Re: [virtio] Feedback: PCI device spec

["Michael S. Tsirkin" <mst@redhat.com>]

On Wed, Jan 29, 2014 at 01:15:18PM +1030, Rusty Russell wrote:
> Pawel Moll <pawel.moll@arm.com> writes:
> > On Tue, 2014-01-28 at 16:58 +0000, Michael S. Tsirkin wrote:
> >> ATM you can put any value that makes sense there I think, drivers will bind
> >> and work.
> >
> > That's true, I'm fine with this.
> >
> > But try to put yourself into such guy's shoes. You know nothing about
> > virtio. You have to implement a virtio PCI device for your custom
> > system. You're looking at the PCI spec, which tells you that you've got
> > to put some values into the device's configuration space. And the virtio
> > spec doesn't tell you what numbers should you use.
> >
> > I appreciate it's convenient for us not to specify such details. Maybe
> > it's the right thing to do. But I also understand my colleague's
> > frustration with this aspect of the spec. Thus the discussion.
> 
> This.
> 
> The spec *exists* so we get more, good implementations.
> 
> What helps this?  Including a non-normative paragraph discussing use of
> PCI class?  Or ignoring it and leaving implementers confused and
> frustrated so we look more rigorous?

Sure. Let's just make it clear that PCI spec
is the normative document here.

> In fact, let me now offer a beer to anyone on the committee who adds a
> decent joke to the spec.

This needs a native English speaker I'm afraid. Let's do this: you
write a joke, I'll add it to the spec and get a beer :)

> Because the spec is for humans, and I want
> readers to like it so much they tell their friends to implement it.
> 
> Thanks,
> Rusty.