Re: [Qemu-devel] [PATCH for-1.6] virtio: clear signalled_used_valid when switching from dataplane

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

On Tue, Aug 13, 2013 at 10:14:11AM +0200, Stefan Hajnoczi wrote:
> On Mon, Aug 12, 2013 at 05:10:11PM +0300, Michael S. Tsirkin wrote:
> > On Mon, Aug 12, 2013 at 03:39:23PM +0200, Stefan Hajnoczi wrote:
> > > On Mon, Aug 12, 2013 at 11:18 AM, Michael S. Tsirkin <mst@redhat.com> wrote:
> > > > On Mon, Aug 12, 2013 at 11:08:09AM +0200, Stefan Hajnoczi wrote:
> > > >> When the dataplane thread stops, its vring.c implementation synchronizes
> > > >> vring state back to virtio.c so we can continue emulating the virtio
> > > >> device.
> > > >>
> > > >> This patch ensures that virtio.c's signalled_used_valid flag is reset so
> > > >> that we do not suppress guest notifications due to stale signalled_used
> > > >> values.
> > > >>
> > > >> Suggested-by: Kevin Wolf <kwolf@redhat.com>
> > > >> Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> > > >
> > > > Good point
> > > >
> > > > Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
> > > >
> > > > and we also need this for vhost.c right?
> > > 
> > > I think vhost is not affected by this because virtio.c clears the
> > > signalled_used_valid flag when the device is reset.  Since vhost_net
> > > does not transition back and forth between userspace virtio and vhost
> > > (without a device reset), there is no need to reset the flag.
> > > 
> > > Stefan
> > 
> > Yes it does transition on vm stop/start.
> 
> I see.  Come to think of it maybe dataplane could use virtio.c's vm
> change state handler instead of registering its own.  Anyway, thanks for
> pointing this out.
> 
> Stefan

It probably should esp if you have a handler for virtio status:
this is the only way to guarantee they are called in a sane order.

-- 
MST