net/hns3: support traffic management
[dpdk.git] / drivers / net / hns3 / hns3_ethdev.c
index 99bcc7a..2bc28ef 100644 (file)
@@ -2,25 +2,9 @@
  * Copyright(c) 2018-2019 Hisilicon Limited.
  */
 
-#include <errno.h>
-#include <stdarg.h>
-#include <stdbool.h>
-#include <stdio.h>
-#include <stdint.h>
-#include <inttypes.h>
-#include <unistd.h>
-#include <rte_atomic.h>
+#include <rte_alarm.h>
 #include <rte_bus_pci.h>
-#include <rte_common.h>
-#include <rte_cycles.h>
-#include <rte_dev.h>
-#include <rte_eal.h>
-#include <rte_ether.h>
-#include <rte_ethdev_driver.h>
 #include <rte_ethdev_pci.h>
-#include <rte_interrupts.h>
-#include <rte_io.h>
-#include <rte_log.h>
 #include <rte_pci.h>
 
 #include "hns3_ethdev.h"
 #define HNS3_RESET_WAIT_MS     100
 #define HNS3_RESET_WAIT_CNT    200
 
+/* FEC mode order defined in HNS3 hardware */
+#define HNS3_HW_FEC_MODE_NOFEC  0
+#define HNS3_HW_FEC_MODE_BASER  1
+#define HNS3_HW_FEC_MODE_RS     2
+
 enum hns3_evt_cause {
        HNS3_VECTOR0_EVENT_RST,
        HNS3_VECTOR0_EVENT_MBX,
@@ -70,6 +59,34 @@ enum hns3_evt_cause {
        HNS3_VECTOR0_EVENT_OTHER,
 };
 
+static const struct rte_eth_fec_capa speed_fec_capa_tbl[] = {
+       { ETH_SPEED_NUM_10G, RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(AUTO) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(BASER) },
+
+       { ETH_SPEED_NUM_25G, RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(AUTO) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(BASER) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(RS) },
+
+       { ETH_SPEED_NUM_40G, RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(AUTO) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(BASER) },
+
+       { ETH_SPEED_NUM_50G, RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(AUTO) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(BASER) |
+                            RTE_ETH_FEC_MODE_CAPA_MASK(RS) },
+
+       { ETH_SPEED_NUM_100G, RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
+                             RTE_ETH_FEC_MODE_CAPA_MASK(AUTO) |
+                             RTE_ETH_FEC_MODE_CAPA_MASK(RS) },
+
+       { ETH_SPEED_NUM_200G, RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC) |
+                             RTE_ETH_FEC_MODE_CAPA_MASK(AUTO) |
+                             RTE_ETH_FEC_MODE_CAPA_MASK(RS) }
+};
+
 static enum hns3_reset_level hns3_get_reset_level(struct hns3_adapter *hns,
                                                 uint64_t *levels);
 static int hns3_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
@@ -81,6 +98,17 @@ static int hns3_add_mc_addr(struct hns3_hw *hw,
                            struct rte_ether_addr *mac_addr);
 static int hns3_remove_mc_addr(struct hns3_hw *hw,
                            struct rte_ether_addr *mac_addr);
+static int hns3_restore_fec(struct hns3_hw *hw);
+static int hns3_query_dev_fec_info(struct hns3_hw *hw);
+
+void hns3_ether_format_addr(char *buf, uint16_t size,
+                           const struct rte_ether_addr *ether_addr)
+{
+       snprintf(buf, size, "%02X:**:**:**:%02X:%02X",
+               ether_addr->addr_bytes[0],
+               ether_addr->addr_bytes[4],
+               ether_addr->addr_bytes[5]);
+}
 
 static void
 hns3_pf_disable_irq0(struct hns3_hw *hw)
@@ -218,7 +246,6 @@ hns3_interrupt_handler(void *param)
        hns3_pf_disable_irq0(hw);
 
        event_cause = hns3_check_event_cause(hns, &clearval);
-
        /* vector 0 interrupt is shared with reset and mailbox source events. */
        if (event_cause == HNS3_VECTOR0_EVENT_ERR) {
                hns3_warn(hw, "Received err interrupt");
@@ -989,7 +1016,7 @@ hns3_init_vlan_config(struct hns3_adapter *hns)
         * ensure that the hardware configuration remains unchanged before and
         * after reset.
         */
-       if (rte_atomic16_read(&hw->reset.resetting) == 0) {
+       if (__atomic_load_n(&hw->reset.resetting, __ATOMIC_RELAXED) == 0) {
                hw->port_base_vlan_cfg.state = HNS3_PORT_BASE_VLAN_DISABLE;
                hw->port_base_vlan_cfg.pvid = HNS3_INVALID_PVID;
        }
@@ -1013,7 +1040,7 @@ hns3_init_vlan_config(struct hns3_adapter *hns)
         * we will restore configurations to hardware in hns3_restore_vlan_table
         * and hns3_restore_vlan_conf later.
         */
-       if (rte_atomic16_read(&hw->reset.resetting) == 0) {
+       if (__atomic_load_n(&hw->reset.resetting, __ATOMIC_RELAXED) == 0) {
                ret = hns3_vlan_pvid_configure(hns, HNS3_INVALID_PVID, 0);
                if (ret) {
                        hns3_err(hw, "pvid set fail in pf, ret =%d", ret);
@@ -1080,8 +1107,8 @@ hns3_dev_configure_vlan(struct rte_eth_dev *dev)
                hns3_warn(hw,
                          "hw_vlan_reject_tagged or hw_vlan_reject_untagged "
                          "configuration is not supported! Ignore these two "
-                         "parameters: hw_vlan_reject_tagged(%d), "
-                         "hw_vlan_reject_untagged(%d)",
+                         "parameters: hw_vlan_reject_tagged(%u), "
+                         "hw_vlan_reject_untagged(%u)",
                          txmode->hw_vlan_reject_tagged,
                          txmode->hw_vlan_reject_untagged);
 
@@ -1105,7 +1132,7 @@ hns3_dev_configure_vlan(struct rte_eth_dev *dev)
        ret = hns3_vlan_pvid_set(dev, txmode->pvid,
                                 txmode->hw_vlan_insert_pvid);
        if (ret)
-               hns3_err(hw, "dev config vlan pvid(%d) failed, ret = %d",
+               hns3_err(hw, "dev config vlan pvid(%u) failed, ret = %d",
                         txmode->pvid, ret);
 
        return ret;
@@ -1430,7 +1457,7 @@ hns3_add_uc_addr_common(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
 
        /* check if mac addr is valid */
        if (!rte_is_valid_assigned_ether_addr(mac_addr)) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "Add unicast mac addr err! addr(%s) invalid",
                         mac_str);
@@ -1470,7 +1497,7 @@ hns3_add_uc_addr_common(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
                return -ENOSPC;
        }
 
-       rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, mac_addr);
+       hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, mac_addr);
 
        /* check if we just hit the duplicate */
        if (ret == 0) {
@@ -1496,7 +1523,7 @@ hns3_add_mc_addr_common(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
                addr = &hw->mc_addrs[i];
                /* Check if there are duplicate addresses */
                if (rte_is_same_ether_addr(addr, mac_addr)) {
-                       rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+                       hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                              addr);
                        hns3_err(hw, "failed to add mc mac addr, same addrs"
                                 "(%s) is added by the set_mc_mac_addr_list "
@@ -1507,7 +1534,7 @@ hns3_add_mc_addr_common(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
 
        ret = hns3_add_mc_addr(hw, mac_addr);
        if (ret) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "failed to add mc mac addr(%s), ret = %d",
                         mac_str, ret);
@@ -1523,7 +1550,7 @@ hns3_remove_mc_addr_common(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
 
        ret = hns3_remove_mc_addr(hw, mac_addr);
        if (ret) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "failed to remove mc mac addr(%s), ret = %d",
                         mac_str, ret);
@@ -1557,7 +1584,7 @@ hns3_add_mac_addr(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
 
        if (ret) {
                rte_spinlock_unlock(&hw->lock);
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "failed to add mac addr(%s), ret = %d", mac_str,
                         ret);
@@ -1580,7 +1607,7 @@ hns3_remove_uc_addr_common(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
 
        /* check if mac addr is valid */
        if (!rte_is_valid_assigned_ether_addr(mac_addr)) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "remove unicast mac addr err! addr(%s) invalid",
                         mac_str);
@@ -1616,7 +1643,7 @@ hns3_remove_mac_addr(struct rte_eth_dev *dev, uint32_t idx)
                ret = hns3_remove_uc_addr_common(hw, mac_addr);
        rte_spinlock_unlock(&hw->lock);
        if (ret) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "failed to remove mac addr(%s), ret = %d", mac_str,
                         ret);
@@ -1647,7 +1674,7 @@ hns3_set_default_mac_addr(struct rte_eth_dev *dev,
        if (default_addr_setted) {
                ret = hns3_remove_uc_addr_common(hw, oaddr);
                if (ret) {
-                       rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+                       hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                              oaddr);
                        hns3_warn(hw, "Remove old uc mac address(%s) fail: %d",
                                  mac_str, ret);
@@ -1658,7 +1685,7 @@ hns3_set_default_mac_addr(struct rte_eth_dev *dev,
 
        ret = hns3_add_uc_addr_common(hw, mac_addr);
        if (ret) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "Failed to set mac addr(%s): %d", mac_str, ret);
                goto err_add_uc_addr;
@@ -1680,7 +1707,7 @@ hns3_set_default_mac_addr(struct rte_eth_dev *dev,
 err_pause_addr_cfg:
        ret_val = hns3_remove_uc_addr_common(hw, mac_addr);
        if (ret_val) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_warn(hw,
                          "Failed to roll back to del setted mac addr(%s): %d",
@@ -1691,7 +1718,7 @@ err_add_uc_addr:
        if (rm_succes) {
                ret_val = hns3_add_uc_addr_common(hw, oaddr);
                if (ret_val) {
-                       rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+                       hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                              oaddr);
                        hns3_warn(hw,
                                  "Failed to restore old uc mac addr(%s): %d",
@@ -1727,7 +1754,7 @@ hns3_configure_all_mac_addr(struct hns3_adapter *hns, bool del)
 
                if (ret) {
                        err = ret;
-                       rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+                       hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                              addr);
                        hns3_err(hw, "failed to %s mac addr(%s) index:%d "
                                 "ret = %d.", del ? "remove" : "restore",
@@ -1776,7 +1803,7 @@ hns3_add_mc_addr(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
 
        /* Check if mac addr is valid */
        if (!rte_is_multicast_ether_addr(mac_addr)) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "failed to add mc mac addr, addr(%s) invalid",
                         mac_str);
@@ -1804,7 +1831,7 @@ hns3_add_mc_addr(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
        if (ret) {
                if (ret == -ENOSPC)
                        hns3_err(hw, "mc mac vlan table is full");
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "failed to add mc mac addr(%s): %d", mac_str, ret);
        }
@@ -1823,7 +1850,7 @@ hns3_remove_mc_addr(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
 
        /* Check if mac addr is valid */
        if (!rte_is_multicast_ether_addr(mac_addr)) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "Failed to rm mc mac addr, addr(%s) invalid",
                         mac_str);
@@ -1851,7 +1878,7 @@ hns3_remove_mc_addr(struct hns3_hw *hw, struct rte_ether_addr *mac_addr)
        }
 
        if (ret) {
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                      mac_addr);
                hns3_err(hw, "Failed to rm mc mac addr(%s): %d", mac_str, ret);
        }
@@ -1870,7 +1897,7 @@ hns3_set_mc_addr_chk_param(struct hns3_hw *hw,
        uint32_t j;
 
        if (nb_mc_addr > HNS3_MC_MACADDR_NUM) {
-               hns3_err(hw, "failed to set mc mac addr, nb_mc_addr(%d) "
+               hns3_err(hw, "failed to set mc mac addr, nb_mc_addr(%u) "
                         "invalid. valid range: 0~%d",
                         nb_mc_addr, HNS3_MC_MACADDR_NUM);
                return -EINVAL;
@@ -1880,7 +1907,7 @@ hns3_set_mc_addr_chk_param(struct hns3_hw *hw,
        for (i = 0; i < nb_mc_addr; i++) {
                addr = &mc_addr_set[i];
                if (!rte_is_multicast_ether_addr(addr)) {
-                       rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+                       hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                              addr);
                        hns3_err(hw,
                                 "failed to set mc mac addr, addr(%s) invalid.",
@@ -1891,7 +1918,7 @@ hns3_set_mc_addr_chk_param(struct hns3_hw *hw,
                /* Check if there are duplicate addresses */
                for (j = i + 1; j < nb_mc_addr; j++) {
                        if (rte_is_same_ether_addr(addr, &mc_addr_set[j])) {
-                               rte_ether_format_addr(mac_str,
+                               hns3_ether_format_addr(mac_str,
                                                      RTE_ETHER_ADDR_FMT_SIZE,
                                                      addr);
                                hns3_err(hw, "failed to set mc mac addr, "
@@ -1908,7 +1935,7 @@ hns3_set_mc_addr_chk_param(struct hns3_hw *hw,
                for (j = 0; j < HNS3_UC_MACADDR_NUM; j++) {
                        if (rte_is_same_ether_addr(addr,
                                                   &hw->data->mac_addrs[j])) {
-                               rte_ether_format_addr(mac_str,
+                               hns3_ether_format_addr(mac_str,
                                                      RTE_ETHER_ADDR_FMT_SIZE,
                                                      addr);
                                hns3_err(hw, "failed to set mc mac addr, "
@@ -2082,7 +2109,7 @@ hns3_configure_all_mc_mac_addr(struct hns3_adapter *hns, bool del)
                        ret = hns3_add_mc_addr(hw, addr);
                if (ret) {
                        err = ret;
-                       rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+                       hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                              addr);
                        hns3_dbg(hw, "%s mc mac addr: %s failed for pf: ret = %d",
                                 del ? "Remove" : "Restore", mac_str, ret);
@@ -2144,7 +2171,7 @@ hns3_check_mq_mode(struct rte_eth_dev *dev)
 
                for (i = 0; i < HNS3_MAX_USER_PRIO; i++) {
                        if (dcb_rx_conf->dcb_tc[i] != dcb_tx_conf->dcb_tc[i]) {
-                               hns3_err(hw, "dcb_tc[%d] = %d in rx direction, "
+                               hns3_err(hw, "dcb_tc[%d] = %u in rx direction, "
                                         "is not equal to one in tx direction.",
                                         i, dcb_rx_conf->dcb_tc[i]);
                                return -EINVAL;
@@ -2218,7 +2245,7 @@ hns3_bind_ring_with_vector(struct hns3_hw *hw, uint8_t vector_id, bool mmap,
        op_str = mmap ? "Map" : "Unmap";
        status = hns3_cmd_send(hw, &desc, 1);
        if (status) {
-               hns3_err(hw, "%s TQP %d fail, vector_id is %d, status is %d.",
+               hns3_err(hw, "%s TQP %u fail, vector_id is %u, status is %d.",
                         op_str, queue_id, req->int_vector_id, status);
                return status;
        }
@@ -2256,13 +2283,17 @@ hns3_init_ring_with_vector(struct hns3_hw *hw)
                hns3_set_queue_intr_gl(hw, i, HNS3_RING_GL_TX,
                                       HNS3_TQP_INTR_GL_DEFAULT);
                hns3_set_queue_intr_rl(hw, i, HNS3_TQP_INTR_RL_DEFAULT);
+               /*
+                * QL(quantity limiter) is not used currently, just set 0 to
+                * close it.
+                */
                hns3_set_queue_intr_ql(hw, i, HNS3_TQP_INTR_QL_DEFAULT);
 
                ret = hns3_bind_ring_with_vector(hw, vec, false,
                                                 HNS3_RING_TYPE_TX, i);
                if (ret) {
                        PMD_INIT_LOG(ERR, "PF fail to unbind TX ring(%d) with "
-                                         "vector: %d, ret=%d", i, vec, ret);
+                                         "vector: %u, ret=%d", i, vec, ret);
                        return ret;
                }
 
@@ -2270,7 +2301,7 @@ hns3_init_ring_with_vector(struct hns3_hw *hw)
                                                 HNS3_RING_TYPE_RX, i);
                if (ret) {
                        PMD_INIT_LOG(ERR, "PF fail to unbind RX ring(%d) with "
-                                         "vector: %d, ret=%d", i, vec, ret);
+                                         "vector: %u, ret=%d", i, vec, ret);
                        return ret;
                }
        }
@@ -2293,20 +2324,25 @@ hns3_dev_configure(struct rte_eth_dev *dev)
        bool gro_en;
        int ret;
 
+       hw->cfg_max_queues = RTE_MAX(nb_rx_q, nb_tx_q);
+
        /*
-        * Hardware does not support individually enable/disable/reset the Tx or
-        * Rx queue in hns3 network engine. Driver must enable/disable/reset Tx
-        * and Rx queues at the same time. When the numbers of Tx queues
-        * allocated by upper applications are not equal to the numbers of Rx
-        * queues, driver needs to setup fake Tx or Rx queues to adjust numbers
-        * of Tx/Rx queues. otherwise, network engine can not work as usual. But
-        * these fake queues are imperceptible, and can not be used by upper
-        * applications.
+        * Some versions of hardware network engine does not support
+        * individually enable/disable/reset the Tx or Rx queue. These devices
+        * must enable/disable/reset Tx and Rx queues at the same time. When the
+        * numbers of Tx queues allocated by upper applications are not equal to
+        * the numbers of Rx queues, driver needs to setup fake Tx or Rx queues
+        * to adjust numbers of Tx/Rx queues. otherwise, network engine can not
+        * work as usual. But these fake queues are imperceptible, and can not
+        * be used by upper applications.
         */
-       ret = hns3_set_fake_rx_or_tx_queues(dev, nb_rx_q, nb_tx_q);
-       if (ret) {
-               hns3_err(hw, "Failed to set rx/tx fake queues: %d", ret);
-               return ret;
+       if (!hns3_dev_indep_txrx_supported(hw)) {
+               ret = hns3_set_fake_rx_or_tx_queues(dev, nb_rx_q, nb_tx_q);
+               if (ret) {
+                       hns3_err(hw, "fail to set Rx/Tx fake queues, ret = %d.",
+                                ret);
+                       return ret;
+               }
        }
 
        hw->adapter_state = HNS3_NIC_CONFIGURING;
@@ -2430,7 +2466,7 @@ hns3_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
        }
 
        rte_spinlock_lock(&hw->lock);
-       is_jumbo_frame = frame_size > RTE_ETHER_MAX_LEN ? true : false;
+       is_jumbo_frame = frame_size > HNS3_DEFAULT_FRAME_LEN ? true : false;
        frame_size = RTE_MAX(frame_size, HNS3_DEFAULT_FRAME_LEN);
 
        /*
@@ -2457,7 +2493,7 @@ hns3_dev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
        return 0;
 }
 
-static int
+int
 hns3_dev_infos_get(struct rte_eth_dev *eth_dev, struct rte_eth_dev_info *info)
 {
        struct hns3_adapter *hns = eth_dev->data->dev_private;
@@ -2504,6 +2540,10 @@ hns3_dev_infos_get(struct rte_eth_dev *eth_dev, struct rte_eth_dev_info *info)
                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE |
                                 hns3_txvlan_cap_get(hw));
 
+       if (hns3_dev_indep_txrx_supported(hw))
+               info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
+                                RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
+
        info->rx_desc_lim = (struct rte_eth_desc_lim) {
                .nb_max = HNS3_MAX_RING_DESC,
                .nb_min = HNS3_MIN_RING_DESC,
@@ -2659,6 +2699,49 @@ hns3_query_function_status(struct hns3_hw *hw)
        return hns3_parse_func_status(hw, req);
 }
 
+static int
+hns3_get_pf_max_tqp_num(struct hns3_hw *hw)
+{
+       struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
+       struct hns3_pf *pf = &hns->pf;
+
+       if (pf->tqp_config_mode == HNS3_FLEX_MAX_TQP_NUM_MODE) {
+               /*
+                * The total_tqps_num obtained from firmware is maximum tqp
+                * numbers of this port, which should be used for PF and VFs.
+                * There is no need for pf to have so many tqp numbers in
+                * most cases. RTE_LIBRTE_HNS3_MAX_TQP_NUM_PER_PF,
+                * coming from config file, is assigned to maximum queue number
+                * for the PF of this port by user. So users can modify the
+                * maximum queue number of PF according to their own application
+                * scenarios, which is more flexible to use. In addition, many
+                * memories can be saved due to allocating queue statistics
+                * room according to the actual number of queues required. The
+                * maximum queue number of PF for network engine with
+                * revision_id greater than 0x30 is assigned by config file.
+                */
+               if (RTE_LIBRTE_HNS3_MAX_TQP_NUM_PER_PF <= 0) {
+                       hns3_err(hw, "RTE_LIBRTE_HNS3_MAX_TQP_NUM_PER_PF(%d) "
+                                "must be greater than 0.",
+                                RTE_LIBRTE_HNS3_MAX_TQP_NUM_PER_PF);
+                       return -EINVAL;
+               }
+
+               hw->tqps_num = RTE_MIN(RTE_LIBRTE_HNS3_MAX_TQP_NUM_PER_PF,
+                                      hw->total_tqps_num);
+       } else {
+               /*
+                * Due to the limitation on the number of PF interrupts
+                * available, the maximum queue number assigned to PF on
+                * the network engine with revision_id 0x21 is 64.
+                */
+               hw->tqps_num = RTE_MIN(hw->total_tqps_num,
+                                      HNS3_MAX_TQP_NUM_HIP08_PF);
+       }
+
+       return 0;
+}
+
 static int
 hns3_query_pf_resource(struct hns3_hw *hw)
 {
@@ -2676,9 +2759,13 @@ hns3_query_pf_resource(struct hns3_hw *hw)
        }
 
        req = (struct hns3_pf_res_cmd *)desc.data;
-       hw->total_tqps_num = rte_le_to_cpu_16(req->tqp_num);
+       hw->total_tqps_num = rte_le_to_cpu_16(req->tqp_num) +
+                            rte_le_to_cpu_16(req->ext_tqp_num);
+       ret = hns3_get_pf_max_tqp_num(hw);
+       if (ret)
+               return ret;
+
        pf->pkt_buf_size = rte_le_to_cpu_16(req->buf_size) << HNS3_BUF_UNIT_S;
-       hw->tqps_num = RTE_MIN(hw->total_tqps_num, HNS3_MAX_TQP_NUM_PER_FUNC);
        pf->func_num = rte_le_to_cpu_16(req->pf_own_fun_number);
 
        if (req->tx_buf_size)
@@ -2709,6 +2796,7 @@ hns3_parse_cfg(struct hns3_cfg *cfg, struct hns3_cmd_desc *desc)
 {
        struct hns3_cfg_param_cmd *req;
        uint64_t mac_addr_tmp_high;
+       uint8_t ext_rss_size_max;
        uint64_t mac_addr_tmp;
        uint32_t i;
 
@@ -2761,6 +2849,21 @@ hns3_parse_cfg(struct hns3_cfg *cfg, struct hns3_cmd_desc *desc)
                                        HNS3_CFG_UMV_TBL_SPACE_S);
        if (!cfg->umv_space)
                cfg->umv_space = HNS3_DEFAULT_UMV_SPACE_PER_PF;
+
+       ext_rss_size_max = hns3_get_field(rte_le_to_cpu_32(req->param[2]),
+                                              HNS3_CFG_EXT_RSS_SIZE_M,
+                                              HNS3_CFG_EXT_RSS_SIZE_S);
+
+       /*
+        * Field ext_rss_size_max obtained from firmware will be more flexible
+        * for future changes and expansions, which is an exponent of 2, instead
+        * of reading out directly. If this field is not zero, hns3 PF PMD
+        * driver uses it as rss_size_max under one TC. Device, whose revision
+        * id is greater than or equal to PCI_REVISION_ID_HIP09_A, obtains the
+        * maximum number of queues supported under a TC through this field.
+        */
+       if (ext_rss_size_max)
+               cfg->rss_size_max = 1U << ext_rss_size_max;
 }
 
 /* hns3_get_board_cfg: query the static parameter from NCL_config file in flash
@@ -2845,6 +2948,7 @@ hns3_set_default_dev_specifications(struct hns3_hw *hw)
        hw->rss_ind_tbl_size = HNS3_RSS_IND_TBL_SIZE;
        hw->rss_key_size = HNS3_RSS_KEY_SIZE;
        hw->max_tm_rate = HNS3_ETHER_MAX_RATE;
+       hw->intr.int_ql_max = HNS3_INTR_QL_NONE;
 }
 
 static void
@@ -2858,6 +2962,7 @@ hns3_parse_dev_specifications(struct hns3_hw *hw, struct hns3_cmd_desc *desc)
        hw->rss_ind_tbl_size = rte_le_to_cpu_16(req0->rss_ind_tbl_size);
        hw->rss_key_size = rte_le_to_cpu_16(req0->rss_key_size);
        hw->max_tm_rate = rte_le_to_cpu_32(req0->max_tm_rate);
+       hw->intr.int_ql_max = rte_le_to_cpu_16(req0->intr_ql_max);
 }
 
 static int
@@ -2886,7 +2991,9 @@ hns3_query_dev_specifications(struct hns3_hw *hw)
 static int
 hns3_get_capability(struct hns3_hw *hw)
 {
+       struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
        struct rte_pci_device *pci_dev;
+       struct hns3_pf *pf = &hns->pf;
        struct rte_eth_dev *eth_dev;
        uint16_t device_id;
        uint8_t revision;
@@ -2915,11 +3022,12 @@ hns3_get_capability(struct hns3_hw *hw)
        if (revision < PCI_REVISION_ID_HIP09_A) {
                hns3_set_default_dev_specifications(hw);
                hw->intr.mapping_mode = HNS3_INTR_MAPPING_VEC_RSV_ONE;
-               hw->intr.coalesce_mode = HNS3_INTR_COALESCE_NON_QL;
                hw->intr.gl_unit = HNS3_INTR_COALESCE_GL_UINT_2US;
                hw->tso_mode = HNS3_TSO_SW_CAL_PSEUDO_H_CSUM;
                hw->vlan_mode = HNS3_SW_SHIFT_AND_DISCARD_MODE;
                hw->min_tx_pkt_len = HNS3_HIP08_MIN_TX_PKT_LEN;
+               pf->tqp_config_mode = HNS3_FIXED_MAX_TQP_NUM_MODE;
+               hw->rss_info.ipv6_sctp_offload_supported = false;
                return 0;
        }
 
@@ -2932,11 +3040,12 @@ hns3_get_capability(struct hns3_hw *hw)
        }
 
        hw->intr.mapping_mode = HNS3_INTR_MAPPING_VEC_ALL;
-       hw->intr.coalesce_mode = HNS3_INTR_COALESCE_QL;
        hw->intr.gl_unit = HNS3_INTR_COALESCE_GL_UINT_1US;
        hw->tso_mode = HNS3_TSO_HW_CAL_PSEUDO_H_CSUM;
        hw->vlan_mode = HNS3_HW_SHIFT_AND_DISCARD_MODE;
        hw->min_tx_pkt_len = HNS3_HIP09_MIN_TX_PKT_LEN;
+       pf->tqp_config_mode = HNS3_FLEX_MAX_TQP_NUM_MODE;
+       hw->rss_info.ipv6_sctp_offload_supported = true;
 
        return 0;
 }
@@ -2974,7 +3083,7 @@ hns3_get_board_configuration(struct hns3_hw *hw)
 
        ret = hns3_parse_speed(cfg.default_speed, &hw->mac.link_speed);
        if (ret) {
-               PMD_INIT_LOG(ERR, "Get wrong speed %d, ret = %d",
+               PMD_INIT_LOG(ERR, "Get wrong speed %u, ret = %d",
                             cfg.default_speed, ret);
                return ret;
        }
@@ -3031,8 +3140,15 @@ hns3_get_configuration(struct hns3_hw *hw)
        }
 
        ret = hns3_get_board_configuration(hw);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "failed to get board configuration: %d", ret);
+               return ret;
+       }
+
+       ret = hns3_query_dev_fec_info(hw);
        if (ret)
-               PMD_INIT_LOG(ERR, "Failed to get board configuration: %d", ret);
+               PMD_INIT_LOG(ERR,
+                            "failed to query FEC information, ret = %d", ret);
 
        return ret;
 }
@@ -3065,29 +3181,18 @@ hns3_map_tqps_to_func(struct hns3_hw *hw, uint16_t func_id, uint16_t tqp_pid,
 static int
 hns3_map_tqp(struct hns3_hw *hw)
 {
-       uint16_t tqps_num = hw->total_tqps_num;
-       uint16_t func_id;
-       uint16_t tqp_id;
-       bool is_pf;
-       int num;
        int ret;
        int i;
 
        /*
-        * In current version VF is not supported when PF is driven by DPDK
-        * driver, so we allocate tqps to PF as much as possible.
+        * In current version, VF is not supported when PF is driven by DPDK
+        * driver, so we assign total tqps_num tqps allocated to this port
+        * to PF.
         */
-       tqp_id = 0;
-       num = DIV_ROUND_UP(hw->total_tqps_num, HNS3_MAX_TQP_NUM_PER_FUNC);
-       for (func_id = HNS3_PF_FUNC_ID; func_id < num; func_id++) {
-               is_pf = func_id == HNS3_PF_FUNC_ID ? true : false;
-               for (i = 0;
-                    i < HNS3_MAX_TQP_NUM_PER_FUNC && tqp_id < tqps_num; i++) {
-                       ret = hns3_map_tqps_to_func(hw, func_id, tqp_id++, i,
-                                                   is_pf);
-                       if (ret)
-                               return ret;
-               }
+       for (i = 0; i < hw->total_tqps_num; i++) {
+               ret = hns3_map_tqps_to_func(hw, HNS3_PF_FUNC_ID, i, i, true);
+               if (ret)
+                       return ret;
        }
 
        return 0;
@@ -3337,8 +3442,8 @@ hns3_is_rx_buf_ok(struct hns3_hw *hw, struct hns3_pkt_buf_alloc *buf_alloc,
                hi_thrd = shared_buf - pf->dv_buf_size;
 
                if (tc_num <= NEED_RESERVE_TC_NUM)
-                       hi_thrd = hi_thrd * BUF_RESERVE_PERCENT
-                                       / BUF_MAX_PERCENT;
+                       hi_thrd = hi_thrd * BUF_RESERVE_PERCENT /
+                                 BUF_MAX_PERCENT;
 
                if (tc_num)
                        hi_thrd = hi_thrd / tc_num;
@@ -3458,9 +3563,7 @@ hns3_drop_pfc_buf_till_fit(struct hns3_hw *hw,
        for (i = HNS3_MAX_TC_NUM - 1; i >= 0; i--) {
                priv = &buf_alloc->priv_buf[i];
                mask = BIT((uint8_t)i);
-
-               if (hw->hw_tc_map & mask &&
-                   hw->dcb_info.hw_pfc_map & mask) {
+               if (hw->hw_tc_map & mask && hw->dcb_info.hw_pfc_map & mask) {
                        /* Reduce the number of pfc TC with private buffer */
                        priv->wl.low = 0;
                        priv->enable = 0;
@@ -3514,7 +3617,6 @@ hns3_only_alloc_priv_buff(struct hns3_hw *hw,
 
        for (i = 0; i < HNS3_MAX_TC_NUM; i++) {
                priv = &buf_alloc->priv_buf[i];
-
                priv->enable = 0;
                priv->wl.low = 0;
                priv->wl.high = 0;
@@ -3830,7 +3932,7 @@ hns3_get_mac_ethertype_cmd_status(uint16_t cmdq_resp, uint8_t resp_code)
 
        if (cmdq_resp) {
                PMD_INIT_LOG(ERR,
-                            "cmdq execute failed for get_mac_ethertype_cmd_status, status=%d.\n",
+                            "cmdq execute failed for get_mac_ethertype_cmd_status, status=%u.\n",
                             cmdq_resp);
                return -EIO;
        }
@@ -3851,7 +3953,7 @@ hns3_get_mac_ethertype_cmd_status(uint16_t cmdq_resp, uint8_t resp_code)
                break;
        default:
                PMD_INIT_LOG(ERR,
-                            "add mac ethertype failed for undefined, code=%d.",
+                            "add mac ethertype failed for undefined, code=%u.",
                             resp_code);
                return_status = -EIO;
                break;
@@ -4009,7 +4111,7 @@ hns3_promisc_init(struct hns3_hw *hw)
                hns3_promisc_param_init(&param, false, false, false, func_id);
                ret = hns3_cmd_set_promisc_mode(hw, &param);
                if (ret) {
-                       PMD_INIT_LOG(ERR, "failed to clear vf:%d promisc mode,"
+                       PMD_INIT_LOG(ERR, "failed to clear vf:%u promisc mode,"
                                        " ret = %d", func_id, ret);
                        return ret;
                }
@@ -4223,6 +4325,7 @@ static int
 hns3_cfg_mac_speed_dup(struct hns3_hw *hw, uint32_t speed, uint8_t duplex)
 {
        struct hns3_mac *mac = &hw->mac;
+       uint32_t cur_speed = mac->link_speed;
        int ret;
 
        duplex = hns3_check_speed_dup(duplex, speed);
@@ -4234,6 +4337,13 @@ hns3_cfg_mac_speed_dup(struct hns3_hw *hw, uint32_t speed, uint8_t duplex)
                return ret;
 
        mac->link_speed = speed;
+       ret = hns3_dcb_port_shaper_cfg(hw);
+       if (ret) {
+               hns3_err(hw, "failed to configure port shaper, ret = %d.", ret);
+               mac->link_speed = cur_speed;
+               return ret;
+       }
+
        mac->link_duplex = duplex;
 
        return 0;
@@ -4542,17 +4652,21 @@ hns3_init_pf(struct rte_eth_dev *eth_dev)
                goto err_get_config;
        }
 
+       ret = hns3_tqp_stats_init(hw);
+       if (ret)
+               goto err_get_config;
+
        ret = hns3_init_hardware(hns);
        if (ret) {
                PMD_INIT_LOG(ERR, "Failed to init hardware: %d", ret);
-               goto err_get_config;
+               goto err_init_hw;
        }
 
        /* Initialize flow director filter list & hash */
        ret = hns3_fdir_filter_init(hns);
        if (ret) {
                PMD_INIT_LOG(ERR, "Failed to alloc hashmap for fdir: %d", ret);
-               goto err_hw_init;
+               goto err_fdir;
        }
 
        hns3_set_default_rss_args(hw);
@@ -4561,16 +4675,19 @@ hns3_init_pf(struct rte_eth_dev *eth_dev)
        if (ret) {
                PMD_INIT_LOG(ERR, "fail to enable hw error interrupts: %d",
                             ret);
-               goto err_fdir;
+               goto err_enable_intr;
        }
 
+       hns3_tm_conf_init(eth_dev);
+
        return 0;
 
-err_fdir:
+err_enable_intr:
        hns3_fdir_filter_uninit(hns);
-err_hw_init:
+err_fdir:
        hns3_uninit_umv_space(hw);
-
+err_init_hw:
+       hns3_tqp_stats_uninit(hw);
 err_get_config:
        hns3_pf_disable_irq0(hw);
        rte_intr_disable(&pci_dev->intr_handle);
@@ -4596,12 +4713,14 @@ hns3_uninit_pf(struct rte_eth_dev *eth_dev)
 
        PMD_INIT_FUNC_TRACE();
 
+       hns3_tm_conf_uninit(eth_dev);
        hns3_enable_hw_error_intr(hns, false);
        hns3_rss_uninit(hns);
        (void)hns3_config_gro(hw, false);
        hns3_promisc_uninit(hw);
        hns3_fdir_filter_uninit(hns);
        hns3_uninit_umv_space(hw);
+       hns3_tqp_stats_uninit(hw);
        hns3_pf_disable_irq0(hw);
        rte_intr_disable(&pci_dev->intr_handle);
        hns3_intr_unregister(&pci_dev->intr_handle, hns3_interrupt_handler,
@@ -4622,23 +4741,38 @@ hns3_do_start(struct hns3_adapter *hns, bool reset_queue)
        if (ret)
                return ret;
 
-       /* Enable queues */
-       ret = hns3_start_queues(hns, reset_queue);
+       /*
+        * The hns3_dcb_cfg_update may configure TM module, so
+        * hns3_tm_conf_update must called later.
+        */
+       ret = hns3_tm_conf_update(hw);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "failed to update tm conf, ret = %d.", ret);
+               return ret;
+       }
+
+       ret = hns3_init_queues(hns, reset_queue);
        if (ret) {
-               PMD_INIT_LOG(ERR, "Failed to start queues: %d", ret);
+               PMD_INIT_LOG(ERR, "failed to init queues, ret = %d.", ret);
                return ret;
        }
 
-       /* Enable MAC */
        ret = hns3_cfg_mac_mode(hw, true);
        if (ret) {
-               PMD_INIT_LOG(ERR, "Failed to enable MAC: %d", ret);
+               PMD_INIT_LOG(ERR, "failed to enable MAC, ret = %d", ret);
                goto err_config_mac_mode;
        }
        return 0;
 
 err_config_mac_mode:
-       hns3_stop_queues(hns, true);
+       hns3_dev_release_mbufs(hns);
+       /*
+        * Here is exception handling, hns3_reset_all_tqps will have the
+        * corresponding error message if it is handled incorrectly, so it is
+        * not necessary to check hns3_reset_all_tqps return value, here keep
+        * ret as the error code causing the exception.
+        */
+       (void)hns3_reset_all_tqps(hns);
        return ret;
 }
 
@@ -4673,7 +4807,7 @@ hns3_map_rx_interrupt(struct rte_eth_dev *dev)
                        rte_zmalloc("intr_vec",
                                    hw->used_rx_queues * sizeof(int), 0);
                if (intr_handle->intr_vec == NULL) {
-                       hns3_err(hw, "Failed to allocate %d rx_queues"
+                       hns3_err(hw, "Failed to allocate %u rx_queues"
                                     " intr_vec", hw->used_rx_queues);
                        ret = -ENOMEM;
                        goto alloc_intr_vec_error;
@@ -4750,7 +4884,7 @@ hns3_dev_start(struct rte_eth_dev *dev)
        int ret;
 
        PMD_INIT_FUNC_TRACE();
-       if (rte_atomic16_read(&hw->reset.resetting))
+       if (__atomic_load_n(&hw->reset.resetting, __ATOMIC_RELAXED))
                return -EBUSY;
 
        rte_spinlock_lock(&hw->lock);
@@ -4769,6 +4903,32 @@ hns3_dev_start(struct rte_eth_dev *dev)
                return ret;
        }
 
+       /*
+        * There are three register used to control the status of a TQP
+        * (contains a pair of Tx queue and Rx queue) in the new version network
+        * engine. One is used to control the enabling of Tx queue, the other is
+        * used to control the enabling of Rx queue, and the last is the master
+        * switch used to control the enabling of the tqp. The Tx register and
+        * TQP register must be enabled at the same time to enable a Tx queue.
+        * The same applies to the Rx queue. For the older network engine, this
+        * function only refresh the enabled flag, and it is used to update the
+        * status of queue in the dpdk framework.
+        */
+       ret = hns3_start_all_txqs(dev);
+       if (ret) {
+               hw->adapter_state = HNS3_NIC_CONFIGURED;
+               rte_spinlock_unlock(&hw->lock);
+               return ret;
+       }
+
+       ret = hns3_start_all_rxqs(dev);
+       if (ret) {
+               hns3_stop_all_txqs(dev);
+               hw->adapter_state = HNS3_NIC_CONFIGURED;
+               rte_spinlock_unlock(&hw->lock);
+               return ret;
+       }
+
        hw->adapter_state = HNS3_NIC_STARTED;
        rte_spinlock_unlock(&hw->lock);
 
@@ -4781,11 +4941,14 @@ hns3_dev_start(struct rte_eth_dev *dev)
 
        /* Enable interrupt of all rx queues before enabling queues */
        hns3_dev_all_rx_queue_intr_enable(hw, true);
+
        /*
-        * When finished the initialization, enable queues to receive/transmit
-        * packets.
+        * After finished the initialization, enable tqps to receive/transmit
+        * packets and refresh all queue status.
         */
-       hns3_enable_all_queues(hw, true);
+       hns3_start_tqps(hw);
+
+       hns3_tm_dev_start_proc(hw);
 
        hns3_info(hw, "hns3 dev start successful!");
        return 0;
@@ -4795,7 +4958,6 @@ static int
 hns3_do_stop(struct hns3_adapter *hns)
 {
        struct hns3_hw *hw = &hns->hw;
-       bool reset_queue;
        int ret;
 
        ret = hns3_cfg_mac_mode(hw, false);
@@ -4805,11 +4967,15 @@ hns3_do_stop(struct hns3_adapter *hns)
 
        if (rte_atomic16_read(&hw->reset.disable_cmd) == 0) {
                hns3_configure_all_mac_addr(hns, true);
-               reset_queue = true;
-       } else
-               reset_queue = false;
+               ret = hns3_reset_all_tqps(hns);
+               if (ret) {
+                       hns3_err(hw, "failed to reset all queues ret = %d.",
+                                ret);
+                       return ret;
+               }
+       }
        hw->mac.default_addr_setted = false;
-       return hns3_stop_queues(hns, reset_queue);
+       return 0;
 }
 
 static void
@@ -4848,13 +5014,14 @@ hns3_unmap_rx_interrupt(struct rte_eth_dev *dev)
        }
 }
 
-static void
+static int
 hns3_dev_stop(struct rte_eth_dev *dev)
 {
        struct hns3_adapter *hns = dev->data->dev_private;
        struct hns3_hw *hw = &hns->hw;
 
        PMD_INIT_FUNC_TRACE();
+       dev->data->dev_started = 0;
 
        hw->adapter_state = HNS3_NIC_STOPPING;
        hns3_set_rxtx_function(dev);
@@ -4865,7 +5032,9 @@ hns3_dev_stop(struct rte_eth_dev *dev)
        rte_delay_ms(hw->tqps_num);
 
        rte_spinlock_lock(&hw->lock);
-       if (rte_atomic16_read(&hw->reset.resetting) == 0) {
+       if (__atomic_load_n(&hw->reset.resetting, __ATOMIC_RELAXED) == 0) {
+               hns3_tm_dev_stop_proc(hw);
+               hns3_stop_tqps(hw);
                hns3_do_stop(hns);
                hns3_unmap_rx_interrupt(dev);
                hns3_dev_release_mbufs(hns);
@@ -4874,22 +5043,25 @@ hns3_dev_stop(struct rte_eth_dev *dev)
        hns3_rx_scattered_reset(dev);
        rte_eal_alarm_cancel(hns3_service_handler, dev);
        rte_spinlock_unlock(&hw->lock);
+
+       return 0;
 }
 
-static void
+static int
 hns3_dev_close(struct rte_eth_dev *eth_dev)
 {
        struct hns3_adapter *hns = eth_dev->data->dev_private;
        struct hns3_hw *hw = &hns->hw;
+       int ret = 0;
 
        if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
                rte_free(eth_dev->process_private);
                eth_dev->process_private = NULL;
-               return;
+               return 0;
        }
 
        if (hw->adapter_state == HNS3_NIC_STARTED)
-               hns3_dev_stop(eth_dev);
+               ret = hns3_dev_stop(eth_dev);
 
        hw->adapter_state = HNS3_NIC_CLOSING;
        hns3_reset_abort(hns);
@@ -4904,7 +5076,9 @@ hns3_dev_close(struct rte_eth_dev *eth_dev)
        rte_free(eth_dev->process_private);
        eth_dev->process_private = NULL;
        hns3_mp_uninit_primary();
-       hns3_warn(hw, "Close port %d finished", hw->data->port_id);
+       hns3_warn(hw, "Close port %u finished", hw->data->port_id);
+
+       return ret;
 }
 
 static int
@@ -4980,7 +5154,7 @@ hns3_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
                return -EINVAL;
        }
        if (!fc_conf->pause_time) {
-               hns3_err(hw, "Invalid pause time %d setting.",
+               hns3_err(hw, "Invalid pause time %u setting.",
                         fc_conf->pause_time);
                return -EINVAL;
        }
@@ -5033,7 +5207,7 @@ hns3_priority_flow_ctrl_set(struct rte_eth_dev *dev,
                return -EINVAL;
        }
        if (pfc_conf->fc.pause_time == 0) {
-               hns3_err(hw, "Invalid pause time %d setting.",
+               hns3_err(hw, "Invalid pause time %u setting.",
                         pfc_conf->fc.pause_time);
                return -EINVAL;
        }
@@ -5103,7 +5277,7 @@ hns3_reinit_dev(struct hns3_adapter *hns)
                return ret;
        }
 
-       ret = hns3_reset_all_queues(hns);
+       ret = hns3_reset_all_tqps(hns);
        if (ret) {
                hns3_err(hw, "Failed to reset all queues: %d", ret);
                return ret;
@@ -5381,6 +5555,7 @@ hns3_stop_service(struct hns3_adapter *hns)
        rte_spinlock_lock(&hw->lock);
        if (hns->hw.adapter_state == HNS3_NIC_STARTED ||
            hw->adapter_state == HNS3_NIC_STOPPING) {
+               hns3_enable_all_queues(hw, false);
                hns3_do_stop(hns);
                hw->reset.mbuf_deferred_free = true;
        } else
@@ -5415,6 +5590,11 @@ hns3_start_service(struct hns3_adapter *hns)
 
                /* Enable interrupt of all rx queues before enabling queues */
                hns3_dev_all_rx_queue_intr_enable(hw, true);
+               /*
+                * Enable state of each rxq and txq will be recovered after
+                * reset, so we need to restore them before enable all tqps;
+                */
+               hns3_restore_tqp_enable_state(hw);
                /*
                 * When finished the initialization, enable queues to receive
                 * and transmit packets.
@@ -5463,6 +5643,10 @@ hns3_restore_conf(struct hns3_adapter *hns)
        if (ret)
                goto err_promisc;
 
+       ret = hns3_restore_fec(hw);
+       if (ret)
+               goto err_promisc;
+
        if (hns->hw.adapter_state == HNS3_NIC_STARTED) {
                ret = hns3_do_start(hns, false);
                if (ret)
@@ -5541,6 +5725,329 @@ hns3_reset_service(void *param)
                hns3_msix_process(hns, reset_level);
 }
 
+static unsigned int
+hns3_get_speed_capa_num(uint16_t device_id)
+{
+       unsigned int num;
+
+       switch (device_id) {
+       case HNS3_DEV_ID_25GE:
+       case HNS3_DEV_ID_25GE_RDMA:
+               num = 2;
+               break;
+       case HNS3_DEV_ID_100G_RDMA_MACSEC:
+       case HNS3_DEV_ID_200G_RDMA:
+               num = 1;
+               break;
+       default:
+               num = 0;
+               break;
+       }
+
+       return num;
+}
+
+static int
+hns3_get_speed_fec_capa(struct rte_eth_fec_capa *speed_fec_capa,
+                       uint16_t device_id)
+{
+       switch (device_id) {
+       case HNS3_DEV_ID_25GE:
+       /* fallthrough */
+       case HNS3_DEV_ID_25GE_RDMA:
+               speed_fec_capa[0].speed = speed_fec_capa_tbl[1].speed;
+               speed_fec_capa[0].capa = speed_fec_capa_tbl[1].capa;
+
+               /* In HNS3 device, the 25G NIC is compatible with 10G rate */
+               speed_fec_capa[1].speed = speed_fec_capa_tbl[0].speed;
+               speed_fec_capa[1].capa = speed_fec_capa_tbl[0].capa;
+               break;
+       case HNS3_DEV_ID_100G_RDMA_MACSEC:
+               speed_fec_capa[0].speed = speed_fec_capa_tbl[4].speed;
+               speed_fec_capa[0].capa = speed_fec_capa_tbl[4].capa;
+               break;
+       case HNS3_DEV_ID_200G_RDMA:
+               speed_fec_capa[0].speed = speed_fec_capa_tbl[5].speed;
+               speed_fec_capa[0].capa = speed_fec_capa_tbl[5].capa;
+               break;
+       default:
+               return -ENOTSUP;
+       }
+
+       return 0;
+}
+
+static int
+hns3_fec_get_capability(struct rte_eth_dev *dev,
+                       struct rte_eth_fec_capa *speed_fec_capa,
+                       unsigned int num)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       uint16_t device_id = pci_dev->id.device_id;
+       unsigned int capa_num;
+       int ret;
+
+       capa_num = hns3_get_speed_capa_num(device_id);
+       if (capa_num == 0) {
+               hns3_err(hw, "device(0x%x) is not supported by hns3 PMD",
+                        device_id);
+               return -ENOTSUP;
+       }
+
+       if (speed_fec_capa == NULL || num < capa_num)
+               return capa_num;
+
+       ret = hns3_get_speed_fec_capa(speed_fec_capa, device_id);
+       if (ret)
+               return -ENOTSUP;
+
+       return capa_num;
+}
+
+static int
+get_current_fec_auto_state(struct hns3_hw *hw, uint8_t *state)
+{
+       struct hns3_config_fec_cmd *req;
+       struct hns3_cmd_desc desc;
+       int ret;
+
+       /*
+        * CMD(HNS3_OPC_CONFIG_FEC_MODE) read is not supported
+        * in device of link speed
+        * below 10 Gbps.
+        */
+       if (hw->mac.link_speed < ETH_SPEED_NUM_10G) {
+               *state = 0;
+               return 0;
+       }
+
+       hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_CONFIG_FEC_MODE, true);
+       req = (struct hns3_config_fec_cmd *)desc.data;
+       ret = hns3_cmd_send(hw, &desc, 1);
+       if (ret) {
+               hns3_err(hw, "get current fec auto state failed, ret = %d",
+                        ret);
+               return ret;
+       }
+
+       *state = req->fec_mode & (1U << HNS3_MAC_CFG_FEC_AUTO_EN_B);
+       return 0;
+}
+
+static int
+hns3_fec_get_internal(struct hns3_hw *hw, uint32_t *fec_capa)
+{
+#define QUERY_ACTIVE_SPEED     1
+       struct hns3_sfp_speed_cmd *resp;
+       uint32_t tmp_fec_capa;
+       uint8_t auto_state;
+       struct hns3_cmd_desc desc;
+       int ret;
+
+       /*
+        * If link is down and AUTO is enabled, AUTO is returned, otherwise,
+        * configured FEC mode is returned.
+        * If link is up, current FEC mode is returned.
+        */
+       if (hw->mac.link_status == ETH_LINK_DOWN) {
+               ret = get_current_fec_auto_state(hw, &auto_state);
+               if (ret)
+                       return ret;
+
+               if (auto_state == 0x1) {
+                       *fec_capa = RTE_ETH_FEC_MODE_CAPA_MASK(AUTO);
+                       return 0;
+               }
+       }
+
+       hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_SFP_GET_SPEED, true);
+       resp = (struct hns3_sfp_speed_cmd *)desc.data;
+       resp->query_type = QUERY_ACTIVE_SPEED;
+
+       ret = hns3_cmd_send(hw, &desc, 1);
+       if (ret == -EOPNOTSUPP) {
+               hns3_err(hw, "IMP do not support get FEC, ret = %d", ret);
+               return ret;
+       } else if (ret) {
+               hns3_err(hw, "get FEC failed, ret = %d", ret);
+               return ret;
+       }
+
+       /*
+        * FEC mode order defined in hns3 hardware is inconsistend with
+        * that defined in the ethdev library. So the sequence needs
+        * to be converted.
+        */
+       switch (resp->active_fec) {
+       case HNS3_HW_FEC_MODE_NOFEC:
+               tmp_fec_capa = RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC);
+               break;
+       case HNS3_HW_FEC_MODE_BASER:
+               tmp_fec_capa = RTE_ETH_FEC_MODE_CAPA_MASK(BASER);
+               break;
+       case HNS3_HW_FEC_MODE_RS:
+               tmp_fec_capa = RTE_ETH_FEC_MODE_CAPA_MASK(RS);
+               break;
+       default:
+               tmp_fec_capa = RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC);
+               break;
+       }
+
+       *fec_capa = tmp_fec_capa;
+       return 0;
+}
+
+static int
+hns3_fec_get(struct rte_eth_dev *dev, uint32_t *fec_capa)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       return hns3_fec_get_internal(hw, fec_capa);
+}
+
+static int
+hns3_set_fec_hw(struct hns3_hw *hw, uint32_t mode)
+{
+       struct hns3_config_fec_cmd *req;
+       struct hns3_cmd_desc desc;
+       int ret;
+
+       hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_CONFIG_FEC_MODE, false);
+
+       req = (struct hns3_config_fec_cmd *)desc.data;
+       switch (mode) {
+       case RTE_ETH_FEC_MODE_CAPA_MASK(NOFEC):
+               hns3_set_field(req->fec_mode, HNS3_MAC_CFG_FEC_MODE_M,
+                               HNS3_MAC_CFG_FEC_MODE_S, HNS3_MAC_FEC_OFF);
+               break;
+       case RTE_ETH_FEC_MODE_CAPA_MASK(BASER):
+               hns3_set_field(req->fec_mode, HNS3_MAC_CFG_FEC_MODE_M,
+                               HNS3_MAC_CFG_FEC_MODE_S, HNS3_MAC_FEC_BASER);
+               break;
+       case RTE_ETH_FEC_MODE_CAPA_MASK(RS):
+               hns3_set_field(req->fec_mode, HNS3_MAC_CFG_FEC_MODE_M,
+                               HNS3_MAC_CFG_FEC_MODE_S, HNS3_MAC_FEC_RS);
+               break;
+       case RTE_ETH_FEC_MODE_CAPA_MASK(AUTO):
+               hns3_set_bit(req->fec_mode, HNS3_MAC_CFG_FEC_AUTO_EN_B, 1);
+               break;
+       default:
+               return 0;
+       }
+       ret = hns3_cmd_send(hw, &desc, 1);
+       if (ret)
+               hns3_err(hw, "set fec mode failed, ret = %d", ret);
+
+       return ret;
+}
+
+static uint32_t
+get_current_speed_fec_cap(struct hns3_hw *hw, struct rte_eth_fec_capa *fec_capa)
+{
+       struct hns3_mac *mac = &hw->mac;
+       uint32_t cur_capa;
+
+       switch (mac->link_speed) {
+       case ETH_SPEED_NUM_10G:
+               cur_capa = fec_capa[1].capa;
+               break;
+       case ETH_SPEED_NUM_25G:
+       case ETH_SPEED_NUM_100G:
+       case ETH_SPEED_NUM_200G:
+               cur_capa = fec_capa[0].capa;
+               break;
+       default:
+               cur_capa = 0;
+               break;
+       }
+
+       return cur_capa;
+}
+
+static bool
+is_fec_mode_one_bit_set(uint32_t mode)
+{
+       int cnt = 0;
+       uint8_t i;
+
+       for (i = 0; i < sizeof(mode); i++)
+               if (mode >> i & 0x1)
+                       cnt++;
+
+       return cnt == 1 ? true : false;
+}
+
+static int
+hns3_fec_set(struct rte_eth_dev *dev, uint32_t mode)
+{
+#define FEC_CAPA_NUM 2
+       struct hns3_adapter *hns = dev->data->dev_private;
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(hns);
+       struct hns3_pf *pf = &hns->pf;
+
+       struct rte_eth_fec_capa fec_capa[FEC_CAPA_NUM];
+       uint32_t cur_capa;
+       uint32_t num = FEC_CAPA_NUM;
+       int ret;
+
+       ret = hns3_fec_get_capability(dev, fec_capa, num);
+       if (ret < 0)
+               return ret;
+
+       /* HNS3 PMD driver only support one bit set mode, e.g. 0x1, 0x4 */
+       if (!is_fec_mode_one_bit_set(mode))
+               hns3_err(hw, "FEC mode(0x%x) not supported in HNS3 PMD,"
+                            "FEC mode should be only one bit set", mode);
+
+       /*
+        * Check whether the configured mode is within the FEC capability.
+        * If not, the configured mode will not be supported.
+        */
+       cur_capa = get_current_speed_fec_cap(hw, fec_capa);
+       if (!(cur_capa & mode)) {
+               hns3_err(hw, "unsupported FEC mode = 0x%x", mode);
+               return -EINVAL;
+       }
+
+       ret = hns3_set_fec_hw(hw, mode);
+       if (ret)
+               return ret;
+
+       pf->fec_mode = mode;
+       return 0;
+}
+
+static int
+hns3_restore_fec(struct hns3_hw *hw)
+{
+       struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
+       struct hns3_pf *pf = &hns->pf;
+       uint32_t mode = pf->fec_mode;
+       int ret;
+
+       ret = hns3_set_fec_hw(hw, mode);
+       if (ret)
+               hns3_err(hw, "restore fec mode(0x%x) failed, ret = %d",
+                        mode, ret);
+
+       return ret;
+}
+
+static int
+hns3_query_dev_fec_info(struct hns3_hw *hw)
+{
+       struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
+       struct hns3_pf *pf = HNS3_DEV_PRIVATE_TO_PF(hns);
+       int ret;
+
+       ret = hns3_fec_get_internal(hw, &pf->fec_mode);
+       if (ret)
+               hns3_err(hw, "query device FEC info failed, ret = %d", ret);
+
+       return ret;
+}
+
 static const struct eth_dev_ops hns3_eth_dev_ops = {
        .dev_configure      = hns3_dev_configure,
        .dev_start          = hns3_dev_start,
@@ -5564,6 +6071,10 @@ static const struct eth_dev_ops hns3_eth_dev_ops = {
        .tx_queue_setup         = hns3_tx_queue_setup,
        .rx_queue_release       = hns3_dev_rx_queue_release,
        .tx_queue_release       = hns3_dev_tx_queue_release,
+       .rx_queue_start         = hns3_dev_rx_queue_start,
+       .rx_queue_stop          = hns3_dev_rx_queue_stop,
+       .tx_queue_start         = hns3_dev_tx_queue_start,
+       .tx_queue_stop          = hns3_dev_tx_queue_stop,
        .rx_queue_intr_enable   = hns3_dev_rx_queue_intr_enable,
        .rx_queue_intr_disable  = hns3_dev_rx_queue_intr_disable,
        .rxq_info_get           = hns3_rxq_info_get,
@@ -5590,6 +6101,10 @@ static const struct eth_dev_ops hns3_eth_dev_ops = {
        .get_reg                = hns3_get_regs,
        .get_dcb_info           = hns3_get_dcb_info,
        .dev_supported_ptypes_get = hns3_dev_supported_ptypes_get,
+       .fec_get_capability     = hns3_fec_get_capability,
+       .fec_get                = hns3_fec_get,
+       .fec_set                = hns3_fec_set,
+       .tm_ops_get             = hns3_tm_ops_get,
 };
 
 static const struct hns3_reset_ops hns3_reset_ops = {
@@ -5626,6 +6141,7 @@ hns3_dev_init(struct rte_eth_dev *eth_dev)
 
        hns3_set_rxtx_function(eth_dev);
        eth_dev->dev_ops = &hns3_eth_dev_ops;
+       eth_dev->rx_queue_count = hns3_rx_queue_count;
        if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
                ret = hns3_mp_init_secondary();
                if (ret) {
@@ -5638,6 +6154,8 @@ hns3_dev_init(struct rte_eth_dev *eth_dev)
                return 0;
        }
 
+       eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
+
        ret = hns3_mp_init_primary();
        if (ret) {
                PMD_INIT_LOG(ERR,
@@ -5683,7 +6201,7 @@ hns3_dev_init(struct rte_eth_dev *eth_dev)
        eth_addr = (struct rte_ether_addr *)hw->mac.mac_addr;
        if (!rte_is_valid_assigned_ether_addr(eth_addr)) {
                rte_eth_random_addr(hw->mac.mac_addr);
-               rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
+               hns3_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE,
                                (struct rte_ether_addr *)hw->mac.mac_addr);
                hns3_warn(hw, "default mac_addr from firmware is an invalid "
                          "unicast address, using random MAC address %s",
@@ -5693,11 +6211,6 @@ hns3_dev_init(struct rte_eth_dev *eth_dev)
                            &eth_dev->data->mac_addrs[0]);
 
        hw->adapter_state = HNS3_NIC_INITIALIZED;
-       /*
-        * Pass the information to the rte_eth_dev_close() that it should also
-        * release the private port resources.
-        */
-       eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
 
        if (rte_atomic16_read(&hns->hw.reset.schedule) == SCHEDULE_PENDING) {
                hns3_err(hw, "Reschedule reset service after dev_init");
@@ -5741,10 +6254,6 @@ hns3_dev_uninit(struct rte_eth_dev *eth_dev)
        if (rte_eal_process_type() != RTE_PROC_PRIMARY)
                return -EPERM;
 
-       eth_dev->dev_ops = NULL;
-       eth_dev->rx_pkt_burst = NULL;
-       eth_dev->tx_pkt_burst = NULL;
-       eth_dev->tx_pkt_prepare = NULL;
        if (hw->adapter_state < HNS3_NIC_CLOSING)
                hns3_dev_close(eth_dev);
 
@@ -5774,7 +6283,7 @@ static const struct rte_pci_id pci_id_hns3_map[] = {
        { RTE_PCI_DEVICE(PCI_VENDOR_ID_HUAWEI, HNS3_DEV_ID_50GE_RDMA) },
        { RTE_PCI_DEVICE(PCI_VENDOR_ID_HUAWEI, HNS3_DEV_ID_100G_RDMA_MACSEC) },
        { RTE_PCI_DEVICE(PCI_VENDOR_ID_HUAWEI, HNS3_DEV_ID_200G_RDMA) },
-       { .vendor_id = 0, /* sentinel */ },
+       { .vendor_id = 0, }, /* sentinel */
 };
 
 static struct rte_pci_driver rte_hns3_pmd = {