Re: [Xen-devel] OASIS Virtual I/O Device (VIRTIO) TC

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

On Thu, 2013-08-15 at 15:24 +0100, Wei Liu wrote:
> On Thu, Aug 15, 2013 at 10:16:41AM -0400, Konrad Rzeszutek Wilk wrote:
> > On Thu, Aug 15, 2013 at 02:37:28PM +0100, Wei Liu wrote:
> > > On Tue, Aug 13, 2013 at 08:41:28PM +0200, Daniel Kiper wrote:
> > > > Hi,
> > > > 
> > > > I was appointed by Xen Advisory Board to OASIS Virtual I/O Device
> > > > (VIRTIO) TC as a memeber. I will oversee its work from Xen point
> > > > of view, however, deliverables will be as much as possible
> > > > "virtualization platform" agnostic.
> > > > 
> > > > According to [1]:
> > > > 
> > > > The goal of the OASIS Virtual I/O Device (VIRTIO) TC is to simplify
> > > > virtual devices, making them more extensible and more recognizable.
> > > > 
> > > > [...]
> > > > 
> > > > The TC intends to define formal specifications for virtual device buses
> > > > (including PCI) for a variety of devices, including network devices.
> > > > Specification development will be based upon the "Virtio PCI Card
> > > > Specification" [2] v0.9.5, seeking solutions that support portability,
> > > > simplicity, least-surprise for driver authors, extensibility, and
> > > > performance. The specification will also document existing
> > > > implementations and practice.
> > > > 
> > > 
> > > How is Xen PV driver fit in this? Xen PV doesn't have virtual PCI bus.
> > 
> > 
> 
> My limited knowledge of Virtio is pretty dated now, that's why I raised
> that question. ;-)
> 
> > One could implement an PCI over XenBus I suppose. Or just write an
> > XenBus PCI driver that would do all the neccessary things to respond
> > to the proper commands.
> > > 
> 
> Back in the date my impression was that XenBus is asynchronuous while
> virtual PCI is synchronuous (i.e. trap-process-return), they are not
> quite compatible.
> 
> Furtuer more using XenBus is mainly used for configuration, while Virtio
> over PCI supports both configuration and sending notifications etc.
> 
> I'm not sure XenBus PCI driver would be a good idea...

I thought this part of virtio was pluggable and that PCI was one option
(although for a long time the only one). On ARM you can also use
virt_mmio instead these days. A shared ring for cfg + evtchn model for
notification doesn't seem too far fetched to me.

A far bigger problem IMHO with virtio is that it basically discards the
Xen security model. I don't know how hard it will be to retrofit gnttab
based access control into the virtio protocols. A lot would be my
guess...

Ian.



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel