Re: [Qemu-devel] [PATCH] dataplane: endian fix in host notification

[Cornelia Huck]

On Thu, 25 Jun 2015 18:34:47 +0200
Cornelia Huck <address@hidden> wrote:

> On Thu, 25 Jun 2015 17:26:21 +0200
> Greg Kurz <address@hidden> wrote:
> 
> > This field comes either LE with virtio 1.0, either guest endian with legacy.
> > It must only be accessed with an accessor that knows about the appropriate
> > endianness.
> > 
> > Signed-off-by: Greg Kurz <address@hidden>
> > ---
> >  hw/virtio/dataplane/vring.c |    2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/hw/virtio/dataplane/vring.c b/hw/virtio/dataplane/vring.c
> > index 3fa421b9d773..a93ee2d338d7 100644
> > --- a/hw/virtio/dataplane/vring.c
> > +++ b/hw/virtio/dataplane/vring.c
> > @@ -117,7 +117,7 @@ void vring_disable_notification(VirtIODevice *vdev, 
> > Vring *vring)
> >  bool vring_enable_notification(VirtIODevice *vdev, Vring *vring)
> >  {
> >      if (virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
> > -        vring_avail_event(&vring->vr) = vring->vr.avail->idx;
> > +        vring_avail_event(&vring->vr) = vring_get_avail_idx(vdev, vring);
> >      } else {
> >          vring_clear_used_flags(vdev, vring, VRING_USED_F_NO_NOTIFY);
> >      }
> > 
> Hm... we should publish in the same endianness as the queue, shouldn't
> we? IOW, this seems fine.

Er, by this I mean "it is fine without your patch". Long day...

> 
> OTOH, this prompted me to check for other places where we touch
> vring_avail_event and it seems to me that we need to convert
> vring->last_avail_idx before we set it in vring_pop().
> 
> I might be confused, though :)
> 
>