amd64 HVM(+PV) domUs: losing 10+ minutes a day

[Hugo Silva]

Hello,

During my tests with 9.0, I have found another issue that I would like
to bring your attention to.


On all three 9.0-BETA1 virtual machines installed (on a NetBSD dom0,
should it matter) I'm encountering the infamous 'runtime went backwards'
messages (the VM's been up for days, it seems to do it only when there's
heavy activity in the system (not necessarily on the same VM, it seems)-
no such messages for 3 days now, even though the system is up and
services are running; still, clock time is being lost every day).

On that same machine, there are NetBSD (pv), OpenIndiana (pv) and
OpenBSD (hvm) vms - none of which has issues keeping track of time
(therefore I don't think it's a hvm problem (openbsd works) nor a
hardware/dom0 problem (dom0 + !freebsd_domU works)

I'm talking about 10-15 minutes a day lost, enough to confuse the
applications running inside the FreeBSD systems.

Eventually ntpd will refuse to adjust the time, saying:

ntpd[20176]: frequency error 2273 PPM exceeds tolerance 500 PPM

I haven't found how to counter this yet (other than stopping ntpd,
running ntpdate, and restarting ntpd), but in any case that seems to me
like sidestepping the real problem.

Haven't tested with HVM FreeBSD, but if it helps troubleshooting I'll
set something up..


Regards,

Hugo