virtio_net: exploit napi_complete_done() return value

Since commit 364b605 ("net: busy-poll: return busypolling status to
drivers"), napi_complete_done() returns a boolean that can be used
by drivers to conditionally rearm interrupts.

This patch changes virtio_net to use this boolean to avoid a bit of
overhead for busy-poll users.

Jason reports about 1.1% improvement for 1 byte TCP_RR (burst 100).

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/virtio_net.c

@@ -1012,11 +1012,12 @@ static int virtnet_poll(struct napi_struct *napi, int budget)
 	/* Out of packets? */
 	if (received < budget) {
 		r = virtqueue_enable_cb_prepare(rq->vq);
-		napi_complete_done(napi, received);
-		if (unlikely(virtqueue_poll(rq->vq, r)) &&
-		    napi_schedule_prep(napi)) {
-			virtqueue_disable_cb(rq->vq);
-			__napi_schedule(napi);
+		if (napi_complete_done(napi, received)) {
+			if (unlikely(virtqueue_poll(rq->vq, r)) &&
+			    napi_schedule_prep(napi)) {
+				virtqueue_disable_cb(rq->vq);
+				__napi_schedule(napi);
+			}
 		}
 	}