[Xen-devel] Re: [Qemu-devel] [PATCH 02/15] xen: Add xen_machine_fv

[Anthony Liguori <anthony@xxxxxxxxxxxxx>]

On 08/16/2010 09:13 AM, Kevin Wolf wrote:
> Am 16.08.2010 16:04, schrieb Stefano Stabellini:
>    
> > On Mon, 16 Aug 2010, Kevin Wolf wrote:
> >      
> > > Am 12.08.2010 16:09, schrieb stefano.stabellini@xxxxxxxxxxxxx:
> > >        
> > > > From: Anthony PERARD<anthony.perard@xxxxxxxxxx>
> > > >
> > > > Add the Xen FV (Fully Virtualized) machine to Qemu;
> > > > this is groundwork to add Xen device model support in Qemu.
> > > >
> > > > Signed-off-by: Anthony PERARD<anthony.perard@xxxxxxxxxx>
> > > > Signed-off-by: Stefano Stabellini<stefano.stabellini@xxxxxxxxxxxxx>
> > > >          
> > > Why does this need its own machine type? Shouldn't an HVM machine really
> > > look like a PC? And indeed most of this code looks like a (slightly
> > > outdated) copy of pc_piix.c with !pci_enabled code paths removed.
> > >        
> > The main reason is that we need some xen specific initializations, as you can
> > see from xen_init_fv.
> >      
> Right, there are some more Xen specific things added later. However, the
> main part of it is duplicated from pc_piix.c. I'm sure that with some
> refactoring we could call these functions instead of copying and
> modifying them. The problem with the latter is that they will inevitably
> diverge even for changes that make sense for both.
>
> I'm not even sure that the machine is the right place to do these Xen
> specific initializations (expect for the Xen platform PCI device). As
> far as I understand, the QEMUMachine is considered guest state whereas
> most of these initializations concern host state.
>    
To be honest, I think we'll need KVM, Xen, and QEMU specific machines.

The right default set of hardware for all three is different.

Going back to an old series of mine, they might share a MachineCore, but 
they'll ultimately need to be different machines.
Regards,

Anthony Liguori

>    
> > But considering that we have been asked to remove target-xen and that
> > will cause a major refactoring of this series, xen_machine_fv could
> > change significantly in the next iterations anyway...
> >      
> Basically, the request to remove target-xen and my comment both aim in
> the same direction, namely making Xen less special and integrate it
> better in existing structures.
>
> Kevin
>
>    

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