Just thought I'd follow up on this. I'm still not positive
but it appears that the problem I'm seeing is a kernel-build
process problem. It appears that on some kernels, if I
build with:
make -j4; make -j4 modules modules_install; make install
then the resultant kernel (often) fails and if I build with:
make; make modules; make modules_install; make install
then the resultant kernel boots.
I reproduced this, but don't really have time to track down what
the failure is. I usually build in a VNC console window and don't
redirect output of make so it's possible I've gotten error messages
that I've missed/ignored. Maybe it occurs only with Xen modules
because otherwise I'm sure some Linux developer would have
seen it. Since the symptom is that the kernel fails looking
for a root disk using exactly the same parameters and same
disk that work fine for a different kernel, it may have
something to do with xenblk.
Weird but apparently true. And now that I have 2.6.34 working
in both 32-bit and 64-bit PV, I really have to quit looking into
this now.
Dan
> -----Original Message-----
> From: Dan Magenheimer
> Sent: Friday, May 14, 2010 12:11 PM
> To: Pasi Kärkkäinen
> Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
> Subject: RE: [Xen-devel] need known working 64-bit upstream kernel
> config file for PV guest
>
> > > To add one more very interesting piece of data,
> > > booting a 2.6.32 32-bit kernel.org kernel in a PV
> > > guest works, while booting a 2.6.32.13 32-bit
> > > kernel.org kernel doesn't... same config file!
> > > And the failure looks very similar to the one
> > > I've been seeing on several x86_64 kernel.org
> > > kernels.
> > >
> > > I'll confirm that I can reproduce this, and then
> > > post as a separate thread.
> > >
> >
> > I tried both 32bit and 64bit Linux 2.6.32.9 upstream
> > kernel.org kernels, and they worked ok for me.
> >
> > That was actually save/restore testing, which also worked OK,
> > for both uni-vcpu and multi-vcpu guests.
> >
> > Dom0 was EL5.
>
> Well... I couldn't reproduce it and, as a result, I discovered
> another difference: 2.6.32 x86 fails also (as well as 2.6.32.9
> x86), when I am booting it in EL5u5. But it succeeds in EL5u4!
> Perhaps there is a different nash version between the two?
>
> Sorry if this appears terribly disorganized and disjointed.
> I am trying to use two machines with not quite identical
> environments so that I can get two other things done during
> long kernel compiles. As a result, I've missed some seemingly
> irrelevant details that are turning out to be relevant.
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home