[Qemu-devel] Re: [PATCH 15/41] virtio: remove save/load_queue for virtio

[Michael S. Tsirkin]

On Wed, Dec 02, 2009 at 07:22:11PM +0100, Juan Quintela wrote:
> "Michael S. Tsirkin" <address@hidden> wrote:
> > On Wed, Dec 02, 2009 at 01:04:13PM +0100, Juan Quintela wrote:
> >> diff --git a/hw/virtio.c b/hw/virtio.c
> >> index c136005..b565bf9 100644
> >> --- a/hw/virtio.c
> >> +++ b/hw/virtio.c
> >> @@ -643,8 +643,10 @@ void virtio_save(VirtIODevice *vdev, QEMUFile *f)
> >>          qemu_put_be32(f, vdev->vq[i].vring.num);
> >>          qemu_put_be64(f, vdev->vq[i].pa);
> >>          qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
> >> -        if (vdev->binding->save_queue)
> >> -            vdev->binding->save_queue(vdev->binding_opaque, i, f);
> >> +        if (vdev->type == VIRTIO_PCI &&
> >> +            virtio_pci_msix_present(vdev->binding_opaque)) {
> >> +            qemu_put_be16s(f, &vdev->vq[i].vector);
> >> +        }
> >>      }
> >>  }
> >> 
> >
> > I think this will break build on systems without PCI
> > because virtio_pci.c is not linked in there.
> > Correct?
> 
> It compiles for syborg (it has pci).  There are no other users.
> 
> > Making generic virtio.c depend on virtio_pci.c looks
> > wrong in any case. We have bindings to
> > resolve exactly this dependency problem.
> 
> There is no way that you throw "this" blob to vmstate and it will know
> what to do there.  if it is needed, we can create an empty
> virtio_pci_msix_present() function for !CONFIG_PCI.
> 
> Later, Juan.

That's not the idea. virtio knows nothing about msix etc.  This belongs
in the binding. If you want to know the number of vectors, please put
something like ->get_nvectors in the binding and call that to figure out
whether virtio has multivector.

-- 
MST