On Wed, Apr 18, 2012 at 04:34:12PM +0200, Paolo Bonzini wrote:
> Il 18/04/2012 16:21, Michael S. Tsirkin ha scritto:
> >> @@ -1872,6 +1864,8 @@ static int virtcons_restore(struct virtio_device
> >> *vdev)
> >> list_for_each_entry(port, &portdev->ports, list) {
> >> port->in_vq = portdev->in_vqs[port->id];
> >> port->out_vq = portdev->out_vqs[port->id];
> >> + port->in_vq->vdev_priv = port;
> >> + port->out_vq->vdev_priv = port;
> >>
> >> fill_queue(port->in_vq, &port->inbuf_lock);
> >>
> >
> > Let's add an API to set this pointer.
> > Document that you must not set it after
> > probe/restore returned.
>
> Why?
How would you prevent races if you do?
> > With an API we can actually have a BUG_ON that checks it's not modified
> > after probe.
> >
> >> diff --git a/include/linux/virtio.h b/include/linux/virtio.h
> >> index c193ccf..6b39c1a 100644
> >> --- a/include/linux/virtio.h
> >> +++ b/include/linux/virtio.h
> >> @@ -14,6 +14,7 @@
> >> * @callback: the function to call when buffers are consumed (can be
> >> NULL).
> >> * @name: the name of this virtqueue (mainly for debugging)
> >> * @vdev: the virtio device this queue was created for.
> >> + * @vdev_priv: a pointer for the virtio device to use.
> >
> > It's for the driver actually.
>
> Right. However...
pointer for a device can also be misunerstood as
'pointer to a device'. Note priv below actually
gets the correct meaning however you interpret 'for'.
Better 'pointer for the virtqueue driver to use'.
>
> >> * @priv: a pointer for the virtqueue implementation to use.
> >> */
> >> struct virtqueue {
> >> @@ -21,6 +22,7 @@ struct virtqueue {
> >> void (*callback)(struct virtqueue *vq);
> >> const char *name;
> >> struct virtio_device *vdev;
> >> + void *vdev_priv;
> >> void *priv;
> >
> > The name is confusing: it seems to imply it's a device pointer.
>
> ... it's private to the driver that owns vdev, hence the name.
I own a car but I'm not called Michael Car :)
driver_priv might be ok too. unfortunately virtio-pci
is also a driver so it can be misunderstood.
> > Maybe we should rename priv to something like __priv and make
> > priv useful for devices?
>
> I wanted to go for the smallest possible changes. Right now we have 1
> user for each member (virtio-ring vs. virtio-console) so neither member
> is really dominating.
>
> Paolo
devices should dominate. ring is an implementation detail.
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
