net/hns3: fix log on flow director clear
[dpdk.git] / drivers / net / hns3 / hns3_mbx.c
index b03a3d6..31ab130 100644 (file)
@@ -1,28 +1,15 @@
 /* SPDX-License-Identifier: BSD-3-Clause
- * Copyright(c) 2018-2019 Hisilicon Limited.
+ * Copyright(c) 2018-2021 HiSilicon Limited.
  */
 
-#include <errno.h>
-#include <stdbool.h>
-#include <stdint.h>
-#include <stdio.h>
-#include <string.h>
-#include <inttypes.h>
-#include <unistd.h>
-#include <rte_byteorder.h>
-#include <rte_common.h>
-#include <rte_cycles.h>
-#include <rte_dev.h>
-#include <rte_ethdev_driver.h>
+#include <ethdev_driver.h>
 #include <rte_io.h>
-#include <rte_spinlock.h>
-#include <rte_pci.h>
-#include <rte_bus_pci.h>
 
 #include "hns3_ethdev.h"
 #include "hns3_regs.h"
 #include "hns3_logs.h"
 #include "hns3_intr.h"
+#include "hns3_rxtx.h"
 
 #define HNS3_CMD_CODE_OFFSET           2
 
@@ -54,49 +41,41 @@ hns3_resp_to_errno(uint16_t resp_code)
 }
 
 static void
-hns3_poll_all_sync_msg(void)
+hns3_mbx_proc_timeout(struct hns3_hw *hw, uint16_t code, uint16_t subcode)
 {
-       struct rte_eth_dev *eth_dev;
-       struct hns3_adapter *adapter;
-       const char *name;
-       uint16_t port_id;
-
-       RTE_ETH_FOREACH_DEV(port_id) {
-               eth_dev = &rte_eth_devices[port_id];
-               name = eth_dev->device->driver->name;
-               if (strcmp(name, "net_hns3") && strcmp(name, "net_hns3_vf"))
-                       continue;
-               adapter = eth_dev->data->dev_private;
-               if (!adapter || adapter->hw.adapter_state == HNS3_NIC_CLOSED)
-                       continue;
-               /* Synchronous msg, the mbx_resp.req_msg_data is non-zero */
-               if (adapter->hw.mbx_resp.req_msg_data)
-                       hns3_dev_handle_mbx_msg(&adapter->hw);
+       if (hw->mbx_resp.matching_scheme ==
+           HNS3_MBX_RESP_MATCHING_SCHEME_OF_ORIGINAL) {
+               hw->mbx_resp.lost++;
+               hns3_err(hw,
+                        "VF could not get mbx(%u,%u) head(%u) tail(%u) "
+                        "lost(%u) from PF",
+                        code, subcode, hw->mbx_resp.head, hw->mbx_resp.tail,
+                        hw->mbx_resp.lost);
+               return;
        }
+
+       hns3_err(hw, "VF could not get mbx(%u,%u) from PF", code, subcode);
 }
 
 static int
-hns3_get_mbx_resp(struct hns3_hw *hw, uint16_t code0, uint16_t code1,
+hns3_get_mbx_resp(struct hns3_hw *hw, uint16_t code, uint16_t subcode,
                  uint8_t *resp_data, uint16_t resp_len)
 {
-#define HNS3_MAX_RETRY_MS      500
+#define HNS3_MAX_RETRY_US      500000
+#define HNS3_WAIT_RESP_US      100
        struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
        struct hns3_mbx_resp_status *mbx_resp;
-       bool in_irq = false;
-       uint64_t now;
-       uint64_t end;
+       uint32_t wait_time = 0;
+       bool received;
 
        if (resp_len > HNS3_MBX_MAX_RESP_DATA_SIZE) {
-               hns3_err(hw, "VF mbx response len(=%d) exceeds maximum(=%d)",
+               hns3_err(hw, "VF mbx response len(=%u) exceeds maximum(=%d)",
                         resp_len, HNS3_MBX_MAX_RESP_DATA_SIZE);
                return -EINVAL;
        }
 
-       now = get_timeofday_ms();
-       end = now + HNS3_MAX_RETRY_MS;
-       while ((hw->mbx_resp.head != hw->mbx_resp.tail + hw->mbx_resp.lost) &&
-              (now < end)) {
-               if (rte_atomic16_read(&hw->reset.disable_cmd)) {
+       while (wait_time < HNS3_MAX_RETRY_US) {
+               if (__atomic_load_n(&hw->reset.disable_cmd, __ATOMIC_RELAXED)) {
                        hns3_err(hw, "Don't wait for mbx respone because of "
                                 "disable_cmd");
                        return -EBUSY;
@@ -109,26 +88,23 @@ hns3_get_mbx_resp(struct hns3_hw *hw, uint16_t code0, uint16_t code1,
                        return -EIO;
                }
 
-               /*
-                * The mbox response is running on the interrupt thread.
-                * Sending mbox in the interrupt thread cannot wait for the
-                * response, so polling the mbox response on the irq thread.
-                */
-               if (pthread_equal(hw->irq_thread_id, pthread_self())) {
-                       in_irq = true;
-                       hns3_poll_all_sync_msg();
-               } else {
-                       rte_delay_ms(HNS3_POLL_RESPONE_MS);
-               }
-               now = get_timeofday_ms();
+               hns3_dev_handle_mbx_msg(hw);
+               rte_delay_us(HNS3_WAIT_RESP_US);
+
+               if (hw->mbx_resp.matching_scheme ==
+                   HNS3_MBX_RESP_MATCHING_SCHEME_OF_ORIGINAL)
+                       received = (hw->mbx_resp.head ==
+                                   hw->mbx_resp.tail + hw->mbx_resp.lost);
+               else
+                       received = hw->mbx_resp.received_match_resp;
+               if (received)
+                       break;
+
+               wait_time += HNS3_WAIT_RESP_US;
        }
        hw->mbx_resp.req_msg_data = 0;
-       if (now >= end) {
-               hw->mbx_resp.lost++;
-               hns3_err(hw,
-                        "VF could not get mbx(%d,%d) head(%d) tail(%d) lost(%d) from PF in_irq:%d",
-                        code0, code1, hw->mbx_resp.head, hw->mbx_resp.tail,
-                        hw->mbx_resp.lost, in_irq);
+       if (wait_time >= HNS3_MAX_RETRY_US) {
+               hns3_mbx_proc_timeout(hw, code, subcode);
                return -ETIME;
        }
        rte_io_rmb();
@@ -143,6 +119,29 @@ hns3_get_mbx_resp(struct hns3_hw *hw, uint16_t code0, uint16_t code1,
        return 0;
 }
 
+static void
+hns3_mbx_prepare_resp(struct hns3_hw *hw, uint16_t code, uint16_t subcode)
+{
+       /*
+        * Init both matching scheme fields because we may not know the exact
+        * scheme will be used when in the initial phase.
+        *
+        * Also, there are OK to init both matching scheme fields even though
+        * we get the exact scheme which is used.
+        */
+       hw->mbx_resp.req_msg_data = (uint32_t)code << 16 | subcode;
+       hw->mbx_resp.head++;
+
+       /* Update match_id and ensure the value of match_id is not zero */
+       hw->mbx_resp.match_id++;
+       if (hw->mbx_resp.match_id == 0)
+               hw->mbx_resp.match_id = 1;
+       hw->mbx_resp.received_match_resp = false;
+
+       hw->mbx_resp.resp_status = 0;
+       memset(hw->mbx_resp.additional_info, 0, HNS3_MBX_MAX_RESP_DATA_SIZE);
+}
+
 int
 hns3_send_mbx_msg(struct hns3_hw *hw, uint16_t code, uint16_t subcode,
                  const uint8_t *msg_data, uint8_t msg_len, bool need_resp,
@@ -159,7 +158,7 @@ hns3_send_mbx_msg(struct hns3_hw *hw, uint16_t code, uint16_t subcode,
        /* first two bytes are reserved for code & subcode */
        if (msg_len > (HNS3_MBX_MAX_MSG_SIZE - HNS3_CMD_CODE_OFFSET)) {
                hns3_err(hw,
-                        "VF send mbx msg fail, msg len %d exceeds max payload len %d",
+                        "VF send mbx msg fail, msg len %u exceeds max payload len %d",
                         msg_len, HNS3_MBX_MAX_MSG_SIZE - HNS3_CMD_CODE_OFFSET);
                return -EINVAL;
        }
@@ -180,10 +179,11 @@ hns3_send_mbx_msg(struct hns3_hw *hw, uint16_t code, uint16_t subcode,
        if (need_resp) {
                req->mbx_need_resp |= HNS3_MBX_NEED_RESP_BIT;
                rte_spinlock_lock(&hw->mbx_resp.lock);
-               hw->mbx_resp.req_msg_data = (uint32_t)code << 16 | subcode;
-               hw->mbx_resp.head++;
+               hns3_mbx_prepare_resp(hw, code, subcode);
+               req->match_id = hw->mbx_resp.match_id;
                ret = hns3_cmd_send(hw, &desc, 1);
                if (ret) {
+                       hw->mbx_resp.head--;
                        rte_spinlock_unlock(&hw->mbx_resp.lock);
                        hns3_err(hw, "VF failed(=%d) to send mbx message to PF",
                                 ret);
@@ -214,48 +214,42 @@ hns3_cmd_crq_empty(struct hns3_hw *hw)
 }
 
 static void
-hns3_mbx_handler(struct hns3_hw *hw)
+hns3vf_handle_link_change_event(struct hns3_hw *hw,
+                               struct hns3_mbx_pf_to_vf_cmd *req)
 {
-       struct hns3_mac *mac = &hw->mac;
-       enum hns3_reset_level reset_level;
-       uint16_t *msg_q;
-       uint32_t tail;
-
-       tail = hw->arq.tail;
-
-       /* process all the async queue messages */
-       while (tail != hw->arq.head) {
-               msg_q = hw->arq.msg_q[hw->arq.head];
-
-               switch (msg_q[0]) {
-               case HNS3_MBX_LINK_STAT_CHANGE:
-                       memcpy(&mac->link_speed, &msg_q[2],
-                                  sizeof(mac->link_speed));
-                       mac->link_status = rte_le_to_cpu_16(msg_q[1]);
-                       mac->link_duplex = (uint8_t)rte_le_to_cpu_16(msg_q[4]);
-                       break;
-               case HNS3_MBX_ASSERTING_RESET:
-                       /* PF has asserted reset hence VF should go in pending
-                        * state and poll for the hardware reset status till it
-                        * has been completely reset. After this stack should
-                        * eventually be re-initialized.
-                        */
-                       reset_level = rte_le_to_cpu_16(msg_q[1]);
-                       hns3_atomic_set_bit(reset_level, &hw->reset.pending);
-
-                       hns3_warn(hw, "PF inform reset level %d", reset_level);
-                       hw->reset.stats.request_cnt++;
-                       hns3_schedule_reset(HNS3_DEV_HW_TO_ADAPTER(hw));
-                       break;
-               default:
-                       hns3_err(hw, "Fetched unsupported(%d) message from arq",
-                                msg_q[0]);
-                       break;
-               }
+       uint8_t link_status, link_duplex;
+       uint16_t *msg_q = req->msg;
+       uint8_t support_push_lsc;
+       uint32_t link_speed;
+
+       memcpy(&link_speed, &msg_q[2], sizeof(link_speed));
+       link_status = rte_le_to_cpu_16(msg_q[1]);
+       link_duplex = (uint8_t)rte_le_to_cpu_16(msg_q[4]);
+       hns3vf_update_link_status(hw, link_status, link_speed,
+                                 link_duplex);
+       support_push_lsc = (*(uint8_t *)&msg_q[5]) & 1u;
+       hns3vf_update_push_lsc_cap(hw, support_push_lsc);
+}
 
-               hns3_mbx_head_ptr_move_arq(hw->arq);
-               msg_q = hw->arq.msg_q[hw->arq.head];
-       }
+static void
+hns3_handle_asserting_reset(struct hns3_hw *hw,
+                           struct hns3_mbx_pf_to_vf_cmd *req)
+{
+       enum hns3_reset_level reset_level;
+       uint16_t *msg_q = req->msg;
+
+       /*
+        * PF has asserted reset hence VF should go in pending
+        * state and poll for the hardware reset status till it
+        * has been completely reset. After this stack should
+        * eventually be re-initialized.
+        */
+       reset_level = rte_le_to_cpu_16(msg_q[1]);
+       hns3_atomic_set_bit(reset_level, &hw->reset.pending);
+
+       hns3_warn(hw, "PF inform reset level %d", reset_level);
+       hw->reset.stats.request_cnt++;
+       hns3_schedule_reset(HNS3_DEV_HW_TO_ADAPTER(hw));
 }
 
 /*
@@ -277,19 +271,59 @@ hns3_update_resp_position(struct hns3_hw *hw, uint32_t resp_msg)
                if (resp->lost)
                        resp->lost--;
                hns3_warn(hw, "Received a mismatched response req_msg(%x) "
-                         "resp_msg(%x) head(%d) tail(%d) lost(%d)",
+                         "resp_msg(%x) head(%u) tail(%u) lost(%u)",
                          resp->req_msg_data, resp_msg, resp->head, tail,
                          resp->lost);
        } else if (tail + resp->lost > resp->head) {
                resp->lost--;
                hns3_warn(hw, "Received a new response again resp_msg(%x) "
-                         "head(%d) tail(%d) lost(%d)", resp_msg,
+                         "head(%u) tail(%u) lost(%u)", resp_msg,
                          resp->head, tail, resp->lost);
        }
        rte_io_wmb();
        resp->tail = tail;
 }
 
+static void
+hns3_handle_mbx_response(struct hns3_hw *hw, struct hns3_mbx_pf_to_vf_cmd *req)
+{
+       struct hns3_mbx_resp_status *resp = &hw->mbx_resp;
+       uint32_t msg_data;
+
+       if (req->match_id != 0) {
+               /*
+                * If match_id is not zero, it means PF support copy request's
+                * match_id to its response. So VF could use the match_id
+                * to match the request.
+                */
+               if (resp->matching_scheme !=
+                   HNS3_MBX_RESP_MATCHING_SCHEME_OF_MATCH_ID) {
+                       resp->matching_scheme =
+                               HNS3_MBX_RESP_MATCHING_SCHEME_OF_MATCH_ID;
+                       hns3_info(hw, "detect mailbox support match id!");
+               }
+               if (req->match_id == resp->match_id) {
+                       resp->resp_status = hns3_resp_to_errno(req->msg[3]);
+                       memcpy(resp->additional_info, &req->msg[4],
+                              HNS3_MBX_MAX_RESP_DATA_SIZE);
+                       rte_io_wmb();
+                       resp->received_match_resp = true;
+               }
+               return;
+       }
+
+       /*
+        * If the below instructions can be executed, it means PF does not
+        * support copy request's match_id to its response. So VF follows the
+        * original scheme to process.
+        */
+       resp->resp_status = hns3_resp_to_errno(req->msg[3]);
+       memcpy(resp->additional_info, &req->msg[4],
+              HNS3_MBX_MAX_RESP_DATA_SIZE);
+       msg_data = (uint32_t)req->msg[1] << 16 | req->msg[2];
+       hns3_update_resp_position(hw, msg_data);
+}
+
 static void
 hns3_link_fail_parse(struct hns3_hw *hw, uint8_t link_fail_code)
 {
@@ -312,8 +346,8 @@ hns3_link_fail_parse(struct hns3_hw *hw, uint8_t link_fail_code)
 }
 
 static void
-hns3_handle_link_change_event(struct hns3_hw *hw,
-                             struct hns3_mbx_pf_to_vf_cmd *req)
+hns3pf_handle_link_change_event(struct hns3_hw *hw,
+                               struct hns3_mbx_vf_to_pf_cmd *req)
 {
 #define LINK_STATUS_OFFSET     1
 #define LINK_FAIL_CODE_OFFSET  2
@@ -321,7 +355,31 @@ hns3_handle_link_change_event(struct hns3_hw *hw,
        if (!req->msg[LINK_STATUS_OFFSET])
                hns3_link_fail_parse(hw, req->msg[LINK_FAIL_CODE_OFFSET]);
 
-       hns3_update_link_status(hw);
+       hns3_update_linkstatus_and_event(hw, true);
+}
+
+static void
+hns3_update_port_base_vlan_info(struct hns3_hw *hw,
+                               struct hns3_mbx_pf_to_vf_cmd *req)
+{
+#define PVID_STATE_OFFSET      1
+       uint16_t new_pvid_state = req->msg[PVID_STATE_OFFSET] ?
+               HNS3_PORT_BASE_VLAN_ENABLE : HNS3_PORT_BASE_VLAN_DISABLE;
+       /*
+        * Currently, hardware doesn't support more than two layers VLAN offload
+        * based on hns3 network engine, which would cause packets loss or wrong
+        * packets for these types of packets. If the hns3 PF kernel ethdev
+        * driver sets the PVID for VF device after initialization of the
+        * related VF device, the PF driver will notify VF driver to update the
+        * PVID configuration state. The VF driver will update the PVID
+        * configuration state immediately to ensure that the VLAN process in Tx
+        * and Rx is correct. But in the window period of this state transition,
+        * packets loss or packets with wrong VLAN may occur.
+        */
+       if (hw->port_base_vlan_cfg.state != new_pvid_state) {
+               hw->port_base_vlan_cfg.state = new_pvid_state;
+               hns3_update_all_queues_pvid_proc_en(hw);
+       }
 }
 
 static void
@@ -339,31 +397,96 @@ hns3_handle_promisc_info(struct hns3_hw *hw, uint16_t promisc_en)
        }
 }
 
+static void
+hns3_handle_mbx_msg_out_intr(struct hns3_hw *hw)
+{
+       struct hns3_cmq_ring *crq = &hw->cmq.crq;
+       struct hns3_mbx_pf_to_vf_cmd *req;
+       struct hns3_cmd_desc *desc;
+       uint32_t tail, next_to_use;
+       uint8_t opcode;
+       uint16_t flag;
+
+       tail = hns3_read_dev(hw, HNS3_CMDQ_RX_TAIL_REG);
+       next_to_use = crq->next_to_use;
+       while (next_to_use != tail) {
+               desc = &crq->desc[next_to_use];
+               req = (struct hns3_mbx_pf_to_vf_cmd *)desc->data;
+               opcode = req->msg[0] & 0xff;
+
+               flag = rte_le_to_cpu_16(crq->desc[next_to_use].flag);
+               if (!hns3_get_bit(flag, HNS3_CMDQ_RX_OUTVLD_B))
+                       goto scan_next;
+
+               if (crq->desc[next_to_use].opcode == 0)
+                       goto scan_next;
+
+               if (opcode == HNS3_MBX_PF_VF_RESP) {
+                       hns3_handle_mbx_response(hw, req);
+                       /*
+                        * Clear opcode to inform intr thread don't process
+                        * again.
+                        */
+                       crq->desc[crq->next_to_use].opcode = 0;
+               }
+
+scan_next:
+               next_to_use = (next_to_use + 1) % hw->cmq.crq.desc_num;
+       }
+}
+
 void
 hns3_dev_handle_mbx_msg(struct hns3_hw *hw)
 {
-       struct hns3_mbx_resp_status *resp = &hw->mbx_resp;
+       struct hns3_adapter *hns = HNS3_DEV_HW_TO_ADAPTER(hw);
        struct hns3_cmq_ring *crq = &hw->cmq.crq;
        struct hns3_mbx_pf_to_vf_cmd *req;
        struct hns3_cmd_desc *desc;
-       uint32_t msg_data;
-       uint16_t *msg_q;
+       bool handle_out;
+       uint8_t opcode;
        uint16_t flag;
-       uint8_t *temp;
-       int i;
+
+       rte_spinlock_lock(&hw->cmq.crq.lock);
+
+       handle_out = (rte_eal_process_type() != RTE_PROC_PRIMARY ||
+                     !rte_thread_is_intr()) && hns->is_vf;
+       if (handle_out) {
+               /*
+                * Currently, any threads in the primary and secondary processes
+                * could send mailbox sync request, so it will need to process
+                * the crq message (which is the HNS3_MBX_PF_VF_RESP) in there
+                * own thread context. It may also process other messages
+                * because it uses the policy of processing all pending messages
+                * at once.
+                * But some messages such as HNS3_MBX_PUSH_LINK_STATUS could
+                * only process within the intr thread in primary process,
+                * otherwise it may lead to report lsc event in secondary
+                * process.
+                * So the threads other than intr thread in primary process
+                * could only process HNS3_MBX_PF_VF_RESP message, if the
+                * message processed, its opcode will rewrite with zero, then
+                * the intr thread in primary process will not process again.
+                */
+               hns3_handle_mbx_msg_out_intr(hw);
+               rte_spinlock_unlock(&hw->cmq.crq.lock);
+               return;
+       }
 
        while (!hns3_cmd_crq_empty(hw)) {
-               if (rte_atomic16_read(&hw->reset.disable_cmd))
+               if (__atomic_load_n(&hw->reset.disable_cmd, __ATOMIC_RELAXED)) {
+                       rte_spinlock_unlock(&hw->cmq.crq.lock);
                        return;
+               }
 
                desc = &crq->desc[crq->next_to_use];
                req = (struct hns3_mbx_pf_to_vf_cmd *)desc->data;
+               opcode = req->msg[0] & 0xff;
 
                flag = rte_le_to_cpu_16(crq->desc[crq->next_to_use].flag);
                if (unlikely(!hns3_get_bit(flag, HNS3_CMDQ_RX_OUTVLD_B))) {
                        hns3_warn(hw,
-                                 "dropped invalid mailbox message, code = %d",
-                                 req->msg[0]);
+                                 "dropped invalid mailbox message, code = %u",
+                                 opcode);
 
                        /* dropping/not processing this invalid message */
                        crq->desc[crq->next_to_use].flag = 0;
@@ -371,29 +494,41 @@ hns3_dev_handle_mbx_msg(struct hns3_hw *hw)
                        continue;
                }
 
-               switch (req->msg[0]) {
-               case HNS3_MBX_PF_VF_RESP:
-                       resp->resp_status = hns3_resp_to_errno(req->msg[3]);
+               handle_out = hns->is_vf && desc->opcode == 0;
+               if (handle_out) {
+                       /* Message already processed by other thread */
+                       crq->desc[crq->next_to_use].flag = 0;
+                       hns3_mbx_ring_ptr_move_crq(crq);
+                       continue;
+               }
 
-                       temp = (uint8_t *)&req->msg[4];
-                       for (i = 0; i < HNS3_MBX_MAX_RESP_DATA_SIZE; i++) {
-                               resp->additional_info[i] = *temp;
-                               temp++;
-                       }
-                       msg_data = (uint32_t)req->msg[1] << 16 | req->msg[2];
-                       hns3_update_resp_position(hw, msg_data);
+               switch (opcode) {
+               case HNS3_MBX_PF_VF_RESP:
+                       hns3_handle_mbx_response(hw, req);
                        break;
                case HNS3_MBX_LINK_STAT_CHANGE:
+                       hns3vf_handle_link_change_event(hw, req);
+                       break;
                case HNS3_MBX_ASSERTING_RESET:
-                       msg_q = hw->arq.msg_q[hw->arq.tail];
-                       memcpy(&msg_q[0], req->msg,
-                              HNS3_MBX_MAX_ARQ_MSG_SIZE * sizeof(uint16_t));
-                       hns3_mbx_tail_ptr_move_arq(hw->arq);
-
-                       hns3_mbx_handler(hw);
+                       hns3_handle_asserting_reset(hw, req);
                        break;
                case HNS3_MBX_PUSH_LINK_STATUS:
-                       hns3_handle_link_change_event(hw, req);
+                       /*
+                        * This message is reported by the firmware and is
+                        * reported in 'struct hns3_mbx_vf_to_pf_cmd' format.
+                        * Therefore, we should cast the req variable to
+                        * 'struct hns3_mbx_vf_to_pf_cmd' and then process it.
+                        */
+                       hns3pf_handle_link_change_event(hw,
+                               (struct hns3_mbx_vf_to_pf_cmd *)req);
+                       break;
+               case HNS3_MBX_PUSH_VLAN_INFO:
+                       /*
+                        * When the PVID configuration status of VF device is
+                        * changed by the hns3 PF kernel driver, VF driver will
+                        * receive this mailbox message from PF driver.
+                        */
+                       hns3_update_port_base_vlan_info(hw, req);
                        break;
                case HNS3_MBX_PUSH_PROMISC_INFO:
                        /*
@@ -404,8 +539,7 @@ hns3_dev_handle_mbx_msg(struct hns3_hw *hw)
                        hns3_handle_promisc_info(hw, req->msg[1]);
                        break;
                default:
-                       hns3_err(hw,
-                                "VF received unsupported(%d) mbx msg from PF",
+                       hns3_err(hw, "received unsupported(%u) mbx msg",
                                 req->msg[0]);
                        break;
                }
@@ -416,4 +550,6 @@ hns3_dev_handle_mbx_msg(struct hns3_hw *hw)
 
        /* Write back CMDQ_RQ header pointer, IMP need this pointer */
        hns3_write_dev(hw, HNS3_CMDQ_RX_HEAD_REG, crq->next_to_use);
+
+       rte_spinlock_unlock(&hw->cmq.crq.lock);
 }