Re: [virtio-comment] [PATCH 1/1] ccw: ccw payload description

[Cornelia Huck <cornelia.huck@de.ibm.com>]

On Sun, 2 Feb 2014 18:27:26 +0200
"Michael S. Tsirkin" <mst@redhat.com> wrote:

> On Fri, Jan 31, 2014 at 03:25:31PM +1030, Rusty Russell wrote:
> > "Michael S. Tsirkin" <mst@redhat.com> writes:
> > > On Thu, Jan 30, 2014 at 01:10:29PM +0100, Cornelia Huck wrote:
> > >> I think what makes this complicated is that we specify datastructures
> > >> that are supposed to work for both legacy and non-legacy, when they
> > >> are in fact different (at least on be).
> > >
> > > So how about something like attribute((virtio_endian)) (a bit too long)
> > > or ve32 (slightly ugly).
> > >
> > >> How about adding "A note on legacy interfaces" for each of these cases,
> > >> where we specify the data structures with native-endian where
> > >> appropriate, and keep the explicit le notation for the standard? We'd
> > >> end up with some duplication, but we'd be very clear on what is
> > >> expected.
> > >
> > > In practice, this will make it harder to write transitional
> > > drivers. Driver authors will have to reverse-engineer field
> > > layouts and note that offsets happen to match.
> > > That's not being nice to readers.
> > >
> > > Take a look at virtio drivers that Rusty wrote:
> > > ring layouts *don't* use le tags because sometimes
> > > they are used with be.
> > 
> > Sure, but it will always be hard to write such drivers.  There shouldn't
> > be many people doing that though.  And they'll be doing it for backwards
> > compatibility, where they'll presumably test...
> > 
> > I'm more concerned about people writing non-legacy drivers and getting
> > endianness wrong.  But if you feel strongly, we could put a comment
> > above each one?
> > 
> > Cheers,
> > Rusty.
> 
> Sounds good, let's do that. And let's add a comment
> where we define leXX.

https://lists.oasis-open.org/archives/virtio/201401/msg00089.html adds
definitions for leXX types, among other things.