Date: Thu, 19 Mar 2009 17:44:27 -0700 (PDT) From: Won De Erick <won.derick@yahoo.com> To: Andre Albsmeier <andre@albsmeier.net>, Oliver Fromme <olli@lurza.secnetix.de>, Andriy Gapon <avg@icyb.net.ua> Cc: freebsd-hardware@freebsd.org Subject: Re: Hardware clock is not SYNC'ed with kernel clock by ntpdate? Message-ID: <402443.1503.qm@web45812.mail.sp1.yahoo.com>
next in thread | raw e-mail | index | archive | help
--- On Thu, 3/19/09, Andriy Gapon <avg@icyb.net.ua> wrote:
> on 19/03/2009 17:09 Andre Albsmeier said the following:
> > On Sat, 14-Feb-2009 at 19:38:58 +0100, Oliver Fromme
> > wrote:
> >> Won De Erick <won.derick@yahoo.com>
> >> wrote:
> >>
> >> ...
> >>
> >> > However, how should I make this
> >> > automatic, something that will update
> >> > the CMOS clock everytime the kernel
> >> > clock is syncronized with a NTP
> >> > server? Do I need to make changes on
> >> > the variables below?
> >>
> >> You seem to misunderstand. The CMOS clock
> >> _is_ always
> >> updated when you run ntpd. You do not have
> >> to change
> >> anything.
>
> Oliver, if this is not misquoted then you are not correct.
> Sorry, I noticed this thread too late, so I have to follow
> up in this fashion.
> RTC ("CMOS clock") is updated via resettodr function and
> you can search kernel
> sources yourself to see how many times and where it
> called.
> I reported the problem myself once upon a time:
> http://lists.freebsd.org/pipermail/freebsd-hackers/2005-August/013383.html
>
> > Hmm, my experience is different here. I have 17
> > FreeBSD machines
> > which are all running ntpd (btw, no
> > /etc/wall_cmos_clock). When
> > being rebooted, they sync the time with ntpdate before
> > starting
> > ntpd. Somtimes I noticed the time being stepped by
> > several seconds
> > (up to 50). I observed that the higher the previous
> > uptime had been,
> > the bigger the stepping during the following reboot
> > was.
> > While ntpd keeps the kernel time perfectly in sync
> > with our
> > stratum-1 server the CMOS clocks slowly drift away.
> > The higher
> > the uptime, the bigger the drift.
> >
> > I am now using this ugly hack to sync the kernel time
> > to the
> > CMOS clock (settimeofday() sets both) before reboot
> > and
> > the problem never appeared again:
> >
> > #include <sys/time.h>
> > #include <stdlib.h>
> > #include <err.h>
> > #include <sysexits.h>
> >
> > int main( void )
> > {
> > struct timeval tv;
> >
> > if( gettimeofday( &tv, NULL ) != 0 )
> > err( EX_DATAERR,
> > "gettimeofday" );
> >
> > if( settimeofday( &tv, NULL ) != 0 )
> > err( EX_DATAERR,
> > "settimeofday" );
> >
> > return 0;
> > }
> >
> >
> > A cleaner way might be to call resettodr() in
> > subr_clock.c
> > just before rebooting.
> >
> > If there are easier/cleaner ways to keep CMOS time in
> > sync,
> > I will be happy about any suggestions.
>
> I think that ideally RTC time should be updated when
> discrepancy between it and
> kernel time exceeds certain threshold.
> Also, we can could have a kernel self-rescheduling callout
> that would update
> hardware clock every half an hour (or whatever,
> configurable).
> Periodic - because we not always have a graceful shutdown
> (crashes, power loss, etc).
>
> For time beng I am using this hack:
> #save OS clock value to time-of-day register (x86 CMOS)
> 10 0-2,4-23
> * *
> *
> /usr/local/sbin/resettodr.sh
>
> $ cat /usr/local/sbin/resettodr.sh
> #!/bin/sh
>
> # ATTENTION: run this script only at time when you are sure
> that no
> # timezone change (daylight savings time) could possibly
> occur.
> # Otherwise you risk race with adjkerntz -e run from
> /etc/crontab.
> # Time between 06:00 and 23:59 should be safe.
>
>
> trigger_sysctl_name="machdep.adjkerntz"
> sysctl_command="/sbin/sysctl"
>
> $sysctl_command `$sysctl_command -e $trigger_sysctl_name`
> > /dev/null
>
> Calling this at shutdown would be a bonus.
>
> --
> Andriy Gapon
>
Cool!
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?402443.1503.qm>