FW: [Planetlab-devel] RE: pl/xen

[David Irwin]

We'll send out a summary once it's actually working :-)  So it looks
like the most recent PL/Xen kernel uses copying and this seems to be
able to kexec and start eth0 properly.  However, it seems to have
introduced another bug----I keep getting messages similar to:

4gb seg fixup, process ntpd (pid 1028), cs:ip 73:0057ceda
printk: 169 messages suppressed.

I get messages like this for hotplug, udev, curl ntpdate, ntpd, udevd,
and PlanetLabConf.p, etc..  The boot hangs printing out these errors.

A Google for this error shows other FC5/Xen users having the same
problem, although currently their solutions haven't worked for me.
Luckily, though, this seems to be a FC5/Xen/PlanetLab problem and not
a kexec problem.

Maybe someone that knows more about PL, Xen, or Fedora has an idea?
We will keep you posted.

-David

On 11/22/06, Marc E. Fiuczynski <mef at cs.princeton.edu> wrote:
> Hi David^2,
>
> So Xen moved from page-flipping to copying between domains for network
> traffic?!  Interesting.
>
> Glad to hear that you got xenU kexec working.  Could you summarize what it
> was you needed to get it to work in the first place.
>
> Thanks,
> Marc
>
> p.s., for the benefit of others, please send email to devel rather than me
> directly.  There are things you are sharing with me that are useful to
> others on the list.  Thanks.
>
> > -----Original Message-----
> > From: deirwin at gmail.com [mailto:deirwin at gmail.com]On Behalf Of David
> > Irwin
> > Sent: Wednesday, November 22, 2006 11:03 AM
> > To: Marc E. Fiuczynski
> > Cc: David Becker; Jeff Chase
> > Subject: Re: FW: [Planetlab-devel] RE: pl/xen
> >
> >
> > Yea, can you send it?  I'd like to know how to build a new kernel.
> >
> > I actually did get the kernel you sent before kexec'ing correctly.
> > The only problem is I can't bring eth0 up on your kernel (trying to
> > bring it up crashes the VM).
> >
> > I think this may be due to the fact that my initial VM was set to use
> > copying (rather than page flipping) for moving packets b/t domU and
> > dom0.  The default for this was changed in xen-unstable sometime in
> > October according to the changelogs.  Your kernel is probably using
> > page-flipping (the old default).  I'm guessing the hypervisor may not
> > like changing that without rebooting.  eth0 seems to work on other
> > more recent kernels that appear to use copying.
> >
> > Thanks,
> > -David
> >
> > On 11/21/06, Marc E. Fiuczynski <mef at cs.princeton.edu> wrote:
> > > Doah... didn't send the build recipe.  Do you still need that?
> > >
> > > Marc
> > >
>
>