scsi: hisi_sas: allocate number of CQ irq vectors according to cpu numbers

Currently if number of CQ irq vectors (fixed 16) are less than nr_cpus,
it will fail to allocate msi interrupt. So allocate number of CQ irq
vectors from 1 to 16.
And also if offline those cpus assocationed with CQ irq vectors, no
need to issue internal abort command to the queue, or internal abort
command will be timeout as there is no cpu processing the CQ interrupt.
Signed-off-by: NXiang Chen <chenxiang66@hisilicon.com>
Signed-off-by: NJohn Garry <john.garry@huawei.com>

Feature or Bugfix:Bugfix
Signed-off-by: Nchenxiang (M) <chenxiang66@hisilicon.com>
Reviewed-by: Ntanxiaofei <tanxiaofei@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

drivers/scsi/hisi_sas/hisi_sas.h

@@ -281,6 +281,8 @@ struct hisi_sas_hw {
 					   int delay_ms, int timeout_ms);
 	void (*snapshot_prepare)(struct hisi_hba *hisi_hba);
 	void (*snapshot_restore)(struct hisi_hba *hisi_hba);
+	const struct cpumask *(*get_managed_irq_aff)(struct hisi_hba
+			*hisi_hba, int queue);
 	int max_command_entries;
 	int complete_hdr_size;
 	struct scsi_host_template *sht;
@@ -369,6 +371,7 @@ struct hisi_hba {
 
 	bool user_ctl_irq;
 	unsigned int reply_map[NR_CPUS];
+	int nvecs;
 };
 
 /* Generic HW DMA host memory structures */

drivers/scsi/hisi_sas/hisi_sas_main.c

@@ -2088,7 +2088,23 @@ hisi_sas_internal_task_abort(struct hisi_hba *hisi_hba,
 		return _hisi_sas_internal_task_abort(hisi_hba, device,
 						     abort_flag, tag, dq);
 	case HISI_SAS_INT_ABT_DEV:
-		for (i = 0; i < hisi_hba->queue_count; i++) {
+		for (i = 0; i < hisi_hba->nvecs; i++) {
+			const struct cpumask *mask = NULL;
+
+			if (hisi_hba->hw->get_managed_irq_aff)
+				mask = hisi_hba->hw->get_managed_irq_aff(
+						hisi_hba, i);
+			/*
+			 * The kernel will not permit unmanaged (MSI are
+			 * managed) IRQ affinity to offline CPUs, so
+			 * always issue internal abort on all queues
+			 * in this case.
+			 * For MSI interrupts, affinity may be set to
+			 * offline CPUs, so ensure that there's an online
+			 * CPU to handle the CQ interrupt.
+			 */
+			if (mask && !cpumask_intersects(cpu_online_mask, mask))
+				continue;
 			dq = &hisi_hba->dq[i];
 			rc = _hisi_sas_internal_task_abort(hisi_hba, device,
 							   abort_flag, tag, dq);
@@ -2171,7 +2187,7 @@ void hisi_sas_kill_tasklets(struct hisi_hba *hisi_hba)
 {
 	int i;
 
-	for (i = 0; i < hisi_hba->queue_count; i++) {
+	for (i = 0; i < hisi_hba->nvecs; i++) {
 		struct hisi_sas_cq *cq = &hisi_hba->cq[i];
 
 		tasklet_kill(&cq->tasklet);

drivers/scsi/hisi_sas/hisi_sas_v2_hw.c

@@ -3416,6 +3416,7 @@ static int interrupt_init_v2_hw(struct hisi_hba *hisi_hba)
 		tasklet_init(t, cq_tasklet_v2_hw, (unsigned long)cq);
 	}
 
+	hisi_hba->nvecs = hisi_hba->queue_count;
 	return 0;
 
 free_cq_int_irqs:

drivers/scsi/hisi_sas/hisi_sas_v3_hw.c

@@ -406,6 +406,7 @@ struct hisi_sas_err_record_v3 {
 #define T10_CHK_MSK_OFF	    16
 
 #define HISI_SAS_CQ_INT_BASE_VECTORS_V3_HW  16
+#define HISI_SAS_MIN_VECTORS_V3_HW  17
 
 #define HISI_SAS_IS_RW_CMD(op) \
 	((op == READ_6) || (op == WRITE_6) || \
@@ -2131,16 +2132,15 @@ static irqreturn_t cq_interrupt_v3_hw(int irq_no, void *p)
 	return IRQ_HANDLED;
 }
 
-static void setup_reply_map_v3_hw(struct hisi_hba *hisi_hba)
+static void setup_reply_map_v3_hw(struct hisi_hba *hisi_hba, int nvecs)
 {
 	const struct cpumask *mask;
 	int queue, cpu;
 
-	for (queue = 0; queue < hisi_hba->queue_count; queue++) {
+	for (queue = 0; queue < nvecs; queue++) {
 		mask = pci_irq_get_affinity(hisi_hba->pci_dev, queue + 16);
 		if (!mask)
 			goto fallback;
-
 		for_each_cpu(cpu, mask)
 			hisi_hba->reply_map[cpu] = queue;
 	}
@@ -2167,16 +2167,21 @@ static int interrupt_init_v3_hw(struct hisi_hba *hisi_hba)
 						max_msi, PCI_IRQ_MSI);
 	} else {
 		vectors = pci_alloc_irq_vectors_affinity(hisi_hba->pci_dev,
-				    max_msi, max_msi,
-				    PCI_IRQ_MSI | PCI_IRQ_AFFINITY, &desc);
-		setup_reply_map_v3_hw(hisi_hba);
+				HISI_SAS_MIN_VECTORS_V3_HW,
+				max_msi,
+				PCI_IRQ_MSI |
+				PCI_IRQ_AFFINITY,
+				&desc);
+		setup_reply_map_v3_hw(hisi_hba, vectors -
+				      HISI_SAS_CQ_INT_BASE_VECTORS_V3_HW);
 	}
 
-	if (vectors < max_msi) {
-		dev_err(dev, "could not allocate all msi (%d)\n", vectors);
+	if (vectors < HISI_SAS_MIN_VECTORS_V3_HW) {
+		dev_err(dev, "allocate msi (%d) not enough\n", vectors);
 		return -ENOENT;
 	}
 
+	hisi_hba->nvecs = vectors - HISI_SAS_CQ_INT_BASE_VECTORS_V3_HW;
 	rc = devm_request_irq(dev, pci_irq_vector(pdev, 1),
 			      int_phy_up_down_bcast_v3_hw, 0,
 			      DRV_NAME " phy", hisi_hba);
@@ -2205,7 +2210,7 @@ static int interrupt_init_v3_hw(struct hisi_hba *hisi_hba)
 	}
 
 	/* Init tasklets for cq only */
-	for (i = 0; i < hisi_hba->queue_count; i++) {
+	for (i = 0; i < hisi_hba->nvecs; i++) {
 		struct hisi_sas_cq *cq = &hisi_hba->cq[i];
 		struct tasklet_struct *t = &cq->tasklet;
 		int nr = hisi_sas_intr_conv ? 16 : 16 + i;
@@ -2701,6 +2706,16 @@ static void debugfs_snapshot_restore_v3_hw(struct hisi_hba *hisi_hba)
 	clear_bit(HISI_SAS_REJECT_CMD_BIT, &hisi_hba->flags);
 }
 
+const struct cpumask *
+get_managed_irq_aff_v3_hw(struct hisi_hba *hisi_hba, int queue)
+{
+	if (user_ctl_irq)
+		return NULL;
+
+	return pci_irq_get_affinity(hisi_hba->pci_dev, queue +
+				    HISI_SAS_CQ_INT_BASE_VECTORS_V3_HW);
+}
+
 struct device_attribute *host_attrs_v3_hw[] = {
 	&dev_attr_phy_event_threshold,
 	&dev_attr_intr_conv,
@@ -2764,6 +2779,7 @@ static const struct hisi_sas_hw hisi_sas_v3_hw = {
 	.debugfs_reg_port = &debugfs_port_reg,
 	.snapshot_prepare = debugfs_snapshot_prepare_v3_hw,
 	.snapshot_restore = debugfs_snapshot_restore_v3_hw,
+	.get_managed_irq_aff = get_managed_irq_aff_v3_hw,
 };
 
 static struct Scsi_Host *
@@ -2937,7 +2953,7 @@ hisi_sas_v3_destroy_irqs(struct pci_dev *pdev, struct hisi_hba *hisi_hba)
 	free_irq(pci_irq_vector(pdev, 1), hisi_hba);
 	free_irq(pci_irq_vector(pdev, 2), hisi_hba);
 	free_irq(pci_irq_vector(pdev, 11), hisi_hba);
-	for (i = 0; i < hisi_hba->queue_count; i++) {
+	for (i = 0; i < hisi_hba->nvecs; i++) {
 		struct hisi_sas_cq *cq = &hisi_hba->cq[i];
 		int nr = hisi_sas_intr_conv ? 16 : 16 + i;