net/hns3: support Rx interrupt
[dpdk.git] / drivers / net / hns3 / hns3_ethdev.c
index efd5f63..8243ade 100644 (file)
@@ -9,6 +9,7 @@
 #include <stdint.h>
 #include <inttypes.h>
 #include <unistd.h>
+#include <rte_atomic.h>
 #include <rte_bus_pci.h>
 #include <rte_common.h>
 #include <rte_cycles.h>
@@ -28,6 +29,7 @@
 #include "hns3_intr.h"
 #include "hns3_regs.h"
 #include "hns3_dcb.h"
+#include "hns3_mp.h"
 
 #define HNS3_DEFAULT_PORT_CONF_BURST_SIZE      32
 #define HNS3_DEFAULT_PORT_CONF_QUEUES_NUM      1
 #define HNS3_FILTER_FE_INGRESS         (HNS3_FILTER_FE_NIC_INGRESS_B \
                                        | HNS3_FILTER_FE_ROCE_INGRESS_B)
 
+/* Reset related Registers */
+#define HNS3_GLOBAL_RESET_BIT          0
+#define HNS3_CORE_RESET_BIT            1
+#define HNS3_IMP_RESET_BIT             2
+#define HNS3_FUN_RST_ING_B             0
+
+#define HNS3_VECTOR0_IMP_RESET_INT_B   1
+
+#define HNS3_RESET_WAIT_MS     100
+#define HNS3_RESET_WAIT_CNT    200
+
 int hns3_logtype_init;
 int hns3_logtype_driver;
 
@@ -59,6 +72,8 @@ enum hns3_evt_cause {
        HNS3_VECTOR0_EVENT_OTHER,
 };
 
+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);
 static int hns3_vlan_pvid_configure(struct hns3_adapter *hns, uint16_t pvid,
                                    int on);
@@ -96,14 +111,34 @@ hns3_check_event_cause(struct hns3_adapter *hns, uint32_t *clearval)
         * from H/W just for the mailbox.
         */
        if (BIT(HNS3_VECTOR0_IMPRESET_INT_B) & vector0_int_stats) { /* IMP */
+               rte_atomic16_set(&hw->reset.disable_cmd, 1);
+               hns3_atomic_set_bit(HNS3_IMP_RESET, &hw->reset.pending);
                val = BIT(HNS3_VECTOR0_IMPRESET_INT_B);
+               if (clearval) {
+                       hw->reset.stats.imp_cnt++;
+                       hns3_warn(hw, "IMP reset detected, clear reset status");
+               } else {
+                       hns3_schedule_delayed_reset(hns);
+                       hns3_warn(hw, "IMP reset detected, don't clear reset status");
+               }
+
                ret = HNS3_VECTOR0_EVENT_RST;
                goto out;
        }
 
        /* Global reset */
        if (BIT(HNS3_VECTOR0_GLOBALRESET_INT_B) & vector0_int_stats) {
+               rte_atomic16_set(&hw->reset.disable_cmd, 1);
+               hns3_atomic_set_bit(HNS3_GLOBAL_RESET, &hw->reset.pending);
                val = BIT(HNS3_VECTOR0_GLOBALRESET_INT_B);
+               if (clearval) {
+                       hw->reset.stats.global_cnt++;
+                       hns3_warn(hw, "Global reset detected, clear reset status");
+               } else {
+                       hns3_schedule_delayed_reset(hns);
+                       hns3_warn(hw, "Global reset detected, don't clear reset status");
+               }
+
                ret = HNS3_VECTOR0_EVENT_RST;
                goto out;
        }
@@ -177,6 +212,15 @@ hns3_interrupt_handler(void *param)
 
        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_handle_msix_error(hns, &hw->reset.request);
+               hns3_schedule_reset(hns);
+       } else if (event_cause == HNS3_VECTOR0_EVENT_RST)
+               hns3_schedule_reset(hns);
+       else
+               hns3_err(hw, "Received unknown event");
+
        hns3_clear_event_cause(hw, event_cause, clearval);
        /* Enable interrupt if it is not cause by reset */
        hns3_pf_enable_irq0(hw);
@@ -238,6 +282,11 @@ hns3_add_dev_vlan_table(struct hns3_adapter *hns, uint16_t vlan_id,
        struct hns3_hw *hw = &hns->hw;
        struct hns3_pf *pf = &hns->pf;
 
+       LIST_FOREACH(vlan_entry, &pf->vlan_list, next) {
+               if (vlan_entry->vlan_id == vlan_id)
+                       return;
+       }
+
        vlan_entry = rte_zmalloc("hns3_vlan_tbl", sizeof(*vlan_entry), 0);
        if (vlan_entry == NULL) {
                hns3_err(hw, "Failed to malloc hns3 vlan table");
@@ -250,6 +299,32 @@ hns3_add_dev_vlan_table(struct hns3_adapter *hns, uint16_t vlan_id,
        LIST_INSERT_HEAD(&pf->vlan_list, vlan_entry, next);
 }
 
+static int
+hns3_restore_vlan_table(struct hns3_adapter *hns)
+{
+       struct hns3_user_vlan_table *vlan_entry;
+       struct hns3_pf *pf = &hns->pf;
+       uint16_t vlan_id;
+       int ret = 0;
+
+       if (pf->port_base_vlan_cfg.state == HNS3_PORT_BASE_VLAN_ENABLE) {
+               ret = hns3_vlan_pvid_configure(hns, pf->port_base_vlan_cfg.pvid,
+                                              1);
+               return ret;
+       }
+
+       LIST_FOREACH(vlan_entry, &pf->vlan_list, next) {
+               if (vlan_entry->hd_tbl_status) {
+                       vlan_id = vlan_entry->vlan_id;
+                       ret = hns3_set_port_vlan_filter(hns, vlan_id, 1);
+                       if (ret)
+                               break;
+               }
+       }
+
+       return ret;
+}
+
 static int
 hns3_vlan_filter_configure(struct hns3_adapter *hns, uint16_t vlan_id, int on)
 {
@@ -874,6 +949,26 @@ hns3_init_vlan_config(struct hns3_adapter *hns)
        return hns3_default_vlan_config(hns);
 }
 
+static int
+hns3_restore_vlan_conf(struct hns3_adapter *hns)
+{
+       struct hns3_pf *pf = &hns->pf;
+       struct hns3_hw *hw = &hns->hw;
+       int ret;
+
+       ret = hns3_set_vlan_rx_offload_cfg(hns, &pf->vtag_config.rx_vcfg);
+       if (ret) {
+               hns3_err(hw, "hns3 restore vlan rx conf fail, ret =%d", ret);
+               return ret;
+       }
+
+       ret = hns3_set_vlan_tx_offload_cfg(hns, &pf->vtag_config.tx_vcfg);
+       if (ret)
+               hns3_err(hw, "hns3 restore vlan tx conf fail, ret =%d", ret);
+
+       return ret;
+}
+
 static int
 hns3_dev_configure_vlan(struct rte_eth_dev *dev)
 {
@@ -1926,6 +2021,40 @@ hns3_check_dcb_cfg(struct rte_eth_dev *dev)
        return hns3_check_mq_mode(dev);
 }
 
+static int
+hns3_bind_ring_with_vector(struct rte_eth_dev *dev, uint8_t vector_id,
+                          bool mmap, uint16_t queue_id)
+{
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct hns3_cmd_desc desc;
+       struct hns3_ctrl_vector_chain_cmd *req =
+               (struct hns3_ctrl_vector_chain_cmd *)desc.data;
+       enum hns3_cmd_status status;
+       enum hns3_opcode_type op;
+       uint16_t tqp_type_and_id = 0;
+
+       op = mmap ? HNS3_OPC_ADD_RING_TO_VECTOR : HNS3_OPC_DEL_RING_TO_VECTOR;
+       hns3_cmd_setup_basic_desc(&desc, op, false);
+       req->int_vector_id = vector_id;
+
+       hns3_set_field(tqp_type_and_id, HNS3_INT_TYPE_M, HNS3_INT_TYPE_S,
+                      HNS3_RING_TYPE_RX);
+       hns3_set_field(tqp_type_and_id, HNS3_TQP_ID_M, HNS3_TQP_ID_S, queue_id);
+       hns3_set_field(tqp_type_and_id, HNS3_INT_GL_IDX_M, HNS3_INT_GL_IDX_S,
+                      HNS3_RING_GL_RX);
+       req->tqp_type_and_id[0] = rte_cpu_to_le_16(tqp_type_and_id);
+
+       req->int_cause_num = 1;
+       status = hns3_cmd_send(hw, &desc, 1);
+       if (status) {
+               hns3_err(hw, "Map TQP %d fail, vector_id is %d, status is %d.",
+                        queue_id, vector_id, status);
+               return -EIO;
+       }
+
+       return 0;
+}
+
 static int
 hns3_dev_configure(struct rte_eth_dev *dev)
 {
@@ -3527,6 +3656,19 @@ hns3_dev_allmulticast_disable(struct rte_eth_dev *dev)
        return ret;
 }
 
+static int
+hns3_dev_promisc_restore(struct hns3_adapter *hns)
+{
+       struct hns3_hw *hw = &hns->hw;
+       bool en_mc_pmc;
+       bool en_uc_pmc;
+
+       en_uc_pmc = (hw->data->promiscuous == 1) ? true : false;
+       en_mc_pmc = (hw->data->all_multicast == 1) ? true : false;
+
+       return hns3_set_promisc_mode(hw, en_uc_pmc, en_mc_pmc);
+}
+
 static int
 hns3_get_sfp_speed(struct hns3_hw *hw, uint32_t *speed)
 {
@@ -3681,8 +3823,11 @@ hns3_service_handler(void *param)
        struct hns3_adapter *hns = eth_dev->data->dev_private;
        struct hns3_hw *hw = &hns->hw;
 
-       hns3_update_speed_duplex(eth_dev);
-       hns3_update_link_status(hw);
+       if (!hns3_is_reset_pending(hns)) {
+               hns3_update_speed_duplex(eth_dev);
+               hns3_update_link_status(hw);
+       } else
+               hns3_warn(hw, "Cancel the query when reset is pending");
 
        rte_eal_alarm_set(HNS3_SERVICE_INTERVAL, hns3_service_handler, eth_dev);
 }
@@ -3909,13 +4054,82 @@ err_config_mac_mode:
 }
 
 static int
-hns3_dev_start(struct rte_eth_dev *eth_dev)
+hns3_map_rx_interrupt(struct rte_eth_dev *dev)
 {
-       struct hns3_adapter *hns = eth_dev->data->dev_private;
-       struct hns3_hw *hw = &hns->hw;
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+       struct hns3_hw *hw = HNS3_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t intr_vector;
+       uint8_t base = 0;
+       uint8_t vec = 0;
+       uint16_t q_id;
        int ret;
 
+       if (dev->data->dev_conf.intr_conf.rxq == 0)
+               return 0;
+
+       /* disable uio/vfio intr/eventfd mapping */
+       rte_intr_disable(intr_handle);
+
+       /* check and configure queue intr-vector mapping */
+       if (rte_intr_cap_multiple(intr_handle) ||
+           !RTE_ETH_DEV_SRIOV(dev).active) {
+               intr_vector = dev->data->nb_rx_queues;
+               /* creates event fd for each intr vector when MSIX is used */
+               if (rte_intr_efd_enable(intr_handle, intr_vector))
+                       return -EINVAL;
+       }
+       if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
+               intr_handle->intr_vec =
+                       rte_zmalloc("intr_vec",
+                                   dev->data->nb_rx_queues * sizeof(int), 0);
+               if (intr_handle->intr_vec == NULL) {
+                       hns3_err(hw, "Failed to allocate %d rx_queues"
+                                    " intr_vec", dev->data->nb_rx_queues);
+                       ret = -ENOMEM;
+                       goto alloc_intr_vec_error;
+               }
+       }
+
+       if (rte_intr_allow_others(intr_handle)) {
+               vec = RTE_INTR_VEC_RXTX_OFFSET;
+               base = RTE_INTR_VEC_RXTX_OFFSET;
+       }
+       if (rte_intr_dp_is_en(intr_handle)) {
+               for (q_id = 0; q_id < dev->data->nb_rx_queues; q_id++) {
+                       ret = hns3_bind_ring_with_vector(dev, vec, true, q_id);
+                       if (ret)
+                               goto bind_vector_error;
+                       intr_handle->intr_vec[q_id] = vec;
+                       if (vec < base + intr_handle->nb_efd - 1)
+                               vec++;
+               }
+       }
+       rte_intr_enable(intr_handle);
+       return 0;
+
+bind_vector_error:
+       rte_intr_efd_disable(intr_handle);
+       if (intr_handle->intr_vec) {
+               free(intr_handle->intr_vec);
+               intr_handle->intr_vec = NULL;
+       }
+       return ret;
+alloc_intr_vec_error:
+       rte_intr_efd_disable(intr_handle);
+       return ret;
+}
+
+static int
+hns3_dev_start(struct rte_eth_dev *dev)
+{
+       struct hns3_adapter *hns = dev->data->dev_private;
+       struct hns3_hw *hw = &hns->hw;
+       int ret = 0;
+
        PMD_INIT_FUNC_TRACE();
+       if (rte_atomic16_read(&hw->reset.resetting))
+               return -EBUSY;
 
        rte_spinlock_lock(&hw->lock);
        hw->adapter_state = HNS3_NIC_STARTING;
@@ -3929,7 +4143,12 @@ hns3_dev_start(struct rte_eth_dev *eth_dev)
 
        hw->adapter_state = HNS3_NIC_STARTED;
        rte_spinlock_unlock(&hw->lock);
-       hns3_set_rxtx_function(eth_dev);
+
+       ret = hns3_map_rx_interrupt(dev);
+       if (ret)
+               return ret;
+       hns3_set_rxtx_function(dev);
+       hns3_mp_req_start_rxtx(dev);
 
        hns3_info(hw, "hns3 dev start successful!");
        return 0;
@@ -3947,29 +4166,71 @@ hns3_do_stop(struct hns3_adapter *hns)
                return ret;
        hw->mac.link_status = ETH_LINK_DOWN;
 
-       hns3_configure_all_mac_addr(hns, true);
-       reset_queue = true;
+       if (rte_atomic16_read(&hw->reset.disable_cmd) == 0) {
+               hns3_configure_all_mac_addr(hns, true);
+               reset_queue = true;
+       } else
+               reset_queue = false;
        hw->mac.default_addr_setted = false;
        return hns3_stop_queues(hns, reset_queue);
 }
 
 static void
-hns3_dev_stop(struct rte_eth_dev *eth_dev)
+hns3_unmap_rx_interrupt(struct rte_eth_dev *dev)
 {
-       struct hns3_adapter *hns = eth_dev->data->dev_private;
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+       uint8_t base = 0;
+       uint8_t vec = 0;
+       uint16_t q_id;
+
+       if (dev->data->dev_conf.intr_conf.rxq == 0)
+               return;
+
+       /* unmap the ring with vector */
+       if (rte_intr_allow_others(intr_handle)) {
+               vec = RTE_INTR_VEC_RXTX_OFFSET;
+               base = RTE_INTR_VEC_RXTX_OFFSET;
+       }
+       if (rte_intr_dp_is_en(intr_handle)) {
+               for (q_id = 0; q_id < dev->data->nb_rx_queues; q_id++) {
+                       (void)hns3_bind_ring_with_vector(dev, vec, false, q_id);
+                       if (vec < base + intr_handle->nb_efd - 1)
+                               vec++;
+               }
+       }
+       /* Clean datapath event and queue/vec mapping */
+       rte_intr_efd_disable(intr_handle);
+       if (intr_handle->intr_vec) {
+               rte_free(intr_handle->intr_vec);
+               intr_handle->intr_vec = NULL;
+       }
+}
+
+static void
+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();
 
        hw->adapter_state = HNS3_NIC_STOPPING;
-       hns3_set_rxtx_function(eth_dev);
+       hns3_set_rxtx_function(dev);
+       rte_wmb();
+       /* Disable datapath on secondary process. */
+       hns3_mp_req_stop_rxtx(dev);
+       /* Prevent crashes when queues are still in use. */
+       rte_delay_ms(hw->tqps_num);
 
        rte_spinlock_lock(&hw->lock);
-
-       hns3_do_stop(hns);
-       hns3_dev_release_mbufs(hns);
-       hw->adapter_state = HNS3_NIC_CONFIGURED;
+       if (rte_atomic16_read(&hw->reset.resetting) == 0) {
+               hns3_do_stop(hns);
+               hns3_dev_release_mbufs(hns);
+               hw->adapter_state = HNS3_NIC_CONFIGURED;
+       }
        rte_spinlock_unlock(&hw->lock);
+       hns3_unmap_rx_interrupt(dev);
 }
 
 static void
@@ -3978,10 +4239,18 @@ hns3_dev_close(struct rte_eth_dev *eth_dev)
        struct hns3_adapter *hns = eth_dev->data->dev_private;
        struct hns3_hw *hw = &hns->hw;
 
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
+               rte_free(eth_dev->process_private);
+               eth_dev->process_private = NULL;
+               return;
+       }
+
        if (hw->adapter_state == HNS3_NIC_STARTED)
                hns3_dev_stop(eth_dev);
 
        hw->adapter_state = HNS3_NIC_CLOSING;
+       hns3_reset_abort(hns);
+       hw->adapter_state = HNS3_NIC_CLOSED;
        rte_eal_alarm_cancel(hns3_service_handler, eth_dev);
 
        hns3_configure_all_mc_mac_addr(hns, true);
@@ -3989,9 +4258,10 @@ hns3_dev_close(struct rte_eth_dev *eth_dev)
        hns3_vlan_txvlan_cfg(hns, HNS3_PORT_BASE_VLAN_DISABLE, 0);
        hns3_uninit_pf(eth_dev);
        hns3_free_all_queues(eth_dev);
+       rte_free(hw->reset.wait_data);
        rte_free(eth_dev->process_private);
        eth_dev->process_private = NULL;
-       hw->adapter_state = HNS3_NIC_CLOSED;
+       hns3_mp_uninit_primary();
        hns3_warn(hw, "Close port %d finished", hw->data->port_id);
 }
 
@@ -4181,6 +4451,420 @@ hns3_get_dcb_info(struct rte_eth_dev *dev, struct rte_eth_dcb_info *dcb_info)
        return 0;
 }
 
+static int
+hns3_reinit_dev(struct hns3_adapter *hns)
+{
+       struct hns3_hw *hw = &hns->hw;
+       int ret;
+
+       ret = hns3_cmd_init(hw);
+       if (ret) {
+               hns3_err(hw, "Failed to init cmd: %d", ret);
+               return ret;
+       }
+
+       ret = hns3_reset_all_queues(hns);
+       if (ret) {
+               hns3_err(hw, "Failed to reset all queues: %d", ret);
+               goto err_init;
+       }
+
+       ret = hns3_init_hardware(hns);
+       if (ret) {
+               hns3_err(hw, "Failed to init hardware: %d", ret);
+               goto err_init;
+       }
+
+       ret = hns3_enable_hw_error_intr(hns, true);
+       if (ret) {
+               hns3_err(hw, "fail to enable hw error interrupts: %d",
+                            ret);
+               goto err_mac_init;
+       }
+       hns3_info(hw, "Reset done, driver initialization finished.");
+
+       return 0;
+
+err_mac_init:
+       hns3_uninit_umv_space(hw);
+err_init:
+       hns3_cmd_uninit(hw);
+
+       return ret;
+}
+
+static bool
+is_pf_reset_done(struct hns3_hw *hw)
+{
+       uint32_t val, reg, reg_bit;
+
+       switch (hw->reset.level) {
+       case HNS3_IMP_RESET:
+               reg = HNS3_GLOBAL_RESET_REG;
+               reg_bit = HNS3_IMP_RESET_BIT;
+               break;
+       case HNS3_GLOBAL_RESET:
+               reg = HNS3_GLOBAL_RESET_REG;
+               reg_bit = HNS3_GLOBAL_RESET_BIT;
+               break;
+       case HNS3_FUNC_RESET:
+               reg = HNS3_FUN_RST_ING;
+               reg_bit = HNS3_FUN_RST_ING_B;
+               break;
+       case HNS3_FLR_RESET:
+       default:
+               hns3_err(hw, "Wait for unsupported reset level: %d",
+                        hw->reset.level);
+               return true;
+       }
+       val = hns3_read_dev(hw, reg);
+       if (hns3_get_bit(val, reg_bit))
+               return false;
+       else
+               return true;
+}
+
+bool
+hns3_is_reset_pending(struct hns3_adapter *hns)
+{
+       struct hns3_hw *hw = &hns->hw;
+       enum hns3_reset_level reset;
+
+       hns3_check_event_cause(hns, NULL);
+       reset = hns3_get_reset_level(hns, &hw->reset.pending);
+       if (hw->reset.level != HNS3_NONE_RESET && hw->reset.level < reset) {
+               hns3_warn(hw, "High level reset %d is pending", reset);
+               return true;
+       }
+       reset = hns3_get_reset_level(hns, &hw->reset.request);
+       if (hw->reset.level != HNS3_NONE_RESET && hw->reset.level < reset) {
+               hns3_warn(hw, "High level reset %d is request", reset);
+               return true;
+       }
+       return false;
+}
+
+static int
+hns3_wait_hardware_ready(struct hns3_adapter *hns)
+{
+       struct hns3_hw *hw = &hns->hw;
+       struct hns3_wait_data *wait_data = hw->reset.wait_data;
+       struct timeval tv;
+
+       if (wait_data->result == HNS3_WAIT_SUCCESS)
+               return 0;
+       else if (wait_data->result == HNS3_WAIT_TIMEOUT) {
+               gettimeofday(&tv, NULL);
+               hns3_warn(hw, "Reset step4 hardware not ready after reset time=%ld.%.6ld",
+                         tv.tv_sec, tv.tv_usec);
+               return -ETIME;
+       } else if (wait_data->result == HNS3_WAIT_REQUEST)
+               return -EAGAIN;
+
+       wait_data->hns = hns;
+       wait_data->check_completion = is_pf_reset_done;
+       wait_data->end_ms = (uint64_t)HNS3_RESET_WAIT_CNT *
+                                     HNS3_RESET_WAIT_MS + get_timeofday_ms();
+       wait_data->interval = HNS3_RESET_WAIT_MS * USEC_PER_MSEC;
+       wait_data->count = HNS3_RESET_WAIT_CNT;
+       wait_data->result = HNS3_WAIT_REQUEST;
+       rte_eal_alarm_set(wait_data->interval, hns3_wait_callback, wait_data);
+       return -EAGAIN;
+}
+
+static int
+hns3_func_reset_cmd(struct hns3_hw *hw, int func_id)
+{
+       struct hns3_cmd_desc desc;
+       struct hns3_reset_cmd *req = (struct hns3_reset_cmd *)desc.data;
+
+       hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_CFG_RST_TRIGGER, false);
+       hns3_set_bit(req->mac_func_reset, HNS3_CFG_RESET_FUNC_B, 1);
+       req->fun_reset_vfid = func_id;
+
+       return hns3_cmd_send(hw, &desc, 1);
+}
+
+static int
+hns3_imp_reset_cmd(struct hns3_hw *hw)
+{
+       struct hns3_cmd_desc desc;
+
+       hns3_cmd_setup_basic_desc(&desc, 0xFFFE, false);
+       desc.data[0] = 0xeedd;
+
+       return hns3_cmd_send(hw, &desc, 1);
+}
+
+static void
+hns3_msix_process(struct hns3_adapter *hns, enum hns3_reset_level reset_level)
+{
+       struct hns3_hw *hw = &hns->hw;
+       struct timeval tv;
+       uint32_t val;
+
+       gettimeofday(&tv, NULL);
+       if (hns3_read_dev(hw, HNS3_GLOBAL_RESET_REG) ||
+           hns3_read_dev(hw, HNS3_FUN_RST_ING)) {
+               hns3_warn(hw, "Don't process msix during resetting time=%ld.%.6ld",
+                         tv.tv_sec, tv.tv_usec);
+               return;
+       }
+
+       switch (reset_level) {
+       case HNS3_IMP_RESET:
+               hns3_imp_reset_cmd(hw);
+               hns3_warn(hw, "IMP Reset requested time=%ld.%.6ld",
+                         tv.tv_sec, tv.tv_usec);
+               break;
+       case HNS3_GLOBAL_RESET:
+               val = hns3_read_dev(hw, HNS3_GLOBAL_RESET_REG);
+               hns3_set_bit(val, HNS3_GLOBAL_RESET_BIT, 1);
+               hns3_write_dev(hw, HNS3_GLOBAL_RESET_REG, val);
+               hns3_warn(hw, "Global Reset requested time=%ld.%.6ld",
+                         tv.tv_sec, tv.tv_usec);
+               break;
+       case HNS3_FUNC_RESET:
+               hns3_warn(hw, "PF Reset requested time=%ld.%.6ld",
+                         tv.tv_sec, tv.tv_usec);
+               /* schedule again to check later */
+               hns3_atomic_set_bit(HNS3_FUNC_RESET, &hw->reset.pending);
+               hns3_schedule_reset(hns);
+               break;
+       default:
+               hns3_warn(hw, "Unsupported reset level: %d", reset_level);
+               return;
+       }
+       hns3_atomic_clear_bit(reset_level, &hw->reset.request);
+}
+
+static enum hns3_reset_level
+hns3_get_reset_level(struct hns3_adapter *hns, uint64_t *levels)
+{
+       struct hns3_hw *hw = &hns->hw;
+       enum hns3_reset_level reset_level = HNS3_NONE_RESET;
+
+       /* Return the highest priority reset level amongst all */
+       if (hns3_atomic_test_bit(HNS3_IMP_RESET, levels))
+               reset_level = HNS3_IMP_RESET;
+       else if (hns3_atomic_test_bit(HNS3_GLOBAL_RESET, levels))
+               reset_level = HNS3_GLOBAL_RESET;
+       else if (hns3_atomic_test_bit(HNS3_FUNC_RESET, levels))
+               reset_level = HNS3_FUNC_RESET;
+       else if (hns3_atomic_test_bit(HNS3_FLR_RESET, levels))
+               reset_level = HNS3_FLR_RESET;
+
+       if (hw->reset.level != HNS3_NONE_RESET && reset_level < hw->reset.level)
+               return HNS3_NONE_RESET;
+
+       return reset_level;
+}
+
+static int
+hns3_prepare_reset(struct hns3_adapter *hns)
+{
+       struct hns3_hw *hw = &hns->hw;
+       uint32_t reg_val;
+       int ret;
+
+       switch (hw->reset.level) {
+       case HNS3_FUNC_RESET:
+               ret = hns3_func_reset_cmd(hw, 0);
+               if (ret)
+                       return ret;
+
+               /*
+                * After performaning pf reset, it is not necessary to do the
+                * mailbox handling or send any command to firmware, because
+                * any mailbox handling or command to firmware is only valid
+                * after hns3_cmd_init is called.
+                */
+               rte_atomic16_set(&hw->reset.disable_cmd, 1);
+               hw->reset.stats.request_cnt++;
+               break;
+       case HNS3_IMP_RESET:
+               reg_val = hns3_read_dev(hw, HNS3_VECTOR0_OTER_EN_REG);
+               hns3_write_dev(hw, HNS3_VECTOR0_OTER_EN_REG, reg_val |
+                              BIT(HNS3_VECTOR0_IMP_RESET_INT_B));
+               break;
+       default:
+               break;
+       }
+       return 0;
+}
+
+static int
+hns3_set_rst_done(struct hns3_hw *hw)
+{
+       struct hns3_pf_rst_done_cmd *req;
+       struct hns3_cmd_desc desc;
+
+       req = (struct hns3_pf_rst_done_cmd *)desc.data;
+       hns3_cmd_setup_basic_desc(&desc, HNS3_OPC_PF_RST_DONE, false);
+       req->pf_rst_done |= HNS3_PF_RESET_DONE_BIT;
+       return hns3_cmd_send(hw, &desc, 1);
+}
+
+static int
+hns3_stop_service(struct hns3_adapter *hns)
+{
+       struct hns3_hw *hw = &hns->hw;
+       struct rte_eth_dev *eth_dev;
+
+       eth_dev = &rte_eth_devices[hw->data->port_id];
+       rte_eal_alarm_cancel(hns3_service_handler, eth_dev);
+       hw->mac.link_status = ETH_LINK_DOWN;
+
+       hns3_set_rxtx_function(eth_dev);
+       rte_wmb();
+       /* Disable datapath on secondary process. */
+       hns3_mp_req_stop_rxtx(eth_dev);
+       rte_delay_ms(hw->tqps_num);
+
+       rte_spinlock_lock(&hw->lock);
+       if (hns->hw.adapter_state == HNS3_NIC_STARTED ||
+           hw->adapter_state == HNS3_NIC_STOPPING) {
+               hns3_do_stop(hns);
+               hw->reset.mbuf_deferred_free = true;
+       } else
+               hw->reset.mbuf_deferred_free = false;
+
+       /*
+        * It is cumbersome for hardware to pick-and-choose entries for deletion
+        * from table space. Hence, for function reset software intervention is
+        * required to delete the entries
+        */
+       if (rte_atomic16_read(&hw->reset.disable_cmd) == 0)
+               hns3_configure_all_mc_mac_addr(hns, true);
+       rte_spinlock_unlock(&hw->lock);
+
+       return 0;
+}
+
+static int
+hns3_start_service(struct hns3_adapter *hns)
+{
+       struct hns3_hw *hw = &hns->hw;
+       struct rte_eth_dev *eth_dev;
+
+       if (hw->reset.level == HNS3_IMP_RESET ||
+           hw->reset.level == HNS3_GLOBAL_RESET)
+               hns3_set_rst_done(hw);
+       eth_dev = &rte_eth_devices[hw->data->port_id];
+       hns3_set_rxtx_function(eth_dev);
+       hns3_mp_req_start_rxtx(eth_dev);
+       hns3_service_handler(eth_dev);
+       return 0;
+}
+
+static int
+hns3_restore_conf(struct hns3_adapter *hns)
+{
+       struct hns3_hw *hw = &hns->hw;
+       int ret;
+
+       ret = hns3_configure_all_mac_addr(hns, false);
+       if (ret)
+               return ret;
+
+       ret = hns3_configure_all_mc_mac_addr(hns, false);
+       if (ret)
+               goto err_mc_mac;
+
+       ret = hns3_dev_promisc_restore(hns);
+       if (ret)
+               goto err_promisc;
+
+       ret = hns3_restore_vlan_table(hns);
+       if (ret)
+               goto err_promisc;
+
+       ret = hns3_restore_vlan_conf(hns);
+       if (ret)
+               goto err_promisc;
+
+       ret = hns3_restore_all_fdir_filter(hns);
+       if (ret)
+               goto err_promisc;
+
+       if (hns->hw.adapter_state == HNS3_NIC_STARTED) {
+               ret = hns3_do_start(hns, false);
+               if (ret)
+                       goto err_promisc;
+               hns3_info(hw, "hns3 dev restart successful!");
+       } else if (hw->adapter_state == HNS3_NIC_STOPPING)
+               hw->adapter_state = HNS3_NIC_CONFIGURED;
+       return 0;
+
+err_promisc:
+       hns3_configure_all_mc_mac_addr(hns, true);
+err_mc_mac:
+       hns3_configure_all_mac_addr(hns, true);
+       return ret;
+}
+
+static void
+hns3_reset_service(void *param)
+{
+       struct hns3_adapter *hns = (struct hns3_adapter *)param;
+       struct hns3_hw *hw = &hns->hw;
+       enum hns3_reset_level reset_level;
+       struct timeval tv_delta;
+       struct timeval tv_start;
+       struct timeval tv;
+       uint64_t msec;
+       int ret;
+
+       /*
+        * The interrupt is not triggered within the delay time.
+        * The interrupt may have been lost. It is necessary to handle
+        * the interrupt to recover from the error.
+        */
+       if (rte_atomic16_read(&hns->hw.reset.schedule) == SCHEDULE_DEFERRED) {
+               rte_atomic16_set(&hns->hw.reset.schedule, SCHEDULE_REQUESTED);
+               hns3_err(hw, "Handling interrupts in delayed tasks");
+               hns3_interrupt_handler(&rte_eth_devices[hw->data->port_id]);
+               reset_level = hns3_get_reset_level(hns, &hw->reset.pending);
+               if (reset_level == HNS3_NONE_RESET) {
+                       hns3_err(hw, "No reset level is set, try IMP reset");
+                       hns3_atomic_set_bit(HNS3_IMP_RESET, &hw->reset.pending);
+               }
+       }
+       rte_atomic16_set(&hns->hw.reset.schedule, SCHEDULE_NONE);
+
+       /*
+        * Check if there is any ongoing reset in the hardware. This status can
+        * be checked from reset_pending. If there is then, we need to wait for
+        * hardware to complete reset.
+        *    a. If we are able to figure out in reasonable time that hardware
+        *       has fully resetted then, we can proceed with driver, client
+        *       reset.
+        *    b. else, we can come back later to check this status so re-sched
+        *       now.
+        */
+       reset_level = hns3_get_reset_level(hns, &hw->reset.pending);
+       if (reset_level != HNS3_NONE_RESET) {
+               gettimeofday(&tv_start, NULL);
+               ret = hns3_reset_process(hns, reset_level);
+               gettimeofday(&tv, NULL);
+               timersub(&tv, &tv_start, &tv_delta);
+               msec = tv_delta.tv_sec * MSEC_PER_SEC +
+                      tv_delta.tv_usec / USEC_PER_MSEC;
+               if (msec > HNS3_RESET_PROCESS_MS)
+                       hns3_err(hw, "%d handle long time delta %" PRIx64
+                                    " ms time=%ld.%.6ld",
+                                hw->reset.level, msec,
+                                tv.tv_sec, tv.tv_usec);
+               if (ret == -EAGAIN)
+                       return;
+       }
+
+       /* Check if we got any *new* reset requests to be honored */
+       reset_level = hns3_get_reset_level(hns, &hw->reset.request);
+       if (reset_level != HNS3_NONE_RESET)
+               hns3_msix_process(hns, reset_level);
+}
+
 static const struct eth_dev_ops hns3_eth_dev_ops = {
        .dev_start          = hns3_dev_start,
        .dev_stop           = hns3_dev_stop,
@@ -4190,12 +4874,21 @@ static const struct eth_dev_ops hns3_eth_dev_ops = {
        .allmulticast_enable  = hns3_dev_allmulticast_enable,
        .allmulticast_disable = hns3_dev_allmulticast_disable,
        .mtu_set            = hns3_dev_mtu_set,
+       .stats_get          = hns3_stats_get,
+       .stats_reset        = hns3_stats_reset,
+       .xstats_get         = hns3_dev_xstats_get,
+       .xstats_get_names   = hns3_dev_xstats_get_names,
+       .xstats_reset       = hns3_dev_xstats_reset,
+       .xstats_get_by_id   = hns3_dev_xstats_get_by_id,
+       .xstats_get_names_by_id = hns3_dev_xstats_get_names_by_id,
        .dev_infos_get          = hns3_dev_infos_get,
        .fw_version_get         = hns3_fw_version_get,
        .rx_queue_setup         = hns3_rx_queue_setup,
        .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_intr_enable   = hns3_dev_rx_queue_intr_enable,
+       .rx_queue_intr_disable  = hns3_dev_rx_queue_intr_disable,
        .dev_configure          = hns3_dev_configure,
        .flow_ctrl_get          = hns3_flow_ctrl_get,
        .flow_ctrl_set          = hns3_flow_ctrl_set,
@@ -4219,6 +4912,16 @@ static const struct eth_dev_ops hns3_eth_dev_ops = {
        .dev_supported_ptypes_get = hns3_dev_supported_ptypes_get,
 };
 
+static const struct hns3_reset_ops hns3_reset_ops = {
+       .reset_service       = hns3_reset_service,
+       .stop_service        = hns3_stop_service,
+       .prepare_reset       = hns3_prepare_reset,
+       .wait_hardware_ready = hns3_wait_hardware_ready,
+       .reinit_dev          = hns3_reinit_dev,
+       .restore_conf        = hns3_restore_conf,
+       .start_service       = hns3_start_service,
+};
+
 static int
 hns3_dev_init(struct rte_eth_dev *eth_dev)
 {
@@ -4243,9 +4946,13 @@ hns3_dev_init(struct rte_eth_dev *eth_dev)
 
        hns3_set_rxtx_function(eth_dev);
        eth_dev->dev_ops = &hns3_eth_dev_ops;
-       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
+               hns3_mp_init_secondary();
+               hw->secondary_cnt++;
                return 0;
+       }
 
+       hns3_mp_init_primary();
        hw->adapter_state = HNS3_NIC_UNINITIALIZED;
 
        if (device_id == HNS3_DEV_ID_25GE_RDMA ||
@@ -4262,6 +4969,11 @@ hns3_dev_init(struct rte_eth_dev *eth_dev)
         */
        hns->pf.mps = hw->data->mtu + HNS3_ETH_OVERHEAD;
 
+       ret = hns3_reset_init(hw);
+       if (ret)
+               goto err_init_reset;
+       hw->reset.ops = &hns3_reset_ops;
+
        ret = hns3_init_pf(eth_dev);
        if (ret) {
                PMD_INIT_LOG(ERR, "Failed to init pf: %d", ret);
@@ -4291,6 +5003,14 @@ hns3_dev_init(struct rte_eth_dev *eth_dev)
         */
        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");
+               hns3_schedule_reset(hns);
+       } else {
+               /* IMP will wait ready flag before reset */
+               hns3_notify_reset_ready(hw, false);
+       }
+
        rte_eal_alarm_set(HNS3_SERVICE_INTERVAL, hns3_service_handler, eth_dev);
        hns3_info(hw, "hns3 dev initialization successful!");
        return 0;
@@ -4299,6 +5019,8 @@ err_rte_zmalloc:
        hns3_uninit_pf(eth_dev);
 
 err_init_pf:
+       rte_free(hw->reset.wait_data);
+err_init_reset:
        eth_dev->dev_ops = NULL;
        eth_dev->rx_pkt_burst = NULL;
        eth_dev->tx_pkt_burst = NULL;