Skip to contentSkip to conversation

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

PATCH Applied Saurabh Sengar 3 participants 3 messages Nov 1, 2022 – Nov 3, 2022

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 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.

---
 drivers/net/ethernet/microsoft/mana/gdma.h    |  1 +
 .../net/ethernet/microsoft/mana/gdma_main.c   | 30 +++++++++++++++++--
 2 files changed, 28 insertions(+), 3 deletions(-)
drivers/net/ethernet/microsoft/mana/gdma.h +1 -0
index 4a6efe6ada08..db340f36ef29 100644
@@ -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 +27 -3
index a6f99b4344d9..726ac94d96ae 100644
@@ -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;
@@ -1228,7 +1230,21 @@ 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;
@@ -1236,13 +1252,17 @@ static int mana_gd_setup_irqs(struct pci_dev *pdev)
 		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;
Signature · 1 line
2.34.1
maintainer Reviewed-by on v1 1/1
#1 lore raw
Quoted 15 lines

-----Original Message-----
From: Saurabh Sengar <[email protected]>
Sent: Tuesday, November 1, 2022 2:06 AM
To: Saurabh Singh Sengar <[email protected]>; KY Srinivasan <[email protected]>; Haiyang Zhang <[email protected]>; [email protected]; Dexuan Cui <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; linux- [email protected]; [email protected]; linux- [email protected]; Michael Kelley (LINUX) <[email protected]>
Subject: [PATCH] net: mana: Assign interrupts to CPUs based on NUMA nodes

Quoted 7 lines

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.

Thank you.

#2 lore raw
In reply to Saurabh Sengar
This patch was applied to netdev/net-next.git (master) by Paolo Abeni <[email protected]>:

Hello:

This patch was applied to netdev/net-next.git (master) by Paolo Abeni <[email protected]>:

Quoted 9 lines · On Mon, 31 Oct 2022 23:06:01 -0700 you wrote:

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.

[...]

Here is the summary with links:
- net: mana: Assign interrupts to CPUs based on NUMA nodes

    https://git.kernel.org/netdev/net-next/c/71fa6887eeca

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html