Recent qemu and timers issue

[Juergen Lock]

On Tue, Apr 07, 2009 at 11:37:37PM +0200, Juergen Lock wrote:
> In article <200904062254.37824.kalinoj1 at iem.pw.edu.pl> you write:
> >Dnia sobota 04 kwietnia 2009 o 00:23:29 Juergen Lock napisa=C5=82(a):
> >> In article <c948bb4de85d1b2a340ac63a7c46f6d9 at iem.pw.edu.pl> you write:
> >> >Hi all,
> >> >
> >> >When running FreeBSD 8-CURRENT system inside a qemu machine (from the
> >> >latest port by nox at freebsd.org) I have strange timer problems in the
> >> >virtual machine.
> >>
> >> What is the host running, also current?
> >
> >Yes, the host is also running current.=20
> >
> >But the same host has no problem with older FreeBSD kernels on the guest, e=
> >g.=20
> >=46reeSBIe 1.1 or FreeNAS (kernel  v.6.4)
> >
> OK I did some tests in the meantime on my newest box that has 4 cpus and
> is running 7-stable: (host)
> >>
> >> >My logs and console are full of messages like this:
> >> >calcru: runtime went backwards from 11248014678717988 usec to 119242 usec
> >> >for pid 60 (ps)
> >> >
> >> >calcru: runtime went backwards from 11248014678921829 usec to 2516605 us=
> >ec
> >> >for pid 60 (ps)
> >> >calcru: runtime went backwards from 7014844 usec to 6270851 usec for pid=
> > 0
> >> >(kernel)
> >> >
> >> >It even happens for some processes to die inside qemu guest.
> >>
> >>  Is this an smp host?  You could try forcing qemu onto one cpu by
> >> doing like:
> >> 	cpuset -l 0 qemu ...
> >>
> >
> >Yes, it's an smp host (8 cpus). But cpuset -l 0 qemu ... (also -clock=20
> >dynticks) didnt't help.
> >
>  Same here with an 8.0-HEAD-20090403-JPSNAP-i386 guest.  (OK I didn't see
> guest processes die, but I also didn't really load the guests so far...)
> 
>  A 7.1-RELENG_7-20090403-JPSNAP-i386 guest was not affected.
> 
> >> >I tried to use all possible timers using sysctl, where I have:
> >> >TSC(800) HPET(900) ACPI-safe(850) i8254(0) dummy(-1000000)
> >> >None of these helped.
> >>
> >>  On the host or in the guest?
> >>
> >
> >It's in the guest machine., where I tried these settings=20
> >
> >One important thing I forgot to mention in my initial posting:
> >When I run 'systat -vm 1' o the guest machine - it's almost fully loaded - =
> >and=20
> >80% (or more) of it's cpu time is spent in interrrupt handling! This is als=
> >o=20
> >not the case with older kernels.
> >
>  Hmm 'systat -vm 1' doesn't really load the 8.0 guest here.
> 
> >> [snip]
> >> >I would greatly appreciate any help or a hint - where should I look for
> >> >help on this issue.
> >>
> >>  If none of these help you could also try the qemu list:
> >> 	http://lists.nongnu.org/mailman/listinfo/qemu-devel
> >> (I think you can still post without being subscribed, altho in that case
> >> it may take a while for the post to get thru.)
> >>
> >>  HTH,
> >> 	Juergen
> >
> >At this moment it's very hard for me to judge if it's qemu or kernel issue..
> >Any thoughts?
> 
>  Yeah, could be one or the other.

I just noticed running any FreeBSD guest with bootverbose (boot -v,
or menu item 5 in the default beastie menu) gives me lots of these:
	t_delta 16.0109662fcd2712a0 too long
	t_delta 16.014a1099ee815ca0 too long
	t_delta 16.0108d9c2ecd404c0 too long
	t_delta 16.0108d032cb334b60 too long
	t_delta 15.f58306afb53419a0 too short
	t_delta 16.010a762538a38680 too long
etc.  My guess is this could be related...

	Juergen