scsi: lpfc: Revise interrupt coalescing for missing scenarios

The existing "auto eq delay" mechanism was sometimes skipping over an EQ,
not ramping the coalescing down under light load fast enough, and in other
cases never kicked in as cpu sharing by multiple vectors didn't quite add
up right.

Tweak the interrupt mechanism such that:

 - Add a flag to the EQ to force checking for colaescing values when being
   serviced in the interrupt handler.  The flag will be set by any CQ bound
   to the EQ whenever the number of CQ elements process in a single scan
   meets or exceeds the hardware queue notify level. E.g. there's a
   significant number of completions happening.

 - In the heartbeat work item that checks coalescing:

   - Replace the structure that was counting the number of EQs that
     interrupted on a single cpu with a new structure that looks at the EQ
     to see whether EQ currently has a coalescing value (thus it should be
     re-evaluate) or was marked by the new flag indicating heavy
     completions.

   - When a cpu, which may be servicing multiple vectors, had at least 1 EQ
     that should be checked, a new coalescing delay is calculated based on
     the number of interrupts that occurred on the cpu.

   - The new coalescing value is then applied to the EQs that had
     interrupted on the cpu.

Link: https://lore.kernel.org/r/20191018211832.7917-11-jsmart2021@gmail.comSigned-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

drivers/scsi/lpfc/lpfc_hw4.h

@@ -210,7 +210,6 @@ struct lpfc_sli_intf {
 #define LPFC_MAX_IMAX          5000000
 #define LPFC_DEF_IMAX          0
 
-#define LPFC_IMAX_THRESHOLD    1000
 #define LPFC_MAX_AUTO_EQ_DELAY 120
 #define LPFC_EQ_DELAY_STEP     15
 #define LPFC_EQD_ISR_TRIGGER   20000

drivers/scsi/lpfc/lpfc_init.c

@@ -1235,10 +1235,9 @@ lpfc_hb_eq_delay_work(struct work_struct *work)
 					     struct lpfc_hba, eq_delay_work);
 	struct lpfc_eq_intr_info *eqi, *eqi_new;
 	struct lpfc_queue *eq, *eq_next;
-	unsigned char *eqcnt = NULL;
+	unsigned char *ena_delay = NULL;
 	uint32_t usdelay;
 	int i;
-	bool update = false;
 
 	if (!phba->cfg_auto_imax || phba->pport->load_flag & FC_UNLOADING)
 		return;
@@ -1247,44 +1246,36 @@ lpfc_hb_eq_delay_work(struct work_struct *work)
 	    phba->pport->fc_flag & FC_OFFLINE_MODE)
 		goto requeue;
 
-	eqcnt = kcalloc(num_possible_cpus(), sizeof(unsigned char),
-			GFP_KERNEL);
-	if (!eqcnt)
+	ena_delay = kcalloc(phba->sli4_hba.num_possible_cpu, sizeof(*ena_delay),
+			    GFP_KERNEL);
+	if (!ena_delay)
 		goto requeue;
 
-	if (phba->cfg_irq_chann > 1) {
-		/* Loop thru all IRQ vectors */
-		for (i = 0; i < phba->cfg_irq_chann; i++) {
-			/* Get the EQ corresponding to the IRQ vector */
-			eq = phba->sli4_hba.hba_eq_hdl[i].eq;
-			if (!eq)
-				continue;
-			if (eq->q_mode) {
-				update = true;
-				break;
-			}
-			if (eqcnt[eq->last_cpu] < 2)
-				eqcnt[eq->last_cpu]++;
+	for (i = 0; i < phba->cfg_irq_chann; i++) {
+		/* Get the EQ corresponding to the IRQ vector */
+		eq = phba->sli4_hba.hba_eq_hdl[i].eq;
+		if (!eq)
+			continue;
+		if (eq->q_mode || eq->q_flag & HBA_EQ_DELAY_CHK) {
+			eq->q_flag &= ~HBA_EQ_DELAY_CHK;
+			ena_delay[eq->last_cpu] = 1;
 		}
-	} else
-		update = true;
+	}
 
 	for_each_present_cpu(i) {
 		eqi = per_cpu_ptr(phba->sli4_hba.eq_info, i);
-		if (!update && eqcnt[i] < 2) {
-			eqi->icnt = 0;
-			continue;
+		if (ena_delay[i]) {
+			usdelay = (eqi->icnt >> 10) * LPFC_EQ_DELAY_STEP;
+			if (usdelay > LPFC_MAX_AUTO_EQ_DELAY)
+				usdelay = LPFC_MAX_AUTO_EQ_DELAY;
+		} else {
+			usdelay = 0;
 		}
 
-		usdelay = (eqi->icnt / LPFC_IMAX_THRESHOLD) *
-			   LPFC_EQ_DELAY_STEP;
-		if (usdelay > LPFC_MAX_AUTO_EQ_DELAY)
-			usdelay = LPFC_MAX_AUTO_EQ_DELAY;
-
 		eqi->icnt = 0;
 
 		list_for_each_entry_safe(eq, eq_next, &eqi->list, cpu_list) {
-			if (eq->last_cpu != i) {
+			if (unlikely(eq->last_cpu != i)) {
 				eqi_new = per_cpu_ptr(phba->sli4_hba.eq_info,
 						      eq->last_cpu);
 				list_move_tail(&eq->cpu_list, &eqi_new->list);
@@ -1296,7 +1287,7 @@ lpfc_hb_eq_delay_work(struct work_struct *work)
 		}
 	}
 
-	kfree(eqcnt);
+	kfree(ena_delay);
 
 requeue:
 	queue_delayed_work(phba->wq, &phba->eq_delay_work,

drivers/scsi/lpfc/lpfc_sli.c

@@ -13640,6 +13640,7 @@ __lpfc_sli4_process_cq(struct lpfc_hba *phba, struct lpfc_queue *cq,
 			phba->sli4_hba.sli4_write_cq_db(phba, cq, consumed,
 						LPFC_QUEUE_NOARM);
 			consumed = 0;
+			cq->assoc_qp->q_flag |= HBA_EQ_DELAY_CHK;
 		}
 
 		if (count == LPFC_NVMET_CQ_NOTIFY)
@@ -14278,7 +14279,7 @@ lpfc_sli4_hba_intr_handler(int irq, void *dev_id)
 	fpeq->last_cpu = raw_smp_processor_id();
 
 	if (icnt > LPFC_EQD_ISR_TRIGGER &&
-	    phba->cfg_irq_chann == 1 &&
+	    fpeq->q_flag & HBA_EQ_DELAY_CHK &&
 	    phba->cfg_auto_imax &&
 	    fpeq->q_mode != LPFC_MAX_AUTO_EQ_DELAY &&
 	    phba->sli.sli_flag & LPFC_SLI_USE_EQDR)

drivers/scsi/lpfc/lpfc_sli4.h

@@ -199,6 +199,7 @@ struct lpfc_queue {
 	uint8_t q_flag;
 #define HBA_NVMET_WQFULL	0x1 /* We hit WQ Full condition for NVMET */
 #define HBA_NVMET_CQ_NOTIFY	0x1 /* LPFC_NVMET_CQ_NOTIFY CQEs this EQE */
+#define HBA_EQ_DELAY_CHK	0x2 /* EQ is a candidate for coalescing */
 #define LPFC_NVMET_CQ_NOTIFY	4
 	void __iomem *db_regaddr;
 	uint16_t dpp_enable;