scsi: pm8001: Expose hardware queues for pm80xx

In commit 05c6c029a44d ("scsi: pm80xx: Increase number of supported
queues"), support for 80xx chip was improved by enabling multiple HW
queues.

In this, like other SCSI MQ HBA drivers at the time, the HW queues were not
exposed to upper layer, and instead the driver managed the queues
internally.

However, this management duplicates blk-mq code. In addition, the HW queue
management is sub-optimal for a system where the number of CPUs exceeds the
HW queues - this is because queues are selected in a round-robin fashion,
when it would be better to make adjacent CPUs submit on the same queue. And
finally, the affinity of the completion queue interrupts is not set to
mirror the cpu<->HQ queue mapping, which is suboptimal.

As such, for when MSIX is supported, expose HW queues to upper layer. We
always use queue index #0 for "internal" commands, i.e. anything which does
not come from the block layer, so omit this from the affinity spreading.

Link: https://lore.kernel.org/r/1654879602-33497-5-git-send-email-john.garry@huawei.com
Tested-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

1 files changed, 39 insertions, 11 deletions

diff --git a/drivers/scsi/pm8001/pm8001_init.c b/drivers/scsi/pm8001/pm8001_init.c
index 171f0a2cf2d6..a3a972c2e532 100644
--- a/drivers/scsi/pm8001/pm8001_init.c
+++ b/drivers/scsi/pm8001/pm8001_init.c
@@ -81,6 +81,18 @@ LIST_HEAD(hba_list);
 
 struct workqueue_struct *pm8001_wq;
 
+static int pm8001_map_queues(struct Scsi_Host *shost)
+{
+	struct sas_ha_struct *sha = SHOST_TO_SAS_HA(shost);
+	struct pm8001_hba_info *pm8001_ha = sha->lldd_ha;
+	struct blk_mq_queue_map *qmap = &shost->tag_set.map[HCTX_TYPE_DEFAULT];
+
+	if (pm8001_ha->number_of_intr > 1)
+		blk_mq_pci_map_queues(qmap, pm8001_ha->pdev, 1);
+
+	return blk_mq_map_queues(qmap);
+}
+
 /*
  * The main structure which LLDD must register for scsi core.
  */
@@ -110,6 +122,7 @@ static struct scsi_host_template pm8001_sht = {
 	.shost_groups		= pm8001_host_groups,
 	.track_queue_depth	= 1,
 	.cmd_per_lun		= 32,
+	.map_queues		= pm8001_map_queues,
 };
 
 /*
@@ -928,31 +941,35 @@ static int pm8001_configure_phy_settings(struct pm8001_hba_info *pm8001_ha)
  */
 static u32 pm8001_setup_msix(struct pm8001_hba_info *pm8001_ha)
 {
-	u32 number_of_intr;
-	int rc, cpu_online_count;
 	unsigned int allocated_irq_vectors;
+	int rc;
 
 	/* SPCv controllers supports 64 msi-x */
 	if (pm8001_ha->chip_id == chip_8001) {
-		number_of_intr = 1;
+		rc = pci_alloc_irq_vectors(pm8001_ha->pdev, 1, 1,
+					   PCI_IRQ_MSIX);
 	} else {
-		number_of_intr = PM8001_MAX_MSIX_VEC;
+		/*
+		 * Queue index #0 is used always for housekeeping, so don't
+		 * include in the affinity spreading.
+		 */
+		struct irq_affinity desc = {
+			.pre_vectors = 1,
+		};
+		rc = pci_alloc_irq_vectors_affinity(
+				pm8001_ha->pdev, 2, PM8001_MAX_MSIX_VEC,
+				PCI_IRQ_MSIX | PCI_IRQ_AFFINITY, &desc);
 	}
 
-	cpu_online_count = num_online_cpus();
-	number_of_intr = min_t(int, cpu_online_count, number_of_intr);
-	rc = pci_alloc_irq_vectors(pm8001_ha->pdev, number_of_intr,
-			number_of_intr, PCI_IRQ_MSIX);
 	allocated_irq_vectors = rc;
 	if (rc < 0)
 		return rc;
 
 	/* Assigns the number of interrupts */
-	number_of_intr = min_t(int, allocated_irq_vectors, number_of_intr);
-	pm8001_ha->number_of_intr = number_of_intr;
+	pm8001_ha->number_of_intr = allocated_irq_vectors;
 
 	/* Maximum queue number updating in HBA structure */
-	pm8001_ha->max_q_num = number_of_intr;
+	pm8001_ha->max_q_num = allocated_irq_vectors;
 
 	pm8001_dbg(pm8001_ha, INIT,
 		   "pci_alloc_irq_vectors request ret:%d no of intr %d\n",
@@ -1123,8 +1140,19 @@ static int pm8001_pci_probe(struct pci_dev *pdev,
 	if (rc)
 		goto err_out_enable;
 
+
 	PM8001_CHIP_DISP->chip_post_init(pm8001_ha);
 
+	if (pm8001_ha->number_of_intr > 1) {
+		shost->nr_hw_queues = pm8001_ha->number_of_intr - 1;
+		/*
+		 * For now, ensure we're not sent too many commands by setting
+		 * host_tagset. This is also required if we start using request
+		 * tag.
+		 */
+		shost->host_tagset = 1;
+	}
+
 	rc = scsi_add_host(shost, &pdev->dev);
 	if (rc)
 		goto err_out_ha_free;