[SCSI] lpfc 8.3.25: Add FCF priority failover functionality

This patch implements a new FCF failover policy for the lpfc driver. It
allows the driver to choose which FCF to failover to based on the FCF
priority. This patch also introduces a new sysfs parameter
(fcf_failover_policy) to allow the user to choose which FCF failover policy
to use.

Signed-off-by: Alex Iannicelli <alex.iannicelli@emulex.com>
Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>

1 files changed, 212 insertions, 8 deletions

diff --git a/drivers/scsi/lpfc/lpfc_hbadisc.c b/drivers/scsi/lpfc/lpfc_hbadisc.c
index bef17e3e419..0b47adf9fee 100644
--- a/drivers/scsi/lpfc/lpfc_hbadisc.c
+++ b/drivers/scsi/lpfc/lpfc_hbadisc.c
@@ -1109,6 +1109,28 @@ out:
 	return;
 }
 
+/**
+ * lpfc_sli4_clear_fcf_rr_bmask
+ * @phba pointer to the struct lpfc_hba for this port.
+ * This fucnction resets the round robin bit mask and clears the
+ * fcf priority list. The list deletions are done while holding the
+ * hbalock. The ON_LIST flag and the FLOGI_FAILED flags are cleared
+ * from the lpfc_fcf_pri record.
+ **/
+void
+lpfc_sli4_clear_fcf_rr_bmask(struct lpfc_hba *phba)
+{
+	struct lpfc_fcf_pri *fcf_pri;
+	struct lpfc_fcf_pri *next_fcf_pri;
+	memset(phba->fcf.fcf_rr_bmask, 0, sizeof(*phba->fcf.fcf_rr_bmask));
+	spin_lock_irq(&phba->hbalock);
+	list_for_each_entry_safe(fcf_pri, next_fcf_pri,
+				&phba->fcf.fcf_pri_list, list) {
+		list_del_init(&fcf_pri->list);
+		fcf_pri->fcf_rec.flag = 0;
+	}
+	spin_unlock_irq(&phba->hbalock);
+}
 static void
 lpfc_mbx_cmpl_reg_fcfi(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq)
 {
@@ -1130,7 +1152,8 @@ lpfc_mbx_cmpl_reg_fcfi(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq)
 	spin_unlock_irq(&phba->hbalock);
 
 	/* If there is a pending FCoE event, restart FCF table scan. */
-	if (lpfc_check_pending_fcoe_event(phba, LPFC_UNREG_FCF))
+	if ((!(phba->hba_flag & FCF_RR_INPROG)) &&
+		lpfc_check_pending_fcoe_event(phba, LPFC_UNREG_FCF))
 		goto fail_out;
 
 	/* Mark successful completion of FCF table scan */
@@ -1250,6 +1273,30 @@ lpfc_vlan_id_match(uint16_t curr_vlan_id, uint16_t new_vlan_id)
 }
 
 /**
+ * lpfc_update_fcf_record - Update driver fcf record
+ * __lpfc_update_fcf_record_pri - update the lpfc_fcf_pri record.
+ * @phba: pointer to lpfc hba data structure.
+ * @fcf_index: Index for the lpfc_fcf_record.
+ * @new_fcf_record: pointer to hba fcf record.
+ *
+ * This routine updates the driver FCF priority record from the new HBA FCF
+ * record. This routine is called with the host lock held.
+ **/
+static void
+__lpfc_update_fcf_record_pri(struct lpfc_hba *phba, uint16_t fcf_index,
+				 struct fcf_record *new_fcf_record
+				 )
+{
+	struct lpfc_fcf_pri *fcf_pri;
+
+	fcf_pri = &phba->fcf.fcf_pri[fcf_index];
+	fcf_pri->fcf_rec.fcf_index = fcf_index;
+	/* FCF record priority */
+	fcf_pri->fcf_rec.priority = new_fcf_record->fip_priority;
+
+}
+
+/**
  * lpfc_copy_fcf_record - Copy fcf information to lpfc_hba.
  * @fcf: pointer to driver fcf record.
  * @new_fcf_record: pointer to fcf record.
@@ -1332,6 +1379,9 @@ __lpfc_update_fcf_record(struct lpfc_hba *phba, struct lpfc_fcf_rec *fcf_rec,
 	fcf_rec->addr_mode = addr_mode;
 	fcf_rec->vlan_id = vlan_id;
 	fcf_rec->flag |= (flag | RECORD_VALID);
+	__lpfc_update_fcf_record_pri(phba,
+		bf_get(lpfc_fcf_record_fcf_index, new_fcf_record),
+				 new_fcf_record);
 }
 
 /**
@@ -1834,6 +1884,8 @@ lpfc_sli4_fcf_record_match(struct lpfc_hba *phba,
 		return false;
 	if (!lpfc_fab_name_match(fcf_rec->fabric_name, new_fcf_record))
 		return false;
+	if (fcf_rec->priority != new_fcf_record->fip_priority)
+		return false;
 	return true;
 }
 
@@ -1897,6 +1949,152 @@ stop_flogi_current_fcf:
 }
 
 /**
+ * lpfc_sli4_fcf_pri_list_del
+ * @phba: pointer to lpfc hba data structure.
+ * @fcf_index the index of the fcf record to delete
+ * This routine checks the on list flag of the fcf_index to be deleted.
+ * If it is one the list then it is removed from the list, and the flag
+ * is cleared. This routine grab the hbalock before removing the fcf
+ * record from the list.
+ **/
+static void lpfc_sli4_fcf_pri_list_del(struct lpfc_hba *phba,
+			uint16_t fcf_index)
+{
+	struct lpfc_fcf_pri *new_fcf_pri;
+
+	new_fcf_pri = &phba->fcf.fcf_pri[fcf_index];
+	lpfc_printf_log(phba, KERN_INFO, LOG_FIP,
+		"3058 deleting idx x%x pri x%x flg x%x\n",
+		fcf_index, new_fcf_pri->fcf_rec.priority,
+		 new_fcf_pri->fcf_rec.flag);
+	spin_lock_irq(&phba->hbalock);
+	if (new_fcf_pri->fcf_rec.flag & LPFC_FCF_ON_PRI_LIST) {
+		if (phba->fcf.current_rec.priority ==
+				new_fcf_pri->fcf_rec.priority)
+			phba->fcf.eligible_fcf_cnt--;
+		list_del_init(&new_fcf_pri->list);
+		new_fcf_pri->fcf_rec.flag &= ~LPFC_FCF_ON_PRI_LIST;
+	}
+	spin_unlock_irq(&phba->hbalock);
+}
+
+/**
+ * lpfc_sli4_set_fcf_flogi_fail
+ * @phba: pointer to lpfc hba data structure.
+ * @fcf_index the index of the fcf record to update
+ * This routine acquires the hbalock and then set the LPFC_FCF_FLOGI_FAILED
+ * flag so the the round robin slection for the particular priority level
+ * will try a different fcf record that does not have this bit set.
+ * If the fcf record is re-read for any reason this flag is cleared brfore
+ * adding it to the priority list.
+ **/
+void
+lpfc_sli4_set_fcf_flogi_fail(struct lpfc_hba *phba, uint16_t fcf_index)
+{
+	struct lpfc_fcf_pri *new_fcf_pri;
+	new_fcf_pri = &phba->fcf.fcf_pri[fcf_index];
+	spin_lock_irq(&phba->hbalock);
+	new_fcf_pri->fcf_rec.flag |= LPFC_FCF_FLOGI_FAILED;
+	spin_unlock_irq(&phba->hbalock);
+}
+
+/**
+ * lpfc_sli4_fcf_pri_list_add
+ * @phba: pointer to lpfc hba data structure.
+ * @fcf_index the index of the fcf record to add
+ * This routine checks the priority of the fcf_index to be added.
+ * If it is a lower priority than the current head of the fcf_pri list
+ * then it is added to the list in the right order.
+ * If it is the same priority as the current head of the list then it
+ * is added to the head of the list and its bit in the rr_bmask is set.
+ * If the fcf_index to be added is of a higher priority than the current
+ * head of the list then the rr_bmask is cleared, its bit is set in the
+ * rr_bmask and it is added to the head of the list.
+ * returns:
+ * 0=success 1=failure
+ **/
+int lpfc_sli4_fcf_pri_list_add(struct lpfc_hba *phba, uint16_t fcf_index,
+	struct fcf_record *new_fcf_record)
+{
+	uint16_t current_fcf_pri;
+	uint16_t last_index;
+	struct lpfc_fcf_pri *fcf_pri;
+	struct lpfc_fcf_pri *next_fcf_pri;
+	struct lpfc_fcf_pri *new_fcf_pri;
+	int ret;
+
+	new_fcf_pri = &phba->fcf.fcf_pri[fcf_index];
+	lpfc_printf_log(phba, KERN_INFO, LOG_FIP,
+		"3059 adding idx x%x pri x%x flg x%x\n",
+		fcf_index, new_fcf_record->fip_priority,
+		 new_fcf_pri->fcf_rec.flag);
+	spin_lock_irq(&phba->hbalock);
+	if (new_fcf_pri->fcf_rec.flag & LPFC_FCF_ON_PRI_LIST)
+		list_del_init(&new_fcf_pri->list);
+	new_fcf_pri->fcf_rec.fcf_index = fcf_index;
+	new_fcf_pri->fcf_rec.priority = new_fcf_record->fip_priority;
+	if (list_empty(&phba->fcf.fcf_pri_list)) {
+		list_add(&new_fcf_pri->list, &phba->fcf.fcf_pri_list);
+		ret = lpfc_sli4_fcf_rr_index_set(phba,
+				new_fcf_pri->fcf_rec.fcf_index);
+		goto out;
+	}
+
+	last_index = find_first_bit(phba->fcf.fcf_rr_bmask,
+				LPFC_SLI4_FCF_TBL_INDX_MAX);
+	if (last_index >= LPFC_SLI4_FCF_TBL_INDX_MAX) {
+		ret = 0; /* Empty rr list */
+		goto out;
+	}
+	current_fcf_pri = phba->fcf.fcf_pri[last_index].fcf_rec.priority;
+	if (new_fcf_pri->fcf_rec.priority <=  current_fcf_pri) {
+		list_add(&new_fcf_pri->list, &phba->fcf.fcf_pri_list);
+		if (new_fcf_pri->fcf_rec.priority <  current_fcf_pri) {
+			memset(phba->fcf.fcf_rr_bmask, 0,
+				sizeof(*phba->fcf.fcf_rr_bmask));
+			/* fcfs_at_this_priority_level = 1; */
+			phba->fcf.eligible_fcf_cnt = 1;
+		} else
+			/* fcfs_at_this_priority_level++; */
+			phba->fcf.eligible_fcf_cnt++;
+		ret = lpfc_sli4_fcf_rr_index_set(phba,
+				new_fcf_pri->fcf_rec.fcf_index);
+		goto out;
+	}
+
+	list_for_each_entry_safe(fcf_pri, next_fcf_pri,
+				&phba->fcf.fcf_pri_list, list) {
+		if (new_fcf_pri->fcf_rec.priority <=
+				fcf_pri->fcf_rec.priority) {
+			if (fcf_pri->list.prev == &phba->fcf.fcf_pri_list)
+				list_add(&new_fcf_pri->list,
+						&phba->fcf.fcf_pri_list);
+			else
+				list_add(&new_fcf_pri->list,
+					 &((struct lpfc_fcf_pri *)
+					fcf_pri->list.prev)->list);
+			ret = 0;
+			goto out;
+		} else if (fcf_pri->list.next == &phba->fcf.fcf_pri_list
+			|| new_fcf_pri->fcf_rec.priority <
+				next_fcf_pri->fcf_rec.priority) {
+			list_add(&new_fcf_pri->list, &fcf_pri->list);
+			ret = 0;
+			goto out;
+		}
+		if (new_fcf_pri->fcf_rec.priority > fcf_pri->fcf_rec.priority)
+			continue;
+
+	}
+	ret = 1;
+out:
+	/* we use = instead of |= to clear the FLOGI_FAILED flag. */
+	new_fcf_pri->fcf_rec.flag = LPFC_FCF_ON_PRI_LIST;
+	spin_unlock_irq(&phba->hbalock);
+	return ret;
+}
+
+/**
  * lpfc_mbx_cmpl_fcf_scan_read_fcf_rec - fcf scan read_fcf mbox cmpl handler.
  * @phba: pointer to lpfc hba data structure.
  * @mboxq: pointer to mailbox object.
@@ -1958,6 +2156,9 @@ lpfc_mbx_cmpl_fcf_scan_read_fcf_rec(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq)
 	 * record for roundrobin FCF failover.
 	 */
 	if (!rc) {
+		lpfc_sli4_fcf_pri_list_del(phba,
+					bf_get(lpfc_fcf_record_fcf_index,
+					       new_fcf_record));
 		lpfc_printf_log(phba, KERN_WARNING, LOG_FIP,
 				"2781 FCF (x%x) failed connection "
 				"list check: (x%x/x%x)\n",
@@ -2005,7 +2206,8 @@ lpfc_mbx_cmpl_fcf_scan_read_fcf_rec(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq)
 		goto read_next_fcf;
 	} else {
 		fcf_index = bf_get(lpfc_fcf_record_fcf_index, new_fcf_record);
-		rc = lpfc_sli4_fcf_rr_index_set(phba, fcf_index);
+		rc = lpfc_sli4_fcf_pri_list_add(phba, fcf_index,
+							new_fcf_record);
 		if (rc)
 			goto read_next_fcf;
 	}
@@ -2018,7 +2220,8 @@ lpfc_mbx_cmpl_fcf_scan_read_fcf_rec(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq)
 	 */
 	spin_lock_irq(&phba->hbalock);
 	if (phba->fcf.fcf_flag & FCF_IN_USE) {
-		if (lpfc_sli4_fcf_record_match(phba, &phba->fcf.current_rec,
+		if (phba->cfg_fcf_failover_policy == LPFC_FCF_FOV &&
+			lpfc_sli4_fcf_record_match(phba, &phba->fcf.current_rec,
 		    new_fcf_record, vlan_id)) {
 			if (bf_get(lpfc_fcf_record_fcf_index, new_fcf_record) ==
 			    phba->fcf.current_rec.fcf_indx) {
@@ -2232,7 +2435,8 @@ read_next_fcf:
 			    (phba->fcf.fcf_flag & FCF_REDISC_PEND))
 				return;
 
-			if (phba->fcf.fcf_flag & FCF_IN_USE) {
+			if (phba->cfg_fcf_failover_policy == LPFC_FCF_FOV &&
+				phba->fcf.fcf_flag & FCF_IN_USE) {
 				/*
 				 * In case the current in-use FCF record no
 				 * longer existed during FCF discovery that
@@ -2423,7 +2627,8 @@ lpfc_mbx_cmpl_read_fcf_rec(struct lpfc_hba *phba, LPFC_MBOXQ_t *mboxq)
 
 	/* Update the eligible FCF record index bmask */
 	fcf_index = bf_get(lpfc_fcf_record_fcf_index, new_fcf_record);
-	rc = lpfc_sli4_fcf_rr_index_set(phba, fcf_index);
+
+	rc = lpfc_sli4_fcf_pri_list_add(phba, fcf_index, new_fcf_record);
 
 out:
 	lpfc_sli4_mbox_cmd_free(phba, mboxq);
@@ -2893,8 +3098,7 @@ lpfc_mbx_process_link_up(struct lpfc_hba *phba, struct lpfc_mbx_read_top *la)
 			goto out;
 		}
 		/* Reset FCF roundrobin bmask for new discovery */
-		memset(phba->fcf.fcf_rr_bmask, 0,
-		       sizeof(*phba->fcf.fcf_rr_bmask));
+		lpfc_sli4_clear_fcf_rr_bmask(phba);
 	}
 
 	return;
@@ -5592,7 +5796,7 @@ lpfc_unregister_fcf_rescan(struct lpfc_hba *phba)
 	spin_unlock_irq(&phba->hbalock);
 
 	/* Reset FCF roundrobin bmask for new discovery */
-	memset(phba->fcf.fcf_rr_bmask, 0, sizeof(*phba->fcf.fcf_rr_bmask));
+	lpfc_sli4_clear_fcf_rr_bmask(phba);
 
 	rc = lpfc_sli4_fcf_scan_read_fcf_rec(phba, LPFC_FCOE_FCF_GET_FIRST);