Re: [Qemu-devel] [v2][RFC][PATCH] virtio: uniform virtio device IDs

[Michael S. Tsirkin]

On Wed, Feb 11, 2015 at 07:10:22PM +0100, Cornelia Huck wrote:
> On Wed, 11 Feb 2015 13:41:29 +0100
> "Michael S. Tsirkin" <address@hidden> wrote:
> 
> > On Mon, Feb 09, 2015 at 09:47:20AM +0100, Cornelia Huck wrote:
> 
> > > Well, we do need the changes in way more than two places, as every host
> > > or guest has to collect the definitions on its own, no?
> > 
> > This has nothing to do with host.
> 
> Hm, but all hosts and all guests need the ids, no?
> 
> > It's just using linux source as main basis for the file
> > simply because linux is a higher visibility project than qemu,
> > they won't borrow code from us but we can borrow from them.
> 
> It still seems restricting to use Linux as the ultimate source - and
> that has nothing to do with visibililty. I'd say pulling from any
> project is restricting: Why shouldn't we want to implement something in
> qemu that Linux is not (yet) interested in, but other guests are?

We can always add such a hypothetical feature in a qemu
specific headers.

But most things are shared, and manual duplication is bad.

> > > (Granted, with
> > > Linux and qemu you get most of the users; but it feels a bit strange
> > > for a host implementation to collect information from one of its
> > > guests. I really think that we should go back to the common root.
> > > Didn't we have a BSD-licenced header in the spec?)
> > 
> > virtio linux headers are also BSD licensed intentionally for this
> > purpose.
> >  * This header is BSD licensed so anyone can use the definitions to
> >  * implement compatible drivers/servers.
> 
> Sure, but I always took this to mean "you can freely copy the
> definitions", not "this is the authorative source".

virtio spec only has a ring header.
We could get it from there but it's easier to
get everything from one place I think.

-- 
MST