[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>

drivers/scsi/lpfc/lpfc.h

@@ -680,6 +680,9 @@ struct lpfc_hba {
 	uint32_t cfg_enable_rrq;
 	uint32_t cfg_topology;
 	uint32_t cfg_link_speed;
+#define LPFC_FCF_FOV 1		/* Fast fcf failover */
+#define LPFC_FCF_PRIORITY 2	/* Priority fcf failover */
+	uint32_t cfg_fcf_failover_policy;
 	uint32_t cfg_cr_delay;
 	uint32_t cfg_cr_count;
 	uint32_t cfg_multi_ring_support;

drivers/scsi/lpfc/lpfc_attr.c

@@ -2193,6 +2193,9 @@ lpfc_param_show(enable_npiv);
 lpfc_param_init(enable_npiv, 1, 0, 1);
 static DEVICE_ATTR(lpfc_enable_npiv, S_IRUGO, lpfc_enable_npiv_show, NULL);
 
+LPFC_ATTR_R(fcf_failover_policy, 1, 1, 2,
+	"FCF Fast failover=1 Priority failover=2");
+
 int lpfc_enable_rrq;
 module_param(lpfc_enable_rrq, int, S_IRUGO);
 MODULE_PARM_DESC(lpfc_enable_rrq, "Enable RRQ functionality");
@@ -3775,6 +3778,7 @@ struct device_attribute *lpfc_hba_attrs[] = {
 	&dev_attr_lpfc_fdmi_on,
 	&dev_attr_lpfc_max_luns,
 	&dev_attr_lpfc_enable_npiv,
+	&dev_attr_lpfc_fcf_failover_policy,
 	&dev_attr_lpfc_enable_rrq,
 	&dev_attr_nport_evt_cnt,
 	&dev_attr_board_mode,
@@ -4995,6 +4999,7 @@ lpfc_get_cfgparam(struct lpfc_hba *phba)
 	lpfc_link_speed_init(phba, lpfc_link_speed);
 	lpfc_poll_tmo_init(phba, lpfc_poll_tmo);
 	lpfc_enable_npiv_init(phba, lpfc_enable_npiv);
+	lpfc_fcf_failover_policy_init(phba, lpfc_fcf_failover_policy);
 	lpfc_enable_rrq_init(phba, lpfc_enable_rrq);
 	lpfc_use_msi_init(phba, lpfc_use_msi);
 	lpfc_fcp_imax_init(phba, lpfc_fcp_imax);

drivers/scsi/lpfc/lpfc_crtn.h

@@ -235,9 +235,11 @@ int lpfc_sli4_redisc_fcf_table(struct lpfc_hba *);
 void lpfc_fcf_redisc_wait_start_timer(struct lpfc_hba *);
 void lpfc_sli4_fcf_dead_failthrough(struct lpfc_hba *);
 uint16_t lpfc_sli4_fcf_rr_next_index_get(struct lpfc_hba *);
+void lpfc_sli4_set_fcf_flogi_fail(struct lpfc_hba *, uint16_t);
 int lpfc_sli4_fcf_rr_index_set(struct lpfc_hba *, uint16_t);
 void lpfc_sli4_fcf_rr_index_clear(struct lpfc_hba *, uint16_t);
 int lpfc_sli4_fcf_rr_next_proc(struct lpfc_vport *, uint16_t);
+void lpfc_sli4_clear_fcf_rr_bmask(struct lpfc_hba *);
 
 int lpfc_mem_alloc(struct lpfc_hba *, int align);
 void lpfc_mem_free(struct lpfc_hba *);

drivers/scsi/lpfc/lpfc_els.c

@@ -874,6 +874,8 @@ lpfc_cmpl_els_flogi(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
 					phba->fcf.current_rec.fcf_indx,
 					irsp->ulpStatus, irsp->un.ulpWord[4],
 					irsp->ulpTimeout);
+			lpfc_sli4_set_fcf_flogi_fail(phba,
+					phba->fcf.current_rec.fcf_indx);
 			fcf_index = lpfc_sli4_fcf_rr_next_index_get(phba);
 			rc = lpfc_sli4_fcf_rr_next_proc(vport, fcf_index);
 			if (rc)

drivers/scsi/lpfc/lpfc_init.c

@@ -3634,8 +3634,7 @@ lpfc_sli4_async_fip_evt(struct lpfc_hba *phba,
 			lpfc_sli4_fcf_dead_failthrough(phba);
 		} else {
 			/* 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);
 			/*
 			 * Handling fast FCF failover to a DEAD FCF event is
 			 * considered equalivant to receiving CVL to all vports.
@@ -3721,8 +3720,7 @@ lpfc_sli4_async_fip_evt(struct lpfc_hba *phba,
 				 * 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);
 		}
 		break;
 	default:
@@ -9046,6 +9044,7 @@ lpfc_pci_probe_one_s4(struct pci_dev *pdev, const struct pci_device_id *pid)
 	}
 
 	INIT_LIST_HEAD(&phba->active_rrq_list);
+	INIT_LIST_HEAD(&phba->fcf.fcf_pri_list);
 
 	/* Set up common device driver resources */
 	error = lpfc_setup_driver_resource_phase2(phba);

drivers/scsi/lpfc/lpfc_sli.c

@@ -14634,6 +14634,92 @@ lpfc_sli4_read_fcf_rec(struct lpfc_hba *phba, uint16_t fcf_index)
 	return error;
 }
 
+/**
+ * lpfc_check_next_fcf_pri
+ * phba pointer to the lpfc_hba struct for this port.
+ * This routine is called from the lpfc_sli4_fcf_rr_next_index_get
+ * routine when the rr_bmask is empty. The FCF indecies are put into the
+ * rr_bmask based on their priority level. Starting from the highest priority
+ * to the lowest. The most likely FCF candidate will be in the highest
+ * priority group. When this routine is called it searches the fcf_pri list for
+ * next lowest priority group and repopulates the rr_bmask with only those
+ * fcf_indexes.
+ * returns:
+ * 1=success 0=failure
+ **/
+int
+lpfc_check_next_fcf_pri_level(struct lpfc_hba *phba)
+{
+	uint16_t next_fcf_pri;
+	uint16_t last_index;
+	struct lpfc_fcf_pri *fcf_pri;
+	int rc;
+	int ret = 0;
+
+	last_index = find_first_bit(phba->fcf.fcf_rr_bmask,
+			LPFC_SLI4_FCF_TBL_INDX_MAX);
+	lpfc_printf_log(phba, KERN_INFO, LOG_FIP,
+			"3060 Last IDX %d\n", last_index);
+	if (list_empty(&phba->fcf.fcf_pri_list)) {
+		lpfc_printf_log(phba, KERN_ERR, LOG_FIP,
+			"3061 Last IDX %d\n", last_index);
+		return 0; /* Empty rr list */
+	}
+	next_fcf_pri = 0;
+	/*
+	 * Clear the rr_bmask and set all of the bits that are at this
+	 * priority.
+	 */
+	memset(phba->fcf.fcf_rr_bmask, 0,
+			sizeof(*phba->fcf.fcf_rr_bmask));
+	spin_lock_irq(&phba->hbalock);
+	list_for_each_entry(fcf_pri, &phba->fcf.fcf_pri_list, list) {
+		if (fcf_pri->fcf_rec.flag & LPFC_FCF_FLOGI_FAILED)
+			continue;
+		/*
+		 * the 1st priority that has not FLOGI failed
+		 * will be the highest.
+		 */
+		if (!next_fcf_pri)
+			next_fcf_pri = fcf_pri->fcf_rec.priority;
+		spin_unlock_irq(&phba->hbalock);
+		if (fcf_pri->fcf_rec.priority == next_fcf_pri) {
+			rc = lpfc_sli4_fcf_rr_index_set(phba,
+						fcf_pri->fcf_rec.fcf_index);
+			if (rc)
+				return 0;
+		}
+		spin_lock_irq(&phba->hbalock);
+	}
+	/*
+	 * if next_fcf_pri was not set above and the list is not empty then
+	 * we have failed flogis on all of them. So reset flogi failed
+	 * and start at the begining.
+	 */
+	if (!next_fcf_pri && !list_empty(&phba->fcf.fcf_pri_list)) {
+		list_for_each_entry(fcf_pri, &phba->fcf.fcf_pri_list, list) {
+			fcf_pri->fcf_rec.flag &= ~LPFC_FCF_FLOGI_FAILED;
+			/*
+			 * the 1st priority that has not FLOGI failed
+			 * will be the highest.
+			 */
+			if (!next_fcf_pri)
+				next_fcf_pri = fcf_pri->fcf_rec.priority;
+			spin_unlock_irq(&phba->hbalock);
+			if (fcf_pri->fcf_rec.priority == next_fcf_pri) {
+				rc = lpfc_sli4_fcf_rr_index_set(phba,
+						fcf_pri->fcf_rec.fcf_index);
+				if (rc)
+					return 0;
+			}
+			spin_lock_irq(&phba->hbalock);
+		}
+	} else
+		ret = 1;
+	spin_unlock_irq(&phba->hbalock);
+
+	return ret;
+}
 /**
  * lpfc_sli4_fcf_rr_next_index_get - Get next eligible fcf record index
  * @phba: pointer to lpfc hba data structure.
@@ -14650,6 +14736,7 @@ lpfc_sli4_fcf_rr_next_index_get(struct lpfc_hba *phba)
 	uint16_t next_fcf_index;
 
 	/* Search start from next bit of currently registered FCF index */
+next_priority:
 	next_fcf_index = (phba->fcf.current_rec.fcf_indx + 1) %
 					LPFC_SLI4_FCF_TBL_INDX_MAX;
 	next_fcf_index = find_next_bit(phba->fcf.fcf_rr_bmask,
@@ -14657,16 +14744,45 @@ lpfc_sli4_fcf_rr_next_index_get(struct lpfc_hba *phba)
 				       next_fcf_index);
 
 	/* Wrap around condition on phba->fcf.fcf_rr_bmask */
-	if (next_fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX)
+	if (next_fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX) {
+		/*
+		 * If we have wrapped then we need to clear the bits that
+		 * have been tested so that we can detect when we should
+		 * change the priority level.
+		 */
 		next_fcf_index = find_next_bit(phba->fcf.fcf_rr_bmask,
 					       LPFC_SLI4_FCF_TBL_INDX_MAX, 0);
+	}
+
 
 	/* Check roundrobin failover list empty condition */
-	if (next_fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX) {
+	if (next_fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX ||
+		next_fcf_index == phba->fcf.current_rec.fcf_indx) {
+		/*
+		 * If next fcf index is not found check if there are lower
+		 * Priority level fcf's in the fcf_priority list.
+		 * Set up the rr_bmask with all of the avaiable fcf bits
+		 * at that level and continue the selection process.
+		 */
+		if (lpfc_check_next_fcf_pri_level(phba))
+			goto next_priority;
 		lpfc_printf_log(phba, KERN_WARNING, LOG_FIP,
 				"2844 No roundrobin failover FCF available\n");
+		if (next_fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX)
 			return LPFC_FCOE_FCF_NEXT_NONE;
+		else {
+			lpfc_printf_log(phba, KERN_WARNING, LOG_FIP,
+				"3063 Only FCF available idx %d, flag %x\n",
+				next_fcf_index,
+			phba->fcf.fcf_pri[next_fcf_index].fcf_rec.flag);
+			return next_fcf_index;
 		}
+	}
+
+	if (next_fcf_index < LPFC_SLI4_FCF_TBL_INDX_MAX &&
+		phba->fcf.fcf_pri[next_fcf_index].fcf_rec.flag &
+		LPFC_FCF_FLOGI_FAILED)
+		goto next_priority;
 
 	lpfc_printf_log(phba, KERN_INFO, LOG_FIP,
 			"2845 Get next roundrobin failover FCF (x%x)\n",
@@ -14719,6 +14835,7 @@ lpfc_sli4_fcf_rr_index_set(struct lpfc_hba *phba, uint16_t fcf_index)
 void
 lpfc_sli4_fcf_rr_index_clear(struct lpfc_hba *phba, uint16_t fcf_index)
 {
+	struct lpfc_fcf_pri *fcf_pri;
 	if (fcf_index >= LPFC_SLI4_FCF_TBL_INDX_MAX) {
 		lpfc_printf_log(phba, KERN_ERR, LOG_FIP,
 				"2762 FCF (x%x) reached driver's book "
@@ -14727,6 +14844,14 @@ lpfc_sli4_fcf_rr_index_clear(struct lpfc_hba *phba, uint16_t fcf_index)
 		return;
 	}
 	/* Clear the eligible FCF record index bmask */
+	spin_lock_irq(&phba->hbalock);
+	list_for_each_entry(fcf_pri, &phba->fcf.fcf_pri_list, list) {
+		if (fcf_pri->fcf_rec.fcf_index == fcf_index) {
+			list_del_init(&fcf_pri->list);
+			break;
+		}
+	}
+	spin_unlock_irq(&phba->hbalock);
 	clear_bit(fcf_index, phba->fcf.fcf_rr_bmask);
 
 	lpfc_printf_log(phba, KERN_INFO, LOG_FIP,

drivers/scsi/lpfc/lpfc_sli4.h

@@ -159,6 +159,25 @@ struct lpfc_fcf_rec {
 #define RECORD_VALID	0x02
 };
 
+struct lpfc_fcf_pri_rec {
+	uint16_t fcf_index;
+#define LPFC_FCF_ON_PRI_LIST 0x0001
+#define LPFC_FCF_FLOGI_FAILED 0x0002
+	uint16_t flag;
+	uint32_t priority;
+};
+
+struct lpfc_fcf_pri {
+	struct list_head list;
+	struct lpfc_fcf_pri_rec fcf_rec;
+};
+
+/*
+ * Maximum FCF table index, it is for driver internal book keeping, it
+ * just needs to be no less than the supported HBA's FCF table size.
+ */
+#define LPFC_SLI4_FCF_TBL_INDX_MAX	32
+
 struct lpfc_fcf {
 	uint16_t fcfi;
 	uint32_t fcf_flag;
@@ -178,15 +197,13 @@ struct lpfc_fcf {
 	uint32_t eligible_fcf_cnt;
 	struct lpfc_fcf_rec current_rec;
 	struct lpfc_fcf_rec failover_rec;
+	struct list_head fcf_pri_list;
+	struct lpfc_fcf_pri fcf_pri[LPFC_SLI4_FCF_TBL_INDX_MAX];
+	uint32_t current_fcf_scan_pri;
 	struct timer_list redisc_wait;
 	unsigned long *fcf_rr_bmask; /* Eligible FCF indexes for RR failover */
 };
 
-/*
- * Maximum FCF table index, it is for driver internal book keeping, it
- * just needs to be no less than the supported HBA's FCF table size.
- */
-#define LPFC_SLI4_FCF_TBL_INDX_MAX	32
 
 #define LPFC_REGION23_SIGNATURE "RG23"
 #define LPFC_REGION23_VERSION	1