Re: [Qemu-devel] [PATCH 3/5] s390: Add new channel I/O based virtio transport.

[Cornelia Huck]

On Thu, 9 Aug 2012 12:34:04 +0100
Stefan Hajnoczi <address@hidden> wrote:

> On Tue, Aug 7, 2012 at 3:52 PM, Cornelia Huck <address@hidden> wrote:
> > Add a new virtio transport that uses channel commands to perform
> > virtio operations.
> >
> > Add a new machine type s390-ccw that uses this virtio-ccw transport
> > and make it the default machine for s390.
> >
> > Signed-off-by: Cornelia Huck <address@hidden>
> > ---
> >  hw/qdev-monitor.c      |   5 +
> >  hw/s390-virtio.c       | 268 ++++++++++----
> >  hw/s390x/Makefile.objs |   1 +
> >  hw/s390x/virtio-ccw.c  | 962 
> > +++++++++++++++++++++++++++++++++++++++++++++++++
> >  hw/s390x/virtio-ccw.h  |  77 ++++
> >  vl.c                   |   1 +
> >  6 files changed, 1243 insertions(+), 71 deletions(-)
> >  create mode 100644 hw/s390x/virtio-ccw.c
> >  create mode 100644 hw/s390x/virtio-ccw.h
> 
> Is the virtqueue still using vring and assuming the hypervisor reaches
> into guest memory?

The virtqueues are guest-allocated and their location is transmitted
via a control-type ccw to the host, which can then use it until
notified otherwise.

> 
> Can existing ccw device types access memory directly (for some reason
> I assumed ccw always copies or send messages)?

Not sure if I understand your question correctly, but read or write
type ccws specify a memory area where the hardware/hypervisor may write
to or read from. These accesses happen while the channel program is
running (any time between the ssch/rsch and ending status present at
the subchannel). The "specify an area that can be used by hardware and
os" approach exists as well; the closed thing to the virtio-ccw
approach is probably qdio.