[10/10] net/virtio: improve batching in standard Rx path

Message ID 20190319064312.13743-11-tiwei.bie@intel.com
State Accepted, archived
Delegated to: Maxime Coquelin
Headers show
Series
  • net/virtio: cleanups and fixes for packed/split ring
Related show

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK

Commit Message

Tiwei Bie March 19, 2019, 6:43 a.m.
This patch improves descriptors refill by using the same
batching strategy as done in in-order and mergeable path.

Signed-off-by: Tiwei Bie <tiwei.bie@intel.com>
---
 drivers/net/virtio/virtio_rxtx.c | 60 ++++++++++++++++++--------------
 1 file changed, 34 insertions(+), 26 deletions(-)

Comments

Jens Freimann March 19, 2019, 10:04 a.m. | #1
On Tue, Mar 19, 2019 at 02:43:12PM +0800, Tiwei Bie wrote:
>This patch improves descriptors refill by using the same
>batching strategy as done in in-order and mergeable path.
>
>Signed-off-by: Tiwei Bie <tiwei.bie@intel.com>
>---
> drivers/net/virtio/virtio_rxtx.c | 60 ++++++++++++++++++--------------
> 1 file changed, 34 insertions(+), 26 deletions(-)
>
Looks good. How much do we gain by this?

Reviewed-by: Jens Freimann <jfreimann@redhat.com>
Tiwei Bie March 19, 2019, 10:28 a.m. | #2
On Tue, Mar 19, 2019 at 11:04:24AM +0100, Jens Freimann wrote:
> On Tue, Mar 19, 2019 at 02:43:12PM +0800, Tiwei Bie wrote:
> > This patch improves descriptors refill by using the same
> > batching strategy as done in in-order and mergeable path.
> > 
> > Signed-off-by: Tiwei Bie <tiwei.bie@intel.com>
> > ---
> > drivers/net/virtio/virtio_rxtx.c | 60 ++++++++++++++++++--------------
> > 1 file changed, 34 insertions(+), 26 deletions(-)
> > 
> Looks good. How much do we gain by this?

The gain is very visible on my side. E.g. for packed ring,
the PPS changed from ~10786973 to ~11636990 in a macfwd test
between two testpmds.

Thanks,
Tiwei

> 
> Reviewed-by: Jens Freimann <jfreimann@redhat.com>
> 
>
Maxime Coquelin March 19, 2019, 11:08 a.m. | #3
On 3/19/19 11:28 AM, Tiwei Bie wrote:
> On Tue, Mar 19, 2019 at 11:04:24AM +0100, Jens Freimann wrote:
>> On Tue, Mar 19, 2019 at 02:43:12PM +0800, Tiwei Bie wrote:
>>> This patch improves descriptors refill by using the same
>>> batching strategy as done in in-order and mergeable path.
>>>
>>> Signed-off-by: Tiwei Bie <tiwei.bie@intel.com>
>>> ---
>>> drivers/net/virtio/virtio_rxtx.c | 60 ++++++++++++++++++--------------
>>> 1 file changed, 34 insertions(+), 26 deletions(-)
>>>
>> Looks good. How much do we gain by this?
> 
> The gain is very visible on my side. E.g. for packed ring,
> the PPS changed from ~10786973 to ~11636990 in a macfwd test
> between two testpmds.

Nice!

> Thanks,
> Tiwei
> 
>>
>> Reviewed-by: Jens Freimann <jfreimann@redhat.com>
>>
>>
Maxime Coquelin March 19, 2019, 2:15 p.m. | #4
On 3/19/19 7:43 AM, Tiwei Bie wrote:
> This patch improves descriptors refill by using the same
> batching strategy as done in in-order and mergeable path.
> 
> Signed-off-by: Tiwei Bie <tiwei.bie@intel.com>
> ---
>   drivers/net/virtio/virtio_rxtx.c | 60 ++++++++++++++++++--------------
>   1 file changed, 34 insertions(+), 26 deletions(-)
> 


Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime

Patch

diff --git a/drivers/net/virtio/virtio_rxtx.c b/drivers/net/virtio/virtio_rxtx.c
index 42d0f533c..5f6796bdb 100644
--- a/drivers/net/virtio/virtio_rxtx.c
+++ b/drivers/net/virtio/virtio_rxtx.c
@@ -1211,7 +1211,7 @@  virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 	struct virtnet_rx *rxvq = rx_queue;
 	struct virtqueue *vq = rxvq->vq;
 	struct virtio_hw *hw = vq->hw;
-	struct rte_mbuf *rxm, *new_mbuf;
+	struct rte_mbuf *rxm;
 	uint16_t nb_used, num, nb_rx;
 	uint32_t len[VIRTIO_MBUF_BURST_SZ];
 	struct rte_mbuf *rcv_pkts[VIRTIO_MBUF_BURST_SZ];
@@ -1281,20 +1281,24 @@  virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 	rxvq->stats.packets += nb_rx;
 
 	/* Allocate new mbuf for the used descriptor */
-	while (likely(!virtqueue_full(vq))) {
-		new_mbuf = rte_mbuf_raw_alloc(rxvq->mpool);
-		if (unlikely(new_mbuf == NULL)) {
-			struct rte_eth_dev *dev
-				= &rte_eth_devices[rxvq->port_id];
-			dev->data->rx_mbuf_alloc_failed++;
-			break;
+	if (likely(!virtqueue_full(vq))) {
+		uint16_t free_cnt = vq->vq_free_cnt;
+		struct rte_mbuf *new_pkts[free_cnt];
+
+		if (likely(rte_pktmbuf_alloc_bulk(rxvq->mpool, new_pkts,
+						free_cnt) == 0)) {
+			error = virtqueue_enqueue_recv_refill(vq, new_pkts,
+					free_cnt);
+			if (unlikely(error)) {
+				for (i = 0; i < free_cnt; i++)
+					rte_pktmbuf_free(new_pkts[i]);
+			}
+			nb_enqueued += free_cnt;
+		} else {
+			struct rte_eth_dev *dev =
+				&rte_eth_devices[rxvq->port_id];
+			dev->data->rx_mbuf_alloc_failed += free_cnt;
 		}
-		error = virtqueue_enqueue_recv_refill(vq, &new_mbuf, 1);
-		if (unlikely(error)) {
-			rte_pktmbuf_free(new_mbuf);
-			break;
-		}
-		nb_enqueued++;
 	}
 
 	if (likely(nb_enqueued)) {
@@ -1316,7 +1320,7 @@  virtio_recv_pkts_packed(void *rx_queue, struct rte_mbuf **rx_pkts,
 	struct virtnet_rx *rxvq = rx_queue;
 	struct virtqueue *vq = rxvq->vq;
 	struct virtio_hw *hw = vq->hw;
-	struct rte_mbuf *rxm, *new_mbuf;
+	struct rte_mbuf *rxm;
 	uint16_t num, nb_rx;
 	uint32_t len[VIRTIO_MBUF_BURST_SZ];
 	struct rte_mbuf *rcv_pkts[VIRTIO_MBUF_BURST_SZ];
@@ -1380,20 +1384,24 @@  virtio_recv_pkts_packed(void *rx_queue, struct rte_mbuf **rx_pkts,
 	rxvq->stats.packets += nb_rx;
 
 	/* Allocate new mbuf for the used descriptor */
-	while (likely(!virtqueue_full(vq))) {
-		new_mbuf = rte_mbuf_raw_alloc(rxvq->mpool);
-		if (unlikely(new_mbuf == NULL)) {
+	if (likely(!virtqueue_full(vq))) {
+		uint16_t free_cnt = vq->vq_free_cnt;
+		struct rte_mbuf *new_pkts[free_cnt];
+
+		if (likely(rte_pktmbuf_alloc_bulk(rxvq->mpool, new_pkts,
+						free_cnt) == 0)) {
+			error = virtqueue_enqueue_recv_refill_packed(vq,
+					new_pkts, free_cnt);
+			if (unlikely(error)) {
+				for (i = 0; i < free_cnt; i++)
+					rte_pktmbuf_free(new_pkts[i]);
+			}
+			nb_enqueued += free_cnt;
+		} else {
 			struct rte_eth_dev *dev =
 				&rte_eth_devices[rxvq->port_id];
-			dev->data->rx_mbuf_alloc_failed++;
-			break;
+			dev->data->rx_mbuf_alloc_failed += free_cnt;
 		}
-		error = virtqueue_enqueue_recv_refill_packed(vq, &new_mbuf, 1);
-		if (unlikely(error)) {
-			rte_pktmbuf_free(new_mbuf);
-			break;
-		}
-		nb_enqueued++;
 	}
 
 	if (likely(nb_enqueued)) {