X-Git-Url: http://git.droids-corp.org/?a=blobdiff_plain;f=drivers%2Fnet%2Fbnxt%2Fbnxt_txr.c;h=47824334ae3e74f31bd273584735fe8337811a5b;hb=bdf4a3c6316b3e7b7a850efe5facf158cd658ee8;hp=4a152aa86f59b5682ed15d409b95fef41c1fcf30;hpb=294ca8e3ad847990eb099417f398a880f4224ccf;p=dpdk.git

diff --git a/drivers/net/bnxt/bnxt_txr.c b/drivers/net/bnxt/bnxt_txr.c
index 4a152aa86f..47824334ae 100644
--- a/drivers/net/bnxt/bnxt_txr.c
+++ b/drivers/net/bnxt/bnxt_txr.c
@@ -1,34 +1,6 @@
-/*-
- *   BSD LICENSE
- *
- *   Copyright(c) Broadcom Limited.
- *   All rights reserved.
- *
- *   Redistribution and use in source and binary forms, with or without
- *   modification, are permitted provided that the following conditions
- *   are met:
- *
- *     * Redistributions of source code must retain the above copyright
- *       notice, this list of conditions and the following disclaimer.
- *     * Redistributions in binary form must reproduce the above copyright
- *       notice, this list of conditions and the following disclaimer in
- *       the documentation and/or other materials provided with the
- *       distribution.
- *     * Neither the name of Broadcom Corporation nor the names of its
- *       contributors may be used to endorse or promote products derived
- *       from this software without specific prior written permission.
- *
- *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2014-2021 Broadcom
+ * All rights reserved.
  */
 
 #include <inttypes.h>
@@ -37,7 +9,7 @@
 #include <rte_malloc.h>
 
 #include "bnxt.h"
-#include "bnxt_cpr.h"
+#include "bnxt_hwrm.h"
 #include "bnxt_ring.h"
 #include "bnxt_txq.h"
 #include "bnxt_txr.h"
@@ -101,12 +73,13 @@ int bnxt_init_tx_ring_struct(struct bnxt_tx_queue *txq, unsigned int socket_id)
 	if (ring == NULL)
 		return -ENOMEM;
 	txr->tx_ring_struct = ring;
-	ring->ring_size = rte_align32pow2(txq->nb_tx_desc + 1);
+	ring->ring_size = rte_align32pow2(txq->nb_tx_desc);
 	ring->ring_mask = ring->ring_size - 1;
 	ring->bd = (void *)txr->tx_desc_ring;
 	ring->bd_dma = txr->tx_desc_mapping;
-	ring->vmem_size = ring->ring_size * sizeof(struct bnxt_sw_tx_bd);
+	ring->vmem_size = ring->ring_size * sizeof(struct rte_mbuf *);
 	ring->vmem = (void **)&txr->tx_buf_ring;
+	ring->fw_ring_id = INVALID_HW_RING_ID;
 
 	cpr = rte_zmalloc_socket("bnxt_tx_ring",
 				 sizeof(struct bnxt_cp_ring_info),
@@ -127,31 +100,42 @@ int bnxt_init_tx_ring_struct(struct bnxt_tx_queue *txq, unsigned int socket_id)
 	ring->bd_dma = cpr->cp_desc_mapping;
 	ring->vmem_size = 0;
 	ring->vmem = NULL;
+	ring->fw_ring_id = INVALID_HW_RING_ID;
 
 	return 0;
 }
 
-static inline uint32_t bnxt_tx_avail(struct bnxt_tx_ring_info *txr)
+static bool
+bnxt_xmit_need_long_bd(struct rte_mbuf *tx_pkt, struct bnxt_tx_queue *txq)
 {
-	/* Tell compiler to fetch tx indices from memory. */
-	rte_compiler_barrier();
-
-	return txr->tx_ring_struct->ring_size -
-		((txr->tx_prod - txr->tx_cons) &
-			txr->tx_ring_struct->ring_mask) - 1;
+	if (tx_pkt->ol_flags & (PKT_TX_TCP_SEG | PKT_TX_TCP_CKSUM |
+				PKT_TX_UDP_CKSUM | PKT_TX_IP_CKSUM |
+				PKT_TX_VLAN_PKT | PKT_TX_OUTER_IP_CKSUM |
+				PKT_TX_TUNNEL_GRE | PKT_TX_TUNNEL_VXLAN |
+				PKT_TX_TUNNEL_GENEVE | PKT_TX_IEEE1588_TMST |
+				PKT_TX_QINQ_PKT) ||
+	     (BNXT_TRUFLOW_EN(txq->bp) &&
+	      (txq->bp->tx_cfa_action || txq->vfr_tx_cfa_action)))
+		return true;
+	return false;
 }
 
 static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
-				struct bnxt_tx_queue *txq)
+				struct bnxt_tx_queue *txq,
+				uint16_t *coal_pkts,
+				struct tx_bd_long **last_txbd)
 {
 	struct bnxt_tx_ring_info *txr = txq->tx_ring;
+	struct bnxt_ring *ring = txr->tx_ring_struct;
+	uint32_t outer_tpid_bd = 0;
 	struct tx_bd_long *txbd;
-	struct tx_bd_long_hi *txbd1;
-	uint32_t vlan_tag_flags, cfa_action;
+	struct tx_bd_long_hi *txbd1 = NULL;
+	uint32_t vlan_tag_flags;
 	bool long_bd = false;
-	uint16_t last_prod = 0;
+	unsigned short nr_bds;
+	uint16_t prod;
 	struct rte_mbuf *m_seg;
-	struct bnxt_sw_tx_bd *tx_buf;
+	struct rte_mbuf **tx_buf;
 	static const uint32_t lhint_arr[4] = {
 		TX_BD_LONG_FLAGS_LHINT_LT512,
 		TX_BD_LONG_FLAGS_LHINT_LT1K,
@@ -159,40 +143,75 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 		TX_BD_LONG_FLAGS_LHINT_LT2K
 	};
 
-	if (tx_pkt->ol_flags & (PKT_TX_TCP_SEG | PKT_TX_TCP_CKSUM |
-				PKT_TX_UDP_CKSUM | PKT_TX_IP_CKSUM |
-				PKT_TX_VLAN_PKT | PKT_TX_OUTER_IP_CKSUM))
-		long_bd = true;
+	if (unlikely(is_bnxt_in_error(txq->bp)))
+		return -EIO;
 
-	tx_buf = &txr->tx_buf_ring[txr->tx_prod];
-	tx_buf->mbuf = tx_pkt;
-	tx_buf->nr_bds = long_bd + tx_pkt->nb_segs;
-	last_prod = (txr->tx_prod + tx_buf->nr_bds - 1) &
-				txr->tx_ring_struct->ring_mask;
+	long_bd = bnxt_xmit_need_long_bd(tx_pkt, txq);
+	nr_bds = long_bd + tx_pkt->nb_segs;
 
-	if (unlikely(bnxt_tx_avail(txr) < tx_buf->nr_bds))
+	if (unlikely(bnxt_tx_avail(txq) < nr_bds))
 		return -ENOMEM;
 
-	txbd = &txr->tx_desc_ring[txr->tx_prod];
-	txbd->opaque = txr->tx_prod;
-	txbd->flags_type = tx_buf->nr_bds << TX_BD_LONG_FLAGS_BD_CNT_SFT;
+	/* Check if number of Tx descriptors is above HW limit */
+	if (unlikely(nr_bds > BNXT_MAX_TSO_SEGS)) {
+		PMD_DRV_LOG(ERR,
+			    "Num descriptors %d exceeds HW limit\n", nr_bds);
+		return -ENOSPC;
+	}
+
+	/* If packet length is less than minimum packet size, pad it */
+	if (unlikely(rte_pktmbuf_pkt_len(tx_pkt) < BNXT_MIN_PKT_SIZE)) {
+		uint8_t pad = BNXT_MIN_PKT_SIZE - rte_pktmbuf_pkt_len(tx_pkt);
+		char *seg = rte_pktmbuf_append(tx_pkt, pad);
+
+		if (!seg) {
+			PMD_DRV_LOG(ERR,
+				    "Failed to pad mbuf by %d bytes\n",
+				    pad);
+			return -ENOMEM;
+		}
+
+		/* Note: data_len, pkt len are updated in rte_pktmbuf_append */
+		memset(seg, 0, pad);
+	}
+
+	/* Check non zero data_len */
+	RTE_VERIFY(tx_pkt->data_len);
+
+	prod = RING_IDX(ring, txr->tx_raw_prod);
+	tx_buf = &txr->tx_buf_ring[prod];
+	*tx_buf = tx_pkt;
+
+	txbd = &txr->tx_desc_ring[prod];
+	txbd->opaque = *coal_pkts;
+	txbd->flags_type = nr_bds << TX_BD_LONG_FLAGS_BD_CNT_SFT;
+	txbd->flags_type |= TX_BD_SHORT_FLAGS_COAL_NOW;
+	txbd->flags_type |= TX_BD_LONG_FLAGS_NO_CMPL;
 	txbd->len = tx_pkt->data_len;
-	if (txbd->len >= 2014)
+	if (tx_pkt->pkt_len >= 2048)
 		txbd->flags_type |= TX_BD_LONG_FLAGS_LHINT_GTE2K;
 	else
-		txbd->flags_type |= lhint_arr[txbd->len >> 9];
-	txbd->addr = rte_cpu_to_le_32(RTE_MBUF_DATA_DMA_ADDR(tx_buf->mbuf));
+		txbd->flags_type |= lhint_arr[tx_pkt->pkt_len >> 9];
+	txbd->address = rte_cpu_to_le_64(rte_mbuf_data_iova(tx_pkt));
+	*last_txbd = txbd;
 
 	if (long_bd) {
 		txbd->flags_type |= TX_BD_LONG_TYPE_TX_BD_LONG;
 		vlan_tag_flags = 0;
-		cfa_action = 0;
-		if (tx_buf->mbuf->ol_flags & PKT_TX_VLAN_PKT) {
+
+		/* HW can accelerate only outer vlan in QinQ mode */
+		if (tx_pkt->ol_flags & PKT_TX_QINQ_PKT) {
+			vlan_tag_flags = TX_BD_LONG_CFA_META_KEY_VLAN_TAG |
+				tx_pkt->vlan_tci_outer;
+			outer_tpid_bd = txq->bp->outer_tpid_bd &
+				BNXT_OUTER_TPID_BD_MASK;
+			vlan_tag_flags |= outer_tpid_bd;
+		} else if (tx_pkt->ol_flags & PKT_TX_VLAN_PKT) {
 			/* shurd: Should this mask at
 			 * TX_BD_LONG_CFA_META_VLAN_VID_MASK?
 			 */
 			vlan_tag_flags = TX_BD_LONG_CFA_META_KEY_VLAN_TAG |
-				tx_buf->mbuf->vlan_tci;
+				tx_pkt->vlan_tci;
 			/* Currently supports 8021Q, 8021AD vlan offloads
 			 * QINQ1, QINQ2, QINQ3 vlan headers are deprecated
 			 */
@@ -201,68 +220,122 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 					TX_BD_LONG_CFA_META_VLAN_TPID_TPID8100;
 		}
 
-		txr->tx_prod = RING_NEXT(txr->tx_ring_struct, txr->tx_prod);
+		txr->tx_raw_prod = RING_NEXT(txr->tx_raw_prod);
 
-		txbd1 = (struct tx_bd_long_hi *)
-					&txr->tx_desc_ring[txr->tx_prod];
+		prod = RING_IDX(ring, txr->tx_raw_prod);
+		txbd1 = (struct tx_bd_long_hi *)&txr->tx_desc_ring[prod];
 		txbd1->lflags = 0;
 		txbd1->cfa_meta = vlan_tag_flags;
-		txbd1->cfa_action = cfa_action;
+		/* Legacy tx_bd_long_hi->mss =
+		 * tx_bd_long_hi->kid_or_ts_high_mss
+		 */
+		txbd1->kid_or_ts_high_mss = 0;
+
+		if (txq->vfr_tx_cfa_action)
+			txbd1->cfa_action = txq->vfr_tx_cfa_action;
+		else
+			txbd1->cfa_action = txq->bp->tx_cfa_action;
 
 		if (tx_pkt->ol_flags & PKT_TX_TCP_SEG) {
+			uint16_t hdr_size;
+
 			/* TSO */
-			txbd1->lflags |= TX_BD_LONG_LFLAGS_LSO;
-			txbd1->hdr_size = tx_pkt->l2_len + tx_pkt->l3_len +
-					tx_pkt->l4_len + tx_pkt->outer_l2_len +
-					tx_pkt->outer_l3_len;
-			txbd1->mss = tx_pkt->tso_segsz;
+			txbd1->lflags |= TX_BD_LONG_LFLAGS_LSO |
+					 TX_BD_LONG_LFLAGS_T_IPID;
+			hdr_size = tx_pkt->l2_len + tx_pkt->l3_len +
+					tx_pkt->l4_len;
+			hdr_size += (tx_pkt->ol_flags & PKT_TX_TUNNEL_MASK) ?
+				    tx_pkt->outer_l2_len +
+				    tx_pkt->outer_l3_len : 0;
+			/* The hdr_size is multiple of 16bit units not 8bit.
+			 * Hence divide by 2.
+			 * Also legacy hdr_size = kid_or_ts_low_hdr_size.
+			 */
+			txbd1->kid_or_ts_low_hdr_size = hdr_size >> 1;
+			txbd1->kid_or_ts_high_mss = tx_pkt->tso_segsz;
+			RTE_VERIFY(txbd1->kid_or_ts_high_mss);
 
 		} else if ((tx_pkt->ol_flags & PKT_TX_OIP_IIP_TCP_UDP_CKSUM) ==
 			   PKT_TX_OIP_IIP_TCP_UDP_CKSUM) {
 			/* Outer IP, Inner IP, Inner TCP/UDP CSO */
 			txbd1->lflags |= TX_BD_FLG_TIP_IP_TCP_UDP_CHKSUM;
-			txbd1->mss = 0;
+		} else if ((tx_pkt->ol_flags & PKT_TX_OIP_IIP_TCP_CKSUM) ==
+			   PKT_TX_OIP_IIP_TCP_CKSUM) {
+			/* Outer IP, Inner IP, Inner TCP/UDP CSO */
+			txbd1->lflags |= TX_BD_FLG_TIP_IP_TCP_UDP_CHKSUM;
+		} else if ((tx_pkt->ol_flags & PKT_TX_OIP_IIP_UDP_CKSUM) ==
+			   PKT_TX_OIP_IIP_UDP_CKSUM) {
+			/* Outer IP, Inner IP, Inner TCP/UDP CSO */
+			txbd1->lflags |= TX_BD_FLG_TIP_IP_TCP_UDP_CHKSUM;
 		} else if ((tx_pkt->ol_flags & PKT_TX_IIP_TCP_UDP_CKSUM) ==
 			   PKT_TX_IIP_TCP_UDP_CKSUM) {
 			/* (Inner) IP, (Inner) TCP/UDP CSO */
 			txbd1->lflags |= TX_BD_FLG_IP_TCP_UDP_CHKSUM;
-			txbd1->mss = 0;
+		} else if ((tx_pkt->ol_flags & PKT_TX_IIP_UDP_CKSUM) ==
+			   PKT_TX_IIP_UDP_CKSUM) {
+			/* (Inner) IP, (Inner) TCP/UDP CSO */
+			txbd1->lflags |= TX_BD_FLG_IP_TCP_UDP_CHKSUM;
+		} else if ((tx_pkt->ol_flags & PKT_TX_IIP_TCP_CKSUM) ==
+			   PKT_TX_IIP_TCP_CKSUM) {
+			/* (Inner) IP, (Inner) TCP/UDP CSO */
+			txbd1->lflags |= TX_BD_FLG_IP_TCP_UDP_CHKSUM;
 		} else if ((tx_pkt->ol_flags & PKT_TX_OIP_TCP_UDP_CKSUM) ==
 			   PKT_TX_OIP_TCP_UDP_CKSUM) {
 			/* Outer IP, (Inner) TCP/UDP CSO */
 			txbd1->lflags |= TX_BD_FLG_TIP_TCP_UDP_CHKSUM;
-			txbd1->mss = 0;
+		} else if ((tx_pkt->ol_flags & PKT_TX_OIP_UDP_CKSUM) ==
+			   PKT_TX_OIP_UDP_CKSUM) {
+			/* Outer IP, (Inner) TCP/UDP CSO */
+			txbd1->lflags |= TX_BD_FLG_TIP_TCP_UDP_CHKSUM;
+		} else if ((tx_pkt->ol_flags & PKT_TX_OIP_TCP_CKSUM) ==
+			   PKT_TX_OIP_TCP_CKSUM) {
+			/* Outer IP, (Inner) TCP/UDP CSO */
+			txbd1->lflags |= TX_BD_FLG_TIP_TCP_UDP_CHKSUM;
 		} else if ((tx_pkt->ol_flags & PKT_TX_OIP_IIP_CKSUM) ==
 			   PKT_TX_OIP_IIP_CKSUM) {
 			/* Outer IP, Inner IP CSO */
 			txbd1->lflags |= TX_BD_FLG_TIP_IP_CHKSUM;
-			txbd1->mss = 0;
 		} else if ((tx_pkt->ol_flags & PKT_TX_TCP_UDP_CKSUM) ==
 			   PKT_TX_TCP_UDP_CKSUM) {
 			/* TCP/UDP CSO */
 			txbd1->lflags |= TX_BD_LONG_LFLAGS_TCP_UDP_CHKSUM;
-			txbd1->mss = 0;
-		} else if (tx_pkt->ol_flags & PKT_TX_IP_CKSUM) {
+		} else if ((tx_pkt->ol_flags & PKT_TX_TCP_CKSUM) ==
+			   PKT_TX_TCP_CKSUM) {
+			/* TCP/UDP CSO */
+			txbd1->lflags |= TX_BD_LONG_LFLAGS_TCP_UDP_CHKSUM;
+		} else if ((tx_pkt->ol_flags & PKT_TX_UDP_CKSUM) ==
+			   PKT_TX_UDP_CKSUM) {
+			/* TCP/UDP CSO */
+			txbd1->lflags |= TX_BD_LONG_LFLAGS_TCP_UDP_CHKSUM;
+		} else if ((tx_pkt->ol_flags & PKT_TX_IP_CKSUM) ==
+			   PKT_TX_IP_CKSUM) {
 			/* IP CSO */
 			txbd1->lflags |= TX_BD_LONG_LFLAGS_IP_CHKSUM;
-			txbd1->mss = 0;
-		} else if (tx_pkt->ol_flags & PKT_TX_OUTER_IP_CKSUM) {
+		} else if ((tx_pkt->ol_flags & PKT_TX_OUTER_IP_CKSUM) ==
+			   PKT_TX_OUTER_IP_CKSUM) {
 			/* IP CSO */
 			txbd1->lflags |= TX_BD_LONG_LFLAGS_T_IP_CHKSUM;
-			txbd1->mss = 0;
+		} else if ((tx_pkt->ol_flags & PKT_TX_IEEE1588_TMST) ==
+			   PKT_TX_IEEE1588_TMST) {
+			/* PTP */
+			txbd1->lflags |= TX_BD_LONG_LFLAGS_STAMP;
 		}
 	} else {
 		txbd->flags_type |= TX_BD_SHORT_TYPE_TX_BD_SHORT;
 	}
 
 	m_seg = tx_pkt->next;
-	/* i is set at the end of the if(long_bd) block */
-	while (txr->tx_prod != last_prod) {
-		txr->tx_prod = RING_NEXT(txr->tx_ring_struct, txr->tx_prod);
-		tx_buf = &txr->tx_buf_ring[txr->tx_prod];
+	while (m_seg) {
+		/* Check non zero data_len */
+		RTE_VERIFY(m_seg->data_len);
+		txr->tx_raw_prod = RING_NEXT(txr->tx_raw_prod);
+
+		prod = RING_IDX(ring, txr->tx_raw_prod);
+		tx_buf = &txr->tx_buf_ring[prod];
+		*tx_buf = m_seg;
 
-		txbd = &txr->tx_desc_ring[txr->tx_prod];
-		txbd->addr = rte_cpu_to_le_32(RTE_MBUF_DATA_DMA_ADDR(m_seg));
+		txbd = &txr->tx_desc_ring[prod];
+		txbd->address = rte_cpu_to_le_64(rte_mbuf_data_iova(m_seg));
 		txbd->flags_type = TX_BD_SHORT_TYPE_TX_BD_SHORT;
 		txbd->len = m_seg->data_len;
 
@@ -271,96 +344,274 @@ static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
 
 	txbd->flags_type |= TX_BD_LONG_FLAGS_PACKET_END;
 
-	txr->tx_prod = RING_NEXT(txr->tx_ring_struct, txr->tx_prod);
+	txr->tx_raw_prod = RING_NEXT(txr->tx_raw_prod);
 
 	return 0;
 }
 
-static void bnxt_tx_cmp(struct bnxt_tx_queue *txq, int nr_pkts)
+/*
+ * Transmit completion function for use when DEV_TX_OFFLOAD_MBUF_FAST_FREE
+ * is enabled.
+ */
+static void bnxt_tx_cmp_fast(struct bnxt_tx_queue *txq, int nr_pkts)
 {
 	struct bnxt_tx_ring_info *txr = txq->tx_ring;
-	uint16_t cons = txr->tx_cons;
+	struct bnxt_ring *ring = txr->tx_ring_struct;
+	struct rte_mbuf **free = txq->free;
+	uint16_t raw_cons = txr->tx_raw_cons;
+	unsigned int blk = 0;
 	int i, j;
 
 	for (i = 0; i < nr_pkts; i++) {
-		struct bnxt_sw_tx_bd *tx_buf;
-		struct rte_mbuf *mbuf;
+		struct rte_mbuf **tx_buf;
+		unsigned short nr_bds;
+
+		tx_buf = &txr->tx_buf_ring[RING_IDX(ring, raw_cons)];
+		nr_bds = (*tx_buf)->nb_segs +
+			 bnxt_xmit_need_long_bd(*tx_buf, txq);
+		for (j = 0; j < nr_bds; j++) {
+			if (*tx_buf) {
+				/* Add mbuf to the bulk free array */
+				free[blk++] = *tx_buf;
+				*tx_buf = NULL;
+			}
+			raw_cons = RING_NEXT(raw_cons);
+			tx_buf = &txr->tx_buf_ring[RING_IDX(ring, raw_cons)];
+		}
+	}
+	if (blk)
+		rte_mempool_put_bulk(free[0]->pool, (void *)free, blk);
 
-		tx_buf = &txr->tx_buf_ring[cons];
-		cons = RING_NEXT(txr->tx_ring_struct, cons);
-		mbuf = tx_buf->mbuf;
-		tx_buf->mbuf = NULL;
+	txr->tx_raw_cons = raw_cons;
+}
 
-		/* EW - no need to unmap DMA memory? */
+static void bnxt_tx_cmp(struct bnxt_tx_queue *txq, int nr_pkts)
+{
+	struct bnxt_tx_ring_info *txr = txq->tx_ring;
+	struct bnxt_ring *ring = txr->tx_ring_struct;
+	struct rte_mempool *pool = NULL;
+	struct rte_mbuf **free = txq->free;
+	uint16_t raw_cons = txr->tx_raw_cons;
+	unsigned int blk = 0;
+	int i, j;
 
-		for (j = 1; j < tx_buf->nr_bds; j++)
-			cons = RING_NEXT(txr->tx_ring_struct, cons);
-		rte_pktmbuf_free(mbuf);
+	for (i = 0; i < nr_pkts; i++) {
+		struct rte_mbuf *mbuf;
+		struct rte_mbuf **tx_buf;
+		unsigned short nr_bds;
+
+		tx_buf = &txr->tx_buf_ring[RING_IDX(ring, raw_cons)];
+		nr_bds = (*tx_buf)->nb_segs +
+			 bnxt_xmit_need_long_bd(*tx_buf, txq);
+		for (j = 0; j < nr_bds; j++) {
+			mbuf = *tx_buf;
+			*tx_buf = NULL;
+			raw_cons = RING_NEXT(raw_cons);
+			tx_buf = &txr->tx_buf_ring[RING_IDX(ring, raw_cons)];
+			if (!mbuf)	/* long_bd's tx_buf ? */
+				continue;
+
+			mbuf = rte_pktmbuf_prefree_seg(mbuf);
+			if (unlikely(!mbuf))
+				continue;
+
+			/* EW - no need to unmap DMA memory? */
+
+			if (likely(mbuf->pool == pool)) {
+				/* Add mbuf to the bulk free array */
+				free[blk++] = mbuf;
+			} else {
+				/* Found an mbuf from a different pool. Free
+				 * mbufs accumulated so far to the previous
+				 * pool
+				 */
+				if (likely(pool != NULL))
+					rte_mempool_put_bulk(pool,
+							     (void *)free,
+							     blk);
+
+				/* Start accumulating mbufs in a new pool */
+				free[0] = mbuf;
+				pool = mbuf->pool;
+				blk = 1;
+			}
+		}
 	}
+	if (blk)
+		rte_mempool_put_bulk(pool, (void *)free, blk);
 
-	txr->tx_cons = cons;
+	txr->tx_raw_cons = raw_cons;
 }
 
 static int bnxt_handle_tx_cp(struct bnxt_tx_queue *txq)
 {
+	uint32_t nb_tx_pkts = 0, cons, ring_mask, opaque;
 	struct bnxt_cp_ring_info *cpr = txq->cp_ring;
 	uint32_t raw_cons = cpr->cp_raw_cons;
-	uint32_t cons;
-	int nb_tx_pkts = 0;
+	struct bnxt_ring *cp_ring_struct;
 	struct tx_cmpl *txcmp;
 
-	if ((txq->tx_ring->tx_ring_struct->ring_size -
-			(bnxt_tx_avail(txq->tx_ring))) >
-			txq->tx_free_thresh) {
-		while (1) {
-			cons = RING_CMP(cpr->cp_ring_struct, raw_cons);
-			txcmp = (struct tx_cmpl *)&cpr->cp_desc_ring[cons];
-
-			if (!CMP_VALID(txcmp, raw_cons, cpr->cp_ring_struct))
-				break;
-			cpr->valid = FLIP_VALID(cons,
-						cpr->cp_ring_struct->ring_mask,
-						cpr->valid);
-
-			if (CMP_TYPE(txcmp) == TX_CMPL_TYPE_TX_L2)
-				nb_tx_pkts++;
-			else
-				RTE_LOG_DP(DEBUG, PMD,
-						"Unhandled CMP type %02x\n",
-						CMP_TYPE(txcmp));
-			raw_cons = NEXT_RAW_CMP(raw_cons);
-		}
-		if (nb_tx_pkts)
+	if (bnxt_tx_bds_in_hw(txq) < txq->tx_free_thresh)
+		return 0;
+
+	cp_ring_struct = cpr->cp_ring_struct;
+	ring_mask = cp_ring_struct->ring_mask;
+
+	do {
+		cons = RING_CMPL(ring_mask, raw_cons);
+		txcmp = (struct tx_cmpl *)&cpr->cp_desc_ring[cons];
+
+		if (!bnxt_cpr_cmp_valid(txcmp, raw_cons, ring_mask + 1))
+			break;
+
+		opaque = rte_le_to_cpu_32(txcmp->opaque);
+
+		if (CMP_TYPE(txcmp) == TX_CMPL_TYPE_TX_L2)
+			nb_tx_pkts += opaque;
+		else
+			RTE_LOG_DP(ERR, PMD,
+					"Unhandled CMP type %02x\n",
+					CMP_TYPE(txcmp));
+		raw_cons = NEXT_RAW_CMP(raw_cons);
+	} while (nb_tx_pkts < ring_mask);
+
+	if (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;
-		B_CP_DIS_DB(cpr, cpr->cp_raw_cons);
+		bnxt_db_cq(cpr);
 	}
+
 	return nb_tx_pkts;
 }
 
 uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 			       uint16_t nb_pkts)
 {
-	struct bnxt_tx_queue *txq = tx_queue;
+	int rc;
 	uint16_t nb_tx_pkts = 0;
-	uint16_t db_mask = txq->tx_ring->tx_ring_struct->ring_size >> 2;
-	uint16_t last_db_mask = 0;
+	uint16_t coal_pkts = 0;
+	struct bnxt_tx_queue *txq = tx_queue;
+	struct tx_bd_long *last_txbd = NULL;
 
 	/* Handle TX completions */
 	bnxt_handle_tx_cp(txq);
 
+	/* Tx queue was stopped; wait for it to be restarted */
+	if (unlikely(!txq->tx_started)) {
+		PMD_DRV_LOG(DEBUG, "Tx q stopped;return\n");
+		return 0;
+	}
+
 	/* Handle TX burst request */
 	for (nb_tx_pkts = 0; nb_tx_pkts < nb_pkts; nb_tx_pkts++) {
-		if (bnxt_start_xmit(tx_pkts[nb_tx_pkts], txq)) {
+		coal_pkts++;
+		rc = bnxt_start_xmit(tx_pkts[nb_tx_pkts], txq,
+				     &coal_pkts, &last_txbd);
+
+		if (unlikely(rc))
 			break;
-		} else if ((nb_tx_pkts & db_mask) != last_db_mask) {
-			B_TX_DB(txq->tx_ring->tx_doorbell,
-					txq->tx_ring->tx_prod);
-			last_db_mask = nb_tx_pkts & db_mask;
-		}
 	}
-	if (nb_tx_pkts)
-		B_TX_DB(txq->tx_ring->tx_doorbell, txq->tx_ring->tx_prod);
+
+	if (likely(nb_tx_pkts)) {
+		/* Request a completion on the last packet */
+		last_txbd->flags_type &= ~TX_BD_LONG_FLAGS_NO_CMPL;
+		bnxt_db_write(&txq->tx_ring->tx_db, txq->tx_ring->tx_raw_prod);
+	}
 
 	return nb_tx_pkts;
 }
+
+/*
+ * Dummy DPDK callback for TX.
+ *
+ * This function is used to temporarily replace the real callback during
+ * unsafe control operations on the queue, or in case of error.
+ */
+uint16_t
+bnxt_dummy_xmit_pkts(void *tx_queue __rte_unused,
+		     struct rte_mbuf **tx_pkts __rte_unused,
+		     uint16_t nb_pkts __rte_unused)
+{
+	return 0;
+}
+
+int bnxt_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+	struct bnxt *bp = dev->data->dev_private;
+	struct bnxt_tx_queue *txq = bp->tx_queues[tx_queue_id];
+	int rc = 0;
+
+	rc = is_bnxt_in_error(bp);
+	if (rc)
+		return rc;
+
+	bnxt_free_hwrm_tx_ring(bp, tx_queue_id);
+	rc = bnxt_alloc_hwrm_tx_ring(bp, tx_queue_id);
+	if (rc)
+		return rc;
+
+	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+	txq->tx_started = true;
+	PMD_DRV_LOG(DEBUG, "Tx queue started\n");
+
+	return 0;
+}
+
+int bnxt_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+	struct bnxt *bp = dev->data->dev_private;
+	struct bnxt_tx_queue *txq = bp->tx_queues[tx_queue_id];
+	int rc = 0;
+
+	rc = is_bnxt_in_error(bp);
+	if (rc)
+		return rc;
+
+	/* Handle TX completions */
+	bnxt_handle_tx_cp(txq);
+
+	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+	txq->tx_started = false;
+	PMD_DRV_LOG(DEBUG, "Tx queue stopped\n");
+
+	return 0;
+}
+
+/* Sweep the Tx completion queue till HWRM_DONE for ring flush is received.
+ * The mbufs will not be freed in this call.
+ * They will be freed during ring free as a part of mem cleanup.
+ */
+int bnxt_flush_tx_cmp(struct bnxt_cp_ring_info *cpr)
+{
+	uint32_t raw_cons = cpr->cp_raw_cons;
+	uint32_t cons;
+	uint32_t nb_tx_pkts = 0;
+	struct tx_cmpl *txcmp;
+	struct cmpl_base *cp_desc_ring = cpr->cp_desc_ring;
+	struct bnxt_ring *cp_ring_struct = cpr->cp_ring_struct;
+	uint32_t ring_mask = cp_ring_struct->ring_mask;
+	uint32_t opaque = 0;
+
+	do {
+		cons = RING_CMPL(ring_mask, raw_cons);
+		txcmp = (struct tx_cmpl *)&cp_desc_ring[cons];
+
+		opaque = rte_cpu_to_le_32(txcmp->opaque);
+		raw_cons = NEXT_RAW_CMP(raw_cons);
+
+		if (CMP_TYPE(txcmp) == TX_CMPL_TYPE_TX_L2)
+			nb_tx_pkts += opaque;
+		else if (CMP_TYPE(txcmp) == HWRM_CMPL_TYPE_HWRM_DONE)
+			return 1;
+	} while (nb_tx_pkts < ring_mask);
+
+	if (nb_tx_pkts) {
+		cpr->cp_raw_cons = raw_cons;
+		bnxt_db_cq(cpr);
+	}
+
+	return 0;
+}