[Fedora-xen] Re: Four ways RH could help with xen-ia64

[Juan Quintela]

Hi Aron

> I posted a message a while back containing a method for absolutely
> matching the hypervisor to the kernel patch:
> 
> https://www.redhat.com/archives/fedora-xen/2006-July/msg00024.html
> 
> Is that approximately what you're doing now?

No :)

My experience is that HV <-> dom0 is too unstable.  Then basically I
update the hypervisor when I found that i386/x86_64 domU kernel don't
work.  Or somebody complains that HVM has become too unstable and that
using unstable HV fixes it.

With the state of flush that is xen at this moment, I preffer that
approach.  But feel free to ask me to update the HV each time that gets
broken for you :)

> > > 	- old domU should run on new hypervisor
> > 
> > Agreed.  I normally test plain fc5 domU on all my new kernels.
> > 
> > > 	- new domU NOT guaranteed to run on old hypervisor
> > 
> > We have found that lately this "normally" works, versus bugs.
> 
> I'm sorry, your response confuses me here.  :-( 

In the past, no guarantees that not matching dom0 <->HV will work.  And
it failed quite a bit.

Lately (for 3.0.2 more or less), HV and dom0 try to agree on a minimum
set of capabilities.  And it has worked "reasonablely well".  I have
seen a couple of bugs in the compatibility, but they were fixed quite
fast.

> Just to be clear: These three bullet points I posted are the stated
> goals of Xen upstream.  Fedora should assume that a new domU will NOT
> run on an old hypervisor, regardless of empirical evidence...

For FC6, this is not a problem.  But for the future, this is a _big_
problem.  When FC7 is about to be out, it is reasonable for fedora users
to expect that installing  a new domU will lets them to test fc7 domU on
its existing fc6 dom0.  My understanding is that XenSource expect that
dom0 <-> domU interface to stay stable.  HV <-> domU interface to be a
"we will not broke on purpose", but no guarntees.

Later, Juan.