RE: [Xen-ia64-devel] Time for hybrid virtualization?

[Alex Williamson <alex.williamson@xxxxxx>]

On Tue, 2008-01-22 at 10:29 +0100, tgingold@xxxxxxx wrote:
> Quoting "Xu, Anthony" <anthony.xu@xxxxxxxxx>:
> 
> > Hi Alex,
> >     I'm a little confused by hybrid mode for IA64.
> > If all priviledged instructions are emulated by VT,
> > Seems we don't need to modify linux kernel. Other modifications except
> > privileged instructions
> > are not performance critical. We can remove it.
> 
> Not sure we can avoid to virtualize iosapic for example.

  Right, plus we need some kind of hypercall and event channel
interfaces.  We also might find a number of things to optimistically
paravirtualize, for instance time keeping maybe.  I don't know exactly
what it would look like, but it might lead to minimal changes to the
core architecture code with modular interfaces to handle interfacing to
the hypervisor.  Anything that's currently handled by a machine vector
on current Linux/ia64 should be easy to paravirtualize if it's
worthwhile.

   BTW, has anyone else done more benchmarks to compare HVM and PV?  A
kernel build isn't necessarily a very good real world benchmark.
Thanks,

        Alex

-- 
Alex Williamson                             HP Open Source & Linux Org.


_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel