net/hns3: rename Rx burst function
[dpdk.git] / drivers / net / hns3 / hns3_rxtx.c
index 3e708b5..9bb30fc 100644 (file)
@@ -1,28 +1,19 @@
 /* SPDX-License-Identifier: BSD-3-Clause
- * Copyright(c) 2018-2019 Hisilicon Limited.
+ * Copyright(c) 2018-2021 HiSilicon Limited.
  */
 
-#include <stdarg.h>
-#include <stdbool.h>
-#include <stdint.h>
-#include <stdio.h>
-#include <unistd.h>
-#include <inttypes.h>
 #include <rte_bus_pci.h>
-#include <rte_byteorder.h>
 #include <rte_common.h>
 #include <rte_cycles.h>
-#include <rte_dev.h>
-#include <rte_eal.h>
-#include <rte_ether.h>
+#include <rte_geneve.h>
 #include <rte_vxlan.h>
-#include <rte_ethdev_driver.h>
+#include <ethdev_driver.h>
 #include <rte_io.h>
-#include <rte_ip.h>
-#include <rte_gre.h>
 #include <rte_net.h>
 #include <rte_malloc.h>
-#include <rte_pci.h>
+#if defined(RTE_ARCH_ARM64)
+#include <rte_cpuflags.h>
+#endif
 
 #include "hns3_ethdev.h"
 #include "hns3_rxtx.h"
@@ -41,9 +32,23 @@ hns3_rx_queue_release_mbufs(struct hns3_rx_queue *rxq)
        if (rxq->sw_ring == NULL)
                return;
 
-       for (i = 0; i < rxq->nb_rx_desc; i++)
-               if (rxq->sw_ring[i].mbuf)
-                       rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
+       if (rxq->rx_rearm_nb == 0) {
+               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;
+                       }
+               }
+       } else {
+               for (i = rxq->next_to_use;
+                    i != rxq->rx_rearm_start;
+                    i = (i + 1) % rxq->nb_rx_desc) {
+                       if (rxq->sw_ring[i].mbuf != NULL) {
+                               rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
+                               rxq->sw_ring[i].mbuf = NULL;
+                       }
+               }
+       }
 
        for (i = 0; i < rxq->bulk_mbuf_num; i++)
                rte_pktmbuf_free_seg(rxq->bulk_mbuf[i]);
@@ -60,7 +65,7 @@ hns3_tx_queue_release_mbufs(struct hns3_tx_queue *txq)
 {
        uint16_t i;
 
-       /* Note: Fake rx queue will not enter here */
+       /* Note: Fake tx queue will not enter here */
        if (txq->sw_ring) {
                for (i = 0; i < txq->nb_tx_desc; i++) {
                        if (txq->sw_ring[i].mbuf) {
@@ -252,7 +257,7 @@ hns3_alloc_rx_queue_mbufs(struct hns3_hw *hw, struct hns3_rx_queue *rxq)
        for (i = 0; i < rxq->nb_rx_desc; i++) {
                mbuf = rte_mbuf_raw_alloc(rxq->mb_pool);
                if (unlikely(mbuf == NULL)) {
-                       hns3_err(hw, "Failed to allocate RXD[%d] for rx queue!",
+                       hns3_err(hw, "Failed to allocate RXD[%u] for rx queue!",
                                 i);
                        hns3_rx_queue_release_mbufs(rxq);
                        return -ENOMEM;
@@ -325,30 +330,43 @@ hns3_init_tx_queue_hw(struct hns3_tx_queue *txq)
 }
 
 void
-hns3_update_all_queues_pvid_state(struct hns3_hw *hw)
+hns3_update_all_queues_pvid_proc_en(struct hns3_hw *hw)
 {
        uint16_t nb_rx_q = hw->data->nb_rx_queues;
        uint16_t nb_tx_q = hw->data->nb_tx_queues;
        struct hns3_rx_queue *rxq;
        struct hns3_tx_queue *txq;
-       int pvid_state;
+       bool pvid_en;
        int i;
 
-       pvid_state = hw->port_base_vlan_cfg.state;
+       pvid_en = hw->port_base_vlan_cfg.state == HNS3_PORT_BASE_VLAN_ENABLE;
        for (i = 0; i < hw->cfg_max_queues; i++) {
                if (i < nb_rx_q) {
                        rxq = hw->data->rx_queues[i];
                        if (rxq != NULL)
-                               rxq->pvid_state = pvid_state;
+                               rxq->pvid_sw_discard_en = pvid_en;
                }
                if (i < nb_tx_q) {
                        txq = hw->data->tx_queues[i];
                        if (txq != NULL)
-                               txq->pvid_state = pvid_state;
+                               txq->pvid_sw_shift_en = pvid_en;
                }
        }
 }
 
+static void
+hns3_stop_unused_queue(void *tqp_base, enum hns3_ring_type queue_type)
+{
+       uint32_t reg_offset;
+       uint32_t reg;
+
+       reg_offset = queue_type == HNS3_RING_TYPE_TX ?
+                                  HNS3_RING_TX_EN_REG : HNS3_RING_RX_EN_REG;
+       reg = hns3_read_reg(tqp_base, reg_offset);
+       reg &= ~BIT(HNS3_RING_EN_B);
+       hns3_write_reg(tqp_base, reg_offset, reg);
+}
+
 void
 hns3_enable_all_queues(struct hns3_hw *hw, bool en)
 {
@@ -357,27 +375,185 @@ hns3_enable_all_queues(struct hns3_hw *hw, bool en)
        struct hns3_rx_queue *rxq;
        struct hns3_tx_queue *txq;
        uint32_t rcb_reg;
+       void *tqp_base;
        int i;
 
        for (i = 0; i < hw->cfg_max_queues; i++) {
-               if (i < nb_rx_q)
-                       rxq = hw->data->rx_queues[i];
-               else
-                       rxq = hw->fkq_data.rx_queues[i - nb_rx_q];
-               if (i < nb_tx_q)
-                       txq = hw->data->tx_queues[i];
-               else
-                       txq = hw->fkq_data.tx_queues[i - nb_tx_q];
-               if (rxq == NULL || txq == NULL ||
-                   (en && (rxq->rx_deferred_start || txq->tx_deferred_start)))
-                       continue;
+               if (hns3_dev_indep_txrx_supported(hw)) {
+                       rxq = i < nb_rx_q ? hw->data->rx_queues[i] : NULL;
+                       txq = i < nb_tx_q ? hw->data->tx_queues[i] : NULL;
+
+                       tqp_base = (void *)((char *)hw->io_base +
+                                       hns3_get_tqp_reg_offset(i));
+                       /*
+                        * If queue struct is not initialized, it means the
+                        * related HW ring has not been initialized yet.
+                        * So, these queues should be disabled before enable
+                        * the tqps to avoid a HW exception since the queues
+                        * are enabled by default.
+                        */
+                       if (rxq == NULL)
+                               hns3_stop_unused_queue(tqp_base,
+                                                       HNS3_RING_TYPE_RX);
+                       if (txq == NULL)
+                               hns3_stop_unused_queue(tqp_base,
+                                                       HNS3_RING_TYPE_TX);
+               } else {
+                       rxq = i < nb_rx_q ? hw->data->rx_queues[i] :
+                             hw->fkq_data.rx_queues[i - nb_rx_q];
 
-               rcb_reg = hns3_read_dev(rxq, HNS3_RING_EN_REG);
+                       tqp_base = rxq->io_base;
+               }
+               /*
+                * This is the master switch that used to control the enabling
+                * of a pair of Tx and Rx queues. Both the Rx and Tx point to
+                * the same register
+                */
+               rcb_reg = hns3_read_reg(tqp_base, HNS3_RING_EN_REG);
                if (en)
                        rcb_reg |= BIT(HNS3_RING_EN_B);
                else
                        rcb_reg &= ~BIT(HNS3_RING_EN_B);
-               hns3_write_dev(rxq, HNS3_RING_EN_REG, rcb_reg);
+               hns3_write_reg(tqp_base, HNS3_RING_EN_REG, rcb_reg);
+       }
+}
+
+static void
+hns3_enable_txq(struct hns3_tx_queue *txq, bool en)
+{
+       struct hns3_hw *hw = &txq->hns->hw;
+       uint32_t reg;
+
+       if (hns3_dev_indep_txrx_supported(hw)) {
+               reg = hns3_read_dev(txq, HNS3_RING_TX_EN_REG);
+               if (en)
+                       reg |= BIT(HNS3_RING_EN_B);
+               else
+                       reg &= ~BIT(HNS3_RING_EN_B);
+               hns3_write_dev(txq, HNS3_RING_TX_EN_REG, reg);
+       }
+       txq->enabled = en;
+}
+
+static void
+hns3_enable_rxq(struct hns3_rx_queue *rxq, bool en)
+{
+       struct hns3_hw *hw = &rxq->hns->hw;
+       uint32_t reg;
+
+       if (hns3_dev_indep_txrx_supported(hw)) {
+               reg = hns3_read_dev(rxq, HNS3_RING_RX_EN_REG);
+               if (en)
+                       reg |= BIT(HNS3_RING_EN_B);
+               else
+                       reg &= ~BIT(HNS3_RING_EN_B);
+               hns3_write_dev(rxq, HNS3_RING_RX_EN_REG, reg);
+       }
+       rxq->enabled = en;
+}
+
+int
+hns3_start_all_txqs(struct rte_eth_dev *dev)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct hns3_tx_queue *txq;
+       uint16_t i, j;
+
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               txq = hw->data->tx_queues[i];
+               if (!txq) {
+                       hns3_err(hw, "Tx queue %u not available or setup.", i);
+                       goto start_txqs_fail;
+               }
+               /*
+                * Tx queue is enabled by default. Therefore, the Tx queues
+                * needs to be disabled when deferred_start is set. There is
+                * another master switch used to control the enabling of a pair
+                * of Tx and Rx queues. And the master switch is disabled by
+                * default.
+                */
+               if (txq->tx_deferred_start)
+                       hns3_enable_txq(txq, false);
+               else
+                       hns3_enable_txq(txq, true);
+       }
+       return 0;
+
+start_txqs_fail:
+       for (j = 0; j < i; j++) {
+               txq = hw->data->tx_queues[j];
+               hns3_enable_txq(txq, false);
+       }
+       return -EINVAL;
+}
+
+int
+hns3_start_all_rxqs(struct rte_eth_dev *dev)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct hns3_rx_queue *rxq;
+       uint16_t i, j;
+
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               rxq = hw->data->rx_queues[i];
+               if (!rxq) {
+                       hns3_err(hw, "Rx queue %u not available or setup.", i);
+                       goto start_rxqs_fail;
+               }
+               /*
+                * Rx queue is enabled by default. Therefore, the Rx queues
+                * needs to be disabled when deferred_start is set. There is
+                * another master switch used to control the enabling of a pair
+                * of Tx and Rx queues. And the master switch is disabled by
+                * default.
+                */
+               if (rxq->rx_deferred_start)
+                       hns3_enable_rxq(rxq, false);
+               else
+                       hns3_enable_rxq(rxq, true);
+       }
+       return 0;
+
+start_rxqs_fail:
+       for (j = 0; j < i; j++) {
+               rxq = hw->data->rx_queues[j];
+               hns3_enable_rxq(rxq, false);
+       }
+       return -EINVAL;
+}
+
+void
+hns3_restore_tqp_enable_state(struct hns3_hw *hw)
+{
+       struct hns3_rx_queue *rxq;
+       struct hns3_tx_queue *txq;
+       uint16_t i;
+
+       for (i = 0; i < hw->data->nb_rx_queues; i++) {
+               rxq = hw->data->rx_queues[i];
+               if (rxq != NULL)
+                       hns3_enable_rxq(rxq, rxq->enabled);
+       }
+
+       for (i = 0; i < hw->data->nb_tx_queues; i++) {
+               txq = hw->data->tx_queues[i];
+               if (txq != NULL)
+                       hns3_enable_txq(txq, txq->enabled);
+       }
+}
+
+void
+hns3_stop_all_txqs(struct rte_eth_dev *dev)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct hns3_tx_queue *txq;
+       uint16_t i;
+
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               txq = hw->data->tx_queues[i];
+               if (!txq)
+                       continue;
+               hns3_enable_txq(txq, false);
        }
 }
 
@@ -391,7 +567,7 @@ hns3_tqp_enable(struct hns3_hw *hw, uint16_t queue_id, bool enable)
        req = (struct hns3_cfg_com_tqp_queue_cmd *)desc.data;
 
        hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_CFG_COM_TQP_QUEUE, false);
-       req->tqp_id = rte_cpu_to_le_16(queue_id & HNS3_RING_ID_MASK);
+       req->tqp_id = rte_cpu_to_le_16(queue_id);
        req->stream_id = 0;
        hns3_set_bit(req->enable, HNS3_TQP_ENABLE_B, enable ? 1 : 0);
 
@@ -412,18 +588,19 @@ hns3_send_reset_tqp_cmd(struct hns3_hw *hw, uint16_t queue_id, bool enable)
        hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RESET_TQP_QUEUE, false);
 
        req = (struct hns3_reset_tqp_queue_cmd *)desc.data;
-       req->tqp_id = rte_cpu_to_le_16(queue_id & HNS3_RING_ID_MASK);
+       req->tqp_id = rte_cpu_to_le_16(queue_id);
        hns3_set_bit(req->reset_req, HNS3_TQP_RESET_B, enable ? 1 : 0);
-
        ret = hns3_cmd_send(hw, &desc, 1);
        if (ret)
-               hns3_err(hw, "Send tqp reset cmd error, ret = %d", ret);
+               hns3_err(hw, "send tqp reset cmd error, queue_id = %u, "
+                            "ret = %d", queue_id, ret);
 
        return ret;
 }
 
 static int
-hns3_get_reset_status(struct hns3_hw *hw, uint16_t queue_id)
+hns3_get_tqp_reset_status(struct hns3_hw *hw, uint16_t queue_id,
+                         uint8_t *reset_status)
 {
        struct hns3_reset_tqp_queue_cmd *req;
        struct hns3_cmd_desc desc;
@@ -432,29 +609,26 @@ hns3_get_reset_status(struct hns3_hw *hw, uint16_t queue_id)
        hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RESET_TQP_QUEUE, true);
 
        req = (struct hns3_reset_tqp_queue_cmd *)desc.data;
-       req->tqp_id = rte_cpu_to_le_16(queue_id & HNS3_RING_ID_MASK);
+       req->tqp_id = rte_cpu_to_le_16(queue_id);
 
        ret = hns3_cmd_send(hw, &desc, 1);
        if (ret) {
-               hns3_err(hw, "Get reset status error, ret =%d", ret);
+               hns3_err(hw, "get tqp reset status error, queue_id = %u, "
+                            "ret = %d.", queue_id, ret);
                return ret;
        }
-
-       return hns3_get_bit(req->ready_to_reset, HNS3_TQP_RESET_B);
+       *reset_status = hns3_get_bit(req->ready_to_reset, HNS3_TQP_RESET_B);
+       return ret;
 }
 
 static int
-hns3_reset_tqp(struct hns3_hw *hw, uint16_t queue_id)
+hns3pf_reset_tqp(struct hns3_hw *hw, uint16_t queue_id)
 {
 #define HNS3_TQP_RESET_TRY_MS  200
-       uint64_t end;
-       int reset_status;
+       uint16_t wait_time = 0;
+       uint8_t reset_status;
        int ret;
 
-       ret = hns3_tqp_enable(hw, queue_id, false);
-       if (ret)
-               return ret;
-
        /*
         * In current version VF is not supported when PF is driven by DPDK
         * driver, all task queue pairs are mapped to PF function, so PF's queue
@@ -465,21 +639,24 @@ hns3_reset_tqp(struct hns3_hw *hw, uint16_t queue_id)
                hns3_err(hw, "Send reset tqp cmd fail, ret = %d", ret);
                return ret;
        }
-       ret = -ETIMEDOUT;
-       end = get_timeofday_ms() + HNS3_TQP_RESET_TRY_MS;
+
        do {
                /* Wait for tqp hw reset */
                rte_delay_ms(HNS3_POLL_RESPONE_MS);
-               reset_status = hns3_get_reset_status(hw, queue_id);
-               if (reset_status) {
-                       ret = 0;
+               wait_time += HNS3_POLL_RESPONE_MS;
+               ret = hns3_get_tqp_reset_status(hw, queue_id, &reset_status);
+               if (ret)
+                       goto tqp_reset_fail;
+
+               if (reset_status)
                        break;
-               }
-       } while (get_timeofday_ms() < end);
+       } while (wait_time < HNS3_TQP_RESET_TRY_MS);
 
-       if (ret) {
-               hns3_err(hw, "Reset TQP fail, ret = %d", ret);
-               return ret;
+       if (!reset_status) {
+               ret = -ETIMEDOUT;
+               hns3_err(hw, "reset tqp timeout, queue_id = %u, ret = %d",
+                            queue_id, ret);
+               goto tqp_reset_fail;
        }
 
        ret = hns3_send_reset_tqp_cmd(hw, queue_id, false);
@@ -487,6 +664,10 @@ hns3_reset_tqp(struct hns3_hw *hw, uint16_t queue_id)
                hns3_err(hw, "Deassert the soft reset fail, ret = %d", ret);
 
        return ret;
+
+tqp_reset_fail:
+       hns3_send_reset_tqp_cmd(hw, queue_id, false);
+       return ret;
 }
 
 static int
@@ -495,41 +676,272 @@ hns3vf_reset_tqp(struct hns3_hw *hw, uint16_t queue_id)
        uint8_t msg_data[2];
        int ret;
 
-       /* Disable VF's queue before send queue reset msg to PF */
-       ret = hns3_tqp_enable(hw, queue_id, false);
+       memcpy(msg_data, &queue_id, sizeof(uint16_t));
+
+       ret = hns3_send_mbx_msg(hw, HNS3_MBX_QUEUE_RESET, 0, msg_data,
+                                sizeof(msg_data), true, NULL, 0);
+       if (ret)
+               hns3_err(hw, "fail to reset tqp, queue_id = %u, ret = %d.",
+                        queue_id, ret);
+       return ret;
+}
+
+static int
+hns3_reset_rcb_cmd(struct hns3_hw *hw, uint8_t *reset_status)
+{
+       struct hns3_reset_cmd *req;
+       struct hns3_cmd_desc desc;
+       int ret;
+
+       hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_CFG_RST_TRIGGER, false);
+       req = (struct hns3_reset_cmd *)desc.data;
+       hns3_set_bit(req->mac_func_reset, HNS3_CFG_RESET_RCB_B, 1);
+
+       /*
+        * The start qid should be the global qid of the first tqp of the
+        * function which should be reset in this port. Since our PF not
+        * support take over of VFs, so we only need to reset function 0,
+        * and its start qid is always 0.
+        */
+       req->fun_reset_rcb_vqid_start = rte_cpu_to_le_16(0);
+       req->fun_reset_rcb_vqid_num = rte_cpu_to_le_16(hw->cfg_max_queues);
+
+       ret = hns3_cmd_send(hw, &desc, 1);
+       if (ret) {
+               hns3_err(hw, "fail to send rcb reset cmd, ret = %d.", ret);
+               return ret;
+       }
+
+       *reset_status = req->fun_reset_rcb_return_status;
+       return 0;
+}
+
+static int
+hns3pf_reset_all_tqps(struct hns3_hw *hw)
+{
+#define HNS3_RESET_RCB_NOT_SUPPORT     0U
+#define HNS3_RESET_ALL_TQP_SUCCESS     1U
+       uint8_t reset_status;
+       int ret;
+       int i;
+
+       ret = hns3_reset_rcb_cmd(hw, &reset_status);
        if (ret)
                return ret;
 
-       memcpy(msg_data, &queue_id, sizeof(uint16_t));
+       /*
+        * If the firmware version is low, it may not support the rcb reset
+        * which means reset all the tqps at a time. In this case, we should
+        * reset tqps one by one.
+        */
+       if (reset_status == HNS3_RESET_RCB_NOT_SUPPORT) {
+               for (i = 0; i < hw->cfg_max_queues; i++) {
+                       ret = hns3pf_reset_tqp(hw, i);
+                       if (ret) {
+                               hns3_err(hw,
+                                 "fail to reset tqp, queue_id = %d, ret = %d.",
+                                 i, ret);
+                               return ret;
+                       }
+               }
+       } else if (reset_status != HNS3_RESET_ALL_TQP_SUCCESS) {
+               hns3_err(hw, "fail to reset all tqps, reset_status = %u.",
+                               reset_status);
+               return -EIO;
+       }
 
-       return hns3_send_mbx_msg(hw, HNS3_MBX_QUEUE_RESET, 0, msg_data,
-                                sizeof(msg_data), true, NULL, 0);
+       return 0;
 }
 
 static int
-hns3_reset_queue(struct hns3_adapter *hns, uint16_t queue_id)
+hns3vf_reset_all_tqps(struct hns3_hw *hw)
 {
-       struct hns3_hw *hw = &hns->hw;
-       if (hns->is_vf)
-               return hns3vf_reset_tqp(hw, queue_id);
-       else
-               return hns3_reset_tqp(hw, queue_id);
+#define HNS3VF_RESET_ALL_TQP_DONE      1U
+       uint8_t reset_status;
+       uint8_t msg_data[2];
+       int ret;
+       int i;
+
+       memset(msg_data, 0, sizeof(uint16_t));
+       ret = hns3_send_mbx_msg(hw, HNS3_MBX_QUEUE_RESET, 0, msg_data,
+                               sizeof(msg_data), true, &reset_status,
+                               sizeof(reset_status));
+       if (ret) {
+               hns3_err(hw, "fail to send rcb reset mbx, ret = %d.", ret);
+               return ret;
+       }
+
+       if (reset_status == HNS3VF_RESET_ALL_TQP_DONE)
+               return 0;
+
+       /*
+        * If the firmware version or kernel PF version is low, it may not
+        * support the rcb reset which means reset all the tqps at a time.
+        * In this case, we should reset tqps one by one.
+        */
+       for (i = 1; i < hw->cfg_max_queues; i++) {
+               ret = hns3vf_reset_tqp(hw, i);
+               if (ret)
+                       return ret;
+       }
+
+       return 0;
 }
 
 int
-hns3_reset_all_queues(struct hns3_adapter *hns)
+hns3_reset_all_tqps(struct hns3_adapter *hns)
 {
        struct hns3_hw *hw = &hns->hw;
        int ret, i;
 
+       /* Disable all queues before reset all queues */
        for (i = 0; i < hw->cfg_max_queues; i++) {
-               ret = hns3_reset_queue(hns, i);
+               ret = hns3_tqp_enable(hw, i, false);
                if (ret) {
-                       hns3_err(hw, "Failed to reset No.%d queue: %d", i, ret);
+                       hns3_err(hw,
+                           "fail to disable tqps before tqps reset, ret = %d.",
+                           ret);
                        return ret;
                }
        }
-       return 0;
+
+       if (hns->is_vf)
+               return hns3vf_reset_all_tqps(hw);
+       else
+               return hns3pf_reset_all_tqps(hw);
+}
+
+static int
+hns3_send_reset_queue_cmd(struct hns3_hw *hw, uint16_t queue_id,
+                         enum hns3_ring_type queue_type, bool enable)
+{
+       struct hns3_reset_tqp_queue_cmd *req;
+       struct hns3_cmd_desc desc;
+       int queue_direction;
+       int ret;
+
+       hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RESET_TQP_QUEUE_INDEP, false);
+
+       req = (struct hns3_reset_tqp_queue_cmd *)desc.data;
+       req->tqp_id = rte_cpu_to_le_16(queue_id);
+       queue_direction = queue_type == HNS3_RING_TYPE_TX ? 0 : 1;
+       req->queue_direction = rte_cpu_to_le_16(queue_direction);
+       hns3_set_bit(req->reset_req, HNS3_TQP_RESET_B, enable ? 1 : 0);
+
+       ret = hns3_cmd_send(hw, &desc, 1);
+       if (ret)
+               hns3_err(hw, "send queue reset cmd error, queue_id = %u, "
+                        "queue_type = %s, ret = %d.", queue_id,
+                        queue_type == HNS3_RING_TYPE_TX ? "Tx" : "Rx", ret);
+       return ret;
+}
+
+static int
+hns3_get_queue_reset_status(struct hns3_hw *hw, uint16_t queue_id,
+                           enum hns3_ring_type queue_type,
+                           uint8_t *reset_status)
+{
+       struct hns3_reset_tqp_queue_cmd *req;
+       struct hns3_cmd_desc desc;
+       int queue_direction;
+       int ret;
+
+       hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_RESET_TQP_QUEUE_INDEP, true);
+
+       req = (struct hns3_reset_tqp_queue_cmd *)desc.data;
+       req->tqp_id = rte_cpu_to_le_16(queue_id);
+       queue_direction = queue_type == HNS3_RING_TYPE_TX ? 0 : 1;
+       req->queue_direction = rte_cpu_to_le_16(queue_direction);
+
+       ret = hns3_cmd_send(hw, &desc, 1);
+       if (ret) {
+               hns3_err(hw, "get queue reset status error, queue_id = %u "
+                        "queue_type = %s, ret = %d.", queue_id,
+                        queue_type == HNS3_RING_TYPE_TX ? "Tx" : "Rx", ret);
+               return ret;
+       }
+
+       *reset_status = hns3_get_bit(req->ready_to_reset, HNS3_TQP_RESET_B);
+       return  ret;
+}
+
+static int
+hns3_reset_queue(struct hns3_hw *hw, uint16_t queue_id,
+                enum hns3_ring_type queue_type)
+{
+#define HNS3_QUEUE_RESET_TRY_MS        200
+       struct hns3_tx_queue *txq;
+       struct hns3_rx_queue *rxq;
+       uint32_t reset_wait_times;
+       uint32_t max_wait_times;
+       uint8_t reset_status;
+       int ret;
+
+       if (queue_type == HNS3_RING_TYPE_TX) {
+               txq = hw->data->tx_queues[queue_id];
+               hns3_enable_txq(txq, false);
+       } else {
+               rxq = hw->data->rx_queues[queue_id];
+               hns3_enable_rxq(rxq, false);
+       }
+
+       ret = hns3_send_reset_queue_cmd(hw, queue_id, queue_type, true);
+       if (ret) {
+               hns3_err(hw, "send reset queue cmd fail, ret = %d.", ret);
+               return ret;
+       }
+
+       reset_wait_times = 0;
+       max_wait_times = HNS3_QUEUE_RESET_TRY_MS / HNS3_POLL_RESPONE_MS;
+       while (reset_wait_times < max_wait_times) {
+               /* Wait for queue hw reset */
+               rte_delay_ms(HNS3_POLL_RESPONE_MS);
+               ret = hns3_get_queue_reset_status(hw, queue_id,
+                                               queue_type, &reset_status);
+               if (ret)
+                       goto queue_reset_fail;
+
+               if (reset_status)
+                       break;
+               reset_wait_times++;
+       }
+
+       if (!reset_status) {
+               hns3_err(hw, "reset queue timeout, queue_id = %u, "
+                            "queue_type = %s", queue_id,
+                            queue_type == HNS3_RING_TYPE_TX ? "Tx" : "Rx");
+               ret = -ETIMEDOUT;
+               goto queue_reset_fail;
+       }
+
+       ret = hns3_send_reset_queue_cmd(hw, queue_id, queue_type, false);
+       if (ret)
+               hns3_err(hw, "deassert queue reset fail, ret = %d.", ret);
+
+       return ret;
+
+queue_reset_fail:
+       hns3_send_reset_queue_cmd(hw, queue_id, queue_type, false);
+       return ret;
+}
+
+uint32_t
+hns3_get_tqp_intr_reg_offset(uint16_t tqp_intr_id)
+{
+       uint32_t reg_offset;
+
+       /* Need an extend offset to config queues > 64 */
+       if (tqp_intr_id < HNS3_MIN_EXT_TQP_INTR_ID)
+               reg_offset = HNS3_TQP_INTR_REG_BASE +
+                            tqp_intr_id * HNS3_TQP_INTR_LOW_ORDER_OFFSET;
+       else
+               reg_offset = HNS3_TQP_INTR_EXT_REG_BASE +
+                            tqp_intr_id / HNS3_MIN_EXT_TQP_INTR_ID *
+                            HNS3_TQP_INTR_HIGH_ORDER_OFFSET +
+                            tqp_intr_id % HNS3_MIN_EXT_TQP_INTR_ID *
+                            HNS3_TQP_INTR_LOW_ORDER_OFFSET;
+
+       return reg_offset;
 }
 
 void
@@ -544,7 +956,7 @@ hns3_set_queue_intr_gl(struct hns3_hw *hw, uint16_t queue_id,
        if (gl_idx >= RTE_DIM(offset) || gl_value > HNS3_TQP_INTR_GL_MAX)
                return;
 
-       addr = offset[gl_idx] + queue_id * HNS3_TQP_INTR_REG_SIZE;
+       addr = offset[gl_idx] + hns3_get_tqp_intr_reg_offset(queue_id);
        if (hw->intr.gl_unit == HNS3_INTR_COALESCE_GL_UINT_1US)
                value = gl_value | HNS3_TQP_INTR_GL_UNIT_1US;
        else
@@ -561,7 +973,7 @@ hns3_set_queue_intr_rl(struct hns3_hw *hw, uint16_t queue_id, uint16_t rl_value)
        if (rl_value > HNS3_TQP_INTR_RL_MAX)
                return;
 
-       addr = HNS3_TQP_INTR_RL_REG + queue_id * HNS3_TQP_INTR_REG_SIZE;
+       addr = HNS3_TQP_INTR_RL_REG + hns3_get_tqp_intr_reg_offset(queue_id);
        value = HNS3_RL_USEC_TO_REG(rl_value);
        if (value > 0)
                value |= HNS3_TQP_INTR_RL_ENABLE_MASK;
@@ -574,13 +986,18 @@ hns3_set_queue_intr_ql(struct hns3_hw *hw, uint16_t queue_id, uint16_t ql_value)
 {
        uint32_t addr;
 
-       if (hw->intr.coalesce_mode == HNS3_INTR_COALESCE_NON_QL)
+       /*
+        * int_ql_max == 0 means the hardware does not support QL,
+        * QL regs config is not permitted if QL is not supported,
+        * here just return.
+        */
+       if (hw->intr.int_ql_max == HNS3_INTR_QL_NONE)
                return;
 
-       addr = HNS3_TQP_INTR_TX_QL_REG + queue_id * HNS3_TQP_INTR_REG_SIZE;
+       addr = HNS3_TQP_INTR_TX_QL_REG + hns3_get_tqp_intr_reg_offset(queue_id);
        hns3_write_dev(hw, addr, ql_value);
 
-       addr = HNS3_TQP_INTR_RX_QL_REG + queue_id * HNS3_TQP_INTR_REG_SIZE;
+       addr = HNS3_TQP_INTR_RX_QL_REG + hns3_get_tqp_intr_reg_offset(queue_id);
        hns3_write_dev(hw, addr, ql_value);
 }
 
@@ -589,7 +1006,7 @@ hns3_queue_intr_enable(struct hns3_hw *hw, uint16_t queue_id, bool en)
 {
        uint32_t addr, value;
 
-       addr = HNS3_TQP_INTR_CTRL_REG + queue_id * HNS3_TQP_INTR_REG_SIZE;
+       addr = HNS3_TQP_INTR_CTRL_REG + hns3_get_tqp_intr_reg_offset(queue_id);
        value = en ? 1 : 0;
 
        hns3_write_dev(hw, addr, value);
@@ -644,7 +1061,7 @@ hns3_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 }
 
 static int
-hns3_dev_rx_queue_start(struct hns3_adapter *hns, uint16_t idx)
+hns3_init_rxq(struct hns3_adapter *hns, uint16_t idx)
 {
        struct hns3_hw *hw = &hns->hw;
        struct hns3_rx_queue *rxq;
@@ -655,22 +1072,25 @@ hns3_dev_rx_queue_start(struct hns3_adapter *hns, uint16_t idx)
        rxq = (struct hns3_rx_queue *)hw->data->rx_queues[idx];
        ret = hns3_alloc_rx_queue_mbufs(hw, rxq);
        if (ret) {
-               hns3_err(hw, "Failed to alloc mbuf for No.%d rx queue: %d",
+               hns3_err(hw, "fail to alloc mbuf for Rx queue %u, ret = %d.",
                         idx, ret);
                return ret;
        }
 
        rxq->next_to_use = 0;
+       rxq->rx_rearm_start = 0;
        rxq->rx_free_hold = 0;
+       rxq->rx_rearm_nb = 0;
        rxq->pkt_first_seg = NULL;
        rxq->pkt_last_seg = NULL;
        hns3_init_rx_queue_hw(rxq);
+       hns3_rxq_vec_setup(rxq);
 
        return 0;
 }
 
 static void
-hns3_fake_rx_queue_start(struct hns3_adapter *hns, uint16_t idx)
+hns3_init_fake_rxq(struct hns3_adapter *hns, uint16_t idx)
 {
        struct hns3_hw *hw = &hns->hw;
        struct hns3_rx_queue *rxq;
@@ -678,13 +1098,14 @@ hns3_fake_rx_queue_start(struct hns3_adapter *hns, uint16_t idx)
        rxq = (struct hns3_rx_queue *)hw->fkq_data.rx_queues[idx];
        rxq->next_to_use = 0;
        rxq->rx_free_hold = 0;
+       rxq->rx_rearm_start = 0;
+       rxq->rx_rearm_nb = 0;
        hns3_init_rx_queue_hw(rxq);
 }
 
 static void
-hns3_init_tx_queue(struct hns3_tx_queue *queue)
+hns3_init_txq(struct hns3_tx_queue *txq)
 {
-       struct hns3_tx_queue *txq = queue;
        struct hns3_desc *desc;
        int i;
 
@@ -701,26 +1122,6 @@ hns3_init_tx_queue(struct hns3_tx_queue *queue)
        hns3_init_tx_queue_hw(txq);
 }
 
-static void
-hns3_dev_tx_queue_start(struct hns3_adapter *hns, uint16_t idx)
-{
-       struct hns3_hw *hw = &hns->hw;
-       struct hns3_tx_queue *txq;
-
-       txq = (struct hns3_tx_queue *)hw->data->tx_queues[idx];
-       hns3_init_tx_queue(txq);
-}
-
-static void
-hns3_fake_tx_queue_start(struct hns3_adapter *hns, uint16_t idx)
-{
-       struct hns3_hw *hw = &hns->hw;
-       struct hns3_tx_queue *txq;
-
-       txq = (struct hns3_tx_queue *)hw->fkq_data.tx_queues[idx];
-       hns3_init_tx_queue(txq);
-}
-
 static void
 hns3_init_tx_ring_tc(struct hns3_adapter *hns)
 {
@@ -747,38 +1148,41 @@ hns3_init_tx_ring_tc(struct hns3_adapter *hns)
 }
 
 static int
-hns3_start_rx_queues(struct hns3_adapter *hns)
+hns3_init_rx_queues(struct hns3_adapter *hns)
 {
        struct hns3_hw *hw = &hns->hw;
        struct hns3_rx_queue *rxq;
-       int i, j;
+       uint16_t i, j;
        int ret;
 
        /* Initialize RSS for queues */
        ret = hns3_config_rss(hns);
        if (ret) {
-               hns3_err(hw, "Failed to configure rss %d", ret);
+               hns3_err(hw, "failed to configure rss, ret = %d.", ret);
                return ret;
        }
 
        for (i = 0; i < hw->data->nb_rx_queues; i++) {
                rxq = (struct hns3_rx_queue *)hw->data->rx_queues[i];
-               if (rxq == NULL || rxq->rx_deferred_start)
-                       continue;
-               ret = hns3_dev_rx_queue_start(hns, i);
-               if (ret) {
-                       hns3_err(hw, "Failed to start No.%d rx queue: %d", i,
-                                ret);
+               if (!rxq) {
+                       hns3_err(hw, "Rx queue %u not available or setup.", i);
                        goto out;
                }
-       }
 
-       for (i = 0; i < hw->fkq_data.nb_fake_rx_queues; i++) {
-               rxq = (struct hns3_rx_queue *)hw->fkq_data.rx_queues[i];
-               if (rxq == NULL || rxq->rx_deferred_start)
+               if (rxq->rx_deferred_start)
                        continue;
-               hns3_fake_rx_queue_start(hns, i);
+
+               ret = hns3_init_rxq(hns, i);
+               if (ret) {
+                       hns3_err(hw, "failed to init Rx queue %u, ret = %d.", i,
+                                ret);
+                       goto out;
+               }
        }
+
+       for (i = 0; i < hw->fkq_data.nb_fake_rx_queues; i++)
+               hns3_init_fake_rxq(hns, i);
+
        return 0;
 
 out:
@@ -790,73 +1194,137 @@ out:
        return ret;
 }
 
-static void
-hns3_start_tx_queues(struct hns3_adapter *hns)
+static int
+hns3_init_tx_queues(struct hns3_adapter *hns)
 {
        struct hns3_hw *hw = &hns->hw;
        struct hns3_tx_queue *txq;
-       int i;
+       uint16_t i;
 
        for (i = 0; i < hw->data->nb_tx_queues; i++) {
                txq = (struct hns3_tx_queue *)hw->data->tx_queues[i];
-               if (txq == NULL || txq->tx_deferred_start)
+               if (!txq) {
+                       hns3_err(hw, "Tx queue %u not available or setup.", i);
+                       return -EINVAL;
+               }
+
+               if (txq->tx_deferred_start)
                        continue;
-               hns3_dev_tx_queue_start(hns, i);
+               hns3_init_txq(txq);
        }
 
        for (i = 0; i < hw->fkq_data.nb_fake_tx_queues; i++) {
                txq = (struct hns3_tx_queue *)hw->fkq_data.tx_queues[i];
-               if (txq == NULL || txq->tx_deferred_start)
-                       continue;
-               hns3_fake_tx_queue_start(hns, i);
+               hns3_init_txq(txq);
        }
-
        hns3_init_tx_ring_tc(hns);
+
+       return 0;
 }
 
 /*
- * Start all queues.
- * Note: just init and setup queues, and don't enable queue rx&tx.
+ * Init all queues.
+ * Note: just init and setup queues, and don't enable tqps.
  */
 int
-hns3_start_queues(struct hns3_adapter *hns, bool reset_queue)
+hns3_init_queues(struct hns3_adapter *hns, bool reset_queue)
 {
        struct hns3_hw *hw = &hns->hw;
        int ret;
 
        if (reset_queue) {
-               ret = hns3_reset_all_queues(hns);
+               ret = hns3_reset_all_tqps(hns);
                if (ret) {
-                       hns3_err(hw, "Failed to reset all queues %d", ret);
+                       hns3_err(hw, "failed to reset all queues, ret = %d.",
+                                ret);
                        return ret;
                }
        }
 
-       ret = hns3_start_rx_queues(hns);
+       ret = hns3_init_rx_queues(hns);
        if (ret) {
-               hns3_err(hw, "Failed to start rx queues: %d", ret);
+               hns3_err(hw, "failed to init rx queues, ret = %d.", ret);
                return ret;
        }
 
-       hns3_start_tx_queues(hns);
+       ret = hns3_init_tx_queues(hns);
+       if (ret) {
+               hns3_dev_release_mbufs(hns);
+               hns3_err(hw, "failed to init tx queues, ret = %d.", ret);
+       }
 
-       return 0;
+       return ret;
+}
+
+void
+hns3_start_tqps(struct hns3_hw *hw)
+{
+       struct hns3_tx_queue *txq;
+       struct hns3_rx_queue *rxq;
+       uint16_t i;
+
+       hns3_enable_all_queues(hw, true);
+
+       for (i = 0; i < hw->data->nb_tx_queues; i++) {
+               txq = hw->data->tx_queues[i];
+               if (txq->enabled)
+                       hw->data->tx_queue_state[i] =
+                               RTE_ETH_QUEUE_STATE_STARTED;
+       }
+
+       for (i = 0; i < hw->data->nb_rx_queues; i++) {
+               rxq = hw->data->rx_queues[i];
+               if (rxq->enabled)
+                       hw->data->rx_queue_state[i] =
+                               RTE_ETH_QUEUE_STATE_STARTED;
+       }
+}
+
+void
+hns3_stop_tqps(struct hns3_hw *hw)
+{
+       uint16_t i;
+
+       hns3_enable_all_queues(hw, false);
+
+       for (i = 0; i < hw->data->nb_tx_queues; i++)
+               hw->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
+
+       for (i = 0; i < hw->data->nb_rx_queues; i++)
+               hw->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 }
 
+/*
+ * Iterate over all Rx Queue, and call the callback() function for each Rx
+ * queue.
+ *
+ * @param[in] dev
+ *   The target eth dev.
+ * @param[in] callback
+ *   The function to call for each queue.
+ *   if callback function return nonzero will stop iterate and return it's value
+ * @param[in] arg
+ *   The arguments to provide the callback function with.
+ *
+ * @return
+ *   0 on success, otherwise with errno set.
+ */
 int
-hns3_stop_queues(struct hns3_adapter *hns, bool reset_queue)
+hns3_rxq_iterate(struct rte_eth_dev *dev,
+                int (*callback)(struct hns3_rx_queue *, void *), void *arg)
 {
-       struct hns3_hw *hw = &hns->hw;
+       uint32_t i;
        int ret;
 
-       hns3_enable_all_queues(hw, false);
-       if (reset_queue) {
-               ret = hns3_reset_all_queues(hns);
-               if (ret) {
-                       hns3_err(hw, "Failed to reset all queues %d", ret);
+       if (dev->data->rx_queues == NULL)
+               return -EINVAL;
+
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               ret = callback(dev->data->rx_queues[i], arg);
+               if (ret != 0)
                        return ret;
-               }
        }
+
        return 0;
 }
 
@@ -872,7 +1340,7 @@ hns3_alloc_rxq_and_dma_zone(struct rte_eth_dev *dev,
        rxq = rte_zmalloc_socket(q_info->type, sizeof(struct hns3_rx_queue),
                                 RTE_CACHE_LINE_SIZE, q_info->socket_id);
        if (rxq == NULL) {
-               hns3_err(hw, "Failed to allocate memory for No.%d rx ring!",
+               hns3_err(hw, "Failed to allocate memory for No.%u rx ring!",
                         q_info->idx);
                return NULL;
        }
@@ -880,12 +1348,18 @@ hns3_alloc_rxq_and_dma_zone(struct rte_eth_dev *dev,
        /* Allocate rx ring hardware descriptors. */
        rxq->queue_id = q_info->idx;
        rxq->nb_rx_desc = q_info->nb_desc;
-       rx_desc = rxq->nb_rx_desc * sizeof(struct hns3_desc);
+
+       /*
+        * Allocate a litter more memory because rx vector functions
+        * don't check boundaries each time.
+        */
+       rx_desc = (rxq->nb_rx_desc + HNS3_DEFAULT_RX_BURST) *
+                       sizeof(struct hns3_desc);
        rx_mz = rte_eth_dma_zone_reserve(dev, q_info->ring_name, q_info->idx,
                                         rx_desc, HNS3_RING_BASE_ALIGN,
                                         q_info->socket_id);
        if (rx_mz == NULL) {
-               hns3_err(hw, "Failed to reserve DMA memory for No.%d rx ring!",
+               hns3_err(hw, "Failed to reserve DMA memory for No.%u rx ring!",
                         q_info->idx);
                hns3_rx_queue_release(rxq);
                return NULL;
@@ -894,7 +1368,7 @@ hns3_alloc_rxq_and_dma_zone(struct rte_eth_dev *dev,
        rxq->rx_ring = (struct hns3_desc *)rx_mz->addr;
        rxq->rx_ring_phys_addr = rx_mz->iova;
 
-       hns3_dbg(hw, "No.%d rx descriptors iova 0x%" PRIx64, q_info->idx,
+       hns3_dbg(hw, "No.%u rx descriptors iova 0x%" PRIx64, q_info->idx,
                 rxq->rx_ring_phys_addr);
 
        return rxq;
@@ -922,7 +1396,7 @@ hns3_fake_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx,
        q_info.ring_name = "rx_fake_ring";
        rxq = hns3_alloc_rxq_and_dma_zone(dev, &q_info);
        if (rxq == NULL) {
-               hns3_err(hw, "Failed to setup No.%d fake rx ring.", idx);
+               hns3_err(hw, "Failed to setup No.%u fake rx ring.", idx);
                return -ENOMEM;
        }
 
@@ -959,7 +1433,7 @@ hns3_alloc_txq_and_dma_zone(struct rte_eth_dev *dev,
        txq = rte_zmalloc_socket(q_info->type, sizeof(struct hns3_tx_queue),
                                 RTE_CACHE_LINE_SIZE, q_info->socket_id);
        if (txq == NULL) {
-               hns3_err(hw, "Failed to allocate memory for No.%d tx ring!",
+               hns3_err(hw, "Failed to allocate memory for No.%u tx ring!",
                         q_info->idx);
                return NULL;
        }
@@ -972,7 +1446,7 @@ hns3_alloc_txq_and_dma_zone(struct rte_eth_dev *dev,
                                         tx_desc, HNS3_RING_BASE_ALIGN,
                                         q_info->socket_id);
        if (tx_mz == NULL) {
-               hns3_err(hw, "Failed to reserve DMA memory for No.%d tx ring!",
+               hns3_err(hw, "Failed to reserve DMA memory for No.%u tx ring!",
                         q_info->idx);
                hns3_tx_queue_release(txq);
                return NULL;
@@ -981,7 +1455,7 @@ hns3_alloc_txq_and_dma_zone(struct rte_eth_dev *dev,
        txq->tx_ring = (struct hns3_desc *)tx_mz->addr;
        txq->tx_ring_phys_addr = tx_mz->iova;
 
-       hns3_dbg(hw, "No.%d tx descriptors iova 0x%" PRIx64, q_info->idx,
+       hns3_dbg(hw, "No.%u tx descriptors iova 0x%" PRIx64, q_info->idx,
                 txq->tx_ring_phys_addr);
 
        /* Clear tx bd */
@@ -1016,7 +1490,7 @@ hns3_fake_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx,
        q_info.ring_name = "tx_fake_ring";
        txq = hns3_alloc_txq_and_dma_zone(dev, &q_info);
        if (txq == NULL) {
-               hns3_err(hw, "Failed to setup No.%d fake tx ring.", idx);
+               hns3_err(hw, "Failed to setup No.%u fake tx ring.", idx);
                return -ENOMEM;
        }
 
@@ -1044,7 +1518,7 @@ hns3_fake_rx_queue_config(struct hns3_hw *hw, uint16_t nb_queues)
 {
        uint16_t old_nb_queues = hw->fkq_data.nb_fake_rx_queues;
        void **rxq;
-       uint8_t i;
+       uint16_t i;
 
        if (hw->fkq_data.rx_queues == NULL && nb_queues != 0) {
                /* first time configuration */
@@ -1091,7 +1565,7 @@ hns3_fake_tx_queue_config(struct hns3_hw *hw, uint16_t nb_queues)
 {
        uint16_t old_nb_queues = hw->fkq_data.nb_fake_tx_queues;
        void **txq;
-       uint8_t i;
+       uint16_t i;
 
        if (hw->fkq_data.tx_queues == NULL && nb_queues != 0) {
                /* first time configuration */
@@ -1143,13 +1617,12 @@ hns3_set_fake_rx_or_tx_queues(struct rte_eth_dev *dev, uint16_t nb_rx_q,
        int ret;
 
        /* Setup new number of fake RX/TX queues and reconfigure device. */
-       hw->cfg_max_queues = RTE_MAX(nb_rx_q, nb_tx_q);
        rx_need_add_nb_q = hw->cfg_max_queues - nb_rx_q;
        tx_need_add_nb_q = hw->cfg_max_queues - nb_tx_q;
        ret = hns3_fake_rx_queue_config(hw, rx_need_add_nb_q);
        if (ret) {
                hns3_err(hw, "Fail to configure fake rx queues: %d", ret);
-               goto cfg_fake_rx_q_fail;
+               return ret;
        }
 
        ret = hns3_fake_tx_queue_config(hw, tx_need_add_nb_q);
@@ -1182,8 +1655,6 @@ setup_fake_rx_q_fail:
        (void)hns3_fake_tx_queue_config(hw, 0);
 cfg_fake_tx_q_fail:
        (void)hns3_fake_rx_queue_config(hw, 0);
-cfg_fake_rx_q_fail:
-       hw->cfg_max_queues = 0;
 
        return ret;
 }
@@ -1199,7 +1670,7 @@ hns3_dev_release_mbufs(struct hns3_adapter *hns)
        if (dev_data->rx_queues)
                for (i = 0; i < dev_data->nb_rx_queues; i++) {
                        rxq = dev_data->rx_queues[i];
-                       if (rxq == NULL || rxq->rx_deferred_start)
+                       if (rxq == NULL)
                                continue;
                        hns3_rx_queue_release_mbufs(rxq);
                }
@@ -1207,7 +1678,7 @@ hns3_dev_release_mbufs(struct hns3_adapter *hns)
        if (dev_data->tx_queues)
                for (i = 0; i < dev_data->nb_tx_queues; i++) {
                        txq = dev_data->tx_queues[i];
-                       if (txq == NULL || txq->tx_deferred_start)
+                       if (txq == NULL)
                                continue;
                        hns3_tx_queue_release_mbufs(txq);
                }
@@ -1241,7 +1712,6 @@ hns3_rx_buf_len_calc(struct rte_mempool *mp, uint16_t *rx_buf_len)
 
        vld_buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
                        RTE_PKTMBUF_HEADROOM);
-
        if (vld_buf_size < HNS3_MIN_BD_BUF_SIZE)
                return -EINVAL;
 
@@ -1255,11 +1725,50 @@ hns3_rx_buf_len_calc(struct rte_mempool *mp, uint16_t *rx_buf_len)
        return 0;
 }
 
+static int
+hns3_rxq_conf_runtime_check(struct hns3_hw *hw, uint16_t buf_size,
+                               uint16_t nb_desc)
+{
+       struct rte_eth_dev *dev = &rte_eth_devices[hw->data->port_id];
+       struct rte_eth_rxmode *rxmode = &hw->data->dev_conf.rxmode;
+       eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
+       uint16_t min_vec_bds;
+
+       /*
+        * HNS3 hardware network engine set scattered as default. If the driver
+        * is not work in scattered mode and the pkts greater than buf_size
+        * but smaller than max_rx_pkt_len will be distributed to multiple BDs.
+        * Driver cannot handle this situation.
+        */
+       if (!hw->data->scattered_rx && rxmode->max_rx_pkt_len > buf_size) {
+               hns3_err(hw, "max_rx_pkt_len is not allowed to be set greater "
+                            "than rx_buf_len if scattered is off.");
+               return -EINVAL;
+       }
+
+       if (pkt_burst == hns3_recv_pkts_vec) {
+               min_vec_bds = HNS3_DEFAULT_RXQ_REARM_THRESH +
+                             HNS3_DEFAULT_RX_BURST;
+               if (nb_desc < min_vec_bds ||
+                   nb_desc % HNS3_DEFAULT_RXQ_REARM_THRESH) {
+                       hns3_err(hw, "if Rx burst mode is vector, "
+                                "number of descriptor is required to be "
+                                "bigger than min vector bds:%u, and could be "
+                                "divided by rxq rearm thresh:%u.",
+                                min_vec_bds, HNS3_DEFAULT_RXQ_REARM_THRESH);
+                       return -EINVAL;
+               }
+       }
+       return 0;
+}
+
 static int
 hns3_rx_queue_conf_check(struct hns3_hw *hw, const struct rte_eth_rxconf *conf,
                         struct rte_mempool *mp, uint16_t nb_desc,
                         uint16_t *buf_size)
 {
+       int ret;
+
        if (nb_desc > HNS3_MAX_RING_DESC || nb_desc < HNS3_MIN_RING_DESC ||
            nb_desc % HNS3_ALIGN_RING_DESC) {
                hns3_err(hw, "Number (%u) of rx descriptors is invalid",
@@ -1279,9 +1788,33 @@ hns3_rx_queue_conf_check(struct hns3_hw *hw, const struct rte_eth_rxconf *conf,
                return -EINVAL;
        }
 
+       if (hw->data->dev_started) {
+               ret = hns3_rxq_conf_runtime_check(hw, *buf_size, nb_desc);
+               if (ret) {
+                       hns3_err(hw, "Rx queue runtime setup fail.");
+                       return ret;
+               }
+       }
+
        return 0;
 }
 
+uint32_t
+hns3_get_tqp_reg_offset(uint16_t queue_id)
+{
+       uint32_t reg_offset;
+
+       /* Need an extend offset to config queue > 1024 */
+       if (queue_id < HNS3_MIN_EXTEND_QUEUE_ID)
+               reg_offset = HNS3_TQP_REG_OFFSET + queue_id * HNS3_TQP_REG_SIZE;
+       else
+               reg_offset = HNS3_TQP_REG_OFFSET + HNS3_TQP_EXT_REG_OFFSET +
+                            (queue_id - HNS3_MIN_EXTEND_QUEUE_ID) *
+                            HNS3_TQP_REG_SIZE;
+
+       return reg_offset;
+}
+
 int
 hns3_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t nb_desc,
                    unsigned int socket_id, const struct rte_eth_rxconf *conf,
@@ -1295,11 +1828,6 @@ hns3_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t nb_desc,
        int rx_entry_len;
        int ret;
 
-       if (dev->data->dev_started) {
-               hns3_err(hw, "rx_queue_setup after dev_start no supported");
-               return -EINVAL;
-       }
-
        ret = hns3_rx_queue_conf_check(hw, conf, mp, nb_desc, &rx_buf_size);
        if (ret)
                return ret;
@@ -1327,9 +1855,15 @@ hns3_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t nb_desc,
        rxq->mb_pool = mp;
        rxq->rx_free_thresh = (conf->rx_free_thresh > 0) ?
                conf->rx_free_thresh : HNS3_DEFAULT_RX_FREE_THRESH;
+
        rxq->rx_deferred_start = conf->rx_deferred_start;
+       if (rxq->rx_deferred_start && !hns3_dev_indep_txrx_supported(hw)) {
+               hns3_warn(hw, "deferred start is not supported.");
+               rxq->rx_deferred_start = false;
+       }
 
-       rx_entry_len = sizeof(struct hns3_entry) * rxq->nb_rx_desc;
+       rx_entry_len = (rxq->nb_rx_desc + HNS3_DEFAULT_RX_BURST) *
+                       sizeof(struct hns3_entry);
        rxq->sw_ring = rte_zmalloc_socket("hns3 RX sw ring", rx_entry_len,
                                          RTE_CACHE_LINE_SIZE, socket_id);
        if (rxq->sw_ring == NULL) {
@@ -1340,22 +1874,37 @@ hns3_rx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t nb_desc,
 
        rxq->next_to_use = 0;
        rxq->rx_free_hold = 0;
+       rxq->rx_rearm_start = 0;
+       rxq->rx_rearm_nb = 0;
        rxq->pkt_first_seg = NULL;
        rxq->pkt_last_seg = NULL;
        rxq->port_id = dev->data->port_id;
-       rxq->pvid_state = hw->port_base_vlan_cfg.state;
+       /*
+        * For hns3 PF device, if the VLAN mode is HW_SHIFT_AND_DISCARD_MODE,
+        * the pvid_sw_discard_en in the queue struct should not be changed,
+        * because PVID-related operations do not need to be processed by PMD
+        * driver. For hns3 VF device, whether it needs to process PVID depends
+        * on the configuration of PF kernel mode netdevice driver. And the
+        * related PF configuration is delivered through the mailbox and finally
+        * reflectd in port_base_vlan_cfg.
+        */
+       if (hns->is_vf || hw->vlan_mode == HNS3_SW_SHIFT_AND_DISCARD_MODE)
+               rxq->pvid_sw_discard_en = hw->port_base_vlan_cfg.state ==
+                                      HNS3_PORT_BASE_VLAN_ENABLE;
+       else
+               rxq->pvid_sw_discard_en = false;
+       rxq->ptype_en = hns3_dev_rxd_adv_layout_supported(hw) ? true : false;
        rxq->configured = true;
        rxq->io_base = (void *)((char *)hw->io_base + HNS3_TQP_REG_OFFSET +
                                idx * HNS3_TQP_REG_SIZE);
+       rxq->io_base = (void *)((char *)hw->io_base +
+                                       hns3_get_tqp_reg_offset(idx));
        rxq->io_head_reg = (volatile void *)((char *)rxq->io_base +
                           HNS3_RING_RX_HEAD_REG);
        rxq->rx_buf_len = rx_buf_size;
-       rxq->l2_errors = 0;
-       rxq->pkt_len_errors = 0;
-       rxq->l3_csum_errors = 0;
-       rxq->l4_csum_errors = 0;
-       rxq->ol3_csum_errors = 0;
-       rxq->ol4_csum_errors = 0;
+       memset(&rxq->basic_stats, 0, sizeof(struct hns3_rx_basic_stats));
+       memset(&rxq->err_stats, 0, sizeof(struct hns3_rx_bd_errors_stats));
+       memset(&rxq->dfx_stats, 0, sizeof(struct hns3_rx_dfx_stats));
 
        /* CRC len set here is used for amending packet length */
        if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC)
@@ -1413,8 +1962,6 @@ hns3_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 {
        static const uint32_t ptypes[] = {
                RTE_PTYPE_L2_ETHER,
-               RTE_PTYPE_L2_ETHER_VLAN,
-               RTE_PTYPE_L2_ETHER_QINQ,
                RTE_PTYPE_L2_ETHER_LLDP,
                RTE_PTYPE_L2_ETHER_ARP,
                RTE_PTYPE_L3_IPV4,
@@ -1427,35 +1974,70 @@ hns3_dev_supported_ptypes_get(struct rte_eth_dev *dev)
                RTE_PTYPE_L4_TCP,
                RTE_PTYPE_L4_UDP,
                RTE_PTYPE_TUNNEL_GRE,
+               RTE_PTYPE_INNER_L2_ETHER,
+               RTE_PTYPE_INNER_L3_IPV4,
+               RTE_PTYPE_INNER_L3_IPV6,
+               RTE_PTYPE_INNER_L3_IPV4_EXT,
+               RTE_PTYPE_INNER_L3_IPV6_EXT,
+               RTE_PTYPE_INNER_L4_UDP,
+               RTE_PTYPE_INNER_L4_TCP,
+               RTE_PTYPE_INNER_L4_SCTP,
+               RTE_PTYPE_INNER_L4_ICMP,
+               RTE_PTYPE_TUNNEL_VXLAN,
+               RTE_PTYPE_TUNNEL_NVGRE,
+               RTE_PTYPE_UNKNOWN
+       };
+       static const uint32_t adv_layout_ptypes[] = {
+               RTE_PTYPE_L2_ETHER,
+               RTE_PTYPE_L2_ETHER_TIMESYNC,
+               RTE_PTYPE_L2_ETHER_LLDP,
+               RTE_PTYPE_L2_ETHER_ARP,
+               RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
+               RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
+               RTE_PTYPE_L4_FRAG,
+               RTE_PTYPE_L4_NONFRAG,
+               RTE_PTYPE_L4_UDP,
+               RTE_PTYPE_L4_TCP,
+               RTE_PTYPE_L4_SCTP,
+               RTE_PTYPE_L4_IGMP,
+               RTE_PTYPE_L4_ICMP,
+               RTE_PTYPE_TUNNEL_GRE,
+               RTE_PTYPE_TUNNEL_GRENAT,
+               RTE_PTYPE_INNER_L2_ETHER,
+               RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
+               RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
+               RTE_PTYPE_INNER_L4_FRAG,
+               RTE_PTYPE_INNER_L4_ICMP,
+               RTE_PTYPE_INNER_L4_NONFRAG,
+               RTE_PTYPE_INNER_L4_UDP,
+               RTE_PTYPE_INNER_L4_TCP,
+               RTE_PTYPE_INNER_L4_SCTP,
+               RTE_PTYPE_INNER_L4_ICMP,
                RTE_PTYPE_UNKNOWN
        };
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
-       if (dev->rx_pkt_burst == hns3_recv_pkts ||
-           dev->rx_pkt_burst == hns3_recv_scattered_pkts)
-               return ptypes;
+       if (dev->rx_pkt_burst == hns3_recv_pkts_simple ||
+           dev->rx_pkt_burst == hns3_recv_scattered_pkts ||
+           dev->rx_pkt_burst == hns3_recv_pkts_vec ||
+           dev->rx_pkt_burst == hns3_recv_pkts_vec_sve) {
+               if (hns3_dev_rxd_adv_layout_supported(hw))
+                       return adv_layout_ptypes;
+               else
+                       return ptypes;
+       }
 
        return NULL;
 }
 
-void
-hns3_init_rx_ptype_tble(struct rte_eth_dev *dev)
+static void
+hns3_init_non_tunnel_ptype_tbl(struct hns3_ptype_table *tbl)
 {
-       struct hns3_adapter *hns = dev->data->dev_private;
-       struct hns3_ptype_table *tbl = &hns->ptype_tbl;
-
-       memset(tbl, 0, sizeof(*tbl));
-
-       tbl->l2table[0] = RTE_PTYPE_L2_ETHER;
-       tbl->l2table[1] = RTE_PTYPE_L2_ETHER_QINQ;
-       tbl->l2table[2] = RTE_PTYPE_L2_ETHER_VLAN;
-       tbl->l2table[3] = RTE_PTYPE_L2_ETHER_VLAN;
-
-       tbl->l3table[0] = RTE_PTYPE_L3_IPV4;
-       tbl->l3table[1] = RTE_PTYPE_L3_IPV6;
+       tbl->l3table[0] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4;
+       tbl->l3table[1] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6;
        tbl->l3table[2] = RTE_PTYPE_L2_ETHER_ARP;
-       tbl->l3table[3] = RTE_PTYPE_L2_ETHER;
-       tbl->l3table[4] = RTE_PTYPE_L3_IPV4_EXT;
-       tbl->l3table[5] = RTE_PTYPE_L3_IPV6_EXT;
+       tbl->l3table[4] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT;
+       tbl->l3table[5] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT;
        tbl->l3table[6] = RTE_PTYPE_L2_ETHER_LLDP;
 
        tbl->l4table[0] = RTE_PTYPE_L4_UDP;
@@ -1464,37 +2046,244 @@ hns3_init_rx_ptype_tble(struct rte_eth_dev *dev)
        tbl->l4table[3] = RTE_PTYPE_L4_SCTP;
        tbl->l4table[4] = RTE_PTYPE_L4_IGMP;
        tbl->l4table[5] = RTE_PTYPE_L4_ICMP;
+}
 
-       tbl->inner_l2table[0] = RTE_PTYPE_INNER_L2_ETHER;
-       tbl->inner_l2table[1] = RTE_PTYPE_INNER_L2_ETHER_VLAN;
-       tbl->inner_l2table[2] = RTE_PTYPE_INNER_L2_ETHER_QINQ;
-
-       tbl->inner_l3table[0] = RTE_PTYPE_INNER_L3_IPV4;
-       tbl->inner_l3table[1] = RTE_PTYPE_INNER_L3_IPV6;
-       tbl->inner_l3table[2] = 0;
-       tbl->inner_l3table[3] = RTE_PTYPE_INNER_L2_ETHER;
-       tbl->inner_l3table[4] = RTE_PTYPE_INNER_L3_IPV4_EXT;
-       tbl->inner_l3table[5] = RTE_PTYPE_INNER_L3_IPV6_EXT;
+static void
+hns3_init_tunnel_ptype_tbl(struct hns3_ptype_table *tbl)
+{
+       tbl->inner_l3table[0] = RTE_PTYPE_INNER_L2_ETHER |
+                               RTE_PTYPE_INNER_L3_IPV4;
+       tbl->inner_l3table[1] = RTE_PTYPE_INNER_L2_ETHER |
+                               RTE_PTYPE_INNER_L3_IPV6;
+       /* There is not a ptype for inner ARP/RARP */
+       tbl->inner_l3table[2] = RTE_PTYPE_UNKNOWN;
+       tbl->inner_l3table[3] = RTE_PTYPE_UNKNOWN;
+       tbl->inner_l3table[4] = RTE_PTYPE_INNER_L2_ETHER |
+                               RTE_PTYPE_INNER_L3_IPV4_EXT;
+       tbl->inner_l3table[5] = RTE_PTYPE_INNER_L2_ETHER |
+                               RTE_PTYPE_INNER_L3_IPV6_EXT;
 
        tbl->inner_l4table[0] = RTE_PTYPE_INNER_L4_UDP;
        tbl->inner_l4table[1] = RTE_PTYPE_INNER_L4_TCP;
-       tbl->inner_l4table[2] = RTE_PTYPE_TUNNEL_GRE;
+       /* There is not a ptype for inner GRE */
+       tbl->inner_l4table[2] = RTE_PTYPE_UNKNOWN;
        tbl->inner_l4table[3] = RTE_PTYPE_INNER_L4_SCTP;
-       tbl->inner_l4table[4] = RTE_PTYPE_L4_IGMP;
+       /* There is not a ptype for inner IGMP */
+       tbl->inner_l4table[4] = RTE_PTYPE_UNKNOWN;
        tbl->inner_l4table[5] = RTE_PTYPE_INNER_L4_ICMP;
 
-       tbl->ol3table[0] = RTE_PTYPE_L3_IPV4;
-       tbl->ol3table[1] = RTE_PTYPE_L3_IPV6;
-       tbl->ol3table[2] = 0;
-       tbl->ol3table[3] = 0;
-       tbl->ol3table[4] = RTE_PTYPE_L3_IPV4_EXT;
-       tbl->ol3table[5] = RTE_PTYPE_L3_IPV6_EXT;
+       tbl->ol3table[0] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4;
+       tbl->ol3table[1] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6;
+       tbl->ol3table[2] = RTE_PTYPE_UNKNOWN;
+       tbl->ol3table[3] = RTE_PTYPE_UNKNOWN;
+       tbl->ol3table[4] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT;
+       tbl->ol3table[5] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT;
 
-       tbl->ol4table[0] = 0;
-       tbl->ol4table[1] = RTE_PTYPE_TUNNEL_VXLAN;
+       tbl->ol4table[0] = RTE_PTYPE_UNKNOWN;
+       tbl->ol4table[1] = RTE_PTYPE_L4_UDP | RTE_PTYPE_TUNNEL_VXLAN;
        tbl->ol4table[2] = RTE_PTYPE_TUNNEL_NVGRE;
 }
 
+static void
+hns3_init_adv_layout_ptype(struct hns3_ptype_table *tbl)
+{
+       uint32_t *ptype = tbl->ptype;
+
+       /* Non-tunnel L2 */
+       ptype[1] = RTE_PTYPE_L2_ETHER_ARP;
+       ptype[3] = RTE_PTYPE_L2_ETHER_LLDP;
+       ptype[8] = RTE_PTYPE_L2_ETHER_TIMESYNC;
+
+       /* Non-tunnel IPv4 */
+       ptype[17] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_L4_FRAG;
+       ptype[18] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_L4_NONFRAG;
+       ptype[19] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_L4_UDP;
+       ptype[20] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_L4_TCP;
+       ptype[21] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRE;
+       ptype[22] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_L4_SCTP;
+       ptype[23] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_L4_IGMP;
+       ptype[24] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_L4_ICMP;
+       /* The next ptype is PTP over IPv4 + UDP */
+       ptype[25] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_L4_UDP;
+
+       /* IPv4 --> GRE/Teredo/VXLAN */
+       ptype[29] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT;
+       /* IPv4 --> GRE/Teredo/VXLAN --> MAC */
+       ptype[30] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER;
+
+       /* IPv4 --> GRE/Teredo/VXLAN --> MAC --> IPv4 */
+       ptype[31] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_FRAG;
+       ptype[32] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_NONFRAG;
+       ptype[33] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_UDP;
+       ptype[34] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_TCP;
+       ptype[35] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_SCTP;
+       /* The next ptype's inner L4 is IGMP */
+       ptype[36] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN;
+       ptype[37] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_ICMP;
+
+       /* IPv4 --> GRE/Teredo/VXLAN --> MAC --> IPv6 */
+       ptype[39] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_FRAG;
+       ptype[40] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_NONFRAG;
+       ptype[41] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_UDP;
+       ptype[42] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_TCP;
+       ptype[43] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_SCTP;
+       /* The next ptype's inner L4 is IGMP */
+       ptype[44] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN;
+       ptype[45] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV4_EXT_UNKNOWN |
+                   RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                   RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                   RTE_PTYPE_INNER_L4_ICMP;
+
+       /* Non-tunnel IPv6 */
+       ptype[111] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_L4_FRAG;
+       ptype[112] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_L4_NONFRAG;
+       ptype[113] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_L4_UDP;
+       ptype[114] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_L4_TCP;
+       ptype[115] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRE;
+       ptype[116] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_L4_SCTP;
+       ptype[117] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_L4_IGMP;
+       ptype[118] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_L4_ICMP;
+       /* Special for PTP over IPv6 + UDP */
+       ptype[119] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_L4_UDP;
+
+       /* IPv6 --> GRE/Teredo/VXLAN */
+       ptype[123] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT;
+       /* IPv6 --> GRE/Teredo/VXLAN --> MAC */
+       ptype[124] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER;
+
+       /* IPv6 --> GRE/Teredo/VXLAN --> MAC --> IPv4 */
+       ptype[125] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_FRAG;
+       ptype[126] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_NONFRAG;
+       ptype[127] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_UDP;
+       ptype[128] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_TCP;
+       ptype[129] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_SCTP;
+       /* The next ptype's inner L4 is IGMP */
+       ptype[130] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN;
+       ptype[131] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_ICMP;
+
+       /* IPv6 --> GRE/Teredo/VXLAN --> MAC --> IPv6 */
+       ptype[133] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_FRAG;
+       ptype[134] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_NONFRAG;
+       ptype[135] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_UDP;
+       ptype[136] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_TCP;
+       ptype[137] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_SCTP;
+       /* The next ptype's inner L4 is IGMP */
+       ptype[138] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN;
+       ptype[139] = RTE_PTYPE_L2_ETHER | RTE_PTYPE_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_TUNNEL_GRENAT | RTE_PTYPE_INNER_L2_ETHER |
+                    RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN |
+                    RTE_PTYPE_INNER_L4_ICMP;
+}
+
+void
+hns3_init_rx_ptype_tble(struct rte_eth_dev *dev)
+{
+       struct hns3_adapter *hns = dev->data->dev_private;
+       struct hns3_ptype_table *tbl = &hns->ptype_tbl;
+
+       memset(tbl, 0, sizeof(*tbl));
+
+       hns3_init_non_tunnel_ptype_tbl(tbl);
+       hns3_init_tunnel_ptype_tbl(tbl);
+       hns3_init_adv_layout_ptype(tbl);
+}
+
 static inline void
 hns3_rxd_to_vlan_tci(struct hns3_rx_queue *rxq, struct rte_mbuf *mb,
                     uint32_t l234_info, const struct hns3_desc *rxd)
@@ -1529,7 +2318,7 @@ hns3_rxd_to_vlan_tci(struct hns3_rx_queue *rxq, struct rte_mbuf *mb,
        };
        strip_status = hns3_get_field(l234_info, HNS3_RXD_STRP_TAGP_M,
                                      HNS3_RXD_STRP_TAGP_S);
-       report_mode = report_type[rxq->pvid_state][strip_status];
+       report_mode = report_type[rxq->pvid_sw_discard_en][strip_status];
        switch (report_mode) {
        case HNS3_NO_STRP_VLAN_VLD:
                mb->vlan_tci = 0;
@@ -1542,6 +2331,9 @@ hns3_rxd_to_vlan_tci(struct hns3_rx_queue *rxq, struct rte_mbuf *mb,
                mb->ol_flags |= PKT_RX_VLAN | PKT_RX_VLAN_STRIPPED;
                mb->vlan_tci = rte_le_to_cpu_16(rxd->rx.ot_vlan_tag);
                return;
+       default:
+               mb->vlan_tci = 0;
+               return;
        }
 }
 
@@ -1579,8 +2371,27 @@ hns3_rx_alloc_buffer(struct hns3_rx_queue *rxq)
                return rte_mbuf_raw_alloc(rxq->mb_pool);
 }
 
+static inline void
+hns3_rx_ptp_timestamp_handle(struct hns3_rx_queue *rxq, struct rte_mbuf *mbuf,
+                 volatile struct hns3_desc *rxd)
+{
+       struct hns3_pf *pf = HNS3_DEV_PRIVATE_TO_PF(rxq->hns);
+       uint64_t timestamp = rte_le_to_cpu_64(rxd->timestamp);
+
+       mbuf->ol_flags |= PKT_RX_IEEE1588_PTP | PKT_RX_IEEE1588_TMST;
+       if (hns3_timestamp_rx_dynflag > 0) {
+               *RTE_MBUF_DYNFIELD(mbuf, hns3_timestamp_dynfield_offset,
+                       rte_mbuf_timestamp_t *) = timestamp;
+               mbuf->ol_flags |= hns3_timestamp_rx_dynflag;
+       }
+
+       pf->rx_timestamp = timestamp;
+}
+
 uint16_t
-hns3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
+hns3_recv_pkts_simple(void *rx_queue,
+                     struct rte_mbuf **rx_pkts,
+                     uint16_t nb_pkts)
 {
        volatile struct hns3_desc *rx_ring;  /* RX ring (desc) */
        volatile struct hns3_desc *rxdp;     /* pointer of the current desc */
@@ -1638,8 +2449,12 @@ hns3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                }
 
                rxm = rxe->mbuf;
+               rxm->ol_flags = 0;
                rxe->mbuf = nmb;
 
+               if (unlikely(bd_base_info & BIT(HNS3_RXD_TS_VLD_B)))
+                       hns3_rx_ptp_timestamp_handle(rxq, rxm, rxdp);
+
                dma_addr = rte_mbuf_data_iova_default(nmb);
                rxdp->addr = rte_cpu_to_le_64(dma_addr);
                rxdp->rx.bd_base_info = 0;
@@ -1650,7 +2465,7 @@ hns3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
                rxm->data_len = rxm->pkt_len;
                rxm->port = rxq->port_id;
                rxm->hash.rss = rte_le_to_cpu_32(rxd.rx.rss_hash);
-               rxm->ol_flags = PKT_RX_RSS_HASH;
+               rxm->ol_flags |= PKT_RX_RSS_HASH;
                if (unlikely(bd_base_info & BIT(HNS3_RXD_LUM_B))) {
                        rxm->hash.fdir.hi =
                                rte_le_to_cpu_16(rxd.rx.fd_id);
@@ -1669,11 +2484,17 @@ hns3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
                rxm->packet_type = hns3_rx_calc_ptype(rxq, l234_info, ol_info);
 
+               if (rxm->packet_type == RTE_PTYPE_L2_ETHER_TIMESYNC)
+                       rxm->ol_flags |= PKT_RX_IEEE1588_PTP;
+
                if (likely(bd_base_info & BIT(HNS3_RXD_L3L4P_B)))
                        hns3_rx_set_cksum_flag(rxm, rxm->packet_type,
                                               cksum_err);
                hns3_rxd_to_vlan_tci(rxq, rxm, l234_info, &rxd);
 
+               /* Increment bytes counter  */
+               rxq->basic_stats.bytes += rxm->pkt_len;
+
                rx_pkts[nb_rx++] = rxm;
                continue;
 pkt_err:
@@ -1894,6 +2715,9 @@ hns3_recv_scattered_pkts(void *rx_queue,
                                               cksum_err);
                hns3_rxd_to_vlan_tci(rxq, first_seg, l234_info, &rxd);
 
+               /* Increment bytes counter */
+               rxq->basic_stats.bytes += first_seg->pkt_len;
+
                rx_pkts[nb_rx++] = first_seg;
                first_seg = NULL;
                continue;
@@ -1915,6 +2739,33 @@ pkt_err:
        return nb_rx;
 }
 
+void __rte_weak
+hns3_rxq_vec_setup(__rte_unused struct hns3_rx_queue *rxq)
+{
+}
+
+int __rte_weak
+hns3_rx_check_vec_support(__rte_unused struct rte_eth_dev *dev)
+{
+       return -ENOTSUP;
+}
+
+uint16_t __rte_weak
+hns3_recv_pkts_vec(__rte_unused void *tx_queue,
+                  __rte_unused struct rte_mbuf **rx_pkts,
+                  __rte_unused uint16_t nb_pkts)
+{
+       return 0;
+}
+
+uint16_t __rte_weak
+hns3_recv_pkts_vec_sve(__rte_unused void *tx_queue,
+                      __rte_unused struct rte_mbuf **rx_pkts,
+                      __rte_unused uint16_t nb_pkts)
+{
+       return 0;
+}
+
 int
 hns3_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
                       struct rte_eth_burst_mode *mode)
@@ -1923,8 +2774,10 @@ hns3_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
                eth_rx_burst_t pkt_burst;
                const char *info;
        } burst_infos[] = {
-               { hns3_recv_pkts,               "Scalar" },
+               { hns3_recv_pkts_simple,        "Scalar Simple" },
                { hns3_recv_scattered_pkts,     "Scalar Scattered" },
+               { hns3_recv_pkts_vec,           "Vector Neon"   },
+               { hns3_recv_pkts_vec_sve,       "Vector Sve"    },
        };
 
        eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
@@ -1943,15 +2796,53 @@ hns3_rx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
        return ret;
 }
 
+static bool
+hns3_get_default_vec_support(void)
+{
+#if defined(RTE_ARCH_ARM64)
+       if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON))
+               return true;
+#endif
+       return false;
+}
+
+static bool
+hns3_get_sve_support(void)
+{
+#if defined(RTE_ARCH_ARM64) && defined(__ARM_FEATURE_SVE)
+       if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_SVE))
+               return true;
+#endif
+       return false;
+}
+
 static eth_rx_burst_t
 hns3_get_rx_function(struct rte_eth_dev *dev)
 {
        struct hns3_adapter *hns = dev->data->dev_private;
        uint64_t offloads = dev->data->dev_conf.rxmode.offloads;
-
-       if (hns->rx_simple_allowed && !dev->data->scattered_rx &&
-           (offloads & DEV_RX_OFFLOAD_TCP_LRO) == 0)
-               return hns3_recv_pkts;
+       bool vec_allowed, sve_allowed, simple_allowed;
+       bool vec_support;
+
+       vec_support = hns3_rx_check_vec_support(dev) == 0;
+       vec_allowed = vec_support && hns3_get_default_vec_support();
+       sve_allowed = vec_support && hns3_get_sve_support();
+       simple_allowed = !dev->data->scattered_rx &&
+                        (offloads & DEV_RX_OFFLOAD_TCP_LRO) == 0;
+
+       if (hns->rx_func_hint == HNS3_IO_FUNC_HINT_VEC && vec_allowed)
+               return hns3_recv_pkts_vec;
+       if (hns->rx_func_hint == HNS3_IO_FUNC_HINT_SVE && sve_allowed)
+               return hns3_recv_pkts_vec_sve;
+       if (hns->rx_func_hint == HNS3_IO_FUNC_HINT_SIMPLE && simple_allowed)
+               return hns3_recv_pkts_simple;
+       if (hns->rx_func_hint == HNS3_IO_FUNC_HINT_COMMON)
+               return hns3_recv_scattered_pkts;
+
+       if (vec_allowed)
+               return hns3_recv_pkts_vec;
+       if (simple_allowed)
+               return hns3_recv_pkts_simple;
 
        return hns3_recv_scattered_pkts;
 }
@@ -1978,8 +2869,8 @@ hns3_tx_queue_conf_check(struct hns3_hw *hw, const struct rte_eth_txconf *conf,
        if (rs_thresh + free_thresh > nb_desc || nb_desc % rs_thresh ||
            rs_thresh >= nb_desc - HNS3_TX_RS_FREE_THRESH_GAP ||
            free_thresh >= nb_desc - HNS3_TX_RS_FREE_THRESH_GAP) {
-               hns3_err(hw, "tx_rs_thresh (%d) tx_free_thresh (%d) nb_desc "
-                        "(%d) of tx descriptors for port=%d queue=%d check "
+               hns3_err(hw, "tx_rs_thresh (%u) tx_free_thresh (%u) nb_desc "
+                        "(%u) of tx descriptors for port=%u queue=%u check "
                         "fail!",
                         rs_thresh, free_thresh, nb_desc, hw->data->port_id,
                         idx);
@@ -2012,11 +2903,6 @@ hns3_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t nb_desc,
        int tx_entry_len;
        int ret;
 
-       if (dev->data->dev_started) {
-               hns3_err(hw, "tx_queue_setup after dev_start no supported");
-               return -EINVAL;
-       }
-
        ret = hns3_tx_queue_conf_check(hw, conf, nb_desc,
                                       &tx_rs_thresh, &tx_free_thresh, idx);
        if (ret)
@@ -2040,6 +2926,11 @@ hns3_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t nb_desc,
        }
 
        txq->tx_deferred_start = conf->tx_deferred_start;
+       if (txq->tx_deferred_start && !hns3_dev_indep_txrx_supported(hw)) {
+               hns3_warn(hw, "deferred start is not supported.");
+               txq->tx_deferred_start = false;
+       }
+
        tx_entry_len = sizeof(struct hns3_entry) * txq->nb_tx_desc;
        txq->sw_ring = rte_zmalloc_socket("hns3 TX sw ring", tx_entry_len,
                                          RTE_CACHE_LINE_SIZE, socket_id);
@@ -2065,19 +2956,32 @@ hns3_tx_queue_setup(struct rte_eth_dev *dev, uint16_t idx, uint16_t nb_desc,
        }
 
        txq->port_id = dev->data->port_id;
-       txq->pvid_state = hw->port_base_vlan_cfg.state;
+       /*
+        * For hns3 PF device, if the VLAN mode is HW_SHIFT_AND_DISCARD_MODE,
+        * the pvid_sw_shift_en in the queue struct should not be changed,
+        * because PVID-related operations do not need to be processed by PMD
+        * driver. For hns3 VF device, whether it needs to process PVID depends
+        * on the configuration of PF kernel mode netdev driver. And the
+        * related PF configuration is delivered through the mailbox and finally
+        * reflectd in port_base_vlan_cfg.
+        */
+       if (hns->is_vf || hw->vlan_mode == HNS3_SW_SHIFT_AND_DISCARD_MODE)
+               txq->pvid_sw_shift_en = hw->port_base_vlan_cfg.state ==
+                                       HNS3_PORT_BASE_VLAN_ENABLE;
+       else
+               txq->pvid_sw_shift_en = false;
+       txq->max_non_tso_bd_num = hw->max_non_tso_bd_num;
        txq->configured = true;
-       txq->io_base = (void *)((char *)hw->io_base + HNS3_TQP_REG_OFFSET +
-                               idx * HNS3_TQP_REG_SIZE);
+       txq->io_base = (void *)((char *)hw->io_base +
+                                               hns3_get_tqp_reg_offset(idx));
        txq->io_tail_reg = (volatile void *)((char *)txq->io_base +
                                             HNS3_RING_TX_TAIL_REG);
        txq->min_tx_pkt_len = hw->min_tx_pkt_len;
-       txq->over_length_pkt_cnt = 0;
-       txq->exceed_limit_bd_pkt_cnt = 0;
-       txq->exceed_limit_bd_reassem_fail = 0;
-       txq->unsupported_tunnel_pkt_cnt = 0;
-       txq->queue_full_cnt = 0;
-       txq->pkt_padding_fail_cnt = 0;
+       txq->tso_mode = hw->tso_mode;
+       txq->udp_cksum_mode = hw->udp_cksum_mode;
+       memset(&txq->basic_stats, 0, sizeof(struct hns3_tx_basic_stats));
+       memset(&txq->dfx_stats, 0, sizeof(struct hns3_tx_dfx_stats));
+
        rte_spinlock_lock(&hw->lock);
        dev->data->tx_queues[idx] = txq;
        rte_spinlock_unlock(&hw->lock);
@@ -2121,44 +3025,6 @@ hns3_tx_free_useless_buffer(struct hns3_tx_queue *txq)
        txq->tx_bd_ready   = tx_bd_ready;
 }
 
-static int
-hns3_tso_proc_tunnel(struct hns3_desc *desc, uint64_t ol_flags,
-                    struct rte_mbuf *rxm, uint8_t *l2_len)
-{
-       uint64_t tun_flags;
-       uint8_t ol4_len;
-       uint32_t otmp;
-
-       tun_flags = ol_flags & PKT_TX_TUNNEL_MASK;
-       if (tun_flags == 0)
-               return 0;
-
-       otmp = rte_le_to_cpu_32(desc->tx.ol_type_vlan_len_msec);
-       switch (tun_flags) {
-       case PKT_TX_TUNNEL_GENEVE:
-       case PKT_TX_TUNNEL_VXLAN:
-               *l2_len = rxm->l2_len - RTE_ETHER_VXLAN_HLEN;
-               break;
-       case PKT_TX_TUNNEL_GRE:
-               /*
-                * OL4 header size, defined in 4 Bytes, it contains outer
-                * L4(GRE) length and tunneling length.
-                */
-               ol4_len = hns3_get_field(otmp, HNS3_TXD_L4LEN_M,
-                                        HNS3_TXD_L4LEN_S);
-               *l2_len = rxm->l2_len - (ol4_len << HNS3_L4_LEN_UNIT);
-               break;
-       default:
-               /* For non UDP / GRE tunneling, drop the tunnel packet */
-               return -EINVAL;
-       }
-       hns3_set_field(otmp, HNS3_TXD_L2LEN_M, HNS3_TXD_L2LEN_S,
-                      rxm->outer_l2_len >> HNS3_L2_LEN_UNIT);
-       desc->tx.ol_type_vlan_len_msec = rte_cpu_to_le_32(otmp);
-
-       return 0;
-}
-
 int
 hns3_config_gro(struct hns3_hw *hw, bool en)
 {
@@ -2203,31 +3069,15 @@ hns3_pkt_is_tso(struct rte_mbuf *m)
 }
 
 static void
-hns3_set_tso(struct hns3_desc *desc, uint64_t ol_flags,
-               uint32_t paylen, struct rte_mbuf *rxm)
+hns3_set_tso(struct hns3_desc *desc, uint32_t paylen, struct rte_mbuf *rxm)
 {
-       uint8_t l2_len = rxm->l2_len;
-       uint32_t tmp;
-
        if (!hns3_pkt_is_tso(rxm))
                return;
 
-       if (hns3_tso_proc_tunnel(desc, ol_flags, rxm, &l2_len))
-               return;
-
        if (paylen <= rxm->tso_segsz)
                return;
 
-       tmp = rte_le_to_cpu_32(desc->tx.type_cs_vlan_tso_len);
-       hns3_set_bit(tmp, HNS3_TXD_TSO_B, 1);
-       hns3_set_bit(tmp, HNS3_TXD_L3CS_B, 1);
-       hns3_set_field(tmp, HNS3_TXD_L4T_M, HNS3_TXD_L4T_S, HNS3_L4T_TCP);
-       hns3_set_bit(tmp, HNS3_TXD_L4CS_B, 1);
-       hns3_set_field(tmp, HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
-                      sizeof(struct rte_tcp_hdr) >> HNS3_L4_LEN_UNIT);
-       hns3_set_field(tmp, HNS3_TXD_L2LEN_M, HNS3_TXD_L2LEN_S,
-                      l2_len >> HNS3_L2_LEN_UNIT);
-       desc->tx.type_cs_vlan_tso_len = rte_cpu_to_le_32(tmp);
+       desc->tx.type_cs_vlan_tso_len |= rte_cpu_to_le_32(BIT(HNS3_TXD_TSO_B));
        desc->tx.mss = rte_cpu_to_le_16(rxm->tso_segsz);
 }
 
@@ -2236,7 +3086,7 @@ hns3_fill_per_desc(struct hns3_desc *desc, struct rte_mbuf *rxm)
 {
        desc->addr = rte_mbuf_data_iova(rxm);
        desc->tx.send_size = rte_cpu_to_le_16(rte_pktmbuf_data_len(rxm));
-       desc->tx.tp_fe_sc_vld_ra_ri = rte_cpu_to_le_16(BIT(HNS3_TXD_VLD_B));
+       desc->tx.tp_fe_sc_vld_ra_ri |= rte_cpu_to_le_16(BIT(HNS3_TXD_VLD_B));
 }
 
 static void
@@ -2251,8 +3101,8 @@ hns3_fill_first_desc(struct hns3_tx_queue *txq, struct hns3_desc *desc,
        hdr_len += (ol_flags & PKT_TX_TUNNEL_MASK) ?
                           rxm->outer_l2_len + rxm->outer_l3_len : 0;
        paylen = rxm->pkt_len - hdr_len;
-       desc->tx.paylen = rte_cpu_to_le_32(paylen);
-       hns3_set_tso(desc, ol_flags, paylen, rxm);
+       desc->tx.paylen_fd_dop_ol4cs |= rte_cpu_to_le_32(paylen);
+       hns3_set_tso(desc, paylen, rxm);
 
        /*
         * Currently, hardware doesn't support more than two layers VLAN offload
@@ -2266,7 +3116,7 @@ hns3_fill_first_desc(struct hns3_tx_queue *txq, struct hns3_desc *desc,
         * To avoid the VLAN of Tx descriptor is overwritten by PVID, it should
         * be added to the position close to the IP header when PVID is enabled.
         */
-       if (!txq->pvid_state && ol_flags & (PKT_TX_VLAN_PKT |
+       if (!txq->pvid_sw_shift_en && ol_flags & (PKT_TX_VLAN_PKT |
                                PKT_TX_QINQ_PKT)) {
                desc->tx.ol_type_vlan_len_msec |=
                                rte_cpu_to_le_32(BIT(HNS3_TXD_OVLAN_B));
@@ -2279,44 +3129,35 @@ hns3_fill_first_desc(struct hns3_tx_queue *txq, struct hns3_desc *desc,
        }
 
        if (ol_flags & PKT_TX_QINQ_PKT ||
-           ((ol_flags & PKT_TX_VLAN_PKT) && txq->pvid_state)) {
+           ((ol_flags & PKT_TX_VLAN_PKT) && txq->pvid_sw_shift_en)) {
                desc->tx.type_cs_vlan_tso_len |=
                                        rte_cpu_to_le_32(BIT(HNS3_TXD_VLAN_B));
                desc->tx.vlan_tag = rte_cpu_to_le_16(rxm->vlan_tci);
        }
+
+       if (ol_flags & PKT_TX_IEEE1588_TMST)
+               desc->tx.tp_fe_sc_vld_ra_ri |=
+                               rte_cpu_to_le_16(BIT(HNS3_TXD_TSYN_B));
 }
 
-static int
-hns3_tx_alloc_mbufs(struct hns3_tx_queue *txq, struct rte_mempool *mb_pool,
-                   uint16_t nb_new_buf, struct rte_mbuf **alloc_mbuf)
+static inline int
+hns3_tx_alloc_mbufs(struct rte_mempool *mb_pool, uint16_t nb_new_buf,
+                       struct rte_mbuf **alloc_mbuf)
 {
-       struct rte_mbuf *new_mbuf = NULL;
-       struct rte_eth_dev *dev;
-       struct rte_mbuf *temp;
-       struct hns3_hw *hw;
+#define MAX_NON_TSO_BD_PER_PKT 18
+       struct rte_mbuf *pkt_segs[MAX_NON_TSO_BD_PER_PKT];
        uint16_t i;
 
        /* Allocate enough mbufs */
-       for (i = 0; i < nb_new_buf; i++) {
-               temp = rte_pktmbuf_alloc(mb_pool);
-               if (unlikely(temp == NULL)) {
-                       dev = &rte_eth_devices[txq->port_id];
-                       hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
-                       hns3_err(hw, "Failed to alloc TX mbuf port_id=%d,"
-                                    "queue_id=%d in reassemble tx pkts.",
-                                    txq->port_id, txq->queue_id);
-                       rte_pktmbuf_free(new_mbuf);
-                       return -ENOMEM;
-               }
-               temp->next = new_mbuf;
-               new_mbuf = temp;
-       }
-
-       if (new_mbuf == NULL)
+       if (rte_mempool_get_bulk(mb_pool, (void **)pkt_segs, nb_new_buf))
                return -ENOMEM;
 
-       new_mbuf->nb_segs = nb_new_buf;
-       *alloc_mbuf = new_mbuf;
+       for (i = 0; i < nb_new_buf - 1; i++)
+               pkt_segs[i]->next = pkt_segs[i + 1];
+
+       pkt_segs[nb_new_buf - 1]->next = NULL;
+       pkt_segs[0]->nb_segs = nb_new_buf;
+       *alloc_mbuf = pkt_segs[0];
 
        return 0;
 }
@@ -2336,10 +3177,9 @@ hns3_pktmbuf_copy_hdr(struct rte_mbuf *new_pkt, struct rte_mbuf *old_pkt)
 }
 
 static int
-hns3_reassemble_tx_pkts(void *tx_queue, struct rte_mbuf *tx_pkt,
-                       struct rte_mbuf **new_pkt)
+hns3_reassemble_tx_pkts(struct rte_mbuf *tx_pkt, struct rte_mbuf **new_pkt,
+                                 uint8_t max_non_tso_bd_num)
 {
-       struct hns3_tx_queue *txq = tx_queue;
        struct rte_mempool *mb_pool;
        struct rte_mbuf *new_mbuf;
        struct rte_mbuf *temp_new;
@@ -2351,7 +3191,6 @@ hns3_reassemble_tx_pkts(void *tx_queue, struct rte_mbuf *tx_pkt,
        uint16_t len_s;
        uint16_t len_d;
        uint16_t len;
-       uint16_t i;
        int ret;
        char *s;
        char *d;
@@ -2359,7 +3198,7 @@ hns3_reassemble_tx_pkts(void *tx_queue, struct rte_mbuf *tx_pkt,
        mb_pool = tx_pkt->pool;
        buf_size = tx_pkt->buf_len - RTE_PKTMBUF_HEADROOM;
        nb_new_buf = (rte_pktmbuf_pkt_len(tx_pkt) - 1) / buf_size + 1;
-       if (nb_new_buf > HNS3_MAX_NON_TSO_BD_PER_PKT)
+       if (nb_new_buf > max_non_tso_bd_num)
                return -EINVAL;
 
        last_buf_len = rte_pktmbuf_pkt_len(tx_pkt) % buf_size;
@@ -2367,7 +3206,7 @@ hns3_reassemble_tx_pkts(void *tx_queue, struct rte_mbuf *tx_pkt,
                last_buf_len = buf_size;
 
        /* Allocate enough mbufs */
-       ret = hns3_tx_alloc_mbufs(txq, mb_pool, nb_new_buf, &new_mbuf);
+       ret = hns3_tx_alloc_mbufs(mb_pool, nb_new_buf, &new_mbuf);
        if (ret)
                return ret;
 
@@ -2376,12 +3215,9 @@ hns3_reassemble_tx_pkts(void *tx_queue, struct rte_mbuf *tx_pkt,
        s = rte_pktmbuf_mtod(temp, char *);
        len_s = rte_pktmbuf_data_len(temp);
        temp_new = new_mbuf;
-       for (i = 0; i < nb_new_buf; i++) {
+       while (temp != NULL && temp_new != NULL) {
                d = rte_pktmbuf_mtod(temp_new, char *);
-               if (i < nb_new_buf - 1)
-                       buf_len = buf_size;
-               else
-                       buf_len = last_buf_len;
+               buf_len = temp_new->next == NULL ? last_buf_len : buf_size;
                len_d = buf_len;
 
                while (len_d) {
@@ -2415,186 +3251,227 @@ hns3_reassemble_tx_pkts(void *tx_queue, struct rte_mbuf *tx_pkt,
 }
 
 static void
-hns3_parse_outer_params(uint64_t ol_flags, uint32_t *ol_type_vlan_len_msec)
+hns3_parse_outer_params(struct rte_mbuf *m, uint32_t *ol_type_vlan_len_msec)
 {
        uint32_t tmp = *ol_type_vlan_len_msec;
+       uint64_t ol_flags = m->ol_flags;
 
        /* (outer) IP header type */
        if (ol_flags & PKT_TX_OUTER_IPV4) {
-               /* OL3 header size, defined in 4 bytes */
-               hns3_set_field(tmp, HNS3_TXD_L3LEN_M, HNS3_TXD_L3LEN_S,
-                              sizeof(struct rte_ipv4_hdr) >> HNS3_L3_LEN_UNIT);
                if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
-                       hns3_set_field(tmp, HNS3_TXD_OL3T_M,
-                                      HNS3_TXD_OL3T_S, HNS3_OL3T_IPV4_CSUM);
+                       tmp |= hns3_gen_field_val(HNS3_TXD_OL3T_M,
+                                       HNS3_TXD_OL3T_S, HNS3_OL3T_IPV4_CSUM);
                else
-                       hns3_set_field(tmp, HNS3_TXD_OL3T_M, HNS3_TXD_OL3T_S,
-                                      HNS3_OL3T_IPV4_NO_CSUM);
+                       tmp |= hns3_gen_field_val(HNS3_TXD_OL3T_M,
+                               HNS3_TXD_OL3T_S, HNS3_OL3T_IPV4_NO_CSUM);
        } else if (ol_flags & PKT_TX_OUTER_IPV6) {
-               hns3_set_field(tmp, HNS3_TXD_OL3T_M, HNS3_TXD_OL3T_S,
-                              HNS3_OL3T_IPV6);
-               /* OL3 header size, defined in 4 bytes */
-               hns3_set_field(tmp, HNS3_TXD_L3LEN_M, HNS3_TXD_L3LEN_S,
-                              sizeof(struct rte_ipv6_hdr) >> HNS3_L3_LEN_UNIT);
+               tmp |= hns3_gen_field_val(HNS3_TXD_OL3T_M, HNS3_TXD_OL3T_S,
+                                       HNS3_OL3T_IPV6);
        }
-
+       /* OL3 header size, defined in 4 bytes */
+       tmp |= hns3_gen_field_val(HNS3_TXD_L3LEN_M, HNS3_TXD_L3LEN_S,
+                               m->outer_l3_len >> HNS3_L3_LEN_UNIT);
        *ol_type_vlan_len_msec = tmp;
 }
 
 static int
-hns3_parse_inner_params(uint64_t ol_flags, uint32_t *ol_type_vlan_len_msec,
-                       struct rte_net_hdr_lens *hdr_lens)
+hns3_parse_inner_params(struct rte_mbuf *m, uint32_t *ol_type_vlan_len_msec,
+                       uint32_t *type_cs_vlan_tso_len)
 {
-       uint32_t tmp = *ol_type_vlan_len_msec;
-       uint8_t l4_len;
-
-       /* OL2 header size, defined in 2 bytes */
-       hns3_set_field(tmp, HNS3_TXD_L2LEN_M, HNS3_TXD_L2LEN_S,
-                      sizeof(struct rte_ether_hdr) >> HNS3_L2_LEN_UNIT);
+#define HNS3_NVGRE_HLEN 8
+       uint32_t tmp_outer = *ol_type_vlan_len_msec;
+       uint32_t tmp_inner = *type_cs_vlan_tso_len;
+       uint64_t ol_flags = m->ol_flags;
+       uint16_t inner_l2_len;
 
-       /* L4TUNT: L4 Tunneling Type */
        switch (ol_flags & PKT_TX_TUNNEL_MASK) {
+       case PKT_TX_TUNNEL_VXLAN_GPE:
        case PKT_TX_TUNNEL_GENEVE:
        case PKT_TX_TUNNEL_VXLAN:
-               /* MAC in UDP tunnelling packet, include VxLAN */
-               hns3_set_field(tmp, HNS3_TXD_TUNTYPE_M, HNS3_TXD_TUNTYPE_S,
-                              HNS3_TUN_MAC_IN_UDP);
+               /* MAC in UDP tunnelling packet, include VxLAN and GENEVE */
+               tmp_outer |= hns3_gen_field_val(HNS3_TXD_TUNTYPE_M,
+                               HNS3_TXD_TUNTYPE_S, HNS3_TUN_MAC_IN_UDP);
                /*
-                * OL4 header size, defined in 4 Bytes, it contains outer
-                * L4(UDP) length and tunneling length.
+                * The inner l2 length of mbuf is the sum of outer l4 length,
+                * tunneling header length and inner l2 length for a tunnel
+                * packect. But in hns3 tx descriptor, the tunneling header
+                * length is contained in the field of outer L4 length.
+                * Therefore, driver need to calculate the outer L4 length and
+                * inner L2 length.
                 */
-               hns3_set_field(tmp, HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
-                              (uint8_t)RTE_ETHER_VXLAN_HLEN >>
-                              HNS3_L4_LEN_UNIT);
+               tmp_outer |= hns3_gen_field_val(HNS3_TXD_L4LEN_M,
+                                               HNS3_TXD_L4LEN_S,
+                                               (uint8_t)RTE_ETHER_VXLAN_HLEN >>
+                                               HNS3_L4_LEN_UNIT);
+
+               inner_l2_len = m->l2_len - RTE_ETHER_VXLAN_HLEN;
                break;
        case PKT_TX_TUNNEL_GRE:
-               hns3_set_field(tmp, HNS3_TXD_TUNTYPE_M, HNS3_TXD_TUNTYPE_S,
-                              HNS3_TUN_NVGRE);
+               tmp_outer |= hns3_gen_field_val(HNS3_TXD_TUNTYPE_M,
+                                       HNS3_TXD_TUNTYPE_S, HNS3_TUN_NVGRE);
                /*
-                * OL4 header size, defined in 4 Bytes, it contains outer
-                * L4(GRE) length and tunneling length.
+                * For NVGRE tunnel packect, the outer L4 is empty. So only
+                * fill the NVGRE header length to the outer L4 field.
                 */
-               l4_len = hdr_lens->l4_len + hdr_lens->tunnel_len;
-               hns3_set_field(tmp, HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
-                              l4_len >> HNS3_L4_LEN_UNIT);
+               tmp_outer |= hns3_gen_field_val(HNS3_TXD_L4LEN_M,
+                               HNS3_TXD_L4LEN_S,
+                               (uint8_t)HNS3_NVGRE_HLEN >> HNS3_L4_LEN_UNIT);
+
+               inner_l2_len = m->l2_len - HNS3_NVGRE_HLEN;
                break;
        default:
                /* For non UDP / GRE tunneling, drop the tunnel packet */
                return -EINVAL;
        }
 
-       *ol_type_vlan_len_msec = tmp;
+       tmp_inner |= hns3_gen_field_val(HNS3_TXD_L2LEN_M, HNS3_TXD_L2LEN_S,
+                                       inner_l2_len >> HNS3_L2_LEN_UNIT);
+       /* OL2 header size, defined in 2 bytes */
+       tmp_outer |= hns3_gen_field_val(HNS3_TXD_L2LEN_M, HNS3_TXD_L2LEN_S,
+                                       m->outer_l2_len >> HNS3_L2_LEN_UNIT);
+
+       *type_cs_vlan_tso_len = tmp_inner;
+       *ol_type_vlan_len_msec = tmp_outer;
 
        return 0;
 }
 
 static int
-hns3_parse_tunneling_params(struct hns3_tx_queue *txq, uint16_t tx_desc_id,
-                           uint64_t ol_flags,
-                           struct rte_net_hdr_lens *hdr_lens)
+hns3_parse_tunneling_params(struct hns3_tx_queue *txq, struct rte_mbuf *m,
+                           uint16_t tx_desc_id)
 {
        struct hns3_desc *tx_ring = txq->tx_ring;
        struct hns3_desc *desc = &tx_ring[tx_desc_id];
-       uint32_t value = 0;
+       uint64_t ol_flags = m->ol_flags;
+       uint32_t tmp_outer = 0;
+       uint32_t tmp_inner = 0;
+       uint32_t tmp_ol4cs;
        int ret;
 
-       hns3_parse_outer_params(ol_flags, &value);
-       ret = hns3_parse_inner_params(ol_flags, &value, hdr_lens);
-       if (ret)
-               return -EINVAL;
+       /*
+        * The tunnel header is contained in the inner L2 header field of the
+        * mbuf, but for hns3 descriptor, it is contained in the outer L4. So,
+        * there is a need that switching between them. To avoid multiple
+        * calculations, the length of the L2 header include the outer and
+        * inner, will be filled during the parsing of tunnel packects.
+        */
+       if (!(ol_flags & PKT_TX_TUNNEL_MASK)) {
+               /*
+                * For non tunnel type the tunnel type id is 0, so no need to
+                * assign a value to it. Only the inner(normal) L2 header length
+                * is assigned.
+                */
+               tmp_inner |= hns3_gen_field_val(HNS3_TXD_L2LEN_M,
+                              HNS3_TXD_L2LEN_S, m->l2_len >> HNS3_L2_LEN_UNIT);
+       } else {
+               /*
+                * If outer csum is not offload, the outer length may be filled
+                * with 0. And the length of the outer header is added to the
+                * inner l2_len. It would lead a cksum error. So driver has to
+                * calculate the header length.
+                */
+               if (unlikely(!(ol_flags &
+                       (PKT_TX_OUTER_IP_CKSUM | PKT_TX_OUTER_UDP_CKSUM)) &&
+                                       m->outer_l2_len == 0)) {
+                       struct rte_net_hdr_lens hdr_len;
+                       (void)rte_net_get_ptype(m, &hdr_len,
+                                       RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK);
+                       m->outer_l3_len = hdr_len.l3_len;
+                       m->outer_l2_len = hdr_len.l2_len;
+                       m->l2_len = m->l2_len - hdr_len.l2_len - hdr_len.l3_len;
+               }
+               hns3_parse_outer_params(m, &tmp_outer);
+               ret = hns3_parse_inner_params(m, &tmp_outer, &tmp_inner);
+               if (ret)
+                       return -EINVAL;
+       }
 
-       desc->tx.ol_type_vlan_len_msec |= rte_cpu_to_le_32(value);
+       desc->tx.ol_type_vlan_len_msec = rte_cpu_to_le_32(tmp_outer);
+       desc->tx.type_cs_vlan_tso_len = rte_cpu_to_le_32(tmp_inner);
+       tmp_ol4cs = ol_flags & PKT_TX_OUTER_UDP_CKSUM ?
+                       BIT(HNS3_TXD_OL4CS_B) : 0;
+       desc->tx.paylen_fd_dop_ol4cs = rte_cpu_to_le_32(tmp_ol4cs);
 
        return 0;
 }
 
 static void
-hns3_parse_l3_cksum_params(uint64_t ol_flags, uint32_t *type_cs_vlan_tso_len)
+hns3_parse_l3_cksum_params(struct rte_mbuf *m, uint32_t *type_cs_vlan_tso_len)
 {
+       uint64_t ol_flags = m->ol_flags;
+       uint32_t l3_type;
        uint32_t tmp;
 
+       tmp = *type_cs_vlan_tso_len;
+       if (ol_flags & PKT_TX_IPV4)
+               l3_type = HNS3_L3T_IPV4;
+       else if (ol_flags & PKT_TX_IPV6)
+               l3_type = HNS3_L3T_IPV6;
+       else
+               l3_type = HNS3_L3T_NONE;
+
+       /* inner(/normal) L3 header size, defined in 4 bytes */
+       tmp |= hns3_gen_field_val(HNS3_TXD_L3LEN_M, HNS3_TXD_L3LEN_S,
+                                       m->l3_len >> HNS3_L3_LEN_UNIT);
+
+       tmp |= hns3_gen_field_val(HNS3_TXD_L3T_M, HNS3_TXD_L3T_S, l3_type);
+
        /* Enable L3 checksum offloads */
-       if (ol_flags & PKT_TX_IPV4) {
-               tmp = *type_cs_vlan_tso_len;
-               hns3_set_field(tmp, HNS3_TXD_L3T_M, HNS3_TXD_L3T_S,
-                              HNS3_L3T_IPV4);
-               /* inner(/normal) L3 header size, defined in 4 bytes */
-               hns3_set_field(tmp, HNS3_TXD_L3LEN_M, HNS3_TXD_L3LEN_S,
-                              sizeof(struct rte_ipv4_hdr) >> HNS3_L3_LEN_UNIT);
-               if (ol_flags & PKT_TX_IP_CKSUM)
-                       hns3_set_bit(tmp, HNS3_TXD_L3CS_B, 1);
-               *type_cs_vlan_tso_len = tmp;
-       } else if (ol_flags & PKT_TX_IPV6) {
-               tmp = *type_cs_vlan_tso_len;
-               /* L3T, IPv6 don't do checksum */
-               hns3_set_field(tmp, HNS3_TXD_L3T_M, HNS3_TXD_L3T_S,
-                              HNS3_L3T_IPV6);
-               /* inner(/normal) L3 header size, defined in 4 bytes */
-               hns3_set_field(tmp, HNS3_TXD_L3LEN_M, HNS3_TXD_L3LEN_S,
-                              sizeof(struct rte_ipv6_hdr) >> HNS3_L3_LEN_UNIT);
-               *type_cs_vlan_tso_len = tmp;
-       }
+       if (ol_flags & PKT_TX_IP_CKSUM)
+               tmp |= BIT(HNS3_TXD_L3CS_B);
+       *type_cs_vlan_tso_len = tmp;
 }
 
 static void
-hns3_parse_l4_cksum_params(uint64_t ol_flags, uint32_t *type_cs_vlan_tso_len)
+hns3_parse_l4_cksum_params(struct rte_mbuf *m, uint32_t *type_cs_vlan_tso_len)
 {
+       uint64_t ol_flags = m->ol_flags;
        uint32_t tmp;
-
        /* Enable L4 checksum offloads */
-       switch (ol_flags & PKT_TX_L4_MASK) {
+       switch (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG)) {
+       case PKT_TX_TCP_CKSUM | PKT_TX_TCP_SEG:
        case PKT_TX_TCP_CKSUM:
+       case PKT_TX_TCP_SEG:
                tmp = *type_cs_vlan_tso_len;
-               hns3_set_field(tmp, HNS3_TXD_L4T_M, HNS3_TXD_L4T_S,
-                              HNS3_L4T_TCP);
-               hns3_set_bit(tmp, HNS3_TXD_L4CS_B, 1);
-               hns3_set_field(tmp, HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
-                              sizeof(struct rte_tcp_hdr) >> HNS3_L4_LEN_UNIT);
-               *type_cs_vlan_tso_len = tmp;
+               tmp |= hns3_gen_field_val(HNS3_TXD_L4T_M, HNS3_TXD_L4T_S,
+                                       HNS3_L4T_TCP);
                break;
        case PKT_TX_UDP_CKSUM:
                tmp = *type_cs_vlan_tso_len;
-               hns3_set_field(tmp, HNS3_TXD_L4T_M, HNS3_TXD_L4T_S,
-                              HNS3_L4T_UDP);
-               hns3_set_bit(tmp, HNS3_TXD_L4CS_B, 1);
-               hns3_set_field(tmp, HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
-                              sizeof(struct rte_udp_hdr) >> HNS3_L4_LEN_UNIT);
-               *type_cs_vlan_tso_len = tmp;
+               tmp |= hns3_gen_field_val(HNS3_TXD_L4T_M, HNS3_TXD_L4T_S,
+                                       HNS3_L4T_UDP);
                break;
        case PKT_TX_SCTP_CKSUM:
                tmp = *type_cs_vlan_tso_len;
-               hns3_set_field(tmp, HNS3_TXD_L4T_M, HNS3_TXD_L4T_S,
-                              HNS3_L4T_SCTP);
-               hns3_set_bit(tmp, HNS3_TXD_L4CS_B, 1);
-               hns3_set_field(tmp, HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
-                              sizeof(struct rte_sctp_hdr) >> HNS3_L4_LEN_UNIT);
-               *type_cs_vlan_tso_len = tmp;
+               tmp |= hns3_gen_field_val(HNS3_TXD_L4T_M, HNS3_TXD_L4T_S,
+                                       HNS3_L4T_SCTP);
                break;
        default:
-               break;
+               return;
        }
+       tmp |= BIT(HNS3_TXD_L4CS_B);
+       tmp |= hns3_gen_field_val(HNS3_TXD_L4LEN_M, HNS3_TXD_L4LEN_S,
+                                       m->l4_len >> HNS3_L4_LEN_UNIT);
+       *type_cs_vlan_tso_len = tmp;
 }
 
 static void
-hns3_txd_enable_checksum(struct hns3_tx_queue *txq, uint16_t tx_desc_id,
-                        uint64_t ol_flags)
+hns3_txd_enable_checksum(struct hns3_tx_queue *txq, struct rte_mbuf *m,
+                        uint16_t tx_desc_id)
 {
        struct hns3_desc *tx_ring = txq->tx_ring;
        struct hns3_desc *desc = &tx_ring[tx_desc_id];
        uint32_t value = 0;
 
-       /* inner(/normal) L2 header size, defined in 2 bytes */
-       hns3_set_field(value, HNS3_TXD_L2LEN_M, HNS3_TXD_L2LEN_S,
-                      sizeof(struct rte_ether_hdr) >> HNS3_L2_LEN_UNIT);
-
-       hns3_parse_l3_cksum_params(ol_flags, &value);
-       hns3_parse_l4_cksum_params(ol_flags, &value);
+       hns3_parse_l3_cksum_params(m, &value);
+       hns3_parse_l4_cksum_params(m, &value);
 
        desc->tx.type_cs_vlan_tso_len |= rte_cpu_to_le_32(value);
 }
 
 static bool
-hns3_pkt_need_linearized(struct rte_mbuf *tx_pkts, uint32_t bd_num)
+hns3_pkt_need_linearized(struct rte_mbuf *tx_pkts, uint32_t bd_num,
+                                uint32_t max_non_tso_bd_num)
 {
        struct rte_mbuf *m_first = tx_pkts;
        struct rte_mbuf *m_last = tx_pkts;
@@ -2609,10 +3486,10 @@ hns3_pkt_need_linearized(struct rte_mbuf *tx_pkts, uint32_t bd_num)
         * frags greater than gso header len + mss, and the remaining 7
         * consecutive frags greater than MSS except the last 7 frags.
         */
-       if (bd_num <= HNS3_MAX_NON_TSO_BD_PER_PKT)
+       if (bd_num <= max_non_tso_bd_num)
                return false;
 
-       for (i = 0; m_last && i < HNS3_MAX_NON_TSO_BD_PER_PKT - 1;
+       for (i = 0; m_last && i < max_non_tso_bd_num - 1;
             i++, m_last = m_last->next)
                tot_len += m_last->data_len;
 
@@ -2630,7 +3507,7 @@ hns3_pkt_need_linearized(struct rte_mbuf *tx_pkts, uint32_t bd_num)
         * ensure the sum of the data length of every 7 consecutive buffer
         * is greater than mss except the last one.
         */
-       for (i = 0; m_last && i < bd_num - HNS3_MAX_NON_TSO_BD_PER_PKT; i++) {
+       for (i = 0; m_last && i < bd_num - max_non_tso_bd_num; i++) {
                tot_len -= m_first->data_len;
                tot_len += m_last->data_len;
 
@@ -2644,30 +3521,80 @@ hns3_pkt_need_linearized(struct rte_mbuf *tx_pkts, uint32_t bd_num)
        return false;
 }
 
+static bool
+hns3_outer_ipv4_cksum_prepared(struct rte_mbuf *m, uint64_t ol_flags,
+                               uint32_t *l4_proto)
+{
+       struct rte_ipv4_hdr *ipv4_hdr;
+       ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
+                                          m->outer_l2_len);
+       if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
+               ipv4_hdr->hdr_checksum = 0;
+       if (ol_flags & PKT_TX_OUTER_UDP_CKSUM) {
+               struct rte_udp_hdr *udp_hdr;
+               /*
+                * If OUTER_UDP_CKSUM is support, HW can caclulate the pseudo
+                * header for TSO packets
+                */
+               if (ol_flags & PKT_TX_TCP_SEG)
+                       return true;
+               udp_hdr = rte_pktmbuf_mtod_offset(m, struct rte_udp_hdr *,
+                               m->outer_l2_len + m->outer_l3_len);
+               udp_hdr->dgram_cksum = rte_ipv4_phdr_cksum(ipv4_hdr, ol_flags);
+
+               return true;
+       }
+       *l4_proto = ipv4_hdr->next_proto_id;
+       return false;
+}
+
+static bool
+hns3_outer_ipv6_cksum_prepared(struct rte_mbuf *m, uint64_t ol_flags,
+                               uint32_t *l4_proto)
+{
+       struct rte_ipv6_hdr *ipv6_hdr;
+       ipv6_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv6_hdr *,
+                                          m->outer_l2_len);
+       if (ol_flags & PKT_TX_OUTER_UDP_CKSUM) {
+               struct rte_udp_hdr *udp_hdr;
+               /*
+                * If OUTER_UDP_CKSUM is support, HW can caclulate the pseudo
+                * header for TSO packets
+                */
+               if (ol_flags & PKT_TX_TCP_SEG)
+                       return true;
+               udp_hdr = rte_pktmbuf_mtod_offset(m, struct rte_udp_hdr *,
+                               m->outer_l2_len + m->outer_l3_len);
+               udp_hdr->dgram_cksum = rte_ipv6_phdr_cksum(ipv6_hdr, ol_flags);
+
+               return true;
+       }
+       *l4_proto = ipv6_hdr->proto;
+       return false;
+}
+
 static void
 hns3_outer_header_cksum_prepare(struct rte_mbuf *m)
 {
        uint64_t ol_flags = m->ol_flags;
-       struct rte_ipv4_hdr *ipv4_hdr;
+       uint32_t paylen, hdr_len, l4_proto;
        struct rte_udp_hdr *udp_hdr;
-       uint32_t paylen, hdr_len;
 
        if (!(ol_flags & (PKT_TX_OUTER_IPV4 | PKT_TX_OUTER_IPV6)))
                return;
 
-       if (ol_flags & PKT_TX_IPV4) {
-               ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
-                                                  m->outer_l2_len);
-
-               if (ol_flags & PKT_TX_IP_CKSUM)
-                       ipv4_hdr->hdr_checksum = 0;
+       if (ol_flags & PKT_TX_OUTER_IPV4) {
+               if (hns3_outer_ipv4_cksum_prepared(m, ol_flags, &l4_proto))
+                       return;
+       } else {
+               if (hns3_outer_ipv6_cksum_prepared(m, ol_flags, &l4_proto))
+                       return;
        }
 
-       if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM &&
-           ol_flags & PKT_TX_TCP_SEG) {
+       /* driver should ensure the outer udp cksum is 0 for TUNNEL TSO */
+       if (l4_proto == IPPROTO_UDP && (ol_flags & PKT_TX_TCP_SEG)) {
                hdr_len = m->l2_len + m->l3_len + m->l4_len;
-               hdr_len += (ol_flags & PKT_TX_TUNNEL_MASK) ?
-                               m->outer_l2_len + m->outer_l3_len : 0;
+               hdr_len += m->outer_l2_len + m->outer_l3_len;
                paylen = m->pkt_len - hdr_len;
                if (paylen <= m->tso_segsz)
                        return;
@@ -2724,7 +3651,7 @@ hns3_vld_vlan_chk(struct hns3_tx_queue *txq, struct rte_mbuf *m)
        struct rte_ether_hdr *eh;
        struct rte_vlan_hdr *vh;
 
-       if (!txq->pvid_state)
+       if (!txq->pvid_sw_shift_en)
                return 0;
 
        /*
@@ -2759,43 +3686,132 @@ hns3_vld_vlan_chk(struct hns3_tx_queue *txq, struct rte_mbuf *m)
 }
 #endif
 
-uint16_t
-hns3_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
-              uint16_t nb_pkts)
+static uint16_t
+hns3_udp_cksum_help(struct rte_mbuf *m)
 {
-       struct rte_mbuf *m;
-       uint16_t i;
-       int ret;
+       uint64_t ol_flags = m->ol_flags;
+       uint16_t cksum = 0;
+       uint32_t l4_len;
 
-       for (i = 0; i < nb_pkts; i++) {
-               m = tx_pkts[i];
+       if (ol_flags & PKT_TX_IPV4) {
+               struct rte_ipv4_hdr *ipv4_hdr = rte_pktmbuf_mtod_offset(m,
+                               struct rte_ipv4_hdr *, m->l2_len);
+               l4_len = rte_be_to_cpu_16(ipv4_hdr->total_length) - m->l3_len;
+       } else {
+               struct rte_ipv6_hdr *ipv6_hdr = rte_pktmbuf_mtod_offset(m,
+                               struct rte_ipv6_hdr *, m->l2_len);
+               l4_len = rte_be_to_cpu_16(ipv6_hdr->payload_len);
+       }
 
-               if (hns3_pkt_is_tso(m) &&
-                   (hns3_pkt_need_linearized(m, m->nb_segs) ||
-                    hns3_check_tso_pkt_valid(m))) {
-                       rte_errno = EINVAL;
-                       return i;
-               }
+       rte_raw_cksum_mbuf(m, m->l2_len + m->l3_len, l4_len, &cksum);
+
+       cksum = ~cksum;
+       /*
+        * RFC 768:If the computed checksum is zero for UDP, it is transmitted
+        * as all ones
+        */
+       if (cksum == 0)
+               cksum = 0xffff;
+
+       return (uint16_t)cksum;
+}
+
+static bool
+hns3_validate_tunnel_cksum(struct hns3_tx_queue *tx_queue, struct rte_mbuf *m)
+{
+       uint64_t ol_flags = m->ol_flags;
+       struct rte_udp_hdr *udp_hdr;
+       uint16_t dst_port;
+
+       if (tx_queue->udp_cksum_mode == HNS3_SPECIAL_PORT_HW_CKSUM_MODE ||
+           ol_flags & PKT_TX_TUNNEL_MASK ||
+           (ol_flags & PKT_TX_L4_MASK) != PKT_TX_UDP_CKSUM)
+               return true;
+       /*
+        * A UDP packet with the same dst_port as VXLAN\VXLAN_GPE\GENEVE will
+        * be recognized as a tunnel packet in HW. In this case, if UDP CKSUM
+        * offload is set and the tunnel mask has not been set, the CKSUM will
+        * be wrong since the header length is wrong and driver should complete
+        * the CKSUM to avoid CKSUM error.
+        */
+       udp_hdr = rte_pktmbuf_mtod_offset(m, struct rte_udp_hdr *,
+                                               m->l2_len + m->l3_len);
+       dst_port = rte_be_to_cpu_16(udp_hdr->dst_port);
+       switch (dst_port) {
+       case RTE_VXLAN_DEFAULT_PORT:
+       case RTE_VXLAN_GPE_DEFAULT_PORT:
+       case RTE_GENEVE_DEFAULT_PORT:
+               udp_hdr->dgram_cksum = hns3_udp_cksum_help(m);
+               m->ol_flags = ol_flags & ~PKT_TX_L4_MASK;
+               return false;
+       default:
+               return true;
+       }
+}
+
+static int
+hns3_prep_pkt_proc(struct hns3_tx_queue *tx_queue, struct rte_mbuf *m)
+{
+       int ret;
 
 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
-               ret = rte_validate_tx_offload(m);
-               if (ret != 0) {
-                       rte_errno = -ret;
-                       return i;
-               }
+       ret = rte_validate_tx_offload(m);
+       if (ret != 0) {
+               rte_errno = -ret;
+               return ret;
+       }
 
-               if (hns3_vld_vlan_chk(tx_queue, m)) {
+       ret = hns3_vld_vlan_chk(tx_queue, m);
+       if (ret != 0) {
+               rte_errno = EINVAL;
+               return ret;
+       }
+#endif
+       if (hns3_pkt_is_tso(m)) {
+               if (hns3_pkt_need_linearized(m, m->nb_segs,
+                                            tx_queue->max_non_tso_bd_num) ||
+                   hns3_check_tso_pkt_valid(m)) {
                        rte_errno = EINVAL;
-                       return i;
+                       return -EINVAL;
                }
-#endif
-               ret = rte_net_intel_cksum_prepare(m);
-               if (ret != 0) {
-                       rte_errno = -ret;
-                       return i;
+
+               if (tx_queue->tso_mode != HNS3_TSO_SW_CAL_PSEUDO_H_CSUM) {
+                       /*
+                        * (tso mode != HNS3_TSO_SW_CAL_PSEUDO_H_CSUM) means
+                        * hardware support recalculate the TCP pseudo header
+                        * checksum of packets that need TSO, so network driver
+                        * software not need to recalculate it.
+                        */
+                       hns3_outer_header_cksum_prepare(m);
+                       return 0;
                }
+       }
 
-               hns3_outer_header_cksum_prepare(m);
+       ret = rte_net_intel_cksum_prepare(m);
+       if (ret != 0) {
+               rte_errno = -ret;
+               return ret;
+       }
+
+       if (!hns3_validate_tunnel_cksum(tx_queue, m))
+               return 0;
+
+       hns3_outer_header_cksum_prepare(m);
+
+       return 0;
+}
+
+uint16_t
+hns3_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
+              uint16_t nb_pkts)
+{
+       struct rte_mbuf *m;
+       uint16_t i;
+
+       for (i = 0; i < nb_pkts; i++) {
+               m = tx_pkts[i];
+               if (hns3_prep_pkt_proc(tx_queue, m))
+                       return i;
        }
 
        return i;
@@ -2803,20 +3819,25 @@ hns3_prep_pkts(__rte_unused void *tx_queue, struct rte_mbuf **tx_pkts,
 
 static int
 hns3_parse_cksum(struct hns3_tx_queue *txq, uint16_t tx_desc_id,
-                const struct rte_mbuf *m, struct rte_net_hdr_lens *hdr_lens)
-{
-       /* Fill in tunneling parameters if necessary */
-       if (m->ol_flags & PKT_TX_TUNNEL_MASK) {
-               (void)rte_net_get_ptype(m, hdr_lens, RTE_PTYPE_ALL_MASK);
-               if (hns3_parse_tunneling_params(txq, tx_desc_id, m->ol_flags,
-                                               hdr_lens)) {
-                       txq->unsupported_tunnel_pkt_cnt++;
-                       return -EINVAL;
+                struct rte_mbuf *m)
+{
+       struct hns3_desc *tx_ring = txq->tx_ring;
+       struct hns3_desc *desc = &tx_ring[tx_desc_id];
+
+       /* Enable checksum offloading */
+       if (m->ol_flags & HNS3_TX_CKSUM_OFFLOAD_MASK) {
+               /* Fill in tunneling parameters if necessary */
+               if (hns3_parse_tunneling_params(txq, m, tx_desc_id)) {
+                       txq->dfx_stats.unsupported_tunnel_pkt_cnt++;
+                               return -EINVAL;
                }
+
+               hns3_txd_enable_checksum(txq, m, tx_desc_id);
+       } else {
+               /* clear the control bit */
+               desc->tx.type_cs_vlan_tso_len  = 0;
+               desc->tx.ol_type_vlan_len_msec = 0;
        }
-       /* Enable checksum offloading */
-       if (m->ol_flags & HNS3_TX_CKSUM_OFFLOAD_MASK)
-               hns3_txd_enable_checksum(txq, tx_desc_id, m->ol_flags);
 
        return 0;
 }
@@ -2825,6 +3846,7 @@ static int
 hns3_check_non_tso_pkt(uint16_t nb_buf, struct rte_mbuf **m_seg,
                      struct rte_mbuf *tx_pkt, struct hns3_tx_queue *txq)
 {
+       uint8_t max_non_tso_bd_num;
        struct rte_mbuf *new_pkt;
        int ret;
 
@@ -2836,15 +3858,17 @@ hns3_check_non_tso_pkt(uint16_t nb_buf, struct rte_mbuf **m_seg,
         * driver support, the packet will be ignored.
         */
        if (unlikely(rte_pktmbuf_pkt_len(tx_pkt) > HNS3_MAX_FRAME_LEN)) {
-               txq->over_length_pkt_cnt++;
+               txq->dfx_stats.over_length_pkt_cnt++;
                return -EINVAL;
        }
 
-       if (unlikely(nb_buf > HNS3_MAX_NON_TSO_BD_PER_PKT)) {
-               txq->exceed_limit_bd_pkt_cnt++;
-               ret = hns3_reassemble_tx_pkts(txq, tx_pkt, &new_pkt);
+       max_non_tso_bd_num = txq->max_non_tso_bd_num;
+       if (unlikely(nb_buf > max_non_tso_bd_num)) {
+               txq->dfx_stats.exceed_limit_bd_pkt_cnt++;
+               ret = hns3_reassemble_tx_pkts(tx_pkt, &new_pkt,
+                                             max_non_tso_bd_num);
                if (ret) {
-                       txq->exceed_limit_bd_reassem_fail++;
+                       txq->dfx_stats.exceed_limit_bd_reassem_fail++;
                        return ret;
                }
                *m_seg = new_pkt;
@@ -2920,7 +3944,7 @@ hns3_tx_setup_4bd(struct hns3_desc *txdp, struct rte_mbuf **pkts)
                dma_addr = rte_mbuf_data_iova(*pkts);
                txdp->addr = rte_cpu_to_le_64(dma_addr);
                txdp->tx.send_size = rte_cpu_to_le_16((*pkts)->data_len);
-               txdp->tx.paylen = 0;
+               txdp->tx.paylen_fd_dop_ol4cs = 0;
                txdp->tx.type_cs_vlan_tso_len = 0;
                txdp->tx.ol_type_vlan_len_msec = 0;
                txdp->tx.tp_fe_sc_vld_ra_ri = rte_cpu_to_le_16(bd_flag);
@@ -2936,7 +3960,7 @@ hns3_tx_setup_1bd(struct hns3_desc *txdp, struct rte_mbuf **pkts)
        dma_addr = rte_mbuf_data_iova(*pkts);
        txdp->addr = rte_cpu_to_le_64(dma_addr);
        txdp->tx.send_size = rte_cpu_to_le_16((*pkts)->data_len);
-       txdp->tx.paylen = 0;
+       txdp->tx.paylen_fd_dop_ol4cs = 0;
        txdp->tx.type_cs_vlan_tso_len = 0;
        txdp->tx.ol_type_vlan_len_msec = 0;
        txdp->tx.tp_fe_sc_vld_ra_ri = rte_cpu_to_le_16(bd_flag);
@@ -2958,6 +3982,11 @@ hns3_tx_fill_hw_ring(struct hns3_tx_queue *txq,
        for (i = 0; i < mainpart; i += PER_LOOP_NUM) {
                hns3_tx_backup_4mbuf(tx_entry + i, pkts + i);
                hns3_tx_setup_4bd(txdp + i, pkts + i);
+
+               /* Increment bytes counter */
+               uint32_t j;
+               for (j = 0; j < PER_LOOP_NUM; j++)
+                       txq->basic_stats.bytes += pkts[i + j]->pkt_len;
        }
        if (unlikely(leftover > 0)) {
                for (i = 0; i < leftover; i++) {
@@ -2965,6 +3994,9 @@ hns3_tx_fill_hw_ring(struct hns3_tx_queue *txq,
                                             pkts + mainpart + i);
                        hns3_tx_setup_1bd(txdp + mainpart + i,
                                          pkts + mainpart + i);
+
+                       /* Increment bytes counter */
+                       txq->basic_stats.bytes += pkts[mainpart + i]->pkt_len;
                }
        }
 }
@@ -2982,7 +4014,7 @@ hns3_xmit_pkts_simple(void *tx_queue,
        nb_pkts = RTE_MIN(txq->tx_bd_ready, nb_pkts);
        if (unlikely(nb_pkts == 0)) {
                if (txq->tx_bd_ready == 0)
-                       txq->queue_full_cnt++;
+                       txq->dfx_stats.queue_full_cnt++;
                return 0;
        }
 
@@ -3004,7 +4036,6 @@ hns3_xmit_pkts_simple(void *tx_queue,
 uint16_t
 hns3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 {
-       struct rte_net_hdr_lens hdr_lens = {0};
        struct hns3_tx_queue *txq = tx_queue;
        struct hns3_entry *tx_bak_pkt;
        struct hns3_desc *tx_ring;
@@ -3035,7 +4066,7 @@ hns3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                nb_buf = tx_pkt->nb_segs;
 
                if (nb_buf > txq->tx_bd_ready) {
-                       txq->queue_full_cnt++;
+                       txq->dfx_stats.queue_full_cnt++;
                        if (nb_tx == 0)
                                return 0;
 
@@ -3056,7 +4087,7 @@ hns3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                                         rte_pktmbuf_pkt_len(tx_pkt);
                        appended = rte_pktmbuf_append(tx_pkt, add_len);
                        if (appended == NULL) {
-                               txq->pkt_padding_fail_cnt++;
+                               txq->dfx_stats.pkt_padding_fail_cnt++;
                                break;
                        }
 
@@ -3068,7 +4099,7 @@ hns3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                if (hns3_check_non_tso_pkt(nb_buf, &m_seg, tx_pkt, txq))
                        goto end_of_tx;
 
-               if (hns3_parse_cksum(txq, tx_next_use, m_seg, &hdr_lens))
+               if (hns3_parse_cksum(txq, tx_next_use, m_seg))
                        goto end_of_tx;
 
                i = 0;
@@ -3104,6 +4135,8 @@ hns3_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
                desc->tx.tp_fe_sc_vld_ra_ri |=
                                 rte_cpu_to_le_16(BIT(HNS3_TXD_FE_B));
 
+               /* Increment bytes counter */
+               txq->basic_stats.bytes += tx_pkt->pkt_len;
                nb_hold += i;
                txq->next_to_use = tx_next_use;
                txq->tx_bd_ready -= i;
@@ -3131,6 +4164,14 @@ hns3_xmit_pkts_vec(__rte_unused void *tx_queue,
        return 0;
 }
 
+uint16_t __rte_weak
+hns3_xmit_pkts_vec_sve(void __rte_unused * tx_queue,
+                      struct rte_mbuf __rte_unused **tx_pkts,
+                      uint16_t __rte_unused nb_pkts)
+{
+       return 0;
+}
+
 int
 hns3_tx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
                       struct rte_eth_burst_mode *mode)
@@ -3144,6 +4185,8 @@ hns3_tx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
                info = "Scalar";
        else if (pkt_burst == hns3_xmit_pkts_vec)
                info = "Vector Neon";
+       else if (pkt_burst == hns3_xmit_pkts_vec_sve)
+               info = "Vector Sve";
 
        if (info == NULL)
                return -EINVAL;
@@ -3153,22 +4196,47 @@ hns3_tx_burst_mode_get(struct rte_eth_dev *dev, __rte_unused uint16_t queue_id,
        return 0;
 }
 
+static bool
+hns3_tx_check_simple_support(struct rte_eth_dev *dev)
+{
+       uint64_t offloads = dev->data->dev_conf.txmode.offloads;
+
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       if (hns3_dev_ptp_supported(hw))
+               return false;
+
+       return (offloads == (offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE));
+}
+
 static eth_tx_burst_t
 hns3_get_tx_function(struct rte_eth_dev *dev, eth_tx_prep_t *prep)
 {
-       uint64_t offloads = dev->data->dev_conf.txmode.offloads;
        struct hns3_adapter *hns = dev->data->dev_private;
+       bool vec_allowed, sve_allowed, simple_allowed;
+       bool vec_support;
+
+       vec_support = hns3_tx_check_vec_support(dev) == 0;
+       vec_allowed = vec_support && hns3_get_default_vec_support();
+       sve_allowed = vec_support && hns3_get_sve_support();
+       simple_allowed = hns3_tx_check_simple_support(dev);
 
-       if (hns->tx_vec_allowed && hns3_tx_check_vec_support(dev) == 0) {
-               *prep = NULL;
+       *prep = NULL;
+
+       if (hns->tx_func_hint == HNS3_IO_FUNC_HINT_VEC && vec_allowed)
                return hns3_xmit_pkts_vec;
+       if (hns->tx_func_hint == HNS3_IO_FUNC_HINT_SVE && sve_allowed)
+               return hns3_xmit_pkts_vec_sve;
+       if (hns->tx_func_hint == HNS3_IO_FUNC_HINT_SIMPLE && simple_allowed)
+               return hns3_xmit_pkts_simple;
+       if (hns->tx_func_hint == HNS3_IO_FUNC_HINT_COMMON) {
+               *prep = hns3_prep_pkts;
+               return hns3_xmit_pkts;
        }
 
-       if (hns->tx_simple_allowed &&
-           offloads == (offloads & DEV_TX_OFFLOAD_MBUF_FAST_FREE)) {
-               *prep = NULL;
+       if (vec_allowed)
+               return hns3_xmit_pkts_vec;
+       if (simple_allowed)
                return hns3_xmit_pkts_simple;
-       }
 
        *prep = hns3_prep_pkts;
        return hns3_xmit_pkts;
@@ -3182,16 +4250,35 @@ hns3_dummy_rxtx_burst(void *dpdk_txq __rte_unused,
        return 0;
 }
 
+static void
+hns3_trace_rxtx_function(struct rte_eth_dev *dev)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct rte_eth_burst_mode rx_mode;
+       struct rte_eth_burst_mode tx_mode;
+
+       memset(&rx_mode, 0, sizeof(rx_mode));
+       memset(&tx_mode, 0, sizeof(tx_mode));
+       (void)hns3_rx_burst_mode_get(dev, 0, &rx_mode);
+       (void)hns3_tx_burst_mode_get(dev, 0, &tx_mode);
+
+       hns3_dbg(hw, "using rx_pkt_burst: %s, tx_pkt_burst: %s.",
+                rx_mode.info, tx_mode.info);
+}
+
 void hns3_set_rxtx_function(struct rte_eth_dev *eth_dev)
 {
        struct hns3_adapter *hns = eth_dev->data->dev_private;
        eth_tx_prep_t prep = NULL;
 
        if (hns->hw.adapter_state == HNS3_NIC_STARTED &&
-           rte_atomic16_read(&hns->hw.reset.resetting) == 0) {
+           __atomic_load_n(&hns->hw.reset.resetting, __ATOMIC_RELAXED) == 0) {
                eth_dev->rx_pkt_burst = hns3_get_rx_function(eth_dev);
+               eth_dev->rx_descriptor_status = hns3_dev_rx_descriptor_status;
                eth_dev->tx_pkt_burst = hns3_get_tx_function(eth_dev, &prep);
                eth_dev->tx_pkt_prepare = prep;
+               eth_dev->tx_descriptor_status = hns3_dev_tx_descriptor_status;
+               hns3_trace_rxtx_function(eth_dev);
        } else {
                eth_dev->rx_pkt_burst = hns3_dummy_rxtx_burst;
                eth_dev->tx_pkt_burst = hns3_dummy_rxtx_burst;
@@ -3208,6 +4295,8 @@ hns3_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
        qinfo->mp = rxq->mb_pool;
        qinfo->nb_desc = rxq->nb_rx_desc;
        qinfo->scattered_rx = dev->data->scattered_rx;
+       /* Report the HW Rx buffer length to user */
+       qinfo->rx_buf_size = rxq->rx_buf_len;
 
        /*
         * If there are no available Rx buffer descriptors, incoming packets
@@ -3231,3 +4320,282 @@ hns3_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
        qinfo->conf.tx_free_thresh = txq->tx_free_thresh;
        qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
 }
+
+int
+hns3_dev_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct hns3_rx_queue *rxq = dev->data->rx_queues[rx_queue_id];
+       struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
+       int ret;
+
+       if (!hns3_dev_indep_txrx_supported(hw))
+               return -ENOTSUP;
+
+       rte_spinlock_lock(&hw->lock);
+       ret = hns3_reset_queue(hw, rx_queue_id, HNS3_RING_TYPE_RX);
+       if (ret) {
+               hns3_err(hw, "fail to reset Rx queue %u, ret = %d.",
+                        rx_queue_id, ret);
+               rte_spinlock_unlock(&hw->lock);
+               return ret;
+       }
+
+       ret = hns3_init_rxq(hns, rx_queue_id);
+       if (ret) {
+               hns3_err(hw, "fail to init Rx queue %u, ret = %d.",
+                        rx_queue_id, ret);
+               rte_spinlock_unlock(&hw->lock);
+               return ret;
+       }
+
+       hns3_enable_rxq(rxq, true);
+       dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+       rte_spinlock_unlock(&hw->lock);
+
+       return ret;
+}
+
+static void
+hns3_reset_sw_rxq(struct hns3_rx_queue *rxq)
+{
+       rxq->next_to_use = 0;
+       rxq->rx_rearm_start = 0;
+       rxq->rx_free_hold = 0;
+       rxq->rx_rearm_nb = 0;
+       rxq->pkt_first_seg = NULL;
+       rxq->pkt_last_seg = NULL;
+       memset(&rxq->rx_ring[0], 0, rxq->nb_rx_desc * sizeof(struct hns3_desc));
+       hns3_rxq_vec_setup(rxq);
+}
+
+int
+hns3_dev_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct hns3_rx_queue *rxq = dev->data->rx_queues[rx_queue_id];
+
+       if (!hns3_dev_indep_txrx_supported(hw))
+               return -ENOTSUP;
+
+       rte_spinlock_lock(&hw->lock);
+       hns3_enable_rxq(rxq, false);
+
+       hns3_rx_queue_release_mbufs(rxq);
+
+       hns3_reset_sw_rxq(rxq);
+       dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+       rte_spinlock_unlock(&hw->lock);
+
+       return 0;
+}
+
+int
+hns3_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct hns3_tx_queue *txq = dev->data->tx_queues[tx_queue_id];
+       int ret;
+
+       if (!hns3_dev_indep_txrx_supported(hw))
+               return -ENOTSUP;
+
+       rte_spinlock_lock(&hw->lock);
+       ret = hns3_reset_queue(hw, tx_queue_id, HNS3_RING_TYPE_TX);
+       if (ret) {
+               hns3_err(hw, "fail to reset Tx queue %u, ret = %d.",
+                        tx_queue_id, ret);
+               rte_spinlock_unlock(&hw->lock);
+               return ret;
+       }
+
+       hns3_init_txq(txq);
+       hns3_enable_txq(txq, true);
+       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+       rte_spinlock_unlock(&hw->lock);
+
+       return ret;
+}
+
+int
+hns3_dev_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct hns3_tx_queue *txq = dev->data->tx_queues[tx_queue_id];
+
+       if (!hns3_dev_indep_txrx_supported(hw))
+               return -ENOTSUP;
+
+       rte_spinlock_lock(&hw->lock);
+       hns3_enable_txq(txq, false);
+       hns3_tx_queue_release_mbufs(txq);
+       /*
+        * All the mbufs in sw_ring are released and all the pointers in sw_ring
+        * are set to NULL. If this queue is still called by upper layer,
+        * residual SW status of this txq may cause these pointers in sw_ring
+        * which have been set to NULL to be released again. To avoid it,
+        * reinit the txq.
+        */
+       hns3_init_txq(txq);
+       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+       rte_spinlock_unlock(&hw->lock);
+
+       return 0;
+}
+
+static int
+hns3_tx_done_cleanup_full(struct hns3_tx_queue *txq, uint32_t free_cnt)
+{
+       uint16_t next_to_clean = txq->next_to_clean;
+       uint16_t next_to_use   = txq->next_to_use;
+       uint16_t tx_bd_ready   = txq->tx_bd_ready;
+       struct hns3_entry *tx_pkt = &txq->sw_ring[next_to_clean];
+       struct hns3_desc *desc = &txq->tx_ring[next_to_clean];
+       uint32_t idx;
+
+       if (free_cnt == 0 || free_cnt > txq->nb_tx_desc)
+               free_cnt = txq->nb_tx_desc;
+
+       for (idx = 0; idx < free_cnt; idx++) {
+               if (next_to_clean == next_to_use)
+                       break;
+
+               if (desc->tx.tp_fe_sc_vld_ra_ri &
+                   rte_cpu_to_le_16(BIT(HNS3_TXD_VLD_B)))
+                       break;
+
+               if (tx_pkt->mbuf != NULL) {
+                       rte_pktmbuf_free_seg(tx_pkt->mbuf);
+                       tx_pkt->mbuf = NULL;
+               }
+
+               next_to_clean++;
+               tx_bd_ready++;
+               tx_pkt++;
+               desc++;
+               if (next_to_clean == txq->nb_tx_desc) {
+                       tx_pkt = txq->sw_ring;
+                       desc = txq->tx_ring;
+                       next_to_clean = 0;
+               }
+       }
+
+       if (idx > 0) {
+               txq->next_to_clean = next_to_clean;
+               txq->tx_bd_ready = tx_bd_ready;
+       }
+
+       return (int)idx;
+}
+
+int
+hns3_tx_done_cleanup(void *txq, uint32_t free_cnt)
+{
+       struct hns3_tx_queue *q = (struct hns3_tx_queue *)txq;
+       struct rte_eth_dev *dev = &rte_eth_devices[q->port_id];
+
+       if (dev->tx_pkt_burst == hns3_xmit_pkts)
+               return hns3_tx_done_cleanup_full(q, free_cnt);
+       else if (dev->tx_pkt_burst == hns3_dummy_rxtx_burst)
+               return 0;
+       else
+               return -ENOTSUP;
+}
+
+int
+hns3_dev_rx_descriptor_status(void *rx_queue, uint16_t offset)
+{
+       volatile struct hns3_desc *rxdp;
+       struct hns3_rx_queue *rxq;
+       struct rte_eth_dev *dev;
+       uint32_t bd_base_info;
+       uint16_t desc_id;
+
+       rxq = (struct hns3_rx_queue *)rx_queue;
+       if (offset >= rxq->nb_rx_desc)
+               return -EINVAL;
+
+       desc_id = (rxq->next_to_use + offset) % rxq->nb_rx_desc;
+       rxdp = &rxq->rx_ring[desc_id];
+       bd_base_info = rte_le_to_cpu_32(rxdp->rx.bd_base_info);
+       dev = &rte_eth_devices[rxq->port_id];
+       if (dev->rx_pkt_burst == hns3_recv_pkts_simple ||
+           dev->rx_pkt_burst == hns3_recv_scattered_pkts) {
+               if (offset >= rxq->nb_rx_desc - rxq->rx_free_hold)
+                       return RTE_ETH_RX_DESC_UNAVAIL;
+       } else if (dev->rx_pkt_burst == hns3_recv_pkts_vec ||
+                  dev->rx_pkt_burst == hns3_recv_pkts_vec_sve) {
+               if (offset >= rxq->nb_rx_desc - rxq->rx_rearm_nb)
+                       return RTE_ETH_RX_DESC_UNAVAIL;
+       } else {
+               return RTE_ETH_RX_DESC_UNAVAIL;
+       }
+
+       if (!(bd_base_info & BIT(HNS3_RXD_VLD_B)))
+               return RTE_ETH_RX_DESC_AVAIL;
+       else
+               return RTE_ETH_RX_DESC_DONE;
+}
+
+int
+hns3_dev_tx_descriptor_status(void *tx_queue, uint16_t offset)
+{
+       volatile struct hns3_desc *txdp;
+       struct hns3_tx_queue *txq;
+       struct rte_eth_dev *dev;
+       uint16_t desc_id;
+
+       txq = (struct hns3_tx_queue *)tx_queue;
+       if (offset >= txq->nb_tx_desc)
+               return -EINVAL;
+
+       dev = &rte_eth_devices[txq->port_id];
+       if (dev->tx_pkt_burst != hns3_xmit_pkts_simple &&
+           dev->tx_pkt_burst != hns3_xmit_pkts &&
+           dev->tx_pkt_burst != hns3_xmit_pkts_vec_sve &&
+           dev->tx_pkt_burst != hns3_xmit_pkts_vec)
+               return RTE_ETH_TX_DESC_UNAVAIL;
+
+       desc_id = (txq->next_to_use + offset) % txq->nb_tx_desc;
+       txdp = &txq->tx_ring[desc_id];
+       if (txdp->tx.tp_fe_sc_vld_ra_ri & rte_cpu_to_le_16(BIT(HNS3_TXD_VLD_B)))
+               return RTE_ETH_TX_DESC_FULL;
+       else
+               return RTE_ETH_TX_DESC_DONE;
+}
+
+uint32_t
+hns3_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+       /*
+        * Number of BDs that have been processed by the driver
+        * but have not been notified to the hardware.
+        */
+       uint32_t driver_hold_bd_num;
+       struct hns3_rx_queue *rxq;
+       uint32_t fbd_num;
+
+       rxq = dev->data->rx_queues[rx_queue_id];
+       fbd_num = hns3_read_dev(rxq, HNS3_RING_RX_FBDNUM_REG);
+       if (dev->rx_pkt_burst == hns3_recv_pkts_vec ||
+           dev->rx_pkt_burst == hns3_recv_pkts_vec_sve)
+               driver_hold_bd_num = rxq->rx_rearm_nb;
+       else
+               driver_hold_bd_num = rxq->rx_free_hold;
+
+       if (fbd_num <= driver_hold_bd_num)
+               return 0;
+       else
+               return fbd_num - driver_hold_bd_num;
+}
+
+void
+hns3_enable_rxd_adv_layout(struct hns3_hw *hw)
+{
+       /*
+        * If the hardware support rxd advanced layout, then driver enable it
+        * default.
+        */
+       if (hns3_dev_rxd_adv_layout_supported(hw))
+               hns3_write_dev(hw, HNS3_RXD_ADV_LAYOUT_EN_REG, 1);
+}