linux-kernel - Re: Decreasing stime running confuses top (was: top displaying 9999% CPU usage)

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: <20071003202446.GQ10199@1wt.eu>
Date:	Wed, 3 Oct 2007 22:24:46 +0200
From:	Willy Tarreau <w@....eu>
To:	Frans Pop <elendil@...net.nl>
Cc:	LKML <linux-kernel@...r.kernel.org>,
	Ilpo Järvinen <ilpo.jarvinen@...sinki.fi>,
	"Alexander E. Patrakov" <patrakov@....usu.ru>
Subject: Re: Decreasing stime running confuses top (was: top displaying 9999% CPU usage)

On Wed, Oct 03, 2007 at 09:27:41PM +0200, Frans Pop wrote:
> On Wednesday 03 October 2007, you wrote:
> > On Wed, 3 Oct 2007, Ilpo Järvinen wrote:
> > > On Wed, 3 Oct 2007, Frans Pop wrote:
> > > > The only change is in 2 consecutive columns: "2911 502" -> "2912
> > > > 500". Is processor usage calculated from those? Can someone explain
> > > > how?
> > >
> > > The latter seems to be utime ...decreasing. No wonder if arithmetics
> > > will give strange results (probably top is using unsigned delta?)...
> >
> > Hmm, minor miscounting from my side, stime seems more appropriate...
> 
> Here is a series showing utime and stime for kontact over 2 minutes.
> 
> Values were obtained using (identical values removed):
> $ while true; do awk '{print $14" "$15}' /proc/5269/stat; sleep 1; done | ts
> 
> Oct 03 21:17:12 12220 1593
> Oct 03 21:17:18 12221 1594
> Oct 03 21:17:26 12222 1593  <--
> Oct 03 21:17:34 12223 1594
> Oct 03 21:17:43 12224 1594
> Oct 03 21:17:51 12224 1595
> Oct 03 21:17:59 12225 1596
> Oct 03 21:18:07 12226 1595  <--
> Oct 03 21:18:15 12227 1596
> Oct 03 21:18:18 12228 1596
> Oct 03 21:18:22 12229 1595  <--
> Oct 03 21:18:31 12230 1596
> Oct 03 21:18:39 12230 1597
> Oct 03 21:18:44 12231 1597
> Oct 03 21:18:48 12232 1596  <--
> Oct 03 21:18:56 12233 1597
> Oct 03 21:19:04 12234 1596  <--
> Oct 03 21:19:11 12235 1597
> 
> So, is it normal that stime decreases sometimes or a kernel bug?
> /me expects the last...

Let me guess... Dual core AMD64 ?

I'm 99.99% sure that if you boot with "notsc", the problem disappears. If
so, you have one of those wonderful AMD64 with unsynced clock and without
HPET to sync with. I wrote a simple program in the past to exhibit the
problem. It would bsimply run "date +%s" in a busy loops and display each
time it would change. Amazing. It could jump back and forth by up to 3
seconds!

Basically, it looked like this :

old=$(date +%s)
while : ; do
   x=$(date +%s)
   if [ $x != $old ]; then
      echo "$old -> $x"
      old=$x
   fi
done
 
Regards,
Willy

-
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/