As Michael point out, We used to limit the max pending DMAs to get better cache
utilization. But it was not done correctly since it was one done when there's no
new buffers submitted from guest. Guest can easily exceeds the limitation by
keeping sending packets.
So this patch moves the check into main loop. Tests shows about 5%-10%
improvement on per cpu throughput for guest tx. But a 5% drop on per cpu
transaction rate for a single session TCP_RR.
Signed-off-by: Jason Wang <jasowang@xxxxxxxxxx>
---
drivers/vhost/net.c | 15 ++++-----------
1 files changed, 4 insertions(+), 11 deletions(-)
diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index d09c17c..592e1f2 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -363,6 +363,10 @@ static void handle_tx(struct vhost_net *net)
if (zcopy)
vhost_zerocopy_signal_used(net, vq);
+ if ((nvq->upend_idx + vq->num - VHOST_MAX_PEND) % UIO_MAXIOV ==
+ nvq->done_idx)
+ break;
+
head = vhost_get_vq_desc(&net->dev, vq, vq->iov,
ARRAY_SIZE(vq->iov),
&out, &in,
@@ -372,17 +376,6 @@ static void handle_tx(struct vhost_net *net)
break;
/* Nothing new? Wait for eventfd to tell us they refilled. */
if (head == vq->num) {
- int num_pends;
-
- /* If more outstanding DMAs, queue the work.
- * Handle upend_idx wrap around
- */
- num_pends = likely(nvq->upend_idx >= nvq->done_idx) ?
- (nvq->upend_idx - nvq->done_idx) :
- (nvq->upend_idx + UIO_MAXIOV -
- nvq->done_idx);
- if (unlikely(num_pends > VHOST_MAX_PEND))
- break;
if (unlikely(vhost_enable_notify(&net->dev, vq))) {
vhost_disable_notify(&net->dev, vq);
continue;
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
