Re: [virtio-dev] Re: [virtio] [PATCH v7 08/11] packed virtqueues: more efficient virtqueue layout

[Cornelia Huck <cohuck@redhat.com>]

On Mon, 5 Feb 2018 20:21:12 +0200
"Michael S. Tsirkin" <mst@redhat.com> wrote:

> On Mon, Feb 05, 2018 at 07:16:54PM +0100, Cornelia Huck wrote:
> > > I agree that this doesn't seem optimal.  This is a much bigger change
> > > than anything between virtio 0.9 and in virtio 1.0, because it affects
> > > the data path directly.  
> > 
> > Nod. It looks in pretty good shape to me, once we fixed up the things I
> > noticed during this round, but I still feel a bit uncomfortable without
> > a prototype for non-pci.
> > 
> > One issue for me is that all work has been done with pci as the
> > transport, and with a view as to what could be helpful for implementers
> > of pci cards. There's nothing inherently bad in that, but it does
> > introduce a chance that other transports may run into problems when
> > they try to implement it.
> > 
> > Not an insurmountable problem, but something that should be kept in
> > mind.  
> 
> What do you suggest then?
> 

Do at least a quick qemu prototype. That gives you some idea where
things won't work, as it's the only way to cover all transports without
having to be aware of architecture details.

[The real solution is to have people with enough time, but that's a
pipe dream.]