Re: [Qemu-devel] [PATCH 2/2] virtio-balloon: resume collecting stats on vmload

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

On Fri, Sep 02, 2016 at 10:21:58AM +0300, Roman Kagan wrote:
> On Thu, Sep 01, 2016 at 10:26:54PM +0300, Michael S. Tsirkin wrote:
> > On Thu, Sep 01, 2016 at 09:14:00PM +0300, Roman Kagan wrote:
> > > Upon save/restore virtio-balloon stats acquisition stops.  The reason is
> > > that the in-use virtqueue element is not saved, and upon restore it's
> > > not released to the guest, making further progress impossible.
> > > 
> > > Saving the information about the used element would introduce
> > > unjustified vmstate incompatibility.
> > > 
> > > However, the number of virtqueue in-use elements can be deduced from the
> > > data available on restore (see bccdef6b1a204db0f41ffb6e24ce373e4d7890d4
> > > "virtio: recalculate vq->inuse after migration").
> > > 
> > > So make the stats virtqueue forget that an element was popped from it,
> > > and start over.  As this tackles the problem on the "load" side, it
> > > is compatible with the state saved by earlier QEMU versions.
> > > 
> > > Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
> > > Cc: "Michael S. Tsirkin" <mst@redhat.com>
> > > Cc: Ladi Prosek <lprosek@redhat.com>
> > > Cc: Stefan Hajnoczi <stefanha@redhat.com>
> > > ---
> > >  hw/virtio/virtio-balloon.c | 21 ++++++++++++++++++++-
> > >  1 file changed, 20 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c
> > > index 5af429a..8660052 100644
> > > --- a/hw/virtio/virtio-balloon.c
> > > +++ b/hw/virtio/virtio-balloon.c
> > > @@ -406,7 +406,13 @@ static void virtio_balloon_save_device(VirtIODevice *vdev, QEMUFile *f)
> > >  
> > >  static int virtio_balloon_load(QEMUFile *f, void *opaque, size_t size)
> > >  {
> > > -    return virtio_load(VIRTIO_DEVICE(opaque), f, 1);
> > > +    VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
> > > +    VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
> > > +    int ret = virtio_load(vdev, f, 1);
> > > +    /* rewind needs vq->inuse populated which happens in virtio_load() after
> > > +     * vdev->load */
> > > +    virtqueue_rewind(s->svq);
> > > +    return ret;
> > >  }
> > >  
> > >  static int virtio_balloon_load_device(VirtIODevice *vdev, QEMUFile *f,
> > > @@ -468,6 +474,18 @@ static void virtio_balloon_device_reset(VirtIODevice *vdev)
> > >      }
> > >  }
> > >  
> > > +static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status)
> > > +{
> > > +    VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
> > > +
> > > +    if (vdev->vm_running && balloon_stats_supported(s) &&
> > > +        (status & VIRTIO_CONFIG_S_DRIVER_OK) && !s->stats_vq_elem) {
> > > +        /* poll stats queue for the element we may have discarded when the VM
> > > +         * was stopped */
> > > +        virtio_balloon_receive_stats(vdev, s->svq);
> > > +    }
> > > +}
> > > +
> > >  static void virtio_balloon_instance_init(Object *obj)
> > >  {
> > >      VirtIOBalloon *s = VIRTIO_BALLOON(obj);
> > > @@ -505,6 +523,7 @@ static void virtio_balloon_class_init(ObjectClass *klass, void *data)
> > >      vdc->get_features = virtio_balloon_get_features;
> > >      vdc->save = virtio_balloon_save_device;
> > >      vdc->load = virtio_balloon_load_device;
> > > +    vdc->set_status = virtio_balloon_set_status;
> > >  }
> > >  
> > >  static const TypeInfo virtio_balloon_info = {
> > 
> > 
> > I'm sorry - I don't like this patch. This means that
> > virtio_balloon_receive_stats will be called and will poke
> > at the ring even if the ring was never kicked.
> 
> I'm not sure I understand what the problem is with that:
> virtio_balloon_receive_stats just returns early if virtio_queue_empty(),
> which is no more poking at the ring than is already done in virtio_load.

Generally we should not look at ring until there was a kick.


> > This is why in my opinion it is cleaner to have
> > virtqueue_rewind return a true/false status,
> > and only process the ring if ring was rewinded.
> > 
> > I think Stefan's patches did something similar.
> 
> I'm fine with Stefan's patches, too.
> 
> Roman.