Merge branch 'hns3-next'

Huazhong Tan says:

====================
net: hns3: some optimizaions related to work task

This series refactors the work task of the HNS3 ethernet driver.

[patch 1/5] uses delayed workqueue to replace the timer for
hclgevf_service task, make the code simpler.

[patch 2/5] & [patch 3/5] unifies current mailbox, reset and
service work into one.

[patch 4/5] allocates a private work queue with WQ_MEM_RECLAIM
for the HNS3 driver.

[patch 5/5] adds a new flag to indicate whether reset fails,
and prevent scheduling service task to handle periodic task
when this flag has been set.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_debugfs.c

@@ -976,6 +976,14 @@ void hclge_dbg_dump_rst_info(struct hclge_dev *hdev)
 	dev_info(&hdev->pdev->dev, "hdev state: 0x%lx\n", hdev->state);
 }
 
+static void hclge_dbg_dump_serv_info(struct hclge_dev *hdev)
+{
+	dev_info(&hdev->pdev->dev, "last_serv_processed: %lu\n",
+		 hdev->last_serv_processed);
+	dev_info(&hdev->pdev->dev, "last_serv_cnt: %lu\n",
+		 hdev->serv_processed_cnt);
+}
+
 static void hclge_dbg_get_m7_stats_info(struct hclge_dev *hdev)
 {
 	struct hclge_desc *desc_src, *desc_tmp;
@@ -1227,6 +1235,8 @@ int hclge_dbg_run_cmd(struct hnae3_handle *handle, const char *cmd_buf)
 		hclge_dbg_dump_reg_cmd(hdev, &cmd_buf[sizeof(DUMP_REG)]);
 	} else if (strncmp(cmd_buf, "dump reset info", 15) == 0) {
 		hclge_dbg_dump_rst_info(hdev);
+	} else if (strncmp(cmd_buf, "dump serv info", 14) == 0) {
+		hclge_dbg_dump_serv_info(hdev);
 	} else if (strncmp(cmd_buf, "dump m7 info", 12) == 0) {
 		hclge_dbg_get_m7_stats_info(hdev);
 	} else if (strncmp(cmd_buf, "dump ncl_config", 15) == 0) {

drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_main.c

@@ -72,6 +72,8 @@ static int hclge_set_default_loopback(struct hclge_dev *hdev);
 
 static struct hnae3_ae_algo ae_algo;
 
+static struct workqueue_struct *hclge_wq;
+
 static const struct pci_device_id ae_algo_pci_tbl[] = {
 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_GE), 0},
 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_25GE), 0},
@@ -416,7 +418,7 @@ static int hclge_mac_update_stats_defective(struct hclge_dev *hdev)
 {
 #define HCLGE_MAC_CMD_NUM 21
 
-	u64 *data = (u64 *)(&hdev->hw_stats.mac_stats);
+	u64 *data = (u64 *)(&hdev->mac_stats);
 	struct hclge_desc desc[HCLGE_MAC_CMD_NUM];
 	__le64 *desc_data;
 	int i, k, n;
@@ -453,7 +455,7 @@ static int hclge_mac_update_stats_defective(struct hclge_dev *hdev)
 
 static int hclge_mac_update_stats_complete(struct hclge_dev *hdev, u32 desc_num)
 {
-	u64 *data = (u64 *)(&hdev->hw_stats.mac_stats);
+	u64 *data = (u64 *)(&hdev->mac_stats);
 	struct hclge_desc *desc;
 	__le64 *desc_data;
 	u16 i, k, n;
@@ -802,7 +804,7 @@ static void hclge_get_stats(struct hnae3_handle *handle, u64 *data)
 	struct hclge_dev *hdev = vport->back;
 	u64 *p;
 
-	p = hclge_comm_get_stats(&hdev->hw_stats.mac_stats, g_mac_stats_string,
+	p = hclge_comm_get_stats(&hdev->mac_stats, g_mac_stats_string,
 				 ARRAY_SIZE(g_mac_stats_string), data);
 	p = hclge_tqps_get_stats(handle, p);
 }
@@ -815,8 +817,8 @@ static void hclge_get_mac_stat(struct hnae3_handle *handle,
 
 	hclge_update_stats(handle, NULL);
 
-	mac_stats->tx_pause_cnt = hdev->hw_stats.mac_stats.mac_tx_mac_pause_num;
-	mac_stats->rx_pause_cnt = hdev->hw_stats.mac_stats.mac_rx_mac_pause_num;
+	mac_stats->tx_pause_cnt = hdev->mac_stats.mac_tx_mac_pause_num;
+	mac_stats->rx_pause_cnt = hdev->mac_stats.mac_rx_mac_pause_num;
 }
 
 static int hclge_parse_func_status(struct hclge_dev *hdev,
@@ -2665,31 +2667,27 @@ static int hclge_mac_init(struct hclge_dev *hdev)
 
 static void hclge_mbx_task_schedule(struct hclge_dev *hdev)
 {
-	if (!test_bit(HCLGE_STATE_CMD_DISABLE, &hdev->state) &&
+	if (!test_bit(HCLGE_STATE_REMOVING, &hdev->state) &&
 	    !test_and_set_bit(HCLGE_STATE_MBX_SERVICE_SCHED, &hdev->state))
-		queue_work_on(cpumask_first(&hdev->affinity_mask), system_wq,
-			      &hdev->mbx_service_task);
+		mod_delayed_work_on(cpumask_first(&hdev->affinity_mask),
+				    hclge_wq, &hdev->service_task, 0);
 }
 
 static void hclge_reset_task_schedule(struct hclge_dev *hdev)
 {
 	if (!test_bit(HCLGE_STATE_REMOVING, &hdev->state) &&
 	    !test_and_set_bit(HCLGE_STATE_RST_SERVICE_SCHED, &hdev->state))
-		queue_work_on(cpumask_first(&hdev->affinity_mask), system_wq,
-			      &hdev->rst_service_task);
+		mod_delayed_work_on(cpumask_first(&hdev->affinity_mask),
+				    hclge_wq, &hdev->service_task, 0);
 }
 
 void hclge_task_schedule(struct hclge_dev *hdev, unsigned long delay_time)
 {
-	if (!test_bit(HCLGE_STATE_DOWN, &hdev->state) &&
-	    !test_bit(HCLGE_STATE_REMOVING, &hdev->state) &&
-	    !test_and_set_bit(HCLGE_STATE_SERVICE_SCHED, &hdev->state)) {
-		hdev->hw_stats.stats_timer++;
-		hdev->fd_arfs_expire_timer++;
+	if (!test_bit(HCLGE_STATE_REMOVING, &hdev->state) &&
+	    !test_bit(HCLGE_STATE_RST_FAIL, &hdev->state))
 		mod_delayed_work_on(cpumask_first(&hdev->affinity_mask),
-				    system_wq, &hdev->service_task,
+				    hclge_wq, &hdev->service_task,
 				    delay_time);
-	}
 }
 
 static int hclge_get_mac_link_status(struct hclge_dev *hdev)
@@ -2748,6 +2746,10 @@ static void hclge_update_link_status(struct hclge_dev *hdev)
 
 	if (!client)
 		return;
+
+	if (test_and_set_bit(HCLGE_STATE_LINK_UPDATING, &hdev->state))
+		return;
+
 	state = hclge_get_mac_phy_link(hdev);
 	if (state != hdev->hw.mac.link) {
 		for (i = 0; i < hdev->num_vmdq_vport + 1; i++) {
@@ -2761,6 +2763,8 @@ static void hclge_update_link_status(struct hclge_dev *hdev)
 		}
 		hdev->hw.mac.link = state;
 	}
+
+	clear_bit(HCLGE_STATE_LINK_UPDATING, &hdev->state);
 }
 
 static void hclge_update_port_capability(struct hclge_mac *mac)
@@ -3352,6 +3356,18 @@ static int hclge_set_all_vf_rst(struct hclge_dev *hdev, bool reset)
 	return 0;
 }
 
+static void hclge_mailbox_service_task(struct hclge_dev *hdev)
+{
+	if (!test_and_clear_bit(HCLGE_STATE_MBX_SERVICE_SCHED, &hdev->state) ||
+	    test_bit(HCLGE_STATE_CMD_DISABLE, &hdev->state) ||
+	    test_and_set_bit(HCLGE_STATE_MBX_HANDLING, &hdev->state))
+		return;
+
+	hclge_mbx_handler(hdev);
+
+	clear_bit(HCLGE_STATE_MBX_HANDLING, &hdev->state);
+}
+
 static int hclge_func_reset_sync_vf(struct hclge_dev *hdev)
 {
 	struct hclge_pf_rst_sync_cmd *req;
@@ -3363,6 +3379,9 @@ static int hclge_func_reset_sync_vf(struct hclge_dev *hdev)
 	hclge_cmd_setup_basic_desc(&desc, HCLGE_OPC_QUERY_VF_RST_RDY, true);
 
 	do {
+		/* vf need to down netdev by mbx during PF or FLR reset */
+		hclge_mailbox_service_task(hdev);
+
 		ret = hclge_cmd_send(&hdev->hw, &desc, 1);
 		/* for compatible with old firmware, wait
 		 * 100 ms for VF to stop IO
@@ -3672,6 +3691,8 @@ static bool hclge_reset_err_handle(struct hclge_dev *hdev)
 
 	hclge_dbg_dump_rst_info(hdev);
 
+	set_bit(HCLGE_STATE_RST_FAIL, &hdev->state);
+
 	return false;
 }
 
@@ -3825,6 +3846,7 @@ static void hclge_reset(struct hclge_dev *hdev)
 	hdev->rst_stats.reset_fail_cnt = 0;
 	hdev->rst_stats.reset_done_cnt++;
 	ae_dev->reset_type = HNAE3_NONE_RESET;
+	clear_bit(HCLGE_STATE_RST_FAIL, &hdev->state);
 
 	/* if default_reset_request has a higher level reset request,
 	 * it should be handled as soon as possible. since some errors
@@ -3939,36 +3961,19 @@ static void hclge_reset_subtask(struct hclge_dev *hdev)
 	hdev->reset_type = HNAE3_NONE_RESET;
 }
 
-static void hclge_reset_service_task(struct work_struct *work)
+static void hclge_reset_service_task(struct hclge_dev *hdev)
 {
-	struct hclge_dev *hdev =
-		container_of(work, struct hclge_dev, rst_service_task);
+	if (!test_and_clear_bit(HCLGE_STATE_RST_SERVICE_SCHED, &hdev->state))
+		return;
 
 	if (test_and_set_bit(HCLGE_STATE_RST_HANDLING, &hdev->state))
 		return;
 
-	clear_bit(HCLGE_STATE_RST_SERVICE_SCHED, &hdev->state);
-
 	hclge_reset_subtask(hdev);
 
 	clear_bit(HCLGE_STATE_RST_HANDLING, &hdev->state);
 }
 
-static void hclge_mailbox_service_task(struct work_struct *work)
-{
-	struct hclge_dev *hdev =
-		container_of(work, struct hclge_dev, mbx_service_task);
-
-	if (test_and_set_bit(HCLGE_STATE_MBX_HANDLING, &hdev->state))
-		return;
-
-	clear_bit(HCLGE_STATE_MBX_SERVICE_SCHED, &hdev->state);
-
-	hclge_mbx_handler(hdev);
-
-	clear_bit(HCLGE_STATE_MBX_HANDLING, &hdev->state);
-}
-
 static void hclge_update_vport_alive(struct hclge_dev *hdev)
 {
 	int i;
@@ -3986,29 +3991,62 @@ static void hclge_update_vport_alive(struct hclge_dev *hdev)
 	}
 }
 
-static void hclge_service_task(struct work_struct *work)
+static void hclge_periodic_service_task(struct hclge_dev *hdev)
 {
-	struct hclge_dev *hdev =
-		container_of(work, struct hclge_dev, service_task.work);
+	unsigned long delta = round_jiffies_relative(HZ);
 
-	clear_bit(HCLGE_STATE_SERVICE_SCHED, &hdev->state);
+	/* Always handle the link updating to make sure link state is
+	 * updated when it is triggered by mbx.
+	 */
+	hclge_update_link_status(hdev);
 
-	if (hdev->hw_stats.stats_timer >= HCLGE_STATS_TIMER_INTERVAL) {
-		hclge_update_stats_for_all(hdev);
-		hdev->hw_stats.stats_timer = 0;
+	if (time_is_after_jiffies(hdev->last_serv_processed + HZ)) {
+		delta = jiffies - hdev->last_serv_processed;
+
+		if (delta < round_jiffies_relative(HZ)) {
+			delta = round_jiffies_relative(HZ) - delta;
+			goto out;
+		}
 	}
 
-	hclge_update_port_info(hdev);
-	hclge_update_link_status(hdev);
+	hdev->serv_processed_cnt++;
 	hclge_update_vport_alive(hdev);
+
+	if (test_bit(HCLGE_STATE_DOWN, &hdev->state)) {
+		hdev->last_serv_processed = jiffies;
+		goto out;
+	}
+
+	if (!(hdev->serv_processed_cnt % HCLGE_STATS_TIMER_INTERVAL))
+		hclge_update_stats_for_all(hdev);
+
+	hclge_update_port_info(hdev);
 	hclge_sync_vlan_filter(hdev);
 
-	if (hdev->fd_arfs_expire_timer >= HCLGE_FD_ARFS_EXPIRE_TIMER_INTERVAL) {
+	if (!(hdev->serv_processed_cnt % HCLGE_ARFS_EXPIRE_INTERVAL))
 		hclge_rfs_filter_expire(hdev);
-		hdev->fd_arfs_expire_timer = 0;
-	}
 
-	hclge_task_schedule(hdev, round_jiffies_relative(HZ));
+	hdev->last_serv_processed = jiffies;
+
+out:
+	hclge_task_schedule(hdev, delta);
+}
+
+static void hclge_service_task(struct work_struct *work)
+{
+	struct hclge_dev *hdev =
+		container_of(work, struct hclge_dev, service_task.work);
+
+	hclge_reset_service_task(hdev);
+	hclge_mailbox_service_task(hdev);
+	hclge_periodic_service_task(hdev);
+
+	/* Handle reset and mbx again in case periodical task delays the
+	 * handling by calling hclge_task_schedule() in
+	 * hclge_periodic_service_task().
+	 */
+	hclge_reset_service_task(hdev);
+	hclge_mailbox_service_task(hdev);
 }
 
 struct hclge_vport *hclge_get_vport(struct hnae3_handle *handle)
@@ -6734,6 +6772,19 @@ static void hclge_reset_tqp_stats(struct hnae3_handle *handle)
 	}
 }
 
+static void hclge_flush_link_update(struct hclge_dev *hdev)
+{
+#define HCLGE_FLUSH_LINK_TIMEOUT	100000
+
+	unsigned long last = hdev->serv_processed_cnt;
+	int i = 0;
+
+	while (test_bit(HCLGE_STATE_LINK_UPDATING, &hdev->state) &&
+	       i++ < HCLGE_FLUSH_LINK_TIMEOUT &&
+	       last == hdev->serv_processed_cnt)
+		usleep_range(1, 1);
+}
+
 static void hclge_set_timer_task(struct hnae3_handle *handle, bool enable)
 {
 	struct hclge_vport *vport = hclge_get_vport(handle);
@@ -6742,12 +6793,12 @@ static void hclge_set_timer_task(struct hnae3_handle *handle, bool enable)
 	if (enable) {
 		hclge_task_schedule(hdev, round_jiffies_relative(HZ));
 	} else {
-		/* Set the DOWN flag here to disable the service to be
-		 * scheduled again
-		 */
+		/* Set the DOWN flag here to disable link updating */
 		set_bit(HCLGE_STATE_DOWN, &hdev->state);
-		cancel_delayed_work_sync(&hdev->service_task);
-		clear_bit(HCLGE_STATE_SERVICE_SCHED, &hdev->state);
+
+		/* flush memory to make sure DOWN is seen by service task */
+		smp_mb__before_atomic();
+		hclge_flush_link_update(hdev);
 	}
 }
 
@@ -9256,6 +9307,7 @@ static void hclge_state_init(struct hclge_dev *hdev)
 	set_bit(HCLGE_STATE_DOWN, &hdev->state);
 	clear_bit(HCLGE_STATE_RST_SERVICE_SCHED, &hdev->state);
 	clear_bit(HCLGE_STATE_RST_HANDLING, &hdev->state);
+	clear_bit(HCLGE_STATE_RST_FAIL, &hdev->state);
 	clear_bit(HCLGE_STATE_MBX_SERVICE_SCHED, &hdev->state);
 	clear_bit(HCLGE_STATE_MBX_HANDLING, &hdev->state);
 }
@@ -9269,10 +9321,6 @@ static void hclge_state_uninit(struct hclge_dev *hdev)
 		del_timer_sync(&hdev->reset_timer);
 	if (hdev->service_task.work.func)
 		cancel_delayed_work_sync(&hdev->service_task);
-	if (hdev->rst_service_task.func)
-		cancel_work_sync(&hdev->rst_service_task);
-	if (hdev->mbx_service_task.func)
-		cancel_work_sync(&hdev->mbx_service_task);
 }
 
 static void hclge_flr_prepare(struct hnae3_ae_dev *ae_dev)
@@ -9477,8 +9525,6 @@ static int hclge_init_ae_dev(struct hnae3_ae_dev *ae_dev)
 
 	timer_setup(&hdev->reset_timer, hclge_reset_timer, 0);
 	INIT_DELAYED_WORK(&hdev->service_task, hclge_service_task);
-	INIT_WORK(&hdev->rst_service_task, hclge_reset_service_task);
-	INIT_WORK(&hdev->mbx_service_task, hclge_mailbox_service_task);
 
 	/* Setup affinity after service timer setup because add_timer_on
 	 * is called in affinity notify.
@@ -9512,6 +9558,8 @@ static int hclge_init_ae_dev(struct hnae3_ae_dev *ae_dev)
 	dev_info(&hdev->pdev->dev, "%s driver initialization finished.\n",
 		 HCLGE_DRIVER_NAME);
 
+	hclge_task_schedule(hdev, round_jiffies_relative(HZ));
+
 	return 0;
 
 err_mdiobus_unreg:
@@ -9534,7 +9582,7 @@ static int hclge_init_ae_dev(struct hnae3_ae_dev *ae_dev)
 
 static void hclge_stats_clear(struct hclge_dev *hdev)
 {
-	memset(&hdev->hw_stats, 0, sizeof(hdev->hw_stats));
+	memset(&hdev->mac_stats, 0, sizeof(hdev->mac_stats));
 }
 
 static int hclge_set_mac_spoofchk(struct hclge_dev *hdev, int vf, bool enable)
@@ -10611,6 +10659,12 @@ static int hclge_init(void)
 {
 	pr_info("%s is initializing\n", HCLGE_NAME);
 
+	hclge_wq = alloc_workqueue("%s", WQ_MEM_RECLAIM, 0, HCLGE_NAME);
+	if (!hclge_wq) {
+		pr_err("%s: failed to create workqueue\n", HCLGE_NAME);
+		return -ENOMEM;
+	}
+
 	hnae3_register_ae_algo(&ae_algo);
 
 	return 0;
@@ -10619,6 +10673,7 @@ static int hclge_init(void)
 static void hclge_exit(void)
 {
 	hnae3_unregister_ae_algo(&ae_algo);
+	destroy_workqueue(hclge_wq);
 }
 module_init(hclge_init);
 module_exit(hclge_exit);

drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_main.h

@@ -208,13 +208,14 @@ enum HCLGE_DEV_STATE {
 	HCLGE_STATE_NIC_REGISTERED,
 	HCLGE_STATE_ROCE_REGISTERED,
 	HCLGE_STATE_SERVICE_INITED,
-	HCLGE_STATE_SERVICE_SCHED,
 	HCLGE_STATE_RST_SERVICE_SCHED,
 	HCLGE_STATE_RST_HANDLING,
 	HCLGE_STATE_MBX_SERVICE_SCHED,
 	HCLGE_STATE_MBX_HANDLING,
 	HCLGE_STATE_STATISTICS_UPDATING,
 	HCLGE_STATE_CMD_DISABLE,
+	HCLGE_STATE_LINK_UPDATING,
+	HCLGE_STATE_RST_FAIL,
 	HCLGE_STATE_MAX
 };
 
@@ -454,11 +455,7 @@ struct hclge_mac_stats {
 	u64 mac_rx_ctrl_pkt_num;
 };
 
-#define HCLGE_STATS_TIMER_INTERVAL	(60 * 5)
-struct hclge_hw_stats {
-	struct hclge_mac_stats      mac_stats;
-	u32 stats_timer;
-};
+#define HCLGE_STATS_TIMER_INTERVAL	300UL
 
 struct hclge_vlan_type_cfg {
 	u16 rx_ot_fst_vlan_type;
@@ -549,7 +546,7 @@ struct key_info {
 
 /* assigned by firmware, the real filter number for each pf may be less */
 #define MAX_FD_FILTER_NUM	4096
-#define HCLGE_FD_ARFS_EXPIRE_TIMER_INTERVAL	5
+#define HCLGE_ARFS_EXPIRE_INTERVAL	5UL
 
 enum HCLGE_FD_ACTIVE_RULE_TYPE {
 	HCLGE_FD_RULE_NONE,
@@ -712,7 +709,7 @@ struct hclge_dev {
 	struct hnae3_ae_dev *ae_dev;
 	struct hclge_hw hw;
 	struct hclge_misc_vector misc_vector;
-	struct hclge_hw_stats hw_stats;
+	struct hclge_mac_stats mac_stats;
 	unsigned long state;
 	unsigned long flr_state;
 	unsigned long last_reset_time;
@@ -774,8 +771,6 @@ struct hclge_dev {
 	unsigned long service_timer_previous;
 	struct timer_list reset_timer;
 	struct delayed_work service_task;
-	struct work_struct rst_service_task;
-	struct work_struct mbx_service_task;
 
 	bool cur_promisc;
 	int num_alloc_vfs;	/* Actual number of VFs allocated */
@@ -811,7 +806,8 @@ struct hclge_dev {
 	struct hlist_head fd_rule_list;
 	spinlock_t fd_rule_lock; /* protect fd_rule_list and fd_bmap */
 	u16 hclge_fd_rule_num;
-	u16 fd_arfs_expire_timer;
+	unsigned long serv_processed_cnt;
+	unsigned long last_serv_processed;
 	unsigned long fd_bmap[BITS_TO_LONGS(MAX_FD_FILTER_NUM)];
 	enum HCLGE_FD_ACTIVE_RULE_TYPE fd_active_type;
 	u8 fd_en;

drivers/net/ethernet/hisilicon/hns3/hns3pf/hclge_mbx.c

@@ -635,7 +635,6 @@ static void hclge_handle_link_change_event(struct hclge_dev *hdev,
 #define LINK_STATUS_OFFSET	1
 #define LINK_FAIL_CODE_OFFSET	2
 
-	clear_bit(HCLGE_STATE_SERVICE_SCHED, &hdev->state);
 	hclge_task_schedule(hdev, 0);
 
 	if (!req->msg[LINK_STATUS_OFFSET])

drivers/net/ethernet/hisilicon/hns3/hns3vf/hclgevf_main.c

@@ -16,6 +16,8 @@
 static int hclgevf_reset_hdev(struct hclgevf_dev *hdev);
 static struct hnae3_ae_algo ae_algovf;
 
+static struct workqueue_struct *hclgevf_wq;
+
 static const struct pci_device_id ae_algovf_pci_tbl[] = {
 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_VF), 0},
 	{PCI_VDEVICE(HUAWEI, HNAE3_DEV_ID_100G_RDMA_DCB_PFC_VF), 0},
@@ -440,6 +442,9 @@ void hclgevf_update_link_status(struct hclgevf_dev *hdev, int link_state)
 	struct hnae3_client *rclient;
 	struct hnae3_client *client;
 
+	if (test_and_set_bit(HCLGEVF_STATE_LINK_UPDATING, &hdev->state))
+		return;
+
 	client = handle->client;
 	rclient = hdev->roce_client;
 
@@ -452,6 +457,8 @@ void hclgevf_update_link_status(struct hclgevf_dev *hdev, int link_state)
 			rclient->ops->link_status_change(rhandle, !!link_state);
 		hdev->hw.mac.link = link_state;
 	}
+
+	clear_bit(HCLGEVF_STATE_LINK_UPDATING, &hdev->state);
 }
 
 static void hclgevf_update_link_mode(struct hclgevf_dev *hdev)
@@ -1591,6 +1598,7 @@ static void hclgevf_reset_err_handle(struct hclgevf_dev *hdev)
 		set_bit(HCLGEVF_RESET_PENDING, &hdev->reset_state);
 		hclgevf_reset_task_schedule(hdev);
 	} else {
+		set_bit(HCLGEVF_STATE_RST_FAIL, &hdev->state);
 		hclgevf_dump_rst_info(hdev);
 	}
 }
@@ -1652,6 +1660,7 @@ static int hclgevf_reset(struct hclgevf_dev *hdev)
 	ae_dev->reset_type = HNAE3_NONE_RESET;
 	hdev->rst_stats.rst_done_cnt++;
 	hdev->rst_stats.rst_fail_cnt = 0;
+	clear_bit(HCLGEVF_STATE_RST_FAIL, &hdev->state);
 
 	return ret;
 err_reset_lock:
@@ -1767,62 +1776,39 @@ static void hclgevf_get_misc_vector(struct hclgevf_dev *hdev)
 
 void hclgevf_reset_task_schedule(struct hclgevf_dev *hdev)
 {
-	if (!test_bit(HCLGEVF_STATE_RST_SERVICE_SCHED, &hdev->state) &&
-	    !test_bit(HCLGEVF_STATE_REMOVING, &hdev->state)) {
-		set_bit(HCLGEVF_STATE_RST_SERVICE_SCHED, &hdev->state);
-		schedule_work(&hdev->rst_service_task);
-	}
+	if (!test_bit(HCLGEVF_STATE_REMOVING, &hdev->state) &&
+	    !test_and_set_bit(HCLGEVF_STATE_RST_SERVICE_SCHED,
+			      &hdev->state))
+		mod_delayed_work(hclgevf_wq, &hdev->service_task, 0);
 }
 
 void hclgevf_mbx_task_schedule(struct hclgevf_dev *hdev)
 {
-	if (!test_bit(HCLGEVF_STATE_MBX_SERVICE_SCHED, &hdev->state) &&
-	    !test_bit(HCLGEVF_STATE_MBX_HANDLING, &hdev->state)) {
-		set_bit(HCLGEVF_STATE_MBX_SERVICE_SCHED, &hdev->state);
-		schedule_work(&hdev->mbx_service_task);
-	}
+	if (!test_bit(HCLGEVF_STATE_REMOVING, &hdev->state) &&
+	    !test_and_set_bit(HCLGEVF_STATE_MBX_SERVICE_SCHED,
+			      &hdev->state))
+		mod_delayed_work(hclgevf_wq, &hdev->service_task, 0);
 }
 
-static void hclgevf_task_schedule(struct hclgevf_dev *hdev)
+static void hclgevf_task_schedule(struct hclgevf_dev *hdev,
+				  unsigned long delay)
 {
-	if (!test_bit(HCLGEVF_STATE_DOWN, &hdev->state)  &&
-	    !test_and_set_bit(HCLGEVF_STATE_SERVICE_SCHED, &hdev->state))
-		schedule_work(&hdev->service_task);
+	if (!test_bit(HCLGEVF_STATE_REMOVING, &hdev->state) &&
+	    !test_bit(HCLGEVF_STATE_RST_FAIL, &hdev->state))
+		mod_delayed_work(hclgevf_wq, &hdev->service_task, delay);
 }
 
-static void hclgevf_deferred_task_schedule(struct hclgevf_dev *hdev)
-{
-	/* if we have any pending mailbox event then schedule the mbx task */
-	if (hdev->mbx_event_pending)
-		hclgevf_mbx_task_schedule(hdev);
-
-	if (test_bit(HCLGEVF_RESET_PENDING, &hdev->reset_state))
-		hclgevf_reset_task_schedule(hdev);
-}
-
-static void hclgevf_service_timer(struct timer_list *t)
-{
-	struct hclgevf_dev *hdev = from_timer(hdev, t, service_timer);
-
-	mod_timer(&hdev->service_timer, jiffies +
-		  HCLGEVF_GENERAL_TASK_INTERVAL * HZ);
-
-	hdev->stats_timer++;
-	hclgevf_task_schedule(hdev);
-}
-
-static void hclgevf_reset_service_task(struct work_struct *work)
+static void hclgevf_reset_service_task(struct hclgevf_dev *hdev)
 {
 #define	HCLGEVF_MAX_RESET_ATTEMPTS_CNT	3
 
-	struct hclgevf_dev *hdev =
-		container_of(work, struct hclgevf_dev, rst_service_task);
 	int ret;
 
-	if (test_and_set_bit(HCLGEVF_STATE_RST_HANDLING, &hdev->state))
+	if (!test_and_clear_bit(HCLGEVF_STATE_RST_SERVICE_SCHED, &hdev->state))
 		return;
 
-	clear_bit(HCLGEVF_STATE_RST_SERVICE_SCHED, &hdev->state);
+	if (test_and_set_bit(HCLGEVF_STATE_RST_HANDLING, &hdev->state))
+		return;
 
 	if (test_and_clear_bit(HCLGEVF_RESET_PENDING,
 			       &hdev->reset_state)) {
@@ -1885,39 +1871,24 @@ static void hclgevf_reset_service_task(struct work_struct *work)
 	clear_bit(HCLGEVF_STATE_RST_HANDLING, &hdev->state);
 }
 
-static void hclgevf_mailbox_service_task(struct work_struct *work)
+static void hclgevf_mailbox_service_task(struct hclgevf_dev *hdev)
 {
-	struct hclgevf_dev *hdev;
-
-	hdev = container_of(work, struct hclgevf_dev, mbx_service_task);
+	if (!test_and_clear_bit(HCLGEVF_STATE_MBX_SERVICE_SCHED, &hdev->state))
+		return;
 
 	if (test_and_set_bit(HCLGEVF_STATE_MBX_HANDLING, &hdev->state))
 		return;
 
-	clear_bit(HCLGEVF_STATE_MBX_SERVICE_SCHED, &hdev->state);
-
 	hclgevf_mbx_async_handler(hdev);
 
 	clear_bit(HCLGEVF_STATE_MBX_HANDLING, &hdev->state);
 }
 
-static void hclgevf_keep_alive_timer(struct timer_list *t)
+static void hclgevf_keep_alive(struct hclgevf_dev *hdev)
 {
-	struct hclgevf_dev *hdev = from_timer(hdev, t, keep_alive_timer);
-
-	schedule_work(&hdev->keep_alive_task);
-	mod_timer(&hdev->keep_alive_timer, jiffies +
-		  HCLGEVF_KEEP_ALIVE_TASK_INTERVAL * HZ);
-}
-
-static void hclgevf_keep_alive_task(struct work_struct *work)
-{
-	struct hclgevf_dev *hdev;
 	u8 respmsg;
 	int ret;
 
-	hdev = container_of(work, struct hclgevf_dev, keep_alive_task);
-
 	if (test_bit(HCLGEVF_STATE_CMD_DISABLE, &hdev->state))
 		return;
 
@@ -1928,19 +1899,32 @@ static void hclgevf_keep_alive_task(struct work_struct *work)
 			"VF sends keep alive cmd failed(=%d)\n", ret);
 }
 
-static void hclgevf_service_task(struct work_struct *work)
+static void hclgevf_periodic_service_task(struct hclgevf_dev *hdev)
 {
-	struct hnae3_handle *handle;
-	struct hclgevf_dev *hdev;
+	unsigned long delta = round_jiffies_relative(HZ);
+	struct hnae3_handle *handle = &hdev->nic;
 
-	hdev = container_of(work, struct hclgevf_dev, service_task);
-	handle = &hdev->nic;
+	if (time_is_after_jiffies(hdev->last_serv_processed + HZ)) {
+		delta = jiffies - hdev->last_serv_processed;
 
-	if (hdev->stats_timer >= HCLGEVF_STATS_TIMER_INTERVAL) {
-		hclgevf_tqps_update_stats(handle);
-		hdev->stats_timer = 0;
+		if (delta < round_jiffies_relative(HZ)) {
+			delta = round_jiffies_relative(HZ) - delta;
+			goto out;
+		}
+	}
+
+	hdev->serv_processed_cnt++;
+	if (!(hdev->serv_processed_cnt % HCLGEVF_KEEP_ALIVE_TASK_INTERVAL))
+		hclgevf_keep_alive(hdev);
+
+	if (test_bit(HCLGEVF_STATE_DOWN, &hdev->state)) {
+		hdev->last_serv_processed = jiffies;
+		goto out;
 	}
 
+	if (!(hdev->serv_processed_cnt % HCLGEVF_STATS_TIMER_INTERVAL))
+		hclgevf_tqps_update_stats(handle);
+
 	/* request the link status from the PF. PF would be able to tell VF
 	 * about such updates in future so we might remove this later
 	 */
@@ -1950,9 +1934,27 @@ static void hclgevf_service_task(struct work_struct *work)
 
 	hclgevf_sync_vlan_filter(hdev);
 
-	hclgevf_deferred_task_schedule(hdev);
+	hdev->last_serv_processed = jiffies;
 
-	clear_bit(HCLGEVF_STATE_SERVICE_SCHED, &hdev->state);
+out:
+	hclgevf_task_schedule(hdev, delta);
+}
+
+static void hclgevf_service_task(struct work_struct *work)
+{
+	struct hclgevf_dev *hdev = container_of(work, struct hclgevf_dev,
+						service_task.work);
+
+	hclgevf_reset_service_task(hdev);
+	hclgevf_mailbox_service_task(hdev);
+	hclgevf_periodic_service_task(hdev);
+
+	/* Handle reset and mbx again in case periodical task delays the
+	 * handling by calling hclgevf_task_schedule() in
+	 * hclgevf_periodic_service_task()
+	 */
+	hclgevf_reset_service_task(hdev);
+	hclgevf_mailbox_service_task(hdev);
 }
 
 static void hclgevf_clear_event_cause(struct hclgevf_dev *hdev, u32 regclr)
@@ -2189,16 +2191,31 @@ static int hclgevf_init_vlan_config(struct hclgevf_dev *hdev)
 				       false);
 }
 
+static void hclgevf_flush_link_update(struct hclgevf_dev *hdev)
+{
+#define HCLGEVF_FLUSH_LINK_TIMEOUT	100000
+
+	unsigned long last = hdev->serv_processed_cnt;
+	int i = 0;
+
+	while (test_bit(HCLGEVF_STATE_LINK_UPDATING, &hdev->state) &&
+	       i++ < HCLGEVF_FLUSH_LINK_TIMEOUT &&
+	       last == hdev->serv_processed_cnt)
+		usleep_range(1, 1);
+}
+
 static void hclgevf_set_timer_task(struct hnae3_handle *handle, bool enable)
 {
 	struct hclgevf_dev *hdev = hclgevf_ae_get_hdev(handle);
 
 	if (enable) {
-		mod_timer(&hdev->service_timer, jiffies + HZ);
+		hclgevf_task_schedule(hdev, 0);
 	} else {
-		del_timer_sync(&hdev->service_timer);
-		cancel_work_sync(&hdev->service_task);
-		clear_bit(HCLGEVF_STATE_SERVICE_SCHED, &hdev->state);
+		set_bit(HCLGEVF_STATE_DOWN, &hdev->state);
+
+		/* flush memory to make sure DOWN is seen by service task */
+		smp_mb__before_atomic();
+		hclgevf_flush_link_update(hdev);
 	}
 }
 
@@ -2245,16 +2262,12 @@ static int hclgevf_set_alive(struct hnae3_handle *handle, bool alive)
 
 static int hclgevf_client_start(struct hnae3_handle *handle)
 {
-	struct hclgevf_dev *hdev = hclgevf_ae_get_hdev(handle);
 	int ret;
 
 	ret = hclgevf_set_alive(handle, true);
 	if (ret)
 		return ret;
 
-	mod_timer(&hdev->keep_alive_timer, jiffies +
-		  HCLGEVF_KEEP_ALIVE_TASK_INTERVAL * HZ);
-
 	return 0;
 }
 
@@ -2267,25 +2280,15 @@ static void hclgevf_client_stop(struct hnae3_handle *handle)
 	if (ret)
 		dev_warn(&hdev->pdev->dev,
 			 "%s failed %d\n", __func__, ret);
-
-	del_timer_sync(&hdev->keep_alive_timer);
-	cancel_work_sync(&hdev->keep_alive_task);
 }
 
 static void hclgevf_state_init(struct hclgevf_dev *hdev)
 {
-	/* setup tasks for the MBX */
-	INIT_WORK(&hdev->mbx_service_task, hclgevf_mailbox_service_task);
 	clear_bit(HCLGEVF_STATE_MBX_SERVICE_SCHED, &hdev->state);
 	clear_bit(HCLGEVF_STATE_MBX_HANDLING, &hdev->state);
+	clear_bit(HCLGEVF_STATE_RST_FAIL, &hdev->state);
 
-	/* setup tasks for service timer */
-	timer_setup(&hdev->service_timer, hclgevf_service_timer, 0);
-
-	INIT_WORK(&hdev->service_task, hclgevf_service_task);
-	clear_bit(HCLGEVF_STATE_SERVICE_SCHED, &hdev->state);
-
-	INIT_WORK(&hdev->rst_service_task, hclgevf_reset_service_task);
+	INIT_DELAYED_WORK(&hdev->service_task, hclgevf_service_task);
 
 	mutex_init(&hdev->mbx_resp.mbx_mutex);
 
@@ -2298,18 +2301,8 @@ static void hclgevf_state_uninit(struct hclgevf_dev *hdev)
 	set_bit(HCLGEVF_STATE_DOWN, &hdev->state);
 	set_bit(HCLGEVF_STATE_REMOVING, &hdev->state);
 
-	if (hdev->keep_alive_timer.function)
-		del_timer_sync(&hdev->keep_alive_timer);
-	if (hdev->keep_alive_task.func)
-		cancel_work_sync(&hdev->keep_alive_task);
-	if (hdev->service_timer.function)
-		del_timer_sync(&hdev->service_timer);
-	if (hdev->service_task.func)
-		cancel_work_sync(&hdev->service_task);
-	if (hdev->mbx_service_task.func)
-		cancel_work_sync(&hdev->mbx_service_task);
-	if (hdev->rst_service_task.func)
-		cancel_work_sync(&hdev->rst_service_task);
+	if (hdev->service_task.work.func)
+		cancel_delayed_work_sync(&hdev->service_task);
 
 	mutex_destroy(&hdev->mbx_resp.mbx_mutex);
 }
@@ -2807,6 +2800,8 @@ static int hclgevf_init_hdev(struct hclgevf_dev *hdev)
 	dev_info(&hdev->pdev->dev, "finished initializing %s driver\n",
 		 HCLGEVF_DRIVER_NAME);
 
+	hclgevf_task_schedule(hdev, round_jiffies_relative(HZ));
+
 	return 0;
 
 err_config:
@@ -2838,7 +2833,6 @@ static void hclgevf_uninit_hdev(struct hclgevf_dev *hdev)
 static int hclgevf_init_ae_dev(struct hnae3_ae_dev *ae_dev)
 {
 	struct pci_dev *pdev = ae_dev->pdev;
-	struct hclgevf_dev *hdev;
 	int ret;
 
 	ret = hclgevf_alloc_hdev(ae_dev);
@@ -2853,10 +2847,6 @@ static int hclgevf_init_ae_dev(struct hnae3_ae_dev *ae_dev)
 		return ret;
 	}
 
-	hdev = ae_dev->priv;
-	timer_setup(&hdev->keep_alive_timer, hclgevf_keep_alive_timer, 0);
-	INIT_WORK(&hdev->keep_alive_task, hclgevf_keep_alive_task);
-
 	return 0;
 }
 
@@ -3213,6 +3203,12 @@ static int hclgevf_init(void)
 {
 	pr_info("%s is initializing\n", HCLGEVF_NAME);
 
+	hclgevf_wq = alloc_workqueue("%s", WQ_MEM_RECLAIM, 0, HCLGEVF_NAME);
+	if (!hclgevf_wq) {
+		pr_err("%s: failed to create workqueue\n", HCLGEVF_NAME);
+		return -ENOMEM;
+	}
+
 	hnae3_register_ae_algo(&ae_algovf);
 
 	return 0;
@@ -3221,6 +3217,7 @@ static int hclgevf_init(void)
 static void hclgevf_exit(void)
 {
 	hnae3_unregister_ae_algo(&ae_algovf);
+	destroy_workqueue(hclgevf_wq);
 }
 module_init(hclgevf_init);
 module_exit(hclgevf_exit);

drivers/net/ethernet/hisilicon/hns3/hns3vf/hclgevf_main.h

@@ -142,12 +142,13 @@ enum hclgevf_states {
 	HCLGEVF_STATE_REMOVING,
 	HCLGEVF_STATE_NIC_REGISTERED,
 	/* task states */
-	HCLGEVF_STATE_SERVICE_SCHED,
 	HCLGEVF_STATE_RST_SERVICE_SCHED,
 	HCLGEVF_STATE_RST_HANDLING,
 	HCLGEVF_STATE_MBX_SERVICE_SCHED,
 	HCLGEVF_STATE_MBX_HANDLING,
 	HCLGEVF_STATE_CMD_DISABLE,
+	HCLGEVF_STATE_LINK_UPDATING,
+	HCLGEVF_STATE_RST_FAIL,
 };
 
 struct hclgevf_mac {
@@ -283,12 +284,7 @@ struct hclgevf_dev {
 	struct hclgevf_mbx_resp_status mbx_resp; /* mailbox response */
 	struct hclgevf_mbx_arq_ring arq; /* mailbox async rx queue */
 
-	struct timer_list service_timer;
-	struct timer_list keep_alive_timer;
-	struct work_struct service_task;
-	struct work_struct keep_alive_task;
-	struct work_struct rst_service_task;
-	struct work_struct mbx_service_task;
+	struct delayed_work service_task;
 
 	struct hclgevf_tqp *htqp;
 
@@ -298,7 +294,8 @@ struct hclgevf_dev {
 	struct hnae3_client *nic_client;
 	struct hnae3_client *roce_client;
 	u32 flag;
-	u32 stats_timer;
+	unsigned long serv_processed_cnt;
+	unsigned long last_serv_processed;
 };
 
 static inline bool hclgevf_is_reset_pending(struct hclgevf_dev *hdev)