net/ice: complete device configure in DCF
[dpdk.git] / drivers / net / atlantic / atl_rxtx.c
index 0fbd930..449ffd4 100644 (file)
  * Copyright(c) 2018 Aquantia Corporation
  */
 
+#include <rte_malloc.h>
+#include <rte_ethdev_driver.h>
+#include <rte_net.h>
+
 #include "atl_ethdev.h"
+#include "atl_hw_regs.h"
+
+#include "atl_logs.h"
+#include "hw_atl/hw_atl_llh.h"
+#include "hw_atl/hw_atl_b0.h"
+#include "hw_atl/hw_atl_b0_internal.h"
+
+#define ATL_TX_CKSUM_OFFLOAD_MASK (                     \
+       PKT_TX_IP_CKSUM |                                \
+       PKT_TX_L4_MASK |                                 \
+       PKT_TX_TCP_SEG)
+
+#define ATL_TX_OFFLOAD_MASK (                           \
+       PKT_TX_VLAN |                                    \
+       PKT_TX_IPV6 |                                    \
+       PKT_TX_IPV4 |                                    \
+       PKT_TX_IP_CKSUM |                                \
+       PKT_TX_L4_MASK |                                 \
+       PKT_TX_TCP_SEG)
+
+#define ATL_TX_OFFLOAD_NOTSUP_MASK \
+       (PKT_TX_OFFLOAD_MASK ^ ATL_TX_OFFLOAD_MASK)
+
+/**
+ * Structure associated with each descriptor of the RX ring of a RX queue.
+ */
+struct atl_rx_entry {
+       struct rte_mbuf *mbuf;
+};
+
+/**
+ * Structure associated with each descriptor of the TX ring of a TX queue.
+ */
+struct atl_tx_entry {
+       struct rte_mbuf *mbuf;
+       uint16_t next_id;
+       uint16_t last_id;
+};
+
+/**
+ * Structure associated with each RX queue.
+ */
+struct atl_rx_queue {
+       struct rte_mempool      *mb_pool;
+       struct hw_atl_rxd_s     *hw_ring;
+       uint64_t                hw_ring_phys_addr;
+       struct atl_rx_entry     *sw_ring;
+       uint16_t                nb_rx_desc;
+       uint16_t                rx_tail;
+       uint16_t                nb_rx_hold;
+       uint16_t                rx_free_thresh;
+       uint16_t                queue_id;
+       uint16_t                port_id;
+       uint16_t                buff_size;
+       bool                    l3_csum_enabled;
+       bool                    l4_csum_enabled;
+};
+
+/**
+ * Structure associated with each TX queue.
+ */
+struct atl_tx_queue {
+       struct hw_atl_txd_s     *hw_ring;
+       uint64_t                hw_ring_phys_addr;
+       struct atl_tx_entry     *sw_ring;
+       uint16_t                nb_tx_desc;
+       uint16_t                tx_tail;
+       uint16_t                tx_head;
+       uint16_t                queue_id;
+       uint16_t                port_id;
+       uint16_t                tx_free_thresh;
+       uint16_t                tx_free;
+};
+
+static inline void
+atl_reset_rx_queue(struct atl_rx_queue *rxq)
+{
+       struct hw_atl_rxd_s *rxd = NULL;
+       int i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < rxq->nb_rx_desc; i++) {
+               rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[i];
+               rxd->buf_addr = 0;
+               rxd->hdr_addr = 0;
+       }
+
+       rxq->rx_tail = 0;
+}
 
 int
-atl_tx_init(struct rte_eth_dev *eth_dev __rte_unused)
+atl_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
+                  uint16_t nb_rx_desc, unsigned int socket_id,
+                  const struct rte_eth_rxconf *rx_conf,
+                  struct rte_mempool *mb_pool)
 {
+       struct atl_rx_queue *rxq;
+       const struct rte_memzone *mz;
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* make sure a valid number of descriptors have been requested */
+       if (nb_rx_desc < AQ_HW_MIN_RX_RING_SIZE ||
+                       nb_rx_desc > AQ_HW_MAX_RX_RING_SIZE) {
+               PMD_INIT_LOG(ERR, "Number of Rx descriptors must be "
+               "less than or equal to %d, "
+               "greater than or equal to %d", AQ_HW_MAX_RX_RING_SIZE,
+               AQ_HW_MIN_RX_RING_SIZE);
+               return -EINVAL;
+       }
+
+       /*
+        * if this queue existed already, free the associated memory. The
+        * queue cannot be reused in case we need to allocate memory on
+        * different socket than was previously used.
+        */
+       if (dev->data->rx_queues[rx_queue_id] != NULL) {
+               atl_rx_queue_release(dev->data->rx_queues[rx_queue_id]);
+               dev->data->rx_queues[rx_queue_id] = NULL;
+       }
+
+       /* allocate memory for the queue structure */
+       rxq = rte_zmalloc_socket("atlantic Rx queue", sizeof(*rxq),
+                                RTE_CACHE_LINE_SIZE, socket_id);
+       if (rxq == NULL) {
+               PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
+               return -ENOMEM;
+       }
+
+       /* setup queue */
+       rxq->mb_pool = mb_pool;
+       rxq->nb_rx_desc = nb_rx_desc;
+       rxq->port_id = dev->data->port_id;
+       rxq->queue_id = rx_queue_id;
+       rxq->rx_free_thresh = rx_conf->rx_free_thresh;
+
+       rxq->l3_csum_enabled = dev->data->dev_conf.rxmode.offloads &
+               DEV_RX_OFFLOAD_IPV4_CKSUM;
+       rxq->l4_csum_enabled = dev->data->dev_conf.rxmode.offloads &
+               (DEV_RX_OFFLOAD_UDP_CKSUM | DEV_RX_OFFLOAD_TCP_CKSUM);
+       if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC)
+               PMD_DRV_LOG(ERR, "PMD does not support KEEP_CRC offload");
+
+       /* allocate memory for the software ring */
+       rxq->sw_ring = rte_zmalloc_socket("atlantic sw rx ring",
+                               nb_rx_desc * sizeof(struct atl_rx_entry),
+                               RTE_CACHE_LINE_SIZE, socket_id);
+       if (rxq->sw_ring == NULL) {
+               PMD_INIT_LOG(ERR,
+                       "Port %d: Cannot allocate software ring for queue %d",
+                       rxq->port_id, rxq->queue_id);
+               rte_free(rxq);
+               return -ENOMEM;
+       }
+
+       /*
+        * allocate memory for the hardware descriptor ring. A memzone large
+        * enough to hold the maximum ring size is requested to allow for
+        * resizing in later calls to the queue setup function.
+        */
+       mz = rte_eth_dma_zone_reserve(dev, "rx hw_ring", rx_queue_id,
+                                     HW_ATL_B0_MAX_RXD *
+                                       sizeof(struct hw_atl_rxd_s),
+                                     128, socket_id);
+       if (mz == NULL) {
+               PMD_INIT_LOG(ERR,
+                       "Port %d: Cannot allocate hardware ring for queue %d",
+                       rxq->port_id, rxq->queue_id);
+               rte_free(rxq->sw_ring);
+               rte_free(rxq);
+               return -ENOMEM;
+       }
+       rxq->hw_ring = mz->addr;
+       rxq->hw_ring_phys_addr = mz->iova;
+
+       atl_reset_rx_queue(rxq);
+
+       dev->data->rx_queues[rx_queue_id] = rxq;
        return 0;
 }
 
+static inline void
+atl_reset_tx_queue(struct atl_tx_queue *txq)
+{
+       struct atl_tx_entry *tx_entry;
+       union hw_atl_txc_s *txc;
+       uint16_t i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (!txq) {
+               PMD_DRV_LOG(ERR, "Pointer to txq is NULL");
+               return;
+       }
+
+       tx_entry = txq->sw_ring;
+
+       for (i = 0; i < txq->nb_tx_desc; i++) {
+               txc = (union hw_atl_txc_s *)&txq->hw_ring[i];
+               txc->flags1 = 0;
+               txc->flags2 = 2;
+       }
+
+       for (i = 0; i < txq->nb_tx_desc; i++) {
+               txq->hw_ring[i].dd = 1;
+               tx_entry[i].mbuf = NULL;
+       }
+
+       txq->tx_tail = 0;
+       txq->tx_head = 0;
+       txq->tx_free = txq->nb_tx_desc - 1;
+}
+
 int
-atl_rx_init(struct rte_eth_dev *eth_dev __rte_unused)
+atl_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
+                  uint16_t nb_tx_desc, unsigned int socket_id,
+                  const struct rte_eth_txconf *tx_conf)
 {
+       struct atl_tx_queue *txq;
+       const struct rte_memzone *mz;
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* make sure a valid number of descriptors have been requested */
+       if (nb_tx_desc < AQ_HW_MIN_TX_RING_SIZE ||
+               nb_tx_desc > AQ_HW_MAX_TX_RING_SIZE) {
+               PMD_INIT_LOG(ERR, "Number of Tx descriptors must be "
+                       "less than or equal to %d, "
+                       "greater than or equal to %d", AQ_HW_MAX_TX_RING_SIZE,
+                       AQ_HW_MIN_TX_RING_SIZE);
+               return -EINVAL;
+       }
+
+       /*
+        * if this queue existed already, free the associated memory. The
+        * queue cannot be reused in case we need to allocate memory on
+        * different socket than was previously used.
+        */
+       if (dev->data->tx_queues[tx_queue_id] != NULL) {
+               atl_tx_queue_release(dev->data->tx_queues[tx_queue_id]);
+               dev->data->tx_queues[tx_queue_id] = NULL;
+       }
+
+       /* allocate memory for the queue structure */
+       txq = rte_zmalloc_socket("atlantic Tx queue", sizeof(*txq),
+                                RTE_CACHE_LINE_SIZE, socket_id);
+       if (txq == NULL) {
+               PMD_INIT_LOG(ERR, "Cannot allocate queue structure");
+               return -ENOMEM;
+       }
+
+       /* setup queue */
+       txq->nb_tx_desc = nb_tx_desc;
+       txq->port_id = dev->data->port_id;
+       txq->queue_id = tx_queue_id;
+       txq->tx_free_thresh = tx_conf->tx_free_thresh;
+
+
+       /* allocate memory for the software ring */
+       txq->sw_ring = rte_zmalloc_socket("atlantic sw tx ring",
+                               nb_tx_desc * sizeof(struct atl_tx_entry),
+                               RTE_CACHE_LINE_SIZE, socket_id);
+       if (txq->sw_ring == NULL) {
+               PMD_INIT_LOG(ERR,
+                       "Port %d: Cannot allocate software ring for queue %d",
+                       txq->port_id, txq->queue_id);
+               rte_free(txq);
+               return -ENOMEM;
+       }
+
+       /*
+        * allocate memory for the hardware descriptor ring. A memzone large
+        * enough to hold the maximum ring size is requested to allow for
+        * resizing in later calls to the queue setup function.
+        */
+       mz = rte_eth_dma_zone_reserve(dev, "tx hw_ring", tx_queue_id,
+                               HW_ATL_B0_MAX_TXD * sizeof(struct hw_atl_txd_s),
+                               128, socket_id);
+       if (mz == NULL) {
+               PMD_INIT_LOG(ERR,
+                       "Port %d: Cannot allocate hardware ring for queue %d",
+                       txq->port_id, txq->queue_id);
+               rte_free(txq->sw_ring);
+               rte_free(txq);
+               return -ENOMEM;
+       }
+       txq->hw_ring = mz->addr;
+       txq->hw_ring_phys_addr = mz->iova;
+
+       atl_reset_tx_queue(txq);
+
+       dev->data->tx_queues[tx_queue_id] = txq;
        return 0;
 }
 
-uint16_t
-atl_prep_pkts(void *tx_queue __rte_unused,
-             struct rte_mbuf **tx_pkts __rte_unused,
-             uint16_t nb_pkts __rte_unused)
+int
+atl_tx_init(struct rte_eth_dev *eth_dev)
+{
+       struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+       struct atl_tx_queue *txq;
+       uint64_t base_addr = 0;
+       int i = 0;
+       int err = 0;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < eth_dev->data->nb_tx_queues; i++) {
+               txq = eth_dev->data->tx_queues[i];
+               base_addr = txq->hw_ring_phys_addr;
+
+               err = hw_atl_b0_hw_ring_tx_init(hw, base_addr,
+                                               txq->queue_id,
+                                               txq->nb_tx_desc, 0,
+                                               txq->port_id);
+
+               if (err) {
+                       PMD_INIT_LOG(ERR,
+                               "Port %d: Cannot init TX queue %d",
+                               txq->port_id, txq->queue_id);
+                       break;
+               }
+       }
+
+       return err;
+}
+
+int
+atl_rx_init(struct rte_eth_dev *eth_dev)
 {
+       struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+       struct aq_rss_parameters *rss_params = &hw->aq_nic_cfg->aq_rss;
+       struct atl_rx_queue *rxq;
+       uint64_t base_addr = 0;
+       int i = 0;
+       int err = 0;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < eth_dev->data->nb_rx_queues; i++) {
+               rxq = eth_dev->data->rx_queues[i];
+               base_addr = rxq->hw_ring_phys_addr;
+
+               /* Take requested pool mbuf size and adapt
+                * descriptor buffer to best fit
+                */
+               int buff_size = rte_pktmbuf_data_room_size(rxq->mb_pool) -
+                               RTE_PKTMBUF_HEADROOM;
+
+               buff_size = RTE_ALIGN_FLOOR(buff_size, 1024);
+               if (buff_size > HW_ATL_B0_RXD_BUF_SIZE_MAX) {
+                       PMD_INIT_LOG(WARNING,
+                               "Port %d queue %d: mem pool buff size is too big\n",
+                               rxq->port_id, rxq->queue_id);
+                       buff_size = HW_ATL_B0_RXD_BUF_SIZE_MAX;
+               }
+               if (buff_size < 1024) {
+                       PMD_INIT_LOG(ERR,
+                               "Port %d queue %d: mem pool buff size is too small\n",
+                               rxq->port_id, rxq->queue_id);
+                       return -EINVAL;
+               }
+               rxq->buff_size = buff_size;
+
+               err = hw_atl_b0_hw_ring_rx_init(hw, base_addr, rxq->queue_id,
+                                               rxq->nb_rx_desc, buff_size, 0,
+                                               rxq->port_id);
+
+               if (err) {
+                       PMD_INIT_LOG(ERR, "Port %d: Cannot init RX queue %d",
+                                    rxq->port_id, rxq->queue_id);
+                       break;
+               }
+       }
+
+       for (i = rss_params->indirection_table_size; i--;)
+               rss_params->indirection_table[i] = i &
+                       (eth_dev->data->nb_rx_queues - 1);
+       hw_atl_b0_hw_rss_set(hw, rss_params);
+       return err;
+}
+
+static int
+atl_alloc_rx_queue_mbufs(struct atl_rx_queue *rxq)
+{
+       struct atl_rx_entry *rx_entry = rxq->sw_ring;
+       struct hw_atl_rxd_s *rxd;
+       uint64_t dma_addr = 0;
+       uint32_t i = 0;
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* fill Rx ring */
+       for (i = 0; i < rxq->nb_rx_desc; i++) {
+               struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
+
+               if (mbuf == NULL) {
+                       PMD_INIT_LOG(ERR,
+                               "Port %d: mbuf alloc failed for rx queue %d",
+                               rxq->port_id, rxq->queue_id);
+                       return -ENOMEM;
+               }
+
+               mbuf->data_off = RTE_PKTMBUF_HEADROOM;
+               mbuf->port = rxq->port_id;
+
+               dma_addr = rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
+               rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[i];
+               rxd->buf_addr = dma_addr;
+               rxd->hdr_addr = 0;
+               rx_entry[i].mbuf = mbuf;
+       }
+
        return 0;
 }
 
-uint16_t
-atl_recv_pkts(void *rx_queue __rte_unused,
-             struct rte_mbuf **rx_pkts __rte_unused,
-             uint16_t nb_pkts __rte_unused)
+static void
+atl_rx_queue_release_mbufs(struct atl_rx_queue *rxq)
+{
+       int i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (rxq->sw_ring != NULL) {
+               for (i = 0; i < rxq->nb_rx_desc; i++) {
+                       if (rxq->sw_ring[i].mbuf != NULL) {
+                               rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
+                               rxq->sw_ring[i].mbuf = NULL;
+                       }
+               }
+       }
+}
+
+int
+atl_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+       struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct atl_rx_queue *rxq = NULL;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (rx_queue_id < dev->data->nb_rx_queues) {
+               rxq = dev->data->rx_queues[rx_queue_id];
+
+               if (atl_alloc_rx_queue_mbufs(rxq) != 0) {
+                       PMD_INIT_LOG(ERR,
+                               "Port %d: Allocate mbufs for queue %d failed",
+                               rxq->port_id, rxq->queue_id);
+                       return -1;
+               }
+
+               hw_atl_b0_hw_ring_rx_start(hw, rx_queue_id);
+
+               rte_wmb();
+               hw_atl_reg_rx_dma_desc_tail_ptr_set(hw, rxq->nb_rx_desc - 1,
+                                                   rx_queue_id);
+               dev->data->rx_queue_state[rx_queue_id] =
+                       RTE_ETH_QUEUE_STATE_STARTED;
+       } else {
+               return -1;
+       }
+
+       return 0;
+}
+
+int
+atl_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 {
+       struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct atl_rx_queue *rxq = NULL;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (rx_queue_id < dev->data->nb_rx_queues) {
+               rxq = dev->data->rx_queues[rx_queue_id];
+
+               hw_atl_b0_hw_ring_rx_stop(hw, rx_queue_id);
+
+               atl_rx_queue_release_mbufs(rxq);
+               atl_reset_rx_queue(rxq);
+
+               dev->data->rx_queue_state[rx_queue_id] =
+                       RTE_ETH_QUEUE_STATE_STOPPED;
+       } else {
+               return -1;
+       }
+
        return 0;
 }
 
-uint16_t
-atl_xmit_pkts(void *tx_queue __rte_unused,
-             struct rte_mbuf **tx_pkts __rte_unused,
-             uint16_t nb_pkts __rte_unused)
+void
+atl_rx_queue_release(void *rx_queue)
+{
+       PMD_INIT_FUNC_TRACE();
+
+       if (rx_queue != NULL) {
+               struct atl_rx_queue *rxq = (struct atl_rx_queue *)rx_queue;
+
+               atl_rx_queue_release_mbufs(rxq);
+               rte_free(rxq->sw_ring);
+               rte_free(rxq);
+       }
+}
+
+static void
+atl_tx_queue_release_mbufs(struct atl_tx_queue *txq)
+{
+       int i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (txq->sw_ring != NULL) {
+               for (i = 0; i < txq->nb_tx_desc; i++) {
+                       if (txq->sw_ring[i].mbuf != NULL) {
+                               rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
+                               txq->sw_ring[i].mbuf = NULL;
+                       }
+               }
+       }
+}
+
+int
+atl_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+       struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (tx_queue_id < dev->data->nb_tx_queues) {
+               hw_atl_b0_hw_ring_tx_start(hw, tx_queue_id);
+
+               rte_wmb();
+               hw_atl_b0_hw_tx_ring_tail_update(hw, 0, tx_queue_id);
+               dev->data->tx_queue_state[tx_queue_id] =
+                       RTE_ETH_QUEUE_STATE_STARTED;
+       } else {
+               return -1;
+       }
+
+       return 0;
+}
+
+int
+atl_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+       struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct atl_tx_queue *txq;
+
+       PMD_INIT_FUNC_TRACE();
+
+       txq = dev->data->tx_queues[tx_queue_id];
+
+       hw_atl_b0_hw_ring_tx_stop(hw, tx_queue_id);
+
+       atl_tx_queue_release_mbufs(txq);
+       atl_reset_tx_queue(txq);
+       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+
+       return 0;
+}
+
+void
+atl_tx_queue_release(void *tx_queue)
 {
+       PMD_INIT_FUNC_TRACE();
+
+       if (tx_queue != NULL) {
+               struct atl_tx_queue *txq = (struct atl_tx_queue *)tx_queue;
+
+               atl_tx_queue_release_mbufs(txq);
+               rte_free(txq->sw_ring);
+               rte_free(txq);
+       }
+}
+
+void
+atl_free_queues(struct rte_eth_dev *dev)
+{
+       unsigned int i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               atl_rx_queue_release(dev->data->rx_queues[i]);
+               dev->data->rx_queues[i] = 0;
+       }
+       dev->data->nb_rx_queues = 0;
+
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               atl_tx_queue_release(dev->data->tx_queues[i]);
+               dev->data->tx_queues[i] = 0;
+       }
+       dev->data->nb_tx_queues = 0;
+}
+
+int
+atl_start_queues(struct rte_eth_dev *dev)
+{
+       int i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               if (atl_tx_queue_start(dev, i) != 0) {
+                       PMD_DRV_LOG(ERR,
+                               "Port %d: Start Tx queue %d failed",
+                               dev->data->port_id, i);
+                       return -1;
+               }
+       }
+
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               if (atl_rx_queue_start(dev, i) != 0) {
+                       PMD_DRV_LOG(ERR,
+                               "Port %d: Start Rx queue %d failed",
+                               dev->data->port_id, i);
+                       return -1;
+               }
+       }
+
        return 0;
 }
 
+int
+atl_stop_queues(struct rte_eth_dev *dev)
+{
+       int i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               if (atl_tx_queue_stop(dev, i) != 0) {
+                       PMD_DRV_LOG(ERR,
+                               "Port %d: Stop Tx queue %d failed",
+                               dev->data->port_id, i);
+                       return -1;
+               }
+       }
+
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               if (atl_rx_queue_stop(dev, i) != 0) {
+                       PMD_DRV_LOG(ERR,
+                               "Port %d: Stop Rx queue %d failed",
+                               dev->data->port_id, i);
+                       return -1;
+               }
+       }
+
+       return 0;
+}
+
+void
+atl_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
+                struct rte_eth_rxq_info *qinfo)
+{
+       struct atl_rx_queue *rxq;
+
+       PMD_INIT_FUNC_TRACE();
+
+       rxq = dev->data->rx_queues[queue_id];
+
+       qinfo->mp = rxq->mb_pool;
+       qinfo->scattered_rx = dev->data->scattered_rx;
+       qinfo->nb_desc = rxq->nb_rx_desc;
+}
+
+void
+atl_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
+                struct rte_eth_txq_info *qinfo)
+{
+       struct atl_tx_queue *txq;
+
+       PMD_INIT_FUNC_TRACE();
+
+       txq = dev->data->tx_queues[queue_id];
+
+       qinfo->nb_desc = txq->nb_tx_desc;
+}
+
+/* Return Rx queue avail count */
+
+uint32_t
+atl_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+       struct atl_rx_queue *rxq;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (rx_queue_id >= dev->data->nb_rx_queues) {
+               PMD_DRV_LOG(ERR, "Invalid RX queue id=%d", rx_queue_id);
+               return 0;
+       }
+
+       rxq = dev->data->rx_queues[rx_queue_id];
+
+       if (rxq == NULL)
+               return 0;
+
+       return rxq->nb_rx_desc - rxq->nb_rx_hold;
+}
+
+int
+atl_dev_rx_descriptor_status(void *rx_queue, uint16_t offset)
+{
+       struct atl_rx_queue *rxq = rx_queue;
+       struct hw_atl_rxd_wb_s *rxd;
+       uint32_t idx;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (unlikely(offset >= rxq->nb_rx_desc))
+               return -EINVAL;
+
+       if (offset >= rxq->nb_rx_desc - rxq->nb_rx_hold)
+               return RTE_ETH_RX_DESC_UNAVAIL;
+
+       idx = rxq->rx_tail + offset;
+
+       if (idx >= rxq->nb_rx_desc)
+               idx -= rxq->nb_rx_desc;
+
+       rxd = (struct hw_atl_rxd_wb_s *)&rxq->hw_ring[idx];
+
+       if (rxd->dd)
+               return RTE_ETH_RX_DESC_DONE;
+
+       return RTE_ETH_RX_DESC_AVAIL;
+}
+
+int
+atl_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
+{
+       struct atl_tx_queue *txq = tx_queue;
+       struct hw_atl_txd_s *txd;
+       uint32_t idx;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (unlikely(offset >= txq->nb_tx_desc))
+               return -EINVAL;
+
+       idx = txq->tx_tail + offset;
+
+       if (idx >= txq->nb_tx_desc)
+               idx -= txq->nb_tx_desc;
+
+       txd = &txq->hw_ring[idx];
+
+       if (txd->dd)
+               return RTE_ETH_TX_DESC_DONE;
+
+       return RTE_ETH_TX_DESC_FULL;
+}
+
+static int
+atl_rx_enable_intr(struct rte_eth_dev *dev, uint16_t queue_id, bool enable)
+{
+       struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct atl_rx_queue *rxq;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (queue_id >= dev->data->nb_rx_queues) {
+               PMD_DRV_LOG(ERR, "Invalid RX queue id=%d", queue_id);
+               return -EINVAL;
+       }
+
+       rxq = dev->data->rx_queues[queue_id];
+
+       if (rxq == NULL)
+               return 0;
+
+       /* Mapping interrupt vector */
+       hw_atl_itr_irq_map_en_rx_set(hw, enable, queue_id);
+
+       return 0;
+}
+
+int
+atl_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev, uint16_t queue_id)
+{
+       return atl_rx_enable_intr(eth_dev, queue_id, true);
+}
+
+int
+atl_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev, uint16_t queue_id)
+{
+       return atl_rx_enable_intr(eth_dev, queue_id, false);
+}
+
+uint16_t
+atl_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
+             uint16_t nb_pkts)
+{
+       int i, ret;
+       uint64_t ol_flags;
+       struct rte_mbuf *m;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < nb_pkts; i++) {
+               m = tx_pkts[i];
+               ol_flags = m->ol_flags;
+
+               if (m->nb_segs > AQ_HW_MAX_SEGS_SIZE) {
+                       rte_errno = EINVAL;
+                       return i;
+               }
+
+               if (ol_flags & ATL_TX_OFFLOAD_NOTSUP_MASK) {
+                       rte_errno = ENOTSUP;
+                       return i;
+               }
+
+#ifdef RTE_LIBRTE_ETHDEV_DEBUG
+               ret = rte_validate_tx_offload(m);
+               if (ret != 0) {
+                       rte_errno = -ret;
+                       return i;
+               }
+#endif
+               ret = rte_net_intel_cksum_prepare(m);
+               if (ret != 0) {
+                       rte_errno = -ret;
+                       return i;
+               }
+       }
+
+       return i;
+}
+
+static uint64_t
+atl_desc_to_offload_flags(struct atl_rx_queue *rxq,
+                         struct hw_atl_rxd_wb_s *rxd_wb)
+{
+       uint64_t mbuf_flags = 0;
+
+       PMD_INIT_FUNC_TRACE();
+
+       /* IPv4 ? */
+       if (rxq->l3_csum_enabled && ((rxd_wb->pkt_type & 0x3) == 0)) {
+               /* IPv4 csum error ? */
+               if (rxd_wb->rx_stat & BIT(1))
+                       mbuf_flags |= PKT_RX_IP_CKSUM_BAD;
+               else
+                       mbuf_flags |= PKT_RX_IP_CKSUM_GOOD;
+       } else {
+               mbuf_flags |= PKT_RX_IP_CKSUM_UNKNOWN;
+       }
+
+       /* CSUM calculated ? */
+       if (rxq->l4_csum_enabled && (rxd_wb->rx_stat & BIT(3))) {
+               if (rxd_wb->rx_stat & BIT(2))
+                       mbuf_flags |= PKT_RX_L4_CKSUM_BAD;
+               else
+                       mbuf_flags |= PKT_RX_L4_CKSUM_GOOD;
+       } else {
+               mbuf_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
+       }
+
+       return mbuf_flags;
+}
+
+static uint32_t
+atl_desc_to_pkt_type(struct hw_atl_rxd_wb_s *rxd_wb)
+{
+       uint32_t type = RTE_PTYPE_UNKNOWN;
+       uint16_t l2_l3_type = rxd_wb->pkt_type & 0x3;
+       uint16_t l4_type = (rxd_wb->pkt_type & 0x1C) >> 2;
+
+       switch (l2_l3_type) {
+       case 0:
+               type = RTE_PTYPE_L3_IPV4;
+               break;
+       case 1:
+               type = RTE_PTYPE_L3_IPV6;
+               break;
+       case 2:
+               type = RTE_PTYPE_L2_ETHER;
+               break;
+       case 3:
+               type = RTE_PTYPE_L2_ETHER_ARP;
+               break;
+       }
+
+       switch (l4_type) {
+       case 0:
+               type |= RTE_PTYPE_L4_TCP;
+               break;
+       case 1:
+               type |= RTE_PTYPE_L4_UDP;
+               break;
+       case 2:
+               type |= RTE_PTYPE_L4_SCTP;
+               break;
+       case 3:
+               type |= RTE_PTYPE_L4_ICMP;
+               break;
+       }
+
+       if (rxd_wb->pkt_type & BIT(5))
+               type |= RTE_PTYPE_L2_ETHER_VLAN;
+
+       return type;
+}
+
+uint16_t
+atl_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+{
+       struct atl_rx_queue *rxq = (struct atl_rx_queue *)rx_queue;
+       struct rte_eth_dev *dev = &rte_eth_devices[rxq->port_id];
+       struct atl_adapter *adapter =
+               ATL_DEV_TO_ADAPTER(&rte_eth_devices[rxq->port_id]);
+       struct aq_hw_s *hw = ATL_DEV_PRIVATE_TO_HW(adapter);
+       struct aq_hw_cfg_s *cfg =
+               ATL_DEV_PRIVATE_TO_CFG(dev->data->dev_private);
+       struct atl_rx_entry *sw_ring = rxq->sw_ring;
+
+       struct rte_mbuf *new_mbuf;
+       struct rte_mbuf *rx_mbuf, *rx_mbuf_prev, *rx_mbuf_first;
+       struct atl_rx_entry *rx_entry;
+       uint16_t nb_rx = 0;
+       uint16_t nb_hold = 0;
+       struct hw_atl_rxd_wb_s rxd_wb;
+       struct hw_atl_rxd_s *rxd = NULL;
+       uint16_t tail = rxq->rx_tail;
+       uint64_t dma_addr;
+       uint16_t pkt_len = 0;
+
+       while (nb_rx < nb_pkts) {
+               uint16_t eop_tail = tail;
+
+               rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[tail];
+               rxd_wb = *(struct hw_atl_rxd_wb_s *)rxd;
+
+               if (!rxd_wb.dd) { /* RxD is not done */
+                       break;
+               }
+
+               PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u tail=%u "
+                          "eop=0x%x pkt_len=%u hash=0x%x hash_type=0x%x",
+                          (unsigned int)rxq->port_id,
+                          (unsigned int)rxq->queue_id,
+                          (unsigned int)tail, (unsigned int)rxd_wb.eop,
+                          (unsigned int)rte_le_to_cpu_16(rxd_wb.pkt_len),
+                       rxd_wb.rss_hash, rxd_wb.rss_type);
+
+               /* RxD is not done */
+               if (!rxd_wb.eop) {
+                       while (true) {
+                               struct hw_atl_rxd_wb_s *eop_rxwbd;
+
+                               eop_tail = (eop_tail + 1) % rxq->nb_rx_desc;
+                               eop_rxwbd = (struct hw_atl_rxd_wb_s *)
+                                       &rxq->hw_ring[eop_tail];
+                               if (!eop_rxwbd->dd) {
+                                       /* no EOP received yet */
+                                       eop_tail = tail;
+                                       break;
+                               }
+                               if (eop_rxwbd->dd && eop_rxwbd->eop)
+                                       break;
+                       }
+                       /* No EOP in ring */
+                       if (eop_tail == tail)
+                               break;
+               }
+               rx_mbuf_prev = NULL;
+               rx_mbuf_first = NULL;
+
+               /* Run through packet segments */
+               while (true) {
+                       new_mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
+                       if (new_mbuf == NULL) {
+                               PMD_RX_LOG(DEBUG,
+                                  "RX mbuf alloc failed port_id=%u "
+                                  "queue_id=%u", (unsigned int)rxq->port_id,
+                                  (unsigned int)rxq->queue_id);
+                               dev->data->rx_mbuf_alloc_failed++;
+                               adapter->sw_stats.rx_nombuf++;
+                               goto err_stop;
+                       }
+
+                       nb_hold++;
+                       rx_entry = &sw_ring[tail];
+
+                       rx_mbuf = rx_entry->mbuf;
+                       rx_entry->mbuf = new_mbuf;
+                       dma_addr = rte_cpu_to_le_64(
+                               rte_mbuf_data_iova_default(new_mbuf));
+
+                       /* setup RX descriptor */
+                       rxd->hdr_addr = 0;
+                       rxd->buf_addr = dma_addr;
+
+                       /*
+                        * Initialize the returned mbuf.
+                        * 1) setup generic mbuf fields:
+                        *        - number of segments,
+                        *        - next segment,
+                        *        - packet length,
+                        *        - RX port identifier.
+                        * 2) integrate hardware offload data, if any:
+                        *      <  - RSS flag & hash,
+                        *        - IP checksum flag,
+                        *        - VLAN TCI, if any,
+                        *        - error flags.
+                        */
+                       pkt_len = (uint16_t)rte_le_to_cpu_16(rxd_wb.pkt_len);
+                       rx_mbuf->data_off = RTE_PKTMBUF_HEADROOM;
+                       rte_prefetch1((char *)rx_mbuf->buf_addr +
+                               rx_mbuf->data_off);
+                       rx_mbuf->nb_segs = 0;
+                       rx_mbuf->next = NULL;
+                       rx_mbuf->pkt_len = pkt_len;
+                       rx_mbuf->data_len = pkt_len;
+                       if (rxd_wb.eop) {
+                               u16 remainder_len = pkt_len % rxq->buff_size;
+                               if (!remainder_len)
+                                       remainder_len = rxq->buff_size;
+                               rx_mbuf->data_len = remainder_len;
+                       } else {
+                               rx_mbuf->data_len = pkt_len > rxq->buff_size ?
+                                               rxq->buff_size : pkt_len;
+                       }
+                       rx_mbuf->port = rxq->port_id;
+
+                       rx_mbuf->hash.rss = rxd_wb.rss_hash;
+
+                       rx_mbuf->vlan_tci = rxd_wb.vlan;
+
+                       rx_mbuf->ol_flags =
+                               atl_desc_to_offload_flags(rxq, &rxd_wb);
+
+                       rx_mbuf->packet_type = atl_desc_to_pkt_type(&rxd_wb);
+
+                       if (rx_mbuf->packet_type & RTE_PTYPE_L2_ETHER_VLAN) {
+                               rx_mbuf->ol_flags |= PKT_RX_VLAN;
+                               rx_mbuf->vlan_tci = rxd_wb.vlan;
+
+                               if (cfg->vlan_strip)
+                                       rx_mbuf->ol_flags |=
+                                               PKT_RX_VLAN_STRIPPED;
+                       }
+
+                       if (!rx_mbuf_first)
+                               rx_mbuf_first = rx_mbuf;
+                       rx_mbuf_first->nb_segs++;
+
+                       if (rx_mbuf_prev)
+                               rx_mbuf_prev->next = rx_mbuf;
+                       rx_mbuf_prev = rx_mbuf;
+
+                       tail = (tail + 1) % rxq->nb_rx_desc;
+                       /* Prefetch next mbufs */
+                       rte_prefetch0(sw_ring[tail].mbuf);
+                       if ((tail & 0x3) == 0) {
+                               rte_prefetch0(&sw_ring[tail]);
+                               rte_prefetch0(&sw_ring[tail]);
+                       }
+
+                       /* filled mbuf_first */
+                       if (rxd_wb.eop)
+                               break;
+                       rxd = (struct hw_atl_rxd_s *)&rxq->hw_ring[tail];
+                       rxd_wb = *(struct hw_atl_rxd_wb_s *)rxd;
+               };
+
+               /*
+                * Store the mbuf address into the next entry of the array
+                * of returned packets.
+                */
+               rx_pkts[nb_rx++] = rx_mbuf_first;
+               adapter->sw_stats.q_ipackets[rxq->queue_id]++;
+               adapter->sw_stats.q_ibytes[rxq->queue_id] +=
+                       rx_mbuf_first->pkt_len;
+
+               PMD_RX_LOG(DEBUG, "add mbuf segs=%d pkt_len=%d",
+                       rx_mbuf_first->nb_segs,
+                       rx_mbuf_first->pkt_len);
+       }
+
+err_stop:
+
+       rxq->rx_tail = tail;
+
+       /*
+        * If the number of free RX descriptors is greater than the RX free
+        * threshold of the queue, advance the Receive Descriptor Tail (RDT)
+        * register.
+        * Update the RDT with the value of the last processed RX descriptor
+        * minus 1, to guarantee that the RDT register is never equal to the
+        * RDH register, which creates a "full" ring situtation from the
+        * hardware point of view...
+        */
+       nb_hold = (uint16_t)(nb_hold + rxq->nb_rx_hold);
+       if (nb_hold > rxq->rx_free_thresh) {
+               PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
+                       "nb_hold=%u nb_rx=%u",
+                       (unsigned int)rxq->port_id, (unsigned int)rxq->queue_id,
+                       (unsigned int)tail, (unsigned int)nb_hold,
+                       (unsigned int)nb_rx);
+               tail = (uint16_t)((tail == 0) ?
+                       (rxq->nb_rx_desc - 1) : (tail - 1));
+
+               hw_atl_reg_rx_dma_desc_tail_ptr_set(hw, tail, rxq->queue_id);
+
+               nb_hold = 0;
+       }
+
+       rxq->nb_rx_hold = nb_hold;
+
+       return nb_rx;
+}
+
+static void
+atl_xmit_cleanup(struct atl_tx_queue *txq)
+{
+       struct atl_tx_entry *sw_ring;
+       struct hw_atl_txd_s *txd;
+       int to_clean = 0;
+
+       if (txq != NULL) {
+               sw_ring = txq->sw_ring;
+               int head = txq->tx_head;
+               int cnt;
+               int i;
+
+               for (i = 0, cnt = head; ; i++) {
+                       txd = &txq->hw_ring[cnt];
+
+                       if (txd->dd)
+                               to_clean++;
+
+                       cnt = (cnt + 1) % txq->nb_tx_desc;
+                       if (cnt == txq->tx_tail)
+                               break;
+               }
+
+               if (to_clean == 0)
+                       return;
+
+               while (to_clean) {
+                       txd = &txq->hw_ring[head];
+
+                       struct atl_tx_entry *rx_entry = &sw_ring[head];
+
+                       if (rx_entry->mbuf) {
+                               rte_pktmbuf_free_seg(rx_entry->mbuf);
+                               rx_entry->mbuf = NULL;
+                       }
+
+                       if (txd->dd)
+                               to_clean--;
+
+                       txd->buf_addr = 0;
+                       txd->flags = 0;
+
+                       head = (head + 1) % txq->nb_tx_desc;
+                       txq->tx_free++;
+               }
+
+               txq->tx_head = head;
+       }
+}
+
+static int
+atl_tso_setup(struct rte_mbuf *tx_pkt, union hw_atl_txc_s *txc)
+{
+       uint32_t tx_cmd = 0;
+       uint64_t ol_flags = tx_pkt->ol_flags;
+
+       if (ol_flags & PKT_TX_TCP_SEG) {
+               tx_cmd |= tx_desc_cmd_lso | tx_desc_cmd_l4cs;
+
+               txc->cmd = 0x4;
+
+               if (ol_flags & PKT_TX_IPV6)
+                       txc->cmd |= 0x2;
+
+               txc->l2_len = tx_pkt->l2_len;
+               txc->l3_len = tx_pkt->l3_len;
+               txc->l4_len = tx_pkt->l4_len;
+
+               txc->mss_len = tx_pkt->tso_segsz;
+       }
+
+       if (ol_flags & PKT_TX_VLAN) {
+               tx_cmd |= tx_desc_cmd_vlan;
+               txc->vlan_tag = tx_pkt->vlan_tci;
+       }
+
+       if (tx_cmd) {
+               txc->type = tx_desc_type_ctx;
+               txc->idx = 0;
+       }
+
+       return tx_cmd;
+}
+
+static inline void
+atl_setup_csum_offload(struct rte_mbuf *mbuf, struct hw_atl_txd_s *txd,
+                      uint32_t tx_cmd)
+{
+       txd->cmd |= tx_desc_cmd_fcs;
+       txd->cmd |= (mbuf->ol_flags & PKT_TX_IP_CKSUM) ? tx_desc_cmd_ipv4 : 0;
+       /* L4 csum requested */
+       txd->cmd |= (mbuf->ol_flags & PKT_TX_L4_MASK) ? tx_desc_cmd_l4cs : 0;
+       txd->cmd |= tx_cmd;
+}
+
+static inline void
+atl_xmit_pkt(struct aq_hw_s *hw, struct atl_tx_queue *txq,
+            struct rte_mbuf *tx_pkt)
+{
+       struct atl_adapter *adapter =
+               ATL_DEV_TO_ADAPTER(&rte_eth_devices[txq->port_id]);
+       uint32_t pay_len = 0;
+       int tail = 0;
+       struct atl_tx_entry *tx_entry;
+       uint64_t buf_dma_addr;
+       struct rte_mbuf *m_seg;
+       union hw_atl_txc_s *txc = NULL;
+       struct hw_atl_txd_s *txd = NULL;
+       u32 tx_cmd = 0U;
+       int desc_count = 0;
+
+       tail = txq->tx_tail;
+
+       txc = (union hw_atl_txc_s *)&txq->hw_ring[tail];
+
+       txc->flags1 = 0U;
+       txc->flags2 = 0U;
+
+       tx_cmd = atl_tso_setup(tx_pkt, txc);
+
+       if (tx_cmd) {
+               /* We've consumed the first desc, adjust counters */
+               tail = (tail + 1) % txq->nb_tx_desc;
+               txq->tx_tail = tail;
+               txq->tx_free -= 1;
+
+               txd = &txq->hw_ring[tail];
+               txd->flags = 0U;
+       } else {
+               txd = (struct hw_atl_txd_s *)txc;
+       }
+
+       txd->ct_en = !!tx_cmd;
+
+       txd->type = tx_desc_type_desc;
+
+       atl_setup_csum_offload(tx_pkt, txd, tx_cmd);
+
+       if (tx_cmd)
+               txd->ct_idx = 0;
+
+       pay_len = tx_pkt->pkt_len;
+
+       txd->pay_len = pay_len;
+
+       for (m_seg = tx_pkt; m_seg; m_seg = m_seg->next) {
+               if (desc_count > 0) {
+                       txd = &txq->hw_ring[tail];
+                       txd->flags = 0U;
+               }
+
+               buf_dma_addr = rte_mbuf_data_iova(m_seg);
+               txd->buf_addr = rte_cpu_to_le_64(buf_dma_addr);
+
+               txd->type = tx_desc_type_desc;
+               txd->len = m_seg->data_len;
+               txd->pay_len = pay_len;
+
+               /* Store mbuf for freeing later */
+               tx_entry = &txq->sw_ring[tail];
+
+               if (tx_entry->mbuf)
+                       rte_pktmbuf_free_seg(tx_entry->mbuf);
+               tx_entry->mbuf = m_seg;
+
+               tail = (tail + 1) % txq->nb_tx_desc;
+
+               desc_count++;
+       }
+
+       // Last descriptor requires EOP and WB
+       txd->eop = 1U;
+       txd->cmd |= tx_desc_cmd_wb;
+
+       hw_atl_b0_hw_tx_ring_tail_update(hw, tail, txq->queue_id);
+
+       txq->tx_tail = tail;
+
+       txq->tx_free -= desc_count;
+
+       adapter->sw_stats.q_opackets[txq->queue_id]++;
+       adapter->sw_stats.q_obytes[txq->queue_id] += pay_len;
+}
+
+uint16_t
+atl_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
+{
+       struct rte_eth_dev *dev = NULL;
+       struct aq_hw_s *hw = NULL;
+       struct atl_tx_queue *txq = tx_queue;
+       struct rte_mbuf *tx_pkt;
+       uint16_t nb_tx;
+
+       dev = &rte_eth_devices[txq->port_id];
+       hw = ATL_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       PMD_TX_LOG(DEBUG,
+               "port %d txq %d pkts: %d tx_free=%d tx_tail=%d tx_head=%d",
+               txq->port_id, txq->queue_id, nb_pkts, txq->tx_free,
+               txq->tx_tail, txq->tx_head);
+
+       for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
+               tx_pkt = *tx_pkts++;
+
+               /* Clean Tx queue if needed */
+               if (txq->tx_free < txq->tx_free_thresh)
+                       atl_xmit_cleanup(txq);
+
+               /* Check if we have enough free descriptors */
+               if (txq->tx_free < tx_pkt->nb_segs)
+                       break;
+
+               /* check mbuf is valid */
+               if ((tx_pkt->nb_segs == 0) ||
+                       ((tx_pkt->nb_segs > 1) && (tx_pkt->next == NULL)))
+                       break;
+
+               /* Send the packet */
+               atl_xmit_pkt(hw, txq, tx_pkt);
+       }
+
+       PMD_TX_LOG(DEBUG, "atl_xmit_pkts %d transmitted", nb_tx);
+
+       return nb_tx;
+}