On Wednesday, September 12, 2012 8:36:47 pm Peter Grehan wrote:
> Author: grehan
> Date: Thu Sep 13 00:36:46 2012
> New Revision: 240427
> URL: http://svn.freebsd.org/changeset/base/240427
>
> Log:
> Relax requirement of certain mb()s
>
> Submitted by: Bryan Venteicher bryanv at daemoninthecloset org
>
> Modified:
> head/sys/dev/virtio/virtqueue.c
>
> Modified: head/sys/dev/virtio/virtqueue.c
>
==============================================================================
> --- head/sys/dev/virtio/virtqueue.c Wed Sep 12 22:54:11 2012
> (r240426)
> +++ head/sys/dev/virtio/virtqueue.c Thu Sep 13 00:36:46 2012
> (r240427)
> @@ -525,7 +525,7 @@ virtqueue_dequeue(struct virtqueue *vq,
> used_idx = vq->vq_used_cons_idx++ & (vq->vq_nentries - 1);
> uep = &vq->vq_ring.used->ring[used_idx];
>
> - mb();
> + rmb();
> desc_idx = (uint16_t) uep->id;
> if (len != NULL)
> *len = uep->len;
> @@ -623,7 +623,7 @@ vq_ring_update_avail(struct virtqueue *v
> avail_idx = vq->vq_ring.avail->idx & (vq->vq_nentries - 1);
> vq->vq_ring.avail->ring[avail_idx] = desc_idx;
>
> - mb();
> + wmb();
> vq->vq_ring.avail->idx++;
>
> /* Keep pending count until virtqueue_notify(). */
Would it be possible to use atomic_load/store() instead of direct
memory barriers? For example:
desc_idx = (uint16_t)atomic_load_acq_int(&uep->id);
and
atomic_store_rel_int(&vq->vq_ring.avail->ring[avail_idx], desc_idx);
--
John Baldwin
_______________________________________________
svn-src-head@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/svn-src-head
To unsubscribe, send any mail to "svn-src-head-unsubscr...@freebsd.org"
