[net-next,2/7] bnxt_en: Enable XPS by default on driver load

Commit Message

Pavan Chebbi April 1, 2024, 3:57 a.m. UTC

From: Somnath Kotur <[e-mail]>

Enable XPS on default during NIC open. The choice of
Tx queue is based on the CPU executing the thread that
submits the Tx request. The pool of Tx queues will be
spread evenly across both device-attached NUMA nodes(local)
and remote NUMA nodes.

Signed-off-by: Somnath Kotur <[e-mail]>
Reviewed-by: Michael Chan <[e-mail]>
Signed-off-by: Pavan Chebbi <[e-mail]>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 45 ++++++++++++++++++++++-
 1 file changed, 44 insertions(+), 1 deletion(-)

Comments

Jakub Kicinski April 2, 2024, 4:53 a.m. UTC | #1

On Sun, 31 Mar 2024 20:57:25 -0700 Pavan Chebbi wrote:
> +	for (i = 0;  i < nr_cpus;  i++) {

double spaces here

> +		map_idx = i % bp->tx_nr_rings_per_tc;
> +		cpu = cpumask_local_spread(i, numa_node);
> +		cpu_mask_ptr = get_cpu_mask(cpu);
> +		cpumask_or(&q_map[map_idx], &q_map[map_idx], cpu_mask_ptr);
> +	}
> +
> +	/* Register CPU mask for each TX queue excluding the ones marked for XDP */

don't go over 80 chars when you don't have to

Patch

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 6e24a341ad28..54955f878b73 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -11804,6 +11804,45 @@  static void bnxt_cfg_usr_fltrs(struct bnxt *bp)
 		bnxt_cfg_one_usr_fltr(bp, usr_fltr);
 }
 
+static int bnxt_set_xps_mapping(struct bnxt *bp)
+{
+	int numa_node = dev_to_node(&bp->pdev->dev);
+	unsigned int q_idx, map_idx, cpu, i;
+	const struct cpumask *cpu_mask_ptr;
+	int nr_cpus = num_online_cpus();
+	cpumask_t *q_map;
+	int rc = 0;
+
+	q_map = kcalloc(bp->tx_nr_rings_per_tc, sizeof(*q_map), GFP_KERNEL);
+	if (!q_map)
+		return -ENOMEM;
+
+	/* Create CPU mask for all TX queues across MQPRIO traffic classes.
+	 * Each TC has the same number of TX queues. The nth TX queue for each
+	 * TC will have the same CPU mask.
+	 */
+	for (i = 0;  i < nr_cpus;  i++) {
+		map_idx = i % bp->tx_nr_rings_per_tc;
+		cpu = cpumask_local_spread(i, numa_node);
+		cpu_mask_ptr = get_cpu_mask(cpu);
+		cpumask_or(&q_map[map_idx], &q_map[map_idx], cpu_mask_ptr);
+	}
+
+	/* Register CPU mask for each TX queue excluding the ones marked for XDP */
+	for (q_idx = 0; q_idx < bp->dev->real_num_tx_queues; q_idx++) {
+		map_idx = q_idx % bp->tx_nr_rings_per_tc;
+		rc = netif_set_xps_queue(bp->dev, &q_map[map_idx], q_idx);
+		if (rc) {
+			netdev_warn(bp->dev, "Error setting XPS for q:%d\n", q_idx);
+			break;
+		}
+	}
+
+	kfree(q_map);
+
+	return rc;
+}
+
 static int __bnxt_open_nic(struct bnxt *bp, bool irq_re_init, bool link_re_init)
 {
 	int rc = 0;
@@ -11866,8 +11905,12 @@  static int __bnxt_open_nic(struct bnxt *bp, bool irq_re_init, bool link_re_init)
 		}
 	}
 
-	if (irq_re_init)
+	if (irq_re_init) {
 		udp_tunnel_nic_reset_ntf(bp->dev);
+		rc = bnxt_set_xps_mapping(bp);
+		if (rc)
+			netdev_warn(bp->dev, "failed to set xps mapping\n");
+	}
 
 	if (bp->tx_nr_rings_xdp < num_possible_cpus()) {
 		if (!static_key_enabled(&bnxt_xdp_locking_key))