Re: [virtio] [PATCH v6 0/5] packed ring layout spec

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

On Wed, Jan 10, 2018 at 11:33:15AM +0100, Cornelia Huck wrote:
> On Wed, 10 Jan 2018 11:47:53 +0200
> "Michael S. Tsirkin" <mst@redhat.com> wrote:
> 
> > OK, this is in a shape where we could include it in the spec.
> > Changes from v5:
> > - scope reductions (see below). We can add more
> >   features down the road, hopefully reduced scope will be enough
> >   to finalize spec soon.
> > - cleanup and integrate in the spec
> > - pseudo-code
> > 
> > Deferred features:
> > - dropped _F_DESC_LIST, 1.0 includes this unconditionally, we
> >   can do same
> > - dropped event structure change notifications - needed for
> >   efficient hardware implementations but let's add this on top
> > 
> > 3 1st patches just move text around so all virtio 1.0
> > things are in the same place. 2 last ones add the new layout
> > 
> > Option to mark descriptors as not generating events isn't
> > yet implemented. Again, let's add this on top.
> > 
> > I also note that for hardware implementations, a different
> > set of memory barriers is needed. Again, let's add this on top
> > 
> > not linking into conformance sections, will add after spec
> > itself is approved.
> > 
> > Michael S. Tsirkin (5):
> >   content: move 1.0 queue format out to a separate section
> >   content: move ring text out to a separate file
> >   content: move virtqueue operation description
> >   packed virtqueues: more efficient virtqueue layout
> >   packed-ring: add in order request support
> > 
> >  conformance.tex |   4 +-
> >  content.tex     | 718 +++++---------------------------------------------------
> >  packed-ring.tex | 692 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  split-ring.tex  | 666 ++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  4 files changed, 1423 insertions(+), 657 deletions(-)
> >  create mode 100644 packed-ring.tex
> >  create mode 100644 split-ring.tex
> > 
> 
> One thing I just noticed: This does not seem to contain any changes for
> transports. For ccw, we use different payloads in the SET_VQ command
> for legacy (pre-virtio-1) and virtio-1 split layout. I think we would
> need a new SET_VQ payload for the packed ring guarded by a new
> revision, but I have not investigated it yet. I have not yet thought
> through how this interacts with feature negotiation, either.

Good point.
There's exactly same number of chunks (descriptor, device and driver
event suppression).

So the idea is to substitute descriptor for descriptor, device event
suppression for available ring and driver event suppression for
used ring.

That's why we don't need to change transports.

I will add text to clarify that.

> Otherwise, I don't see any reason why the packed ring would not work
> fine for ccw as well.