Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 31 May 2006 00:32:26 +0200
From:      Jeremie Le Hen <jeremie@le-hen.org>
To:        freebsd-current@FreeBSD.org
Subject:   Re: [fbsd] timer goes sluggish
Message-ID:  <20060530223226.GT1890@obiwan.tataz.chchile.org>
In-Reply-To: <20060529153103.GA17178@obiwan.tataz.chchile.org>
References:  <20060529153103.GA17178@obiwan.tataz.chchile.org>

next in thread | previous in thread | raw e-mail | index | archive | help
Hi,

On Mon, May 29, 2006 at 05:31:03PM +0200, Jeremie Le Hen wrote:
> Hi,
> 
> I won't write too much, the following session will tell much more:
> 
> % jarjarbinks:root# ntpdate -b chronos.cru.fr
> % 29 May 17:08:44 ntpdate[4254]: step time server 195.220.94.163 offset 0.000273 sec
> % jarjarbinks:root# sysctl kern.timecounter.hardware
> % kern.timecounter.hardware: ACPI-fast
> % jarjarbinks:root# while : ; do date ; sleep 1 ; done
> % Mon May 29 17:08:56 CEST 2006
> % Mon May 29 17:09:03 CEST 2006
> % Mon May 29 17:09:09 CEST 2006
> % Mon May 29 17:09:16 CEST 2006
> % Mon May 29 17:09:22 CEST 2006
> % ^C

Sometimes, I feel dumb.

I noticed that the clock sped up when the CPU makes heavy computing,
which meant there was some power saving on idleness.

I dug a bit I found that hw.acpi.cpu.cx_lowest=C4.
Setting it hw.acpi.cpu.cx_lowest=C1 makes the clock goes ``in time'' :).

> % jarjarbinks:root# ntpdate -b chronos.cru.fr
> % 29 May 17:09:32 ntpdate[4266]: step time server 195.220.94.163 offset 0.000937 sec
> % jarjarbinks:root# sysctl kern.timecounter.hardware=TSC
> % kern.timecounter.hardware: ACPI-fast -> TSC
> % jarjarbinks:root# while : ; do date ; sleep 1 ; done
> % Mon May 29 17:09:41 CEST 2006
> % Mon May 29 17:09:42 CEST 2006
> % Mon May 29 17:09:43 CEST 2006
> % Mon May 29 17:09:44 CEST 2006
> % Mon May 29 17:09:45 CEST 2006
> % ^C
> % jarjarbinks:root# ntpdate -b chronos.cru.fr
> % 29 May 17:10:16 ntpdate[4278]: step time server 195.220.94.163 offset 29.503343 sec

Likewise, we the above sysctl set, this behaviour disappears.  However,
I think humbly that this breaks POLA.  I am pretty confident there is
a good technical reason for this, but it is quite puzzling to have
the computer clock drifting lagging that much.


As an additional question, is there a reason to have
performance_cx_lowest="LOW" in etc/defaults/rc.conf ?


Thank you !
Best regards,
-- 
Jeremie Le Hen
< jeremie at le-hen dot org >< ttz at chchile dot org >



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20060530223226.GT1890>