On Fri, 2011-02-04 at 03:05 +0000, Prasad B wrote:> On Wed, 2010-11-24, Stefano Stabellini wrote:
>
>
> >> Another option would be to just use the pci backend. As I
understand it
>
> >> nowdays pci is available even in pv guests for pci device
passthrough,
>
> >> so in theory this should work fine for both pv and hvm
guests ...
>
>
> > For PV guests the pci bus is only available through the
pcifront driver,
> > it wouldn''t interact with qemu at all. But it should
work fine for PV on
>
> > HVM guests.
> We are exploring the usage of virtio in PV guests (for MIPS
> architecture)
Do you have a working port of Xen to MIPS processors ?
> and stumbled upon the above comment by Stefano. Is it inherently
> difficult to make PV guests talk to qemu (at least in the PCI context)
> or is it just that it has not been attempted ?
Just not attempted I think, you''d ''just'' need to add
some smarts to
pciback and qemu to get them to coordinate around mapping the emulated
PCI device through to the PV guest.
However, my thinking was that for PV guests it would be better to
implement a Xen virtio discovery+transport mechanism in terms of xenbus,
event channels and grant tables. AIUI these mechanisms are abstracted in
virtio and the PCI based scheme is just one possible implementation. At
least I thought this was one of the original design goals -- since PCI
has been, TTBOMK, the only actual implementation perhaps it didn''t
actually end up that way.
Ian.
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel