On Tue, 8 May 2012 13:18:39 +0300, "Michael S. Tsirkin" <mst@xxxxxxxxxx> wrote:
> On Tue, May 08, 2012 at 11:45:55AM +0200, Paolo Bonzini wrote:
> > --- a/drivers/virtio/virtio_ring.c
> > +++ b/drivers/virtio/virtio_ring.c
> > @@ -76,8 +76,6 @@
> >
> > struct vring_virtqueue
> > {
> > - struct virtqueue vq;
> > -
> > /* Actual memory layout for this queue */
> > struct vring vring;
> >
> > @@ -106,6 +104,9 @@ struct vring_virtqueue
> > /* How to notify other side. FIXME: commonalize hcalls! */
> > void (*notify)(struct virtqueue *vq);
> >
> > + /* Tokens for callbacks. */
> > + void **data;
> > +
> > #ifdef DEBUG
> > /* They're supposed to lock for us. */
> > unsigned int in_use;
> > @@ -115,8 +116,9 @@ struct vring_virtqueue
> > ktime_t last_add_time;
> > #endif
> >
> > - /* Tokens for callbacks. */
> > - void *data[];
> > + struct virtqueue vq;
> > +
> > + /* Bus-specific virtqueue data goes here. */
> > };
> >
> > #define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
>
>
> This moves the data out of line and the bus specific stuff inline.
> But bus accesses only happen on io and interrupt which are already
> slow, while data accesses happen on fast path.
>
> >From that POV this looks like a wrong thing to do.
(Resend to all)
Most of it was on a different cacheline anyway though, so it's probably
not measurable. We avoid the pci_vq->vq indirection, but add the
vq->data indirection.
I share your discomfort with the offset arg method, too.
So unless benchmarks show otherwise, let's do it the vanilla way?
I think that making vring_virtqueue public looks like the way to go,
too.
Of course, a nice series would be great as well :)
Thanks!
Rusty.
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
