[Xen-devel] PV domain breakage in xen-unstable tip?

[Dan Magenheimer <dan.magenheimer@xxxxxxxxxx>]

I just updated to latest tip (20943) of all hypervisor/tools/dom0 for
the first time in quite awhile and I'm seeing some strange behavior:

Sometimes, but not always, booting a PV guest that has booted
dozens of times successfully in my previous hv/tools/dom0,
fails to startup properly, seeming to hang at:

"Determining IP information for eth0..."
 
Sometimes I've seen the guest eventually continue (with no networking
enabled), other times it would crash and reboot.  Setting
"on_crash" to preserve, I took an "xm core-dump", but got a couple hundred

(XEN) mm.c:797:d0 Non-privileged (6) attempt to map I/O space ffffffff

on the console.

With some detective work, I think I discovered the circumstances
of the guest crash... Dom0 doesn't have a network driver for
my box so I have to modprobe it manually after building a new
dom0.  If I manually bring up networking and then launch a guest,
that guest will crash.  If I reboot so that dom0 boots with
networking -- or manually stop/start xend -- the guest boots fine.
I'm not absolutely positive this is new behavior, but I've been
using this box for over two years with the same driver-missing-from-dom0
problem and have never seen a guest crash this way before.

It's also possible this has nothing to do with networking,
just that bringing up networking in a domU with no networking
alive in dom0 uncovers another problem.

And clearly the use-model that uncovers this problem is
bogus... networking in dom0 is in a half-up state until
xend has been restarted... but still it shouldn't cause
a guest to crash, should it?

So I thought it best to report it in case others have seen
similar kinds of issues/crashes.

Dan

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel