[dpdk-dev] [PATCH 13/40] bnxt: initial Tx ring code

[Stephen Hurd <stephen.hurd@broadcom.com>]

Initial implementation of rx_pkt_burst
Add code to allocate rings to bnxt_ring.c

Signed-off-by: Stephen Hurd <stephen.hurd@broadcom.com>
Reviewed-by: Ajit Kumar Khaparde <ajit.khaparde@broadcom.com>
---
 drivers/net/bnxt/Makefile              |   1 +
 drivers/net/bnxt/bnxt_cpr.h            |   4 +-
 drivers/net/bnxt/bnxt_ethdev.c         |   5 +-
 drivers/net/bnxt/bnxt_ring.c           | 140 +++++++++
 drivers/net/bnxt/bnxt_ring.h           |   8 +
 drivers/net/bnxt/bnxt_txq.c            |  42 ++-
 drivers/net/bnxt/bnxt_txr.c            | 314 ++++++++++++++++++++
 drivers/net/bnxt/bnxt_txr.h            |  71 +++++
 drivers/net/bnxt/hsi_struct_def_dpdk.h | 512 +++++++++++++++++++++++++++++++++
 9 files changed, 1086 insertions(+), 11 deletions(-)
 create mode 100644 drivers/net/bnxt/bnxt_txr.c
 create mode 100644 drivers/net/bnxt/bnxt_txr.h

diff --git a/drivers/net/bnxt/Makefile b/drivers/net/bnxt/Makefile
index de9ce76..009ddad 100644
--- a/drivers/net/bnxt/Makefile
+++ b/drivers/net/bnxt/Makefile
@@ -56,6 +56,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_BNXT_PMD) += bnxt_ring.c
 SRCS-$(CONFIG_RTE_LIBRTE_BNXT_PMD) += bnxt_rxq.c
 SRCS-$(CONFIG_RTE_LIBRTE_BNXT_PMD) += bnxt_stats.c
 SRCS-$(CONFIG_RTE_LIBRTE_BNXT_PMD) += bnxt_txq.c
+SRCS-$(CONFIG_RTE_LIBRTE_BNXT_PMD) += bnxt_txr.c
 SRCS-$(CONFIG_RTE_LIBRTE_BNXT_PMD) += bnxt_vnic.c
 
 #
diff --git a/drivers/net/bnxt/bnxt_cpr.h b/drivers/net/bnxt/bnxt_cpr.h
index e6333fc..f104281 100644
--- a/drivers/net/bnxt/bnxt_cpr.h
+++ b/drivers/net/bnxt/bnxt_cpr.h
@@ -51,11 +51,11 @@
 
 #define B_CP_DB_REARM(cpr, raw_cons)					\
 		(*(uint32_t *)((cpr)->cp_doorbell) = (DB_CP_REARM_FLAGS | \
-				RING_CMP(&cpr->cp_ring_struct, raw_cons)))
+				RING_CMP(cpr->cp_ring_struct, raw_cons)))
 
 #define B_CP_DIS_DB(cpr, raw_cons)					\
 		(*(uint32_t *)((cpr)->cp_doorbell) = (DB_CP_FLAGS |	\
-				RING_CMP(&cpr->cp_ring_struct, raw_cons)))
+				RING_CMP(cpr->cp_ring_struct, raw_cons)))
 
 struct bnxt_ring_struct;
 struct bnxt_cp_ring_info {
diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c
index 786318c..61e856a 100644
--- a/drivers/net/bnxt/bnxt_ethdev.c
+++ b/drivers/net/bnxt/bnxt_ethdev.c
@@ -44,6 +44,7 @@
 #include "bnxt_rxq.h"
 #include "bnxt_stats.h"
 #include "bnxt_txq.h"
+#include "bnxt_txr.h"
 
 #define DRV_MODULE_NAME		"bnxt"
 static const char bnxt_version[] =
@@ -259,10 +260,8 @@ bnxt_dev_init(struct rte_eth_dev *eth_dev)
 		goto error;
 	}
 	eth_dev->dev_ops = &bnxt_dev_ops;
-	/*
-	eth_dev->rx_pkt_burst = &bnxt_recv_pkts;
+	/* eth_dev->rx_pkt_burst = &bnxt_recv_pkts; */
 	eth_dev->tx_pkt_burst = &bnxt_xmit_pkts;
-	 */
 
 	rc = bnxt_alloc_hwrm_resources(bp);
 	if (rc) {
diff --git a/drivers/net/bnxt/bnxt_ring.c b/drivers/net/bnxt/bnxt_ring.c
index 0434b07..bb20806 100644
--- a/drivers/net/bnxt/bnxt_ring.c
+++ b/drivers/net/bnxt/bnxt_ring.c
@@ -31,8 +31,14 @@
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <rte_memzone.h>
+
 #include "bnxt.h"
+#include "bnxt_cpr.h"
 #include "bnxt_ring.h"
+#include "bnxt_txr.h"
+
+#include "hsi_struct_def_dpdk.h"
 
 /*
  * Generic ring handling
@@ -49,3 +55,137 @@ void bnxt_free_ring(struct bnxt_ring_struct *ring)
 		*ring->vmem = NULL;
 	}
 }
+
+/*
+ * Allocates a completion ring with vmem and stats optionally also allocating
+ * a TX and/or RX ring.  Passing NULL as tx_ring_info and/or rx_ring_info
+ * to not allocate them.
+ *
+ * Order in the allocation is:
+ * stats - Always non-zero length
+ * cp vmem - Always zero-length, supported for the bnxt_ring_struct abstraction
+ * tx vmem - Only non-zero length if tx_ring_info is not NULL
+ * rx vmem - Only non-zero length if rx_ring_info is not NULL
+ * cp bd ring - Always non-zero length
+ * tx bd ring - Only non-zero length if tx_ring_info is not NULL
+ * rx bd ring - Only non-zero length if rx_ring_info is not NULL
+ */
+int bnxt_alloc_rings(struct bnxt *bp, uint16_t qidx,
+			    struct bnxt_tx_ring_info *tx_ring_info,
+			    struct bnxt_rx_ring_info *rx_ring_info,
+			    struct bnxt_cp_ring_info *cp_ring_info,
+			    const char *suffix)
+{
+	struct bnxt_ring_struct *cp_ring = cp_ring_info->cp_ring_struct;
+	struct bnxt_ring_struct *tx_ring;
+	/* TODO: RX ring */
+	/* struct bnxt_ring_struct *rx_ring; */
+	struct rte_pci_device *pdev = bp->pdev;
+	const struct rte_memzone *mz = NULL;
+	char mz_name[RTE_MEMZONE_NAMESIZE];
+
+	int stats_len = (tx_ring_info || rx_ring_info) ?
+	    RTE_CACHE_LINE_ROUNDUP(sizeof(struct ctx_hw_stats64)) : 0;
+
+	int cp_vmem_start = stats_len;
+	int cp_vmem_len = RTE_CACHE_LINE_ROUNDUP(cp_ring->vmem_size);
+
+	int tx_vmem_start = cp_vmem_start + cp_vmem_len;
+	int tx_vmem_len =
+	    tx_ring_info ? RTE_CACHE_LINE_ROUNDUP(tx_ring_info->
+						tx_ring_struct->vmem_size) : 0;
+
+	int rx_vmem_start = tx_vmem_start + tx_vmem_len;
+	/* TODO: RX ring */
+	int rx_vmem_len = /*
+	    rx_ring_info ? RTE_CACHE_LINE_ROUNDUP(rx_ring_info->
+					rx_ring_struct->vmem_size) : */ 0;
+
+	int cp_ring_start = rx_vmem_start + rx_vmem_len;
+	int cp_ring_len = RTE_CACHE_LINE_ROUNDUP(cp_ring->ring_size *
+						 sizeof(struct cmpl_base));
+
+	int tx_ring_start = cp_ring_start + cp_ring_len;
+	int tx_ring_len = tx_ring_info ?
+	    RTE_CACHE_LINE_ROUNDUP(tx_ring_info->tx_ring_struct->ring_size *
+				   sizeof(struct tx_bd_long)) : 0;
+
+	int rx_ring_start = tx_ring_start + tx_ring_len;
+	/* TODO: RX ring */
+	int rx_ring_len = /* rx_ring_info ?
+	    RTE_CACHE_LINE_ROUNDUP(rx_ring_info->rx_ring_struct->ring_size *
+				   sizeof(struct rx_prod_pkt_bd)) : */ 0;
+
+	int total_alloc_len = rx_ring_start + rx_ring_len;
+
+	snprintf(mz_name, RTE_MEMZONE_NAMESIZE,
+		 "bnxt_%04x:%02x:%02x:%02x-%04x_%s", pdev->addr.domain,
+		 pdev->addr.bus, pdev->addr.devid, pdev->addr.function, qidx,
+		 suffix);
+	mz_name[RTE_MEMZONE_NAMESIZE - 1] = 0;
+	mz = rte_memzone_lookup(mz_name);
+	if (!mz) {
+		mz = rte_memzone_reserve(mz_name, total_alloc_len,
+					 SOCKET_ID_ANY,
+					 RTE_MEMZONE_2MB |
+					 RTE_MEMZONE_SIZE_HINT_ONLY);
+		if (mz == NULL)
+			return -ENOMEM;
+	}
+	memset(mz->addr, 0, mz->len);
+
+	if (tx_ring_info) {
+		tx_ring = tx_ring_info->tx_ring_struct;
+
+		tx_ring->bd = ((char *)mz->addr + tx_ring_start);
+		tx_ring_info->tx_desc_ring = (struct tx_bd_long *)tx_ring->bd;
+		tx_ring->bd_dma = mz->phys_addr + tx_ring_start;
+		tx_ring_info->tx_desc_mapping = tx_ring->bd_dma;
+
+		if (!tx_ring->bd)
+			return -ENOMEM;
+		if (tx_ring->vmem_size) {
+			tx_ring->vmem =
+			    (void **)((char *)mz->addr + tx_vmem_start);
+			tx_ring_info->tx_buf_ring =
+			    (struct bnxt_sw_tx_bd *)tx_ring->vmem;
+		}
+	}
+
+/*
+	if (rx_ring_info) {
+		rx_ring = &rx_ring_info->rx_ring_struct;
+
+		rx_ring->bd = ((char *)mz->addr + rx_ring_start);
+		rx_ring_info->rx_desc_ring =
+		    (struct rx_prod_pkt_bd *)rx_ring->bd;
+		rx_ring->bd_dma = mz->phys_addr + rx_ring_start;
+		rx_ring_info->rx_desc_mapping = rx_ring->bd_dma;
+
+		if (!rx_ring->bd)
+			return -ENOMEM;
+		if (rx_ring->vmem_size) {
+			rx_ring->vmem =
+			    (void **)((char *)mz->addr + rx_vmem_start);
+			rx_ring_info->rx_buf_ring =
+			    (struct bnxt_sw_rx_bd *)rx_ring->vmem;
+		}
+	}
+*/
+
+	cp_ring->bd = ((char *)mz->addr + cp_ring_start);
+	cp_ring->bd_dma = mz->phys_addr + cp_ring_start;
+	cp_ring_info->cp_desc_ring = cp_ring->bd;
+	cp_ring_info->cp_desc_mapping = cp_ring->bd_dma;
+
+	if (!cp_ring->bd)
+		return -ENOMEM;
+	if (cp_ring->vmem_size)
+		*cp_ring->vmem = ((char *)mz->addr + stats_len);
+	if (stats_len) {
+		cp_ring_info->hw_stats = mz->addr;
+		cp_ring_info->hw_stats_map = mz->phys_addr;
+	}
+	cp_ring_info->hw_stats_ctx_id = HWRM_NA_SIGNATURE;
+	return 0;
+}
diff --git a/drivers/net/bnxt/bnxt_ring.h b/drivers/net/bnxt/bnxt_ring.h
index f44025c..dfa0401 100644
--- a/drivers/net/bnxt/bnxt_ring.h
+++ b/drivers/net/bnxt/bnxt_ring.h
@@ -87,6 +87,14 @@ struct bnxt_ring_grp_info {
 };
 
 struct bnxt;
+struct bnxt_tx_ring_info;
+struct bnxt_rx_ring_info;
+struct bnxt_cp_ring_info;
 void bnxt_free_ring(struct bnxt_ring_struct *ring);
+int bnxt_alloc_rings(struct bnxt *bp, uint16_t qidx,
+			    struct bnxt_tx_ring_info *tx_ring_info,
+			    struct bnxt_rx_ring_info *rx_ring_info,
+			    struct bnxt_cp_ring_info *cp_ring_info,
+			    const char *suffix);
 
 #endif
diff --git a/drivers/net/bnxt/bnxt_txq.c b/drivers/net/bnxt/bnxt_txq.c
index a3648c2..7aba199 100644
--- a/drivers/net/bnxt/bnxt_txq.c
+++ b/drivers/net/bnxt/bnxt_txq.c
@@ -39,6 +39,7 @@
 #include "bnxt_cpr.h"
 #include "bnxt_ring.h"
 #include "bnxt_txq.h"
+#include "bnxt_txr.h"
 
 /*
  * TX Queues
@@ -55,9 +56,20 @@ void bnxt_free_txq_stats(struct bnxt_tx_queue *txq)
 		cpr->hw_stats = NULL;
 }
 
-static void bnxt_tx_queue_release_mbufs(struct bnxt_tx_queue *txq __rte_unused)
+static void bnxt_tx_queue_release_mbufs(struct bnxt_tx_queue *txq)
 {
-	/* TODO: Requires interaction with TX ring */
+	struct bnxt_sw_tx_bd *sw_ring;
+	uint16_t i;
+
+	sw_ring = txq->tx_ring->tx_buf_ring;
+	if (sw_ring) {
+		for (i = 0; i < txq->tx_ring->tx_ring_struct->ring_size; i++) {
+			if (sw_ring[i].mbuf) {
+				rte_pktmbuf_free(sw_ring[i].mbuf);
+				sw_ring[i].mbuf = NULL;
+			}
+		}
+	}
 }
 
 void bnxt_free_tx_mbufs(struct bnxt *bp)
@@ -76,7 +88,15 @@ void bnxt_tx_queue_release_op(void *tx_queue)
 	struct bnxt_tx_queue *txq = (struct bnxt_tx_queue *)tx_queue;
 
 	if (txq) {
-		/* TODO: Free ring and stats here */
+		/* Free TX ring hardware descriptors */
+		bnxt_tx_queue_release_mbufs(txq);
+		bnxt_free_ring(txq->tx_ring->tx_ring_struct);
+
+		/* Free TX completion ring hardware descriptors */
+		bnxt_free_ring(txq->cp_ring->cp_ring_struct);
+
+		bnxt_free_txq_stats(txq);
+
 		rte_free(txq);
 	}
 }
@@ -112,14 +132,24 @@ int bnxt_tx_queue_setup_op(struct rte_eth_dev *eth_dev,
 	txq->nb_tx_desc = nb_desc;
 	txq->tx_free_thresh = tx_conf->tx_free_thresh;
 
-	/* TODO: Initialize ring structure */
+	bnxt_init_tx_ring_struct(txq);
 
 	txq->queue_id = queue_idx;
 	txq->port_id = eth_dev->data->port_id;
 
-	/* TODO: Allocate TX ring hardware descriptors */
+	/* Allocate TX ring hardware descriptors */
+	if (bnxt_alloc_rings(bp, queue_idx, txq->tx_ring, NULL, txq->cp_ring,
+			"bnxt_tx_ring")) {
+		RTE_LOG(ERR, PMD, "ring_dma_zone_reserve for tx_ring failed!");
+		bnxt_tx_queue_release_op(txq);
+		return -ENOMEM;
+	}
 
-	/* TODO: Initialize the ring */
+	if (bnxt_init_one_tx_ring(txq)) {
+		RTE_LOG(ERR, PMD, "bnxt_init_one_tx_ring failed!");
+		bnxt_tx_queue_release_op(txq);
+		return -ENOMEM;
+	}
 
 	eth_dev->data->tx_queues[queue_idx] = txq;
 	return 0;
diff --git a/drivers/net/bnxt/bnxt_txr.c b/drivers/net/bnxt/bnxt_txr.c
new file mode 100644
index 0000000..2314410
--- /dev/null
+++ b/drivers/net/bnxt/bnxt_txr.c
@@ -0,0 +1,314 @@
+/*-
+ *   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.
+ */
+
+#include <inttypes.h>
+
+#include <rte_byteorder.h>
+#include <rte_malloc.h>
+
+#include "bnxt.h"
+#include "bnxt_cpr.h"
+#include "bnxt_ring.h"
+#include "bnxt_txq.h"
+#include "bnxt_txr.h"
+#include "hsi_struct_def_dpdk.h"
+#include <stdbool.h>
+
+/*
+ * TX Ring handling
+ */
+
+void bnxt_free_tx_rings(struct bnxt *bp)
+{
+	int i;
+
+	for (i = 0; i < (int)bp->tx_nr_rings; i++) {
+		struct bnxt_tx_queue *txq = bp->tx_queues[i];
+
+		if (!txq)
+			continue;
+
+		bnxt_free_ring(txq->tx_ring->tx_ring_struct);
+		/* TODO: free() txq->tx_ring and txq->tx_ring->tx_ring_struct */
+		bnxt_free_ring(txq->cp_ring->cp_ring_struct);
+		/* TODO: free() txq->cp_ring and txq->cp_ring->cp_ring_struct */
+
+		rte_free(txq);
+		bp->tx_queues[i] = NULL;
+	}
+}
+
+int bnxt_init_one_tx_ring(struct bnxt_tx_queue *txq)
+{
+	struct bnxt_tx_ring_info *txr = txq->tx_ring;
+	struct bnxt_ring_struct *ring = txr->tx_ring_struct;
+
+	txq->tx_wake_thresh = ring->ring_size / 2;
+	ring->fw_ring_id = INVALID_HW_RING_ID;
+
+	return 0;
+}
+
+void bnxt_init_tx_ring_struct(struct bnxt_tx_queue *txq)
+{
+	struct bnxt_cp_ring_info *cpr;
+	struct bnxt_tx_ring_info *txr;
+	struct bnxt_ring_struct *ring;
+
+	/* TODO: These need to be allocated */
+	txr = txq->tx_ring;
+	ring = txr->tx_ring_struct;
+	ring->ring_size = rte_align32pow2(txq->nb_tx_desc + 1);
+	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 = (void **)&txr->tx_buf_ring;
+
+	/* TODO: These need to be allocated */
+	cpr = txq->cp_ring;
+	ring = cpr->cp_ring_struct;
+	ring->ring_size = txr->tx_ring_struct->ring_size;
+	ring->ring_mask = ring->ring_size - 1;
+	ring->bd = (void *)cpr->cp_desc_ring;
+	ring->bd_dma = cpr->cp_desc_mapping;
+	ring->vmem_size = 0;
+	ring->vmem = NULL;
+}
+
+static inline uint32_t bnxt_tx_avail(struct bnxt_tx_ring_info *txr)
+{
+	/* 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;
+}
+
+static uint16_t bnxt_start_xmit(struct rte_mbuf *tx_pkt,
+				struct bnxt_tx_queue *txq)
+{
+	struct bnxt_tx_ring_info *txr = txq->tx_ring;
+	struct tx_bd_long *txbd;
+	struct tx_bd_long_hi *txbd1;
+	uint32_t vlan_tag_flags, cfa_action;
+	bool long_bd = false;
+	uint16_t last_prod = 0;
+	struct rte_mbuf *m_seg;
+	struct bnxt_sw_tx_bd *tx_buf;
+	static const uint32_t lhint_arr[4] = {
+		TX_BD_LONG_FLAGS_LHINT_LT512,
+		TX_BD_LONG_FLAGS_LHINT_LT1K,
+		TX_BD_LONG_FLAGS_LHINT_LT2K,
+		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))
+		long_bd = true;
+
+	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;
+
+	if (unlikely(bnxt_tx_avail(txr) < tx_buf->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;
+	txbd->len = tx_pkt->data_len;
+	if (txbd->len >= 2014)
+		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));
+
+	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) {
+			/* 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;
+			/* Currently supports 8021Q, 8021AD vlan offloads
+			 * QINQ1, QINQ2, QINQ3 vlan headers are deprecated
+			 */
+			/* DPDK only supports 802.11q VLAN packets */
+			vlan_tag_flags |=
+					TX_BD_LONG_CFA_META_VLAN_TPID_TPID8100;
+		}
+
+		txr->tx_prod = RING_NEXT(txr->tx_ring_struct, txr->tx_prod);
+
+		txbd1 = (struct tx_bd_long_hi *)
+					&txr->tx_desc_ring[txr->tx_prod];
+		txbd1->lflags = 0;
+		txbd1->cfa_meta = vlan_tag_flags;
+		txbd1->cfa_action = cfa_action;
+
+		if (tx_pkt->ol_flags & PKT_TX_TCP_SEG) {
+			/* TSO */
+			txbd1->lflags = TX_BD_LONG_LFLAGS_LSO;
+			txbd1->hdr_size = tx_pkt->l2_len + tx_pkt->l3_len +
+					tx_pkt->l4_len;
+			txbd1->mss = tx_pkt->tso_segsz;
+
+		} else if (tx_pkt->ol_flags & (PKT_TX_TCP_CKSUM |
+					PKT_TX_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) {
+			/* IP CSO */
+			txbd1->lflags = TX_BD_LONG_LFLAGS_IP_CHKSUM;
+			txbd1->mss = 0;
+		}
+	} 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];
+
+		txbd = &txr->tx_desc_ring[txr->tx_prod];
+		txbd->addr = rte_cpu_to_le_32(RTE_MBUF_DATA_DMA_ADDR(m_seg));
+		txbd->flags_type = TX_BD_SHORT_TYPE_TX_BD_SHORT;
+		txbd->len = m_seg->data_len;
+
+		m_seg = m_seg->next;
+	}
+
+	txbd->flags_type |= TX_BD_LONG_FLAGS_PACKET_END;
+
+	txr->tx_prod = RING_NEXT(txr->tx_ring_struct, txr->tx_prod);
+
+	return 0;
+}
+
+static void bnxt_tx_cmp(struct bnxt_tx_queue *txq, int nr_pkts)
+{
+	struct bnxt_tx_ring_info *txr = txq->tx_ring;
+	uint16_t cons = txr->tx_cons;
+	int i, j;
+
+	for (i = 0; i < nr_pkts; i++) {
+		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 = tx_buf->mbuf;
+		tx_buf->mbuf = NULL;
+
+		/* EW - no need to unmap DMA memory? */
+
+		for (j = 1; j < tx_buf->nr_bds; j++)
+			cons = RING_NEXT(txr->tx_ring_struct, cons);
+		rte_pktmbuf_free(mbuf);
+	}
+
+	txr->tx_cons = cons;
+}
+
+static int bnxt_handle_tx_cp(struct bnxt_tx_queue *txq)
+{
+	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 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;
+
+			if (CMP_TYPE(txcmp) == TX_CMPL_TYPE_TX_L2)
+				nb_tx_pkts++;
+			else
+				RTE_LOG(DEBUG, PMD,
+						"Unhandled CMP type %02x\n",
+						CMP_TYPE(txcmp));
+			raw_cons = NEXT_RAW_CMP(raw_cons);
+		}
+		if (nb_tx_pkts)
+			bnxt_tx_cmp(txq, nb_tx_pkts);
+		cpr->cp_raw_cons = raw_cons;
+		B_CP_DIS_DB(cpr, cpr->cp_raw_cons);
+	}
+	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;
+	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;
+
+	/* Handle TX completions */
+	bnxt_handle_tx_cp(txq);
+
+	/* 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)) {
+			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);
+
+	return nb_tx_pkts;
+}
diff --git a/drivers/net/bnxt/bnxt_txr.h b/drivers/net/bnxt/bnxt_txr.h
new file mode 100644
index 0000000..1797a3d
--- /dev/null
+++ b/drivers/net/bnxt/bnxt_txr.h
@@ -0,0 +1,71 @@
+/*-
+ *   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.
+ */
+
+#ifndef _BNXT_TXR_H_
+#define _BNXT_TXR_H_
+
+#define MAX_TX_RINGS	16
+#define BNXT_TX_PUSH_THRESH 92
+
+#define B_TX_DB(db, prod)						\
+		(*(uint32_t *)db = (DB_KEY_TX | prod))
+
+struct bnxt_tx_ring_info {
+	uint16_t		tx_prod;
+	uint16_t		tx_cons;
+	void			*tx_doorbell;
+
+	struct tx_bd_long	*tx_desc_ring;
+	struct bnxt_sw_tx_bd	*tx_buf_ring;
+
+	phys_addr_t		tx_desc_mapping;
+
+#define BNXT_DEV_STATE_CLOSING	0x1
+	uint32_t		dev_state;
+
+	struct bnxt_ring_struct	*tx_ring_struct;
+};
+
+struct bnxt_sw_tx_bd {
+	struct rte_mbuf		*mbuf; /* mbuf associated with TX descriptor */
+	uint8_t			is_gso;
+	unsigned short		nr_bds;
+};
+
+void bnxt_free_tx_rings(struct bnxt *bp);
+int bnxt_init_one_tx_ring(struct bnxt_tx_queue *txq);
+void bnxt_init_tx_ring_struct(struct bnxt_tx_queue *txq);
+uint16_t bnxt_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
+			       uint16_t nb_pkts);
+
+#endif
diff --git a/drivers/net/bnxt/hsi_struct_def_dpdk.h b/drivers/net/bnxt/hsi_struct_def_dpdk.h
index 989f533..ff02a29 100644
--- a/drivers/net/bnxt/hsi_struct_def_dpdk.h
+++ b/drivers/net/bnxt/hsi_struct_def_dpdk.h
@@ -99,6 +99,518 @@ typedef struct ctx_hw_stats64 {
 #define HWRM_ERR_CODE_INVALID_PARAMS                      (UINT32_C(0x2))
 #define HWRM_ERR_CODE_RESOURCE_ACCESS_DENIED              (UINT32_C(0x3))
 
+/* Short TX BD (16 bytes) */
+struct tx_bd_short {
+	/*
+	 * All bits in this field must be valid on the first BD of a packet.
+	 * Only the packet_end bit must be valid for the remaining BDs of a
+	 * packet.
+	 */
+	/* This value identifies the type of buffer descriptor. */
+	#define TX_BD_SHORT_TYPE_MASK			UINT32_C(0x3f)
+	#define TX_BD_SHORT_TYPE_SFT			0
+		/*
+		 * Indicates that this BD is 16B long and is used for normal L2
+		 * packet transmission.
+		 */
+	#define TX_BD_SHORT_TYPE_TX_BD_SHORT		(UINT32_C(0x0) << 0)
+	/*
+	 * If set to 1, the packet ends with the data in the buffer pointed to
+	 * by this descriptor. This flag must be valid on every BD.
+	 */
+	#define TX_BD_SHORT_FLAGS_PACKET_END		UINT32_C(0x40)
+	/*
+	 * If set to 1, the device will not generate a completion for this
+	 * transmit packet unless there is an error in it's processing. If this
+	 * bit is set to 0, then the packet will be completed normally. This bit
+	 * must be valid only on the first BD of a packet.
+	 */
+	#define TX_BD_SHORT_FLAGS_NO_CMPL		UINT32_C(0x80)
+	/*
+	 * This value indicates how many 16B BD locations are consumed in the
+	 * ring by this packet. A value of 1 indicates that this BD is the only
+	 * BD (and that the it is a short BD). A value of 3 indicates either 3
+	 * short BDs or 1 long BD and one short BD in the packet. A value of 0
+	 * indicates that there are 32 BD locations in the packet (the maximum).
+	 * This field is valid only on the first BD of a packet.
+	 */
+	#define TX_BD_SHORT_FLAGS_BD_CNT_MASK		UINT32_C(0x1f00)
+	#define TX_BD_SHORT_FLAGS_BD_CNT_SFT		8
+	/*
+	 * This value is a hint for the length of the entire packet. It is used
+	 * by the chip to optimize internal processing. The packet will be
+	 * dropped if the hint is too short. This field is valid only on the
+	 * first BD of a packet.
+	 */
+	#define TX_BD_SHORT_FLAGS_LHINT_MASK		UINT32_C(0x6000)
+	#define TX_BD_SHORT_FLAGS_LHINT_SFT		13
+		/* indicates packet length < 512B */
+	#define TX_BD_SHORT_FLAGS_LHINT_LT512		(UINT32_C(0x0) << 13)
+		/* indicates 512 <= packet length < 1KB */
+	#define TX_BD_SHORT_FLAGS_LHINT_LT1K		(UINT32_C(0x1) << 13)
+		/* indicates 1KB <= packet length < 2KB */
+	#define TX_BD_SHORT_FLAGS_LHINT_LT2K		(UINT32_C(0x2) << 13)
+		/* indicates packet length >= 2KB */
+	#define TX_BD_SHORT_FLAGS_LHINT_GTE2K		(UINT32_C(0x3) << 13)
+	#define TX_BD_SHORT_FLAGS_LHINT_LAST	TX_BD_SHORT_FLAGS_LHINT_GTE2K
+	/*
+	 * If set to 1, the device immediately updates the Send Consumer Index
+	 * after the buffer associated with this descriptor has been transferred
+	 * via DMA to NIC memory from host memory. An interrupt may or may not
+	 * be generated according to the state of the interrupt avoidance
+	 * mechanisms. If this bit is set to 0, then the Consumer Index is only
+	 * updated as soon as one of the host interrupt coalescing conditions
+	 * has been met. This bit must be valid on the first BD of a packet.
+	 */
+	#define TX_BD_SHORT_FLAGS_COAL_NOW		UINT32_C(0x8000)
+	/*
+	 * All bits in this field must be valid on the first BD of a packet.
+	 * Only the packet_end bit must be valid for the remaining BDs of a
+	 * packet.
+	 */
+	#define TX_BD_SHORT_FLAGS_MASK			UINT32_C(0xffc0)
+	#define TX_BD_SHORT_FLAGS_SFT			6
+	uint16_t flags_type;
+
+	/*
+	 * This is the length of the host physical buffer this BD describes in
+	 * bytes. This field must be valid on all BDs of a packet.
+	 */
+	uint16_t len;
+	/*
+	 * The opaque data field is pass through to the completion and can be
+	 * used for any data that the driver wants to associate with the
+	 * transmit BD. This field must be valid on the first BD of a packet.
+	 */
+	uint32_t opaque;
+
+	/*
+	 * This is the host physical address for the portion of the packet
+	 * described by this TX BD. This value must be valid on all BDs of a
+	 * packet.
+	 */
+	uint64_t addr;
+} __attribute__((packed));
+
+/* Long TX BD (32 bytes split to 2 16-byte struct) */
+struct tx_bd_long {
+	/*
+	 * All bits in this field must be valid on the first BD of a packet.
+	 * Only the packet_end bit must be valid for the remaining BDs of a
+	 * packet.
+	 */
+	/* This value identifies the type of buffer descriptor. */
+	#define TX_BD_LONG_TYPE_MASK			UINT32_C(0x3f)
+	#define TX_BD_LONG_TYPE_SFT			0
+		/*
+		 * Indicates that this BD is 32B long and is used for normal L2
+		 * packet transmission.
+		 */
+	#define TX_BD_LONG_TYPE_TX_BD_LONG		(UINT32_C(0x10) << 0)
+	/*
+	 * If set to 1, the packet ends with the data in the buffer pointed to
+	 * by this descriptor. This flag must be valid on every BD.
+	 */
+	#define TX_BD_LONG_FLAGS_PACKET_END		UINT32_C(0x40)
+	/*
+	 * If set to 1, the device will not generate a completion for this
+	 * transmit packet unless there is an error in it's processing. If this
+	 * bit is set to 0, then the packet will be completed normally. This bit
+	 * must be valid only on the first BD of a packet.
+	 */
+	#define TX_BD_LONG_FLAGS_NO_CMPL		UINT32_C(0x80)
+	/*
+	 * This value indicates how many 16B BD locations are consumed in the
+	 * ring by this packet. A value of 1 indicates that this BD is the only
+	 * BD (and that the it is a short BD). A value of 3 indicates either 3
+	 * short BDs or 1 long BD and one short BD in the packet. A value of 0
+	 * indicates that there are 32 BD locations in the packet (the maximum).
+	 * This field is valid only on the first BD of a packet.
+	 */
+	#define TX_BD_LONG_FLAGS_BD_CNT_MASK		UINT32_C(0x1f00)
+	#define TX_BD_LONG_FLAGS_BD_CNT_SFT		8
+	/*
+	 * This value is a hint for the length of the entire packet. It is used
+	 * by the chip to optimize internal processing. The packet will be
+	 * dropped if the hint is too short. This field is valid only on the
+	 * first BD of a packet.
+	 */
+	#define TX_BD_LONG_FLAGS_LHINT_MASK		UINT32_C(0x6000)
+	#define TX_BD_LONG_FLAGS_LHINT_SFT		13
+		/* indicates packet length < 512B */
+	#define TX_BD_LONG_FLAGS_LHINT_LT512		(UINT32_C(0x0) << 13)
+		/* indicates 512 <= packet length < 1KB */
+	#define TX_BD_LONG_FLAGS_LHINT_LT1K		(UINT32_C(0x1) << 13)
+		/* indicates 1KB <= packet length < 2KB */
+	#define TX_BD_LONG_FLAGS_LHINT_LT2K		(UINT32_C(0x2) << 13)
+		/* indicates packet length >= 2KB */
+	#define TX_BD_LONG_FLAGS_LHINT_GTE2K		(UINT32_C(0x3) << 13)
+	#define TX_BD_LONG_FLAGS_LHINT_LAST	TX_BD_LONG_FLAGS_LHINT_GTE2K
+	/*
+	 * If set to 1, the device immediately updates the Send Consumer Index
+	 * after the buffer associated with this descriptor has been transferred
+	 * via DMA to NIC memory from host memory. An interrupt may or may not
+	 * be generated according to the state of the interrupt avoidance
+	 * mechanisms. If this bit is set to 0, then the Consumer Index is only
+	 * updated as soon as one of the host interrupt coalescing conditions
+	 * has been met. This bit must be valid on the first BD of a packet.
+	 */
+	#define TX_BD_LONG_FLAGS_COAL_NOW		UINT32_C(0x8000)
+	/*
+	 * All bits in this field must be valid on the first BD of a packet.
+	 * Only the packet_end bit must be valid for the remaining BDs of a
+	 * packet.
+	 */
+	#define TX_BD_LONG_FLAGS_MASK			UINT32_C(0xffc0)
+	#define TX_BD_LONG_FLAGS_SFT			6
+	uint16_t flags_type;
+
+	/*
+	 * This is the length of the host physical buffer this BD describes in
+	 * bytes. This field must be valid on all BDs of a packet.
+	 */
+	uint16_t len;
+
+	/*
+	 * The opaque data field is pass through to the completion and can be
+	 * used for any data that the driver wants to associate with the
+	 * transmit BD. This field must be valid on the first BD of a packet.
+	 */
+	uint32_t opaque;
+
+	/*
+	 * This is the host physical address for the portion of the packet
+	 * described by this TX BD. This value must be valid on all BDs of a
+	 * packet.
+	 */
+	uint64_t addr;
+} __attribute__((packed));
+
+/* last 16 bytes of Long TX BD */
+
+struct tx_bd_long_hi {
+	/*
+	 * All bits in this field must be valid on the first BD of a packet.
+	 * Their value on other BDs of the packet will be ignored.
+	 */
+	/*
+	 * If set to 1, the controller replaces the TCP/UPD checksum fields of
+	 * normal TCP/UPD checksum, or the inner TCP/UDP checksum field of the
+	 * encapsulated TCP/UDP packets with the hardware calculated TCP/UDP
+	 * checksum for the packet associated with this descriptor. This bit
+	 * must be valid on the first BD of a packet.
+	 */
+	#define TX_BD_LONG_LFLAGS_TCP_UDP_CHKSUM	UINT32_C(0x1)
+	/*
+	 * If set to 1, the controller replaces the IP checksum of the normal
+	 * packets, or the inner IP checksum of the encapsulated packets with
+	 * the hardware calculated IP checksum for the packet associated with
+	 * this descriptor. This bit must be valid on the first BD of a packet.
+	 */
+	#define TX_BD_LONG_LFLAGS_IP_CHKSUM		UINT32_C(0x2)
+	/*
+	 * If set to 1, the controller will not append an Ethernet CRC to the
+	 * end of the frame. This bit must be valid on the first BD of a packet.
+	 * Packet must be 64B or longer when this flag is set. It is not useful
+	 * to use this bit with any form of TX offload such as CSO or LSO. The
+	 * intent is that the packet from the host already has a valid Ethernet
+	 * CRC on the packet.
+	 */
+	#define TX_BD_LONG_LFLAGS_NOCRC			UINT32_C(0x4)
+	/*
+	 * If set to 1, the device will record the time at which the packet was
+	 * actually transmitted at the TX MAC. This bit must be valid on the
+	 * first BD of a packet.
+	 */
+	#define TX_BD_LONG_LFLAGS_STAMP			UINT32_C(0x8)
+	/*
+	 * If set to 1, The controller replaces the tunnel IP checksum field
+	 * with hardware calculated IP checksum for the IP header of the packet
+	 * associated with this descriptor. In case of VXLAN, the controller
+	 * also replaces the outer header UDP checksum with hardware calculated
+	 * UDP checksum for the packet associated with this descriptor.
+	 */
+	#define TX_BD_LONG_LFLAGS_T_IP_CHKSUM		UINT32_C(0x10)
+	/*
+	 * If set to 1, the device will treat this packet with LSO(Large Send
+	 * Offload) processing for both normal or encapsulated packets, which is
+	 * a form of TCP segmentation. When this bit is 1, the hdr_size and mss
+	 * fields must be valid. The driver doesn't need to set t_ip_chksum,
+	 * ip_chksum, and tcp_udp_chksum flags since the controller will replace
+	 * the appropriate checksum fields for segmented packets. When this bit
+	 * is 1, the hdr_size and mss fields must be valid.
+	 */
+	#define TX_BD_LONG_LFLAGS_LSO			UINT32_C(0x20)
+	/*
+	 * If set to zero when LSO is '1', then the IPID will be treated as a
+	 * 16b number and will be wrapped if it exceeds a value of 0xffff. If
+	 * set to one when LSO is '1', then the IPID will be treated as a 15b
+	 * number and will be wrapped if it exceeds a value 0f 0x7fff.
+	 */
+	#define TX_BD_LONG_LFLAGS_IPID_FMT		UINT32_C(0x40)
+	/*
+	 * If set to zero when LSO is '1', then the IPID of the tunnel IP header
+	 * will not be modified during LSO operations. If set to one when LSO is
+	 * '1', then the IPID of the tunnel IP header will be incremented for
+	 * each subsequent segment of an LSO operation.
+	 */
+	#define TX_BD_LONG_LFLAGS_T_IPID		UINT32_C(0x80)
+	/*
+	 * If set to '1', then the RoCE ICRC will be appended to the packet.
+	 * Packet must be a valid RoCE format packet.
+	 */
+	#define TX_BD_LONG_LFLAGS_ROCE_CRC		UINT32_C(0x100)
+	/*
+	 * If set to '1', then the FCoE CRC will be appended to the packet.
+	 * Packet must be a valid FCoE format packet.
+	 */
+	#define TX_BD_LONG_LFLAGS_FCOE_CRC		UINT32_C(0x200)
+	uint16_t lflags;
+
+	/*
+	 * When LSO is '1', this field must contain the offset of the TCP
+	 * payload from the beginning of the packet in as 16b words. In case of
+	 * encapsulated/tunneling packet, this field contains the offset of the
+	 * inner TCP payload from beginning of the packet as 16-bit words. This
+	 * value must be valid on the first BD of a packet.
+	 */
+	#define TX_BD_LONG_HDR_SIZE_MASK		UINT32_C(0x1ff)
+	#define TX_BD_LONG_HDR_SIZE_SFT			0
+	uint16_t hdr_size;
+
+	/*
+	 * This is the MSS value that will be used to do the LSO processing. The
+	 * value is the length in bytes of the TCP payload for each segment
+	 * generated by the LSO operation. This value must be valid on the first
+	 * BD of a packet.
+	 */
+	#define TX_BD_LONG_MSS_MASK			UINT32_C(0x7fff)
+	#define TX_BD_LONG_MSS_SFT			0
+	uint32_t mss;
+
+	uint16_t unused_2;
+
+	/*
+	 * This value selects a CFA action to perform on the packet. Set this
+	 * value to zero if no CFA action is desired. This value must be valid
+	 * on the first BD of a packet.
+	 */
+	uint16_t cfa_action;
+
+	/*
+	 * This value is action meta-data that defines CFA edit operations that
+	 * are done in addition to any action editing.
+	 */
+	/* When key=1, This is the VLAN tag VID value. */
+	#define TX_BD_LONG_CFA_META_VLAN_VID_MASK	UINT32_C(0xfff)
+	#define TX_BD_LONG_CFA_META_VLAN_VID_SFT	0
+	/* When key=1, This is the VLAN tag DE value. */
+	#define TX_BD_LONG_CFA_META_VLAN_DE		UINT32_C(0x1000)
+	/* When key=1, This is the VLAN tag PRI value. */
+	#define TX_BD_LONG_CFA_META_VLAN_PRI_MASK	UINT32_C(0xe000)
+	#define TX_BD_LONG_CFA_META_VLAN_PRI_SFT	13
+	/* When key=1, This is the VLAN tag TPID select value. */
+	#define TX_BD_LONG_CFA_META_VLAN_TPID_MASK	UINT32_C(0x70000)
+	#define TX_BD_LONG_CFA_META_VLAN_TPID_SFT	16
+		/* 0x88a8 */
+	#define TX_BD_LONG_CFA_META_VLAN_TPID_TPID88A8	(UINT32_C(0x0) << 16)
+		/* 0x8100 */
+	#define TX_BD_LONG_CFA_META_VLAN_TPID_TPID8100	(UINT32_C(0x1) << 16)
+		/* 0x9100 */
+	#define TX_BD_LONG_CFA_META_VLAN_TPID_TPID9100	(UINT32_C(0x2) << 16)
+		/* 0x9200 */
+	#define TX_BD_LONG_CFA_META_VLAN_TPID_TPID9200	(UINT32_C(0x3) << 16)
+		/* 0x9300 */
+	#define TX_BD_LONG_CFA_META_VLAN_TPID_TPID9300	(UINT32_C(0x4) << 16)
+		/* Value programmed in CFA VLANTPID register. */
+	#define TX_BD_LONG_CFA_META_VLAN_TPID_TPIDCFG	(UINT32_C(0x5) << 16)
+	#define TX_BD_LONG_CFA_META_VLAN_TPID_LAST \
+					TX_BD_LONG_CFA_META_VLAN_TPID_TPIDCFG
+	/* When key=1, This is the VLAN tag TPID select value. */
+	#define TX_BD_LONG_CFA_META_VLAN_RESERVED_MASK	UINT32_C(0xff80000)
+	#define TX_BD_LONG_CFA_META_VLAN_RESERVED_SFT	19
+	/*
+	 * This field identifies the type of edit to be performed on the packet.
+	 * This value must be valid on the first BD of a packet.
+	 */
+	#define TX_BD_LONG_CFA_META_KEY_MASK		UINT32_C(0xf0000000)
+	#define TX_BD_LONG_CFA_META_KEY_SFT		28
+		/* No editing */
+	#define TX_BD_LONG_CFA_META_KEY_NONE		(UINT32_C(0x0) << 28)
+		/*
+		 * - meta[17:16] - TPID select value (0 = 0x8100). - meta[15:12]
+		 * - PRI/DE value. - meta[11:0] - VID value.
+		 */
+	#define TX_BD_LONG_CFA_META_KEY_VLAN_TAG	(UINT32_C(0x1) << 28)
+	#define TX_BD_LONG_CFA_META_KEY_LAST	TX_BD_LONG_CFA_META_KEY_VLAN_TAG
+	uint32_t cfa_meta;
+} __attribute__((packed));
+
+/* Completion Ring Structures */
+/* Note: This structure is used by the HWRM to communicate HWRM Error. */
+/* Base Completion Record (16 bytes) */
+struct cmpl_base {
+	/* unused is 10 b */
+	/*
+	 * This field indicates the exact type of the completion. By convention,
+	 * the LSB identifies the length of the record in 16B units. Even values
+	 * indicate 16B records. Odd values indicate 32B records.
+	 */
+	#define CMPL_BASE_TYPE_MASK			UINT32_C(0x3f)
+	#define CMPL_BASE_TYPE_SFT			0
+		/* TX L2 completion: Completion of TX packet. Length = 16B */
+	#define CMPL_BASE_TYPE_TX_L2			(UINT32_C(0x0) << 0)
+		/*
+		 * RX L2 completion: Completion of and L2 RX packet.
+		 * Length = 32B
+		*/
+	#define CMPL_BASE_TYPE_RX_L2			(UINT32_C(0x11) << 0)
+		/*
+		 * RX Aggregation Buffer completion : Completion of an L2
+		 * aggregation buffer in support of TPA, HDS, or Jumbo packet
+		 * completion. Length = 16B
+		 */
+	#define CMPL_BASE_TYPE_RX_AGG			(UINT32_C(0x12) << 0)
+		/*
+		 * RX L2 TPA Start Completion: Completion at the beginning of a
+		 * TPA operation. Length = 32B
+		 */
+	#define CMPL_BASE_TYPE_RX_TPA_START		(UINT32_C(0x13) << 0)
+		/*
+		 * RX L2 TPA End Completion: Completion at the end of a TPA
+		 * operation. Length = 32B
+		 */
+	#define CMPL_BASE_TYPE_RX_TPA_END		(UINT32_C(0x15) << 0)
+		/*
+		 * Statistics Ejection Completion: Completion of statistics data
+		 * ejection buffer. Length = 16B
+		 */
+	#define CMPL_BASE_TYPE_STAT_EJECT		(UINT32_C(0x1a) << 0)
+		/* HWRM Command Completion: Completion of an HWRM command. */
+	#define CMPL_BASE_TYPE_HWRM_DONE		(UINT32_C(0x20) << 0)
+		/* Forwarded HWRM Request */
+	#define CMPL_BASE_TYPE_HWRM_FWD_REQ		(UINT32_C(0x22) << 0)
+		/* Forwarded HWRM Response */
+	#define CMPL_BASE_TYPE_HWRM_FWD_RESP		(UINT32_C(0x24) << 0)
+		/* HWRM Asynchronous Event Information */
+	#define CMPL_BASE_TYPE_HWRM_ASYNC_EVENT		(UINT32_C(0x2e) << 0)
+		/* CQ Notification */
+	#define CMPL_BASE_TYPE_CQ_NOTIFICATION		(UINT32_C(0x30) << 0)
+		/* SRQ Threshold Event */
+	#define CMPL_BASE_TYPE_SRQ_EVENT		(UINT32_C(0x32) << 0)
+		/* DBQ Threshold Event */
+	#define CMPL_BASE_TYPE_DBQ_EVENT		(UINT32_C(0x34) << 0)
+		/* QP Async Notification */
+	#define CMPL_BASE_TYPE_QP_EVENT			(UINT32_C(0x38) << 0)
+		/* Function Async Notification */
+	#define CMPL_BASE_TYPE_FUNC_EVENT		(UINT32_C(0x3a) << 0)
+	uint16_t type;
+
+	uint16_t info1;
+	uint32_t info2;
+
+	/*
+	 * This value is written by the NIC such that it will be different for
+	 * each pass through the completion queue. The even passes will write 1.
+	 * The odd passes will write 0.
+	 */
+	#define CMPL_BASE_V				UINT32_C(0x1)
+	/* info3 is 31 b */
+	#define CMPL_BASE_INFO3_MASK			UINT32_C(0xfffffffe)
+	#define CMPL_BASE_INFO3_SFT			1
+	uint32_t info3_v;
+
+	uint32_t info4;
+} __attribute__((packed));
+
+/* TX Completion Record (16 bytes) */
+struct tx_cmpl {
+	/*
+	 * This field indicates the exact type of the completion. By convention,
+	 * the LSB identifies the length of the record in 16B units. Even values
+	 * indicate 16B records. Odd values indicate 32B records.
+	 */
+	#define TX_CMPL_TYPE_MASK			UINT32_C(0x3f)
+	#define TX_CMPL_TYPE_SFT			0
+		/* TX L2 completion: Completion of TX packet. Length = 16B */
+	#define TX_CMPL_TYPE_TX_L2			(UINT32_C(0x0) << 0)
+	/*
+	 * When this bit is '1', it indicates a packet that has an error of some
+	 * type. Type of error is indicated in error_flags.
+	 */
+	#define TX_CMPL_FLAGS_ERROR			UINT32_C(0x40)
+	/*
+	 * When this bit is '1', it indicates that the packet completed was
+	 * transmitted using the push acceleration data provided by the driver.
+	 * When this bit is '0', it indicates that the packet had not push
+	 * acceleration data written or was executed as a normal packet even
+	 * though push data was provided.
+	 */
+	#define TX_CMPL_FLAGS_PUSH			UINT32_C(0x80)
+	#define TX_CMPL_FLAGS_MASK			UINT32_C(0xffc0)
+	#define TX_CMPL_FLAGS_SFT			6
+	uint16_t flags_type;
+
+	uint16_t unused_0;
+
+	/*
+	 * This is a copy of the opaque field from the first TX BD of this
+	 * transmitted packet.
+	 */
+	uint32_t opaque;
+
+	/*
+	 * This value is written by the NIC such that it will be different for
+	 * each pass through the completion queue. The even passes will write 1.
+	 * The odd passes will write 0.
+	 */
+	#define TX_CMPL_V				UINT32_C(0x1)
+	/*
+	 * This error indicates that there was some sort of problem with the BDs
+	 * for the packet.
+	 */
+	#define TX_CMPL_ERRORS_BUFFER_ERROR_MASK	UINT32_C(0xe)
+	#define TX_CMPL_ERRORS_BUFFER_ERROR_SFT		1
+		/* No error */
+	#define TX_CMPL_ERRORS_BUFFER_ERROR_NO_ERROR	(UINT32_C(0x0) << 1)
+		/* Bad Format: BDs were not formatted correctly. */
+	#define TX_CMPL_ERRORS_BUFFER_ERROR_BAD_FMT	(UINT32_C(0x2) << 1)
+	#define TX_CMPL_ERRORS_BUFFER_ERROR_LAST \
+					TX_CMPL_ERRORS_BUFFER_ERROR_BAD_FMT
+	/*
+	 * When this bit is '1', it indicates that the length of the packet was
+	 * zero. No packet was transmitted.
+	 */
+	#define TX_CMPL_ERRORS_ZERO_LENGTH_PKT		UINT32_C(0x10)
+	/*
+	 * When this bit is '1', it indicates that the packet was longer than
+	 * the programmed limit in TDI. No packet was transmitted.
+	 */
+	#define TX_CMPL_ERRORS_EXCESSIVE_BD_LENGTH	UINT32_C(0x20)
+	/*
+	 * When this bit is '1', it indicates that one or more of the BDs
+	 * associated with this packet generated a PCI error. This probably
+	 * means the address was not valid.
+	 */
+	#define TX_CMPL_ERRORS_DMA_ERROR		UINT32_C(0x40)
+	/*
+	 * When this bit is '1', it indicates that the packet was longer than
+	 * indicated by the hint. No packet was transmitted.
+	 */
+	#define TX_CMPL_ERRORS_HINT_TOO_SHORT		UINT32_C(0x80)
+	/*
+	 * When this bit is '1', it indicates that the packet was dropped due to
+	 * Poison TLP error on one or more of the TLPs in the PXP completion.
+	 */
+	#define TX_CMPL_ERRORS_POISON_TLP_ERROR		UINT32_C(0x100)
+	#define TX_CMPL_ERRORS_MASK			UINT32_C(0xfffe)
+	#define TX_CMPL_ERRORS_SFT			1
+	uint16_t errors_v;
+
+	uint16_t unused_1;
+	uint32_t unused_2;
+} __attribute__((packed)) tx_cmpl_t, *ptx_cmpl_t;
+
 /* HWRM Forwarded Request (16 bytes) */
 struct hwrm_fwd_req_cmpl {
 	/* Length of forwarded request in bytes. */
-- 
1.9.1