xen-devel.lists.xenproject.org archive mirror
 help / color / mirror / Atom feed
From: Juergen Gross <jgross@suse.com>
To: Steven Haigh <netwiz@crc.id.au>, xen-devel@lists.xenproject.org
Subject: Re: RFE: Detect NUMA misconfigurations and prevent machine freezes
Date: Wed, 29 Aug 2018 07:49:40 +0200	[thread overview]
Message-ID: <29fc38c6-8908-2f6f-e496-a1644f56c59e@suse.com> (raw)
In-Reply-To: <6564259.GaOmOO5kt2@wopr.lan.crc.id.au>

On 29/08/18 07:33, Steven Haigh wrote:
> When playing with NUMA support recently, I noticed a host would always hang 
> when trying to create a cpupool for the second NUMA node in the system.
> 
> I was using the following commands:
> # xl cpupool-create name=\"Pool-1\" sched=\"credit2\
> # xl cpupool-cpu-remove Pool-0 node:1
> # xl cpupool-cpu-add Pool-1 node:1
> 
> After the last command, the system would hang - requiring a hard reset of the 
> machine to fix.
> 
> I tried a different variation with the same result:
> # xl cpupool-create name=\"Pool-1\" sched=\"credit2\
> # xl cpupool-cpu-remove Pool-0 node:1
> # xl cpupool-cpu-add Pool-1 12
> 
> It turns out that the RAM was installed sub-optimally in this machine. A 
> partial output from 'xl info -n' shows:
> numa_info              :
> node:    memsize    memfree    distances
>   0:     67584      62608      10,21
>   1:             0              0      21,10
> 
> A machine where we could get this working every time shows:
> node:    memsize    memfree    distances
>   0:     34816      30483      10,21
>   1:     32768      32125      21,10
> 
> As we can deduce RAM misconfigurations in this scenario, I believe we should 
> check to ensure that RAM configuration / layout is sane *before* attempting to 
> split the system and print a warning.
> 
> This would prevent a hard system freeze in this scenario.

RAM placement should not matter here. As the name already suggests
cpupools do assignment of cpus. RAM allocated will be preferred taken
from a local node, but this shouldn't be mandatory for success.

Would it be possible to use a debug hypervisor (e.g. 4.12-unstable) for
generating a verbose log (hypervisor boot parameter "loglvl=all") and
sending the complete hypervisor log?


Juergen

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

  reply	other threads:[~2018-08-29  5:49 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-08-29  5:33 RFE: Detect NUMA misconfigurations and prevent machine freezes Steven Haigh
2018-08-29  5:49 ` Juergen Gross [this message]
2018-08-30  4:01   ` Steven Haigh
2018-08-30  7:13     ` Juergen Gross
2018-08-30  8:33     ` Jan Beulich
2018-08-30  8:49       ` BUG: sched=credit2 crashes system when using cpupools Steven Haigh
2018-09-12 15:11         ` Dario Faggioli
2018-09-12 15:27           ` Steven Haigh
2018-09-12 15:13     ` RFE: Detect NUMA misconfigurations and prevent machine freezes Dario Faggioli
2018-09-13 11:49       ` Dario Faggioli
2018-08-29  6:39 ` Jan Beulich

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=29fc38c6-8908-2f6f-e496-a1644f56c59e@suse.com \
    --to=jgross@suse.com \
    --cc=netwiz@crc.id.au \
    --cc=xen-devel@lists.xenproject.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;
as well as URLs for NNTP newsgroup(s).