Am Sonntag, den 01.10.2006, 21:09 +0100 schrieb Ian Pratt:
> That's an old issue, not relevant on 3.0.3.
Well, it turns out that was only one way it could crash. I was able to
reproduces this several times. Most of the time I got a "bad page state"
followed by hitting a BUG in rmap.c, or things like that. Then, most of
the time one or two CPUs would lock up, and somewhat later the whole
system.
> > The machine is very similar, it's a machine with two dual-core
> opterons,
> > running one of the latest xen-3.0.3-unstable (20060926 hypervisor, and
> a
> > vanilla 2.6.18 + xen patch from Fedora from 20060915).
>
> Can you repro using the 2.6.16 kernel that came with 3.0.3 rather than
> the Fedora one? I suspect not.
Well, I can not reproduce these bugs, but the same test case is able to
kill the whole machine as well. CPU lockups on Dom0 or any DomU
(depending on where the load was) that spread to the other domains until
everything locks up. At some point the Dom0 even stops answering to
pings.
The only thing that still works is 3x Ctrl-A to get the message that the
serial console was switched, but even `h' didn't give the help text
anymore.
I would like to think that this is a memory problem, but the machine is
brand new and survived memtest86. And as long as I wasn't running
anything except Dom0, I was able to compile a whole Gentoo system four
hours, and once I started adding some DomU's the problems show up within
minutes.
The best way to reproduce this was to run a rsync on lots of files from
one DomU to another (via a bridge in Dom0, filesystems on exported
physical block devices) and start a compile job in any of the machines.
After 5-10 minutes boom.
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home