[RndTbl] system loses time

[Gilles Detillieux]

On 12/11/2012 02:35 PM, Trevor Cordes wrote:
> On 2012-12-10 Trevor Cordes wrote:
>> I have another system (not the same one with RTC problems).  It loses
>> about 1s every 1m in system time!!
> I think I solved my own problem, after spending another hour on it :-(
>
> A clue that I hadn't noticed (or noted in my post) is that ntpstat
> showed "unsynchronised" instead of the normal:
> synchronised to NTP server (208.73.56.29) at stratum 3
>
> Also, the ntpq output indicated in its own obtuse way that it wasn't
> sync'd because there was no server with a * to the left of it.
>
> No matter what I did, I couldn't get ntp to show "synchronised".  I
> even would run ntpdate immediately followed by service ntp start and
> it wouldn't sync.  And I confirmed ntpdate was indeed setting the
> clock perfectly.  It made no sense!  I knew it was *not* a ntp config
> error since it's config'd the same way as 20 other good boxes.
>
> Anyhow, I started playing with kernel time sources and noticed this CPU
> (Petnium D) supports HPET.
>
> cat /sys/devices/system/clocksource/clocksource0/available_clocksource
> tsc hpet acpi_pm
>
> Current setting was tsc
>
> So I changed it to hpet
> echo hpet>>  /sys/devices/system/clocksource/clocksource0/current_clocksource
>
> Then ntpdate again
>
> Then start up ntpd
>
> BOOM!  Instantly ntpd syncs to a source and ntpstat shows sync'd!
> WTF?  Oh well, at least it seems fixed.  Been 10 mins now and no time
> loss.
>
> To get grub2 to change the kernel line so my hpet change survives
> reboot was another story... Grub2 looks nothing like grub!  Can you say
> "over-engineered"?

Rather than changing the grub2 configuration, I think you can make that 
clocksource change permanent by adding the following line to 
/etc/sysctl.conf:

devices.system.clocksource.clocksource0.current_clocksource = "hpet"

There's a similar issue, apparently, under Xen, as described in this post:

http://forums.epicgames.com/threads/629267-Negative-Delta-Time-Linux-Xen-Problem-SOLVED

-- 
Gilles R. Detillieux              E-mail:<grdetil at scrc.umanitoba.ca>
Spinal Cord Research Centre       WWW:    http://www.scrc.umanitoba.ca/
Dept. Physiology, U. of Manitoba  Winnipeg, MB  R3E 0J9  (Canada)