bnxt_en: Improve VF/PF link change logic.

Link status query firmware messages originating from the VFs are forwarded
to the PF.  The driver handles these interactions in a workqueue for the
VF and PF.  The VF driver waits for the response from the PF in the
workqueue.  If the PF and VF driver are running on the same host and the
work for both PF and VF are queued on the same workqueue, the VF driver
may not get the response if the PF work item is queued behind it on the
same workqueue.  This will lead to the VF link query message timing out.

To prevent this, we create a private workqueue for PFs instead of using
the common workqueue.  The VF query and PF response will never be on
the same workqueue.

Fixes: c0c050c5 ("bnxt_en: New Broadcom ethernet driver.")
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/broadcom/bnxt/bnxt.c

@@ -214,6 +214,8 @@ static const u16 bnxt_async_events_arr[] = {
 	ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE,
 };
 
+static struct workqueue_struct *bnxt_pf_wq;
+
 static bool bnxt_vf_pciid(enum board_idx idx)
 {
 	return (idx == NETXTREME_C_VF || idx == NETXTREME_E_VF);
@@ -1024,12 +1026,28 @@ static int bnxt_discard_rx(struct bnxt *bp, struct bnxt_napi *bnapi,
 	return 0;
 }
 
+static void bnxt_queue_sp_work(struct bnxt *bp)
+{
+	if (BNXT_PF(bp))
+		queue_work(bnxt_pf_wq, &bp->sp_task);
+	else
+		schedule_work(&bp->sp_task);
+}
+
+static void bnxt_cancel_sp_work(struct bnxt *bp)
+{
+	if (BNXT_PF(bp))
+		flush_workqueue(bnxt_pf_wq);
+	else
+		cancel_work_sync(&bp->sp_task);
+}
+
 static void bnxt_sched_reset(struct bnxt *bp, struct bnxt_rx_ring_info *rxr)
 {
 	if (!rxr->bnapi->in_reset) {
 		rxr->bnapi->in_reset = true;
 		set_bit(BNXT_RESET_TASK_SP_EVENT, &bp->sp_event);
-		schedule_work(&bp->sp_task);
+		bnxt_queue_sp_work(bp);
 	}
 	rxr->rx_next_cons = 0xffff;
 }
@@ -1717,7 +1735,7 @@ static int bnxt_async_event_process(struct bnxt *bp,
 	default:
 		goto async_event_process_exit;
 	}
-	schedule_work(&bp->sp_task);
+	bnxt_queue_sp_work(bp);
 async_event_process_exit:
 	bnxt_ulp_async_events(bp, cmpl);
 	return 0;
@@ -1751,7 +1769,7 @@ static int bnxt_hwrm_handler(struct bnxt *bp, struct tx_cmp *txcmp)
 
 		set_bit(vf_id - bp->pf.first_vf_id, bp->pf.vf_event_bmap);
 		set_bit(BNXT_HWRM_EXEC_FWD_REQ_SP_EVENT, &bp->sp_event);
-		schedule_work(&bp->sp_task);
+		bnxt_queue_sp_work(bp);
 		break;
 
 	case CMPL_BASE_TYPE_HWRM_ASYNC_EVENT:
@@ -6647,7 +6665,7 @@ static void bnxt_set_rx_mode(struct net_device *dev)
 		vnic->rx_mask = mask;
 
 		set_bit(BNXT_RX_MASK_SP_EVENT, &bp->sp_event);
-		schedule_work(&bp->sp_task);
+		bnxt_queue_sp_work(bp);
 	}
 }
 
@@ -6920,7 +6938,7 @@ static void bnxt_tx_timeout(struct net_device *dev)
 
 	netdev_err(bp->dev,  "TX timeout detected, starting reset task!\n");
 	set_bit(BNXT_RESET_TASK_SP_EVENT, &bp->sp_event);
-	schedule_work(&bp->sp_task);
+	bnxt_queue_sp_work(bp);
 }
 
 #ifdef CONFIG_NET_POLL_CONTROLLER
@@ -6952,7 +6970,7 @@ static void bnxt_timer(unsigned long data)
 	if (bp->link_info.link_up && (bp->flags & BNXT_FLAG_PORT_STATS) &&
 	    bp->stats_coal_ticks) {
 		set_bit(BNXT_PERIODIC_STATS_SP_EVENT, &bp->sp_event);
-		schedule_work(&bp->sp_task);
+		bnxt_queue_sp_work(bp);
 	}
 bnxt_restart_timer:
 	mod_timer(&bp->timer, jiffies + bp->current_interval);
@@ -7433,7 +7451,7 @@ static int bnxt_rx_flow_steer(struct net_device *dev, const struct sk_buff *skb,
 	spin_unlock_bh(&bp->ntp_fltr_lock);
 
 	set_bit(BNXT_RX_NTP_FLTR_SP_EVENT, &bp->sp_event);
-	schedule_work(&bp->sp_task);
+	bnxt_queue_sp_work(bp);
 
 	return new_fltr->sw_id;
 
@@ -7516,7 +7534,7 @@ static void bnxt_udp_tunnel_add(struct net_device *dev,
 		if (bp->vxlan_port_cnt == 1) {
 			bp->vxlan_port = ti->port;
 			set_bit(BNXT_VXLAN_ADD_PORT_SP_EVENT, &bp->sp_event);
-			schedule_work(&bp->sp_task);
+			bnxt_queue_sp_work(bp);
 		}
 		break;
 	case UDP_TUNNEL_TYPE_GENEVE:
@@ -7533,7 +7551,7 @@ static void bnxt_udp_tunnel_add(struct net_device *dev,
 		return;
 	}
 
-	schedule_work(&bp->sp_task);
+	bnxt_queue_sp_work(bp);
 }
 
 static void bnxt_udp_tunnel_del(struct net_device *dev,
@@ -7572,7 +7590,7 @@ static void bnxt_udp_tunnel_del(struct net_device *dev,
 		return;
 	}
 
-	schedule_work(&bp->sp_task);
+	bnxt_queue_sp_work(bp);
 }
 
 static int bnxt_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
@@ -7720,7 +7738,7 @@ static void bnxt_remove_one(struct pci_dev *pdev)
 	pci_disable_pcie_error_reporting(pdev);
 	unregister_netdev(dev);
 	bnxt_shutdown_tc(bp);
-	cancel_work_sync(&bp->sp_task);
+	bnxt_cancel_sp_work(bp);
 	bp->sp_event = 0;
 
 	bnxt_clear_int_mode(bp);
@@ -8138,8 +8156,17 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 	else
 		device_set_wakeup_capable(&pdev->dev, false);
 
-	if (BNXT_PF(bp))
+	if (BNXT_PF(bp)) {
+		if (!bnxt_pf_wq) {
+			bnxt_pf_wq =
+				create_singlethread_workqueue("bnxt_pf_wq");
+			if (!bnxt_pf_wq) {
+				dev_err(&pdev->dev, "Unable to create workqueue.\n");
+				goto init_err_pci_clean;
+			}
+		}
 		bnxt_init_tc(bp);
+	}
 
 	rc = register_netdev(dev);
 	if (rc)
@@ -8375,4 +8402,17 @@ static struct pci_driver bnxt_pci_driver = {
 #endif
 };
 
-module_pci_driver(bnxt_pci_driver);
+static int __init bnxt_init(void)
+{
+	return pci_register_driver(&bnxt_pci_driver);
+}
+
+static void __exit bnxt_exit(void)
+{
+	pci_unregister_driver(&bnxt_pci_driver);
+	if (bnxt_pf_wq)
+		destroy_workqueue(bnxt_pf_wq);
+}
+
+module_init(bnxt_init);
+module_exit(bnxt_exit);