DPDK patches and discussions
 help / color / mirror / Atom feed
From: Lance Richardson <lance.richardson@broadcom.com>
To: Ajit Khaparde <ajit.khaparde@broadcom.com>,
	Somnath Kotur <somnath.kotur@broadcom.com>
Cc: dev@dpdk.org
Subject: [dpdk-dev] [PATCH v3] net/bnxt: support fast mbuf free
Date: Tue,  6 Oct 2020 13:37:52 -0400
Message-ID: <20201006173752.104002-1-lance.richardson@broadcom.com> (raw)
In-Reply-To: <20201002134002.83845-1-lance.richardson@broadcom.com>

Add support for DEV_TX_OFFLOAD_MBUF_FAST_FREE to bnxt
vector mode transmit. This offload may be enabled
only when multi-segment transmit is not needed, all
transmitted mbufs for a given queue will be allocated
from the same pool, and all transmitted mbufs will
have a reference count of 1.

Signed-off-by: Lance Richardson <lance.richardson@broadcom.com>
---
- v2:
  Fixed mangling of advertised tx offload capabilities.
- v3:
  Reduced duplicated code.
  Enable for non-vector transmit path as well.
 
 doc/guides/nics/features/bnxt.ini       |  1 +
 drivers/net/bnxt/bnxt_ethdev.c          | 11 ++++---
 drivers/net/bnxt/bnxt_rxtx_vec_common.h | 32 ++++++++++++++++++++
 drivers/net/bnxt/bnxt_rxtx_vec_neon.c   | 38 ++++--------------------
 drivers/net/bnxt/bnxt_rxtx_vec_sse.c    | 39 ++++---------------------
 drivers/net/bnxt/bnxt_txq.c             |  2 ++
 drivers/net/bnxt/bnxt_txq.h             |  1 +
 drivers/net/bnxt/bnxt_txr.c             |  5 +++-
 drivers/net/bnxt/bnxt_txr.h             | 27 +++++++++++++++++
 9 files changed, 84 insertions(+), 72 deletions(-)

diff --git a/doc/guides/nics/features/bnxt.ini b/doc/guides/nics/features/bnxt.ini
index f1f300c8b..e75cfc44d 100644
--- a/doc/guides/nics/features/bnxt.ini
+++ b/doc/guides/nics/features/bnxt.ini
@@ -8,6 +8,7 @@ Speed capabilities   = Y
 Link status          = Y
 Link status event    = Y
 Rx interrupt         = Y
+Fast mbuf free       = Y
 Queue start/stop     = Y
 Burst mode info      = Y
 MTU update           = Y
diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c
index 624cb2031..c2a9f80e2 100644
--- a/drivers/net/bnxt/bnxt_ethdev.c
+++ b/drivers/net/bnxt/bnxt_ethdev.c
@@ -920,7 +920,9 @@ static int bnxt_dev_info_get_op(struct rte_eth_dev *eth_dev,
 	dev_info->rx_offload_capa = BNXT_DEV_RX_OFFLOAD_SUPPORT;
 	if (bp->flags & BNXT_FLAG_PTP_SUPPORTED)
 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TIMESTAMP;
-	dev_info->tx_offload_capa = BNXT_DEV_TX_OFFLOAD_SUPPORT;
+	dev_info->tx_queue_offload_capa = DEV_TX_OFFLOAD_MBUF_FAST_FREE;
+	dev_info->tx_offload_capa = BNXT_DEV_TX_OFFLOAD_SUPPORT |
+				    dev_info->tx_queue_offload_capa;
 	dev_info->flow_type_rss_offloads = BNXT_ETH_RSS_SUPPORT;
 
 	dev_info->speed_capa = bnxt_get_speed_capabilities(bp);
@@ -1191,6 +1193,7 @@ bnxt_transmit_function(__rte_unused struct rte_eth_dev *eth_dev)
 {
 #if defined(RTE_ARCH_X86) || defined(RTE_ARCH_ARM64)
 #ifndef RTE_LIBRTE_IEEE1588
+	uint64_t offloads = eth_dev->data->dev_conf.txmode.offloads;
 	struct bnxt *bp = eth_dev->data->dev_private;
 
 	/*
@@ -1198,7 +1201,7 @@ bnxt_transmit_function(__rte_unused struct rte_eth_dev *eth_dev)
 	 * or tx offloads.
 	 */
 	if (!eth_dev->data->scattered_rx &&
-	    !eth_dev->data->dev_conf.txmode.offloads &&
+	    !(offloads & ~DEV_TX_OFFLOAD_MBUF_FAST_FREE) &&
 	    !BNXT_TRUFLOW_EN(bp)) {
 		PMD_DRV_LOG(INFO, "Using vector mode transmit for port %d\n",
 			    eth_dev->data->port_id);
@@ -1210,7 +1213,7 @@ bnxt_transmit_function(__rte_unused struct rte_eth_dev *eth_dev)
 		    "Port %d scatter: %d tx offload: %" PRIX64 "\n",
 		    eth_dev->data->port_id,
 		    eth_dev->data->scattered_rx,
-		    eth_dev->data->dev_conf.txmode.offloads);
+		    offloads);
 #endif
 #endif
 	return bnxt_xmit_pkts;
@@ -2685,7 +2688,7 @@ bnxt_txq_info_get_op(struct rte_eth_dev *dev, uint16_t queue_id,
 	qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
 	qinfo->conf.tx_rs_thresh = 0;
 	qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
-	qinfo->conf.offloads = dev->data->dev_conf.txmode.offloads;
+	qinfo->conf.offloads = txq->offloads;
 }
 
 static const struct {
diff --git a/drivers/net/bnxt/bnxt_rxtx_vec_common.h b/drivers/net/bnxt/bnxt_rxtx_vec_common.h
index 4a48152fc..33ac53568 100644
--- a/drivers/net/bnxt/bnxt_rxtx_vec_common.h
+++ b/drivers/net/bnxt/bnxt_rxtx_vec_common.h
@@ -94,4 +94,36 @@ bnxt_rxq_rearm(struct bnxt_rx_queue *rxq, struct bnxt_rx_ring_info *rxr)
 
 	rxq->rxrearm_nb -= nb;
 }
+
+static inline void
+bnxt_tx_cmp_vec(struct bnxt_tx_queue *txq, int nr_pkts)
+{
+	struct bnxt_tx_ring_info *txr = txq->tx_ring;
+	struct rte_mbuf **free = txq->free;
+	uint16_t cons = txr->tx_cons;
+	unsigned int blk = 0;
+	uint32_t ring_mask = txr->tx_ring_struct->ring_mask;
+
+	while (nr_pkts--) {
+		struct bnxt_sw_tx_bd *tx_buf;
+		struct rte_mbuf *mbuf;
+
+		tx_buf = &txr->tx_buf_ring[cons];
+		cons = (cons + 1) & ring_mask;
+		mbuf = rte_pktmbuf_prefree_seg(tx_buf->mbuf);
+		if (unlikely(mbuf == NULL))
+			continue;
+		tx_buf->mbuf = NULL;
+
+		if (blk && mbuf->pool != free[0]->pool) {
+			rte_mempool_put_bulk(free[0]->pool, (void **)free, blk);
+			blk = 0;
+		}
+		free[blk++] = mbuf;
+	}
+	if (blk)
+		rte_mempool_put_bulk(free[0]->pool, (void **)free, blk);
+
+	txr->tx_cons = cons;
+}
 #endif /* _BNXT_RXTX_VEC_COMMON_H_ */
diff --git a/drivers/net/bnxt/bnxt_rxtx_vec_neon.c b/drivers/net/bnxt/bnxt_rxtx_vec_neon.c
index 299b6b86d..4c04cc43a 100644
--- a/drivers/net/bnxt/bnxt_rxtx_vec_neon.c
+++ b/drivers/net/bnxt/bnxt_rxtx_vec_neon.c
@@ -13,10 +13,10 @@
 #include "bnxt.h"
 #include "bnxt_cpr.h"
 #include "bnxt_ring.h"
-#include "bnxt_rxtx_vec_common.h"
 
 #include "bnxt_txq.h"
 #include "bnxt_txr.h"
+#include "bnxt_rxtx_vec_common.h"
 
 /*
  * RX Ring handling
@@ -338,37 +338,6 @@ bnxt_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 	return nb_rx_pkts;
 }
 
-static void
-bnxt_tx_cmp_vec(struct bnxt_tx_queue *txq, int nr_pkts)
-{
-	struct bnxt_tx_ring_info *txr = txq->tx_ring;
-	struct rte_mbuf **free = txq->free;
-	uint16_t cons = txr->tx_cons;
-	unsigned int blk = 0;
-
-	while (nr_pkts--) {
-		struct bnxt_sw_tx_bd *tx_buf;
-		struct rte_mbuf *mbuf;
-
-		tx_buf = &txr->tx_buf_ring[cons];
-		cons = RING_NEXT(txr->tx_ring_struct, cons);
-		mbuf = rte_pktmbuf_prefree_seg(tx_buf->mbuf);
-		if (unlikely(mbuf == NULL))
-			continue;
-		tx_buf->mbuf = NULL;
-
-		if (blk && mbuf->pool != free[0]->pool) {
-			rte_mempool_put_bulk(free[0]->pool, (void **)free, blk);
-			blk = 0;
-		}
-		free[blk++] = mbuf;
-	}
-	if (blk)
-		rte_mempool_put_bulk(free[0]->pool, (void **)free, blk);
-
-	txr->tx_cons = cons;
-}
-
 static void
 bnxt_handle_tx_cp_vec(struct bnxt_tx_queue *txq)
 {
@@ -399,7 +368,10 @@ bnxt_handle_tx_cp_vec(struct bnxt_tx_queue *txq)
 
 	cpr->valid = !!(raw_cons & cp_ring_struct->ring_size);
 	if (nb_tx_pkts) {
-		bnxt_tx_cmp_vec(txq, nb_tx_pkts);
+		if (txq->offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
+			bnxt_tx_cmp_fast(txq, nb_tx_pkts);
+		else
+			bnxt_tx_cmp_vec(txq, nb_tx_pkts);
 		cpr->cp_raw_cons = raw_cons;
 		bnxt_db_cq(cpr);
 	}
diff --git a/drivers/net/bnxt/bnxt_rxtx_vec_sse.c b/drivers/net/bnxt/bnxt_rxtx_vec_sse.c
index 271e7908e..f71f46985 100644
--- a/drivers/net/bnxt/bnxt_rxtx_vec_sse.c
+++ b/drivers/net/bnxt/bnxt_rxtx_vec_sse.c
@@ -13,10 +13,10 @@
 #include "bnxt.h"
 #include "bnxt_cpr.h"
 #include "bnxt_ring.h"
-#include "bnxt_rxtx_vec_common.h"
 
 #include "bnxt_txq.h"
 #include "bnxt_txr.h"
+#include "bnxt_rxtx_vec_common.h"
 
 /*
  * RX Ring handling
@@ -309,38 +309,6 @@ bnxt_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 	return nb_rx_pkts;
 }
 
-static void
-bnxt_tx_cmp_vec(struct bnxt_tx_queue *txq, int nr_pkts)
-{
-	struct bnxt_tx_ring_info *txr = txq->tx_ring;
-	struct rte_mbuf **free = txq->free;
-	uint16_t cons = txr->tx_cons;
-	unsigned int blk = 0;
-	uint32_t ring_mask = txr->tx_ring_struct->ring_mask;
-
-	while (nr_pkts--) {
-		struct bnxt_sw_tx_bd *tx_buf;
-		struct rte_mbuf *mbuf;
-
-		tx_buf = &txr->tx_buf_ring[cons];
-		cons = (cons + 1) & ring_mask;
-		mbuf = rte_pktmbuf_prefree_seg(tx_buf->mbuf);
-		if (unlikely(mbuf == NULL))
-			continue;
-		tx_buf->mbuf = NULL;
-
-		if (blk && mbuf->pool != free[0]->pool) {
-			rte_mempool_put_bulk(free[0]->pool, (void **)free, blk);
-			blk = 0;
-		}
-		free[blk++] = mbuf;
-	}
-	if (blk)
-		rte_mempool_put_bulk(free[0]->pool, (void **)free, blk);
-
-	txr->tx_cons = cons;
-}
-
 static void
 bnxt_handle_tx_cp_vec(struct bnxt_tx_queue *txq)
 {
@@ -371,7 +339,10 @@ bnxt_handle_tx_cp_vec(struct bnxt_tx_queue *txq)
 
 	cpr->valid = !!(raw_cons & cp_ring_struct->ring_size);
 	if (nb_tx_pkts) {
-		bnxt_tx_cmp_vec(txq, nb_tx_pkts);
+		if (txq->offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
+			bnxt_tx_cmp_fast(txq, nb_tx_pkts);
+		else
+			bnxt_tx_cmp_vec(txq, nb_tx_pkts);
 		cpr->cp_raw_cons = raw_cons;
 		bnxt_db_cq(cpr);
 	}
diff --git a/drivers/net/bnxt/bnxt_txq.c b/drivers/net/bnxt/bnxt_txq.c
index bdc7ffaaa..c8d75ac95 100644
--- a/drivers/net/bnxt/bnxt_txq.c
+++ b/drivers/net/bnxt/bnxt_txq.c
@@ -131,6 +131,8 @@ int bnxt_tx_queue_setup_op(struct rte_eth_dev *eth_dev,
 	txq->nb_tx_desc = nb_desc;
 	txq->tx_free_thresh =
 		RTE_MIN(rte_align32pow2(nb_desc) / 4, RTE_BNXT_MAX_TX_BURST);
+	txq->offloads = eth_dev->data->dev_conf.txmode.offloads |
+			tx_conf->offloads;
 
 	txq->tx_deferred_start = tx_conf->tx_deferred_start;
 
diff --git a/drivers/net/bnxt/bnxt_txq.h b/drivers/net/bnxt/bnxt_txq.h
index 9f849fc71..8033e241e 100644
--- a/drivers/net/bnxt/bnxt_txq.h
+++ b/drivers/net/bnxt/bnxt_txq.h
@@ -39,6 +39,7 @@ struct bnxt_tx_queue {
 	struct bnxt_cp_ring_info	*cp_ring;
 	const struct rte_memzone *mz;
 	struct rte_mbuf **free;
+	uint64_t offloads;
 };
 
 void bnxt_free_txq_stats(struct bnxt_tx_queue *txq);
diff --git a/drivers/net/bnxt/bnxt_txr.c b/drivers/net/bnxt/bnxt_txr.c
index c55497960..125ac8341 100644
--- a/drivers/net/bnxt/bnxt_txr.c
+++ b/drivers/net/bnxt/bnxt_txr.c
@@ -428,7 +428,10 @@ static int bnxt_handle_tx_cp(struct bnxt_tx_queue *txq)
 	} while (nb_tx_pkts < ring_mask);
 
 	if (nb_tx_pkts) {
-		bnxt_tx_cmp(txq, nb_tx_pkts);
+		if (txq->offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
+			bnxt_tx_cmp_fast(txq, nb_tx_pkts);
+		else
+			bnxt_tx_cmp(txq, nb_tx_pkts);
 		cpr->cp_raw_cons = raw_cons;
 		bnxt_db_cq(cpr);
 	}
diff --git a/drivers/net/bnxt/bnxt_txr.h b/drivers/net/bnxt/bnxt_txr.h
index 7715c11b8..026a65a72 100644
--- a/drivers/net/bnxt/bnxt_txr.h
+++ b/drivers/net/bnxt/bnxt_txr.h
@@ -52,6 +52,33 @@ static inline uint32_t bnxt_tx_avail(struct bnxt_tx_queue *txq)
 		 bnxt_tx_bds_in_hw(txq)) - 1);
 }
 
+/*
+ * Transmit completion function for use when DEV_TX_OFFLOAD_MBUF_FAST_FREE
+ * is enabled.
+ */
+static inline void
+bnxt_tx_cmp_fast(struct bnxt_tx_queue *txq, int nr_pkts)
+{
+	struct bnxt_tx_ring_info *txr = txq->tx_ring;
+	uint32_t ring_mask = txr->tx_ring_struct->ring_mask;
+	struct rte_mbuf **free = txq->free;
+	uint16_t cons = txr->tx_cons;
+	unsigned int blk = 0;
+
+	while (nr_pkts--) {
+		struct bnxt_sw_tx_bd *tx_buf;
+
+		tx_buf = &txr->tx_buf_ring[cons];
+		cons = (cons + 1) & ring_mask;
+		free[blk++] = tx_buf->mbuf;
+		tx_buf->mbuf = NULL;
+	}
+	if (blk)
+		rte_mempool_put_bulk(free[0]->pool, (void **)free, blk);
+
+	txr->tx_cons = cons;
+}
+
 void bnxt_free_tx_rings(struct bnxt *bp);
 int bnxt_init_one_tx_ring(struct bnxt_tx_queue *txq);
 int bnxt_init_tx_ring_struct(struct bnxt_tx_queue *txq, unsigned int socket_id);
-- 
2.25.1


  reply	other threads:[~2020-10-06 17:38 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-10-02 13:40 [dpdk-dev] [PATCH v2] " Lance Richardson
2020-10-06 17:37 ` Lance Richardson [this message]
2020-10-06 21:26   ` [dpdk-dev] [PATCH v3] " Ajit Khaparde

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=20201006173752.104002-1-lance.richardson@broadcom.com \
    --to=lance.richardson@broadcom.com \
    --cc=ajit.khaparde@broadcom.com \
    --cc=dev@dpdk.org \
    --cc=somnath.kotur@broadcom.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

DPDK patches and discussions

This inbox may be cloned and mirrored by anyone:

	git clone --mirror https://inbox.dpdk.org/dev/0 dev/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 dev dev/ https://inbox.dpdk.org/dev \
		dev@dpdk.org
	public-inbox-index dev

Example config snippet for mirrors.
Newsgroup available over NNTP:
	nntp://inbox.dpdk.org/inbox.dpdk.dev


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git