DPDK patches and discussions
 help / color / mirror / Atom feed
From: Michal Krawczyk <mk@semihalf.com>
To: dev@dpdk.org
Cc: mw@semihalf.com, mba@semihalf.com, gtzalik@amazon.com,
	evgenys@amazon.com, igorch@amazon.com,
	Michal Krawczyk <mk@semihalf.com>
Subject: [dpdk-dev] [PATCH v2 26/29] net/ena: reuse 0 length Rx descriptor
Date: Wed,  1 Apr 2020 16:21:24 +0200	[thread overview]
Message-ID: <20200401142127.13715-27-mk@semihalf.com> (raw)
In-Reply-To: <20200401142127.13715-1-mk@semihalf.com>

Some ENA devices can pass to the driver descriptor with length 0. To
avoid extra allocation, the descriptor can be reused by simply putting
it back to the device.

Signed-off-by: Michal Krawczyk <mk@semihalf.com>
Reviewed-by: Igor Chauskin <igorch@amazon.com>
Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
---
v2:
  * Compare rc (error code) to 0 value explicitely

 drivers/net/ena/ena_ethdev.c | 75 ++++++++++++++++++++++++++++--------
 1 file changed, 60 insertions(+), 15 deletions(-)

diff --git a/drivers/net/ena/ena_ethdev.c b/drivers/net/ena/ena_ethdev.c
index 601b4cb831..34f910c9c4 100644
--- a/drivers/net/ena/ena_ethdev.c
+++ b/drivers/net/ena/ena_ethdev.c
@@ -195,6 +195,8 @@ static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
 				    uint8_t offset);
 static uint16_t eth_ena_recv_pkts(void *rx_queue,
 				  struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
+static int ena_add_single_rx_desc(struct ena_com_io_sq *io_sq,
+				  struct rte_mbuf *mbuf, uint16_t id);
 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
 static void ena_init_rings(struct ena_adapter *adapter,
 			   bool disable_meta_caching);
@@ -1414,6 +1416,24 @@ static int ena_rx_queue_setup(struct rte_eth_dev *dev,
 	return 0;
 }
 
+static int ena_add_single_rx_desc(struct ena_com_io_sq *io_sq,
+				  struct rte_mbuf *mbuf, uint16_t id)
+{
+	struct ena_com_buf ebuf;
+	int rc;
+
+	/* prepare physical address for DMA transaction */
+	ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
+	ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
+
+	/* pass resource to device */
+	rc = ena_com_add_single_rx_desc(io_sq, &ebuf, id);
+	if (unlikely(rc != 0))
+		PMD_DRV_LOG(WARNING, "failed adding rx desc\n");
+
+	return rc;
+}
+
 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
 {
 	unsigned int i;
@@ -1441,7 +1461,6 @@ static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
 
 	for (i = 0; i < count; i++) {
 		struct rte_mbuf *mbuf = mbufs[i];
-		struct ena_com_buf ebuf;
 		struct ena_rx_buffer *rx_info;
 
 		if (likely((i + 4) < count))
@@ -1454,16 +1473,10 @@ static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
 
 		rx_info = &rxq->rx_buffer_info[req_id];
 
-		/* prepare physical address for DMA transaction */
-		ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
-		ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
-		/* pass resource to device */
-		rc = ena_com_add_single_rx_desc(rxq->ena_com_io_sq,
-						&ebuf, req_id);
-		if (unlikely(rc)) {
-			PMD_DRV_LOG(WARNING, "failed adding rx desc\n");
+		rc = ena_add_single_rx_desc(rxq->ena_com_io_sq, mbuf, req_id);
+		if (unlikely(rc != 0))
 			break;
-		}
+
 		rx_info->mbuf = mbuf;
 		next_to_use = ENA_IDX_NEXT_MASKED(next_to_use, rxq->size_mask);
 	}
@@ -2079,6 +2092,7 @@ static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
 	struct rte_mbuf *mbuf;
 	struct rte_mbuf *mbuf_head;
 	struct ena_rx_buffer *rx_info;
+	int rc;
 	uint16_t ntc, len, req_id, buf = 0;
 
 	if (unlikely(descs == 0))
@@ -2121,13 +2135,44 @@ static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
 		rx_info = &rx_ring->rx_buffer_info[req_id];
 		RTE_ASSERT(rx_info->mbuf != NULL);
 
-		/* Create an mbuf chain. */
-		mbuf->next = rx_info->mbuf;
-		mbuf = mbuf->next;
+		if (unlikely(len == 0)) {
+			/*
+			 * Some devices can pass descriptor with the length 0.
+			 * To avoid confusion, the PMD is simply putting the
+			 * descriptor back, as it was never used. We'll avoid
+			 * mbuf allocation that way.
+			 */
+			rc = ena_add_single_rx_desc(rx_ring->ena_com_io_sq,
+				rx_info->mbuf, req_id);
+			if (unlikely(rc != 0)) {
+				/* Free the mbuf in case of an error. */
+				rte_mbuf_raw_free(rx_info->mbuf);
+			} else {
+				/*
+				 * If there was no error, just exit the loop as
+				 * 0 length descriptor is always the last one.
+				 */
+				break;
+			}
+		} else {
+			/* Create an mbuf chain. */
+			mbuf->next = rx_info->mbuf;
+			mbuf = mbuf->next;
 
-		ena_init_rx_mbuf(mbuf, len);
-		mbuf_head->pkt_len += len;
+			ena_init_rx_mbuf(mbuf, len);
+			mbuf_head->pkt_len += len;
+		}
 
+		/*
+		 * Mark the descriptor as depleted and perform necessary
+		 * cleanup.
+		 * This code will execute in two cases:
+		 *  1. Descriptor len was greater than 0 - normal situation.
+		 *  2. Descriptor len was 0 and we failed to add the descriptor
+		 *     to the device. In that situation, we should try to add
+		 *     the mbuf again in the populate routine and mark the
+		 *     descriptor as used up by the device.
+		 */
 		rx_info->mbuf = NULL;
 		rx_ring->empty_rx_reqs[ntc] = req_id;
 		ntc = ENA_IDX_NEXT_MASKED(ntc, rx_ring->size_mask);
-- 
2.20.1


  parent reply	other threads:[~2020-04-01 14:26 UTC|newest]

Thread overview: 46+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-04-01 14:20 [dpdk-dev] [PATCH v2 00/29] Update ENA driver to v2.1.0 Michal Krawczyk
2020-04-01 14:20 ` [dpdk-dev] [PATCH v2 01/29] net/ena: check if size of buffer is at least 1400B Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 02/29] net/ena/base: make allocation macros thread-safe Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 03/29] net/ena/base: prevent allocation of 0-sized memory Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 04/29] net/ena/base: set default hash key Michal Krawczyk
2020-04-02 12:36   ` Ferruh Yigit
2020-04-03 11:10     ` Michał Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 05/29] net/ena/base: rework interrupt moderation Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 06/29] net/ena/base: remove extra properties strings Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 07/29] net/ena/base: add accelerated LLQ mode Michal Krawczyk
2020-04-02 12:41   ` Ferruh Yigit
2020-04-03 11:15     ` Michał Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 08/29] net/ena/base: fix documentation of the functions Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 09/29] net/ena/base: fix indentation in cq polling Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 10/29] net/ena/base: add error logs when preparing Tx Michal Krawczyk
2020-04-02 12:53   ` Ferruh Yigit
2020-04-03 11:31     ` Michał Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 11/29] net/ena/base: use 48-bit memory addresses in ena_com Michal Krawczyk
2020-04-02 12:55   ` Ferruh Yigit
2020-04-03 12:14     ` Michał Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 12/29] net/ena/base: fix types for printing timestamps Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 13/29] net/ena/base: fix indentation of multiple defines Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 14/29] net/ena/base: update gen date and commit Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 15/29] net/ena: set IO ring size to the valid value Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 16/29] net/ena: refactor getting IO queues capabilities Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 17/29] net/ena: add support for large LLQ headers Michal Krawczyk
2020-04-02 13:02   ` Ferruh Yigit
2020-04-03 12:15     ` Michał Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 18/29] net/ena: remove memory barriers before doorbells Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 19/29] net/ena: add Tx drops statistic Michal Krawczyk
2020-04-02 13:05   ` Ferruh Yigit
2020-04-03 12:30     ` Michał Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 20/29] net/ena: disable meta caching Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 21/29] net/ena: refactor Rx path Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 22/29] net/ena: rework getting number of available descs Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 23/29] net/ena: limit refill threshold by fixed value Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 24/29] net/ena: use macros for ring idx operations Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 25/29] net/ena: refactor Tx path Michal Krawczyk
2020-04-01 14:21 ` Michal Krawczyk [this message]
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 27/29] doc: add notes on ENA usage on metal instances Michal Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 28/29] net/ena: update copyright date Michal Krawczyk
2020-04-02 15:51   ` Ferruh Yigit
2020-04-03 12:31     ` Michał Krawczyk
2020-04-01 14:21 ` [dpdk-dev] [PATCH v2 29/29] net/ena: update version of the driver to v2.1.0 Michal Krawczyk
2020-04-02 15:53   ` Ferruh Yigit
2020-04-03 12:32     ` Michał Krawczyk

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200401142127.13715-27-mk@semihalf.com \
    --to=mk@semihalf.com \
    --cc=dev@dpdk.org \
    --cc=evgenys@amazon.com \
    --cc=gtzalik@amazon.com \
    --cc=igorch@amazon.com \
    --cc=mba@semihalf.com \
    --cc=mw@semihalf.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).