qede: Error recovery process

This patch adds the error recovery process in the qede driver.
The process includes a partial/customized driver unload and load, which
allows it to look like a short suspend period to the kernel while
preserving the net devices' state.
Signed-off-by: Tomer Tayar <tomer.tayar@cavium.com>
Signed-off-by: Ariel Elior <ariel.elior@cavium.com>
Signed-off-by: Michal Kalderon <michal.kalderon@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/qlogic/qede/qede.h

@@ -162,6 +162,7 @@ struct qede_rdma_dev {
 	struct list_head entry;
 	struct list_head rdma_event_list;
 	struct workqueue_struct *rdma_wq;
+	bool exp_recovery;
 };
 
 struct qede_ptp;
@@ -264,6 +265,7 @@ struct qede_dev {
 enum QEDE_STATE {
 	QEDE_STATE_CLOSED,
 	QEDE_STATE_OPEN,
+	QEDE_STATE_RECOVERY,
 };
 
 #define HILO_U64(hi, lo)		((((u64)(hi)) << 32) + (lo))
@@ -462,6 +464,7 @@ struct qede_fastpath {
 #define QEDE_CSUM_UNNECESSARY		BIT(1)
 #define QEDE_TUNN_CSUM_UNNECESSARY	BIT(2)
 
+#define QEDE_SP_RECOVERY		0
 #define QEDE_SP_RX_MODE			1
 
 #ifdef CONFIG_RFS_ACCEL

drivers/net/ethernet/qlogic/qede/qede_rdma.c

@@ -50,6 +50,8 @@ static void _qede_rdma_dev_add(struct qede_dev *edev)
 	if (!qedr_drv)
 		return;
 
+	/* Leftovers from previous error recovery */
+	edev->rdma_info.exp_recovery = false;
 	edev->rdma_info.qedr_dev = qedr_drv->add(edev->cdev, edev->pdev,
 						 edev->ndev);
 }
@@ -87,11 +89,17 @@ static void qede_rdma_destroy_wq(struct qede_dev *edev)
 	destroy_workqueue(edev->rdma_info.rdma_wq);
 }
 
-int qede_rdma_dev_add(struct qede_dev *edev)
+int qede_rdma_dev_add(struct qede_dev *edev, bool recovery)
 {
-	int rc = 0;
+	int rc;
+
+	if (!qede_rdma_supported(edev))
+		return 0;
+
+	/* Cannot start qedr while recovering since it wasn't fully stopped */
+	if (recovery)
+		return 0;
 
-	if (qede_rdma_supported(edev)) {
 	rc = qede_rdma_create_wq(edev);
 	if (rc)
 		return rc;
@@ -101,7 +109,6 @@ int qede_rdma_dev_add(struct qede_dev *edev)
 	list_add_tail(&edev->rdma_info.entry, &qedr_dev_list);
 	_qede_rdma_dev_add(edev);
 	mutex_unlock(&qedr_dev_list_lock);
-	}
 
 	return rc;
 }
@@ -110,19 +117,30 @@ static void _qede_rdma_dev_remove(struct qede_dev *edev)
 {
 	if (qedr_drv && qedr_drv->remove && edev->rdma_info.qedr_dev)
 		qedr_drv->remove(edev->rdma_info.qedr_dev);
-	edev->rdma_info.qedr_dev = NULL;
 }
 
-void qede_rdma_dev_remove(struct qede_dev *edev)
+void qede_rdma_dev_remove(struct qede_dev *edev, bool recovery)
 {
 	if (!qede_rdma_supported(edev))
 		return;
 
+	/* Cannot remove qedr while recovering since it wasn't fully stopped */
+	if (!recovery) {
 		qede_rdma_destroy_wq(edev);
 		mutex_lock(&qedr_dev_list_lock);
+		if (!edev->rdma_info.exp_recovery)
 			_qede_rdma_dev_remove(edev);
+		edev->rdma_info.qedr_dev = NULL;
 		list_del(&edev->rdma_info.entry);
 		mutex_unlock(&qedr_dev_list_lock);
+	} else {
+		if (!edev->rdma_info.exp_recovery) {
+			mutex_lock(&qedr_dev_list_lock);
+			_qede_rdma_dev_remove(edev);
+			mutex_unlock(&qedr_dev_list_lock);
+		}
+		edev->rdma_info.exp_recovery = true;
+	}
 }
 
 static void _qede_rdma_dev_open(struct qede_dev *edev)
@@ -204,7 +222,8 @@ void qede_rdma_unregister_driver(struct qedr_driver *drv)
 
 	mutex_lock(&qedr_dev_list_lock);
 	list_for_each_entry(edev, &qedr_dev_list, rdma_info.entry) {
-		if (edev->rdma_info.qedr_dev)
+		/* If device has experienced recovery it was already removed */
+		if (edev->rdma_info.qedr_dev && !edev->rdma_info.exp_recovery)
 			_qede_rdma_dev_remove(edev);
 	}
 	qedr_drv = NULL;
@@ -284,6 +303,10 @@ static void qede_rdma_add_event(struct qede_dev *edev,
 {
 	struct qede_rdma_event_work *event_node;
 
+	/* If a recovery was experienced avoid adding the event */
+	if (edev->rdma_info.exp_recovery)
+		return;
+
 	if (!edev->rdma_info.qedr_dev)
 		return;
 

include/linux/qed/qede_rdma.h

@@ -74,21 +74,23 @@ void qede_rdma_unregister_driver(struct qedr_driver *drv);
 bool qede_rdma_supported(struct qede_dev *dev);
 
 #if IS_ENABLED(CONFIG_QED_RDMA)
-int qede_rdma_dev_add(struct qede_dev *dev);
+int qede_rdma_dev_add(struct qede_dev *dev, bool recovery);
 void qede_rdma_dev_event_open(struct qede_dev *dev);
 void qede_rdma_dev_event_close(struct qede_dev *dev);
-void qede_rdma_dev_remove(struct qede_dev *dev);
+void qede_rdma_dev_remove(struct qede_dev *dev, bool recovery);
 void qede_rdma_event_changeaddr(struct qede_dev *edr);
 
 #else
-static inline int qede_rdma_dev_add(struct qede_dev *dev)
+static inline int qede_rdma_dev_add(struct qede_dev *dev,
+				    bool recovery)
 {
 	return 0;
 }
 
 static inline void qede_rdma_dev_event_open(struct qede_dev *dev) {}
 static inline void qede_rdma_dev_event_close(struct qede_dev *dev) {}
-static inline void qede_rdma_dev_remove(struct qede_dev *dev) {}
+static inline void qede_rdma_dev_remove(struct qede_dev *dev,
+					bool recovery) {}
 static inline void qede_rdma_event_changeaddr(struct qede_dev *edr) {}
 #endif
 #endif