_F_LOCAL really looks completely broken.
> 
> Maybe people expected that anybody who uses MPOL_F_LOCAL will also
> bind all threads to one single node?
> 
> Could we perhaps make that "MPOL_PREFERRED / MPOL_F_LOCAL" case just
> do the MPOL_F_MORON policy, which *does* use that "should I migrate to
> the local node" filter?
> 
> IOW, we've been looking at the waiters (because the problem shows up due
> to the excessive wait queues), but maybe the source of the problem comes
> from the numa balancing code just insanely bouncing pages back-and-forth if
> you use that "always balance to local node" thing.
> 
> Untested (as always) patch attached.

The patch doesn’t work.

Thanks,
Kan