linux-kernel - Re: 5.11-rc4+git: Shortest NUMA path spans too many nodes

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]

Message-ID: <jhjmtx22uv7.mognet@arm.com>
Date:   Thu, 21 Jan 2021 15:05:32 +0000
From:   Valentin Schneider <valentin.schneider@....com>
To:     Meelis Roos <mroos@...ux.ee>, LKML <linux-kernel@...r.kernel.org>
Cc:     Peter Zijlstra <peterz@...radead.org>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        Barry Song <song.bao.hua@...ilicon.com>,
        Mel Gorman <mgorman@...e.de>,
        Dietmar Eggemann <dietmar.eggemann@....com>
Subject: Re: 5.11-rc4+git: Shortest NUMA path spans too many nodes

(+Cc relevant folks)

Hi,

On 21/01/21 15:41, Meelis Roos wrote:
> This happens on Sun Fire X4600 M2 - 32 cores in 8 CPU slots. 5.10 was silent. Current git and
> 5.10.0-13256-g5814bc2d4cc2 exhibit this message in dmesg but otherwise seem to work fine
> (kernel compilation succeeds).
>

b5b217346de8 ("sched/topology: Warn when NUMA diameter > 2") was added in
5.11-rc1, and I believe was marked for stable.

It doesn't come with a scheduler behaviour change, it only catches
topologies that end up being silently (unless run with SCHED_DEBUG=y)
misrepresented / misinterpreted by the scheduler.

Up until now I had only seen it fire on a single, somewhat unusual
topology. As fixing it is far from trivial, I figured adding this warning
would let us build a case for actually fixing it if we get some more
reports.

Could you paste the output of the below?

  $ cat /sys/devices/system/node/node*/distance

Additionally, booting your system with CONFIG_SCHED_DEBUG=y and
appending 'sched_debug' to your cmdline should yield some extra data.