net: mana: Assign interrupts to CPUs based on NUMA nodes

Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2168970

commit 71fa688
Author: Saurabh Sengar <ssengar@linux.microsoft.com>
Date:   Mon Oct 31 23:06:01 2022 -0700

    net: mana: Assign interrupts to CPUs based on NUMA nodes

    In large VMs with multiple NUMA nodes, network performance is usually
    best if network interrupts are all assigned to the same virtual NUMA
    node. This patch assigns online CPU according to a numa aware policy,
    local cpus are returned first, followed by non-local ones, then it wraps
    around.

    Signed-off-by: Saurabh Sengar <ssengar@linux.microsoft.com>
    Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
    Link: https://lore.kernel.org/r/1667282761-11547-1-git-send-email-ssengar@linux.microsoft.com
    Signed-off-by: Paolo Abeni <pabeni@redhat.com>

Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>

Author: esposem

drivers/net/ethernet/microsoft/mana/gdma.h

@@ -353,6 +353,7 @@ struct gdma_context {
 	void __iomem		*shm_base;
 	void __iomem		*db_page_base;
 	u32 db_page_size;
+	int                     numa_node;
 
 	/* Shared memory chanenl (used to bootstrap HWC) */
 	struct shm_channel	shm_channel;

drivers/net/ethernet/microsoft/mana/gdma_main.c

@@ -1208,8 +1208,10 @@ static int mana_gd_setup_irqs(struct pci_dev *pdev)
 	struct gdma_context *gc = pci_get_drvdata(pdev);
 	struct gdma_irq_context *gic;
 	unsigned int max_irqs;
+	u16 *cpus;
+	cpumask_var_t req_mask;
 	int nvec, irq;
-	int err, i, j;
+	int err, i = 0, j;
 
 	if (max_queues_per_port > MANA_MAX_NUM_QUEUES)
 		max_queues_per_port = MANA_MAX_NUM_QUEUES;
@@ -1228,21 +1230,39 @@ static int mana_gd_setup_irqs(struct pci_dev *pdev)
 		goto free_irq_vector;
 	}
 
+	if (!zalloc_cpumask_var(&req_mask, GFP_KERNEL)) {
+		err = -ENOMEM;
+		goto free_irq;
+	}
+
+	cpus = kcalloc(nvec, sizeof(*cpus), GFP_KERNEL);
+	if (!cpus) {
+		err = -ENOMEM;
+		goto free_mask;
+	}
+	for (i = 0; i < nvec; i++)
+		cpus[i] = cpumask_local_spread(i, gc->numa_node);
+
 	for (i = 0; i < nvec; i++) {
+		cpumask_set_cpu(cpus[i], req_mask);
 		gic = &gc->irq_contexts[i];
 		gic->handler = NULL;
 		gic->arg = NULL;
 
 		irq = pci_irq_vector(pdev, i);
 		if (irq < 0) {
 			err = irq;
-			goto free_irq;
+			goto free_mask;
 		}
 
 		err = request_irq(irq, mana_gd_intr, 0, "mana_intr", gic);
 		if (err)
-			goto free_irq;
+			goto free_mask;
+		irq_set_affinity_and_hint(irq, req_mask);
+		cpumask_clear(req_mask);
 	}
+	free_cpumask_var(req_mask);
+	kfree(cpus);
 
 	err = mana_gd_alloc_res_map(nvec, &gc->msix_resource);
 	if (err)
@@ -1253,6 +1273,9 @@ static int mana_gd_setup_irqs(struct pci_dev *pdev)
 
 	return 0;
 
+free_mask:
+	free_cpumask_var(req_mask);
+	kfree(cpus);
 free_irq:
 	for (j = i - 1; j >= 0; j--) {
 		irq = pci_irq_vector(pdev, j);
@@ -1382,6 +1405,7 @@ static int mana_gd_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	if (!bar0_va)
 		goto free_gc;
 
+	gc->numa_node = dev_to_node(&pdev->dev);
 	gc->is_pf = mana_is_pf(pdev->device);
 	gc->bar0_va = bar0_va;
 	gc->dev = &pdev->dev;