net/mlx4_en: Use affinity hint

The “affinity hint” mechanism is used by the user space
daemon, irqbalancer, to indicate a preferred CPU mask for irqs.
Irqbalancer can use this hint to balance the irqs between the
cpus indicated by the mask.

We wish the HCA to preferentially map the IRQs it uses to numa cores
close to it.  To accomplish this, we use cpumask_set_cpu_local_first(), that
sets the affinity hint according the following policy:
First it maps IRQs to “close” numa cores.  If these are exhausted, the
remaining IRQs are mapped to “far” numa cores.

Signed-off-by: Yuval Atias <yuvala@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/mellanox/mlx4/en_cq.c

@@ -163,6 +163,13 @@ int mlx4_en_activate_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq *cq,
 		netif_napi_add(cq->dev, &cq->napi, mlx4_en_poll_tx_cq,
 			       NAPI_POLL_WEIGHT);
 	} else {
+		struct mlx4_en_rx_ring *ring = priv->rx_ring[cq->ring];
+
+		err = irq_set_affinity_hint(cq->mcq.irq,
+					    ring->affinity_mask);
+		if (err)
+			mlx4_warn(mdev, "Failed setting affinity hint\n");
+
 		netif_napi_add(cq->dev, &cq->napi, mlx4_en_poll_rx_cq, 64);
 		napi_hash_add(&cq->napi);
 	}
@@ -179,8 +186,11 @@ void mlx4_en_destroy_cq(struct mlx4_en_priv *priv, struct mlx4_en_cq **pcq)
 
 	mlx4_en_unmap_buffer(&cq->wqres.buf);
 	mlx4_free_hwq_res(mdev->dev, &cq->wqres, cq->buf_size);
-	if (priv->mdev->dev->caps.comp_pool && cq->vector)
+	if (priv->mdev->dev->caps.comp_pool && cq->vector) {
+		if (!cq->is_tx)
+			irq_set_affinity_hint(cq->mcq.irq, NULL);
 		mlx4_release_eq(priv->mdev->dev, cq->vector);
+	}
 	cq->vector = 0;
 	cq->buf_size = 0;
 	cq->buf = NULL;

drivers/net/ethernet/mellanox/mlx4/en_netdev.c

@@ -1526,6 +1526,27 @@ static void mlx4_en_linkstate(struct work_struct *work)
 	mutex_unlock(&mdev->state_lock);
 }
 
+static int mlx4_en_init_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
+{
+	struct mlx4_en_rx_ring *ring = priv->rx_ring[ring_idx];
+	int numa_node = priv->mdev->dev->numa_node;
+	int ret = 0;
+
+	if (!zalloc_cpumask_var(&ring->affinity_mask, GFP_KERNEL))
+		return -ENOMEM;
+
+	ret = cpumask_set_cpu_local_first(ring_idx, numa_node,
+					  ring->affinity_mask);
+	if (ret)
+		free_cpumask_var(ring->affinity_mask);
+
+	return ret;
+}
+
+static void mlx4_en_free_affinity_hint(struct mlx4_en_priv *priv, int ring_idx)
+{
+	free_cpumask_var(priv->rx_ring[ring_idx]->affinity_mask);
+}
 
 int mlx4_en_start_port(struct net_device *dev)
 {
@@ -1567,9 +1588,16 @@ int mlx4_en_start_port(struct net_device *dev)
 
 		mlx4_en_cq_init_lock(cq);
 
+		err = mlx4_en_init_affinity_hint(priv, i);
+		if (err) {
+			en_err(priv, "Failed preparing IRQ affinity hint\n");
+			goto cq_err;
+		}
+
 		err = mlx4_en_activate_cq(priv, cq, i);
 		if (err) {
 			en_err(priv, "Failed activating Rx CQ\n");
+			mlx4_en_free_affinity_hint(priv, i);
 			goto cq_err;
 		}
 		for (j = 0; j < cq->size; j++)
@@ -1578,6 +1606,7 @@ int mlx4_en_start_port(struct net_device *dev)
 		if (err) {
 			en_err(priv, "Failed setting cq moderation parameters\n");
 			mlx4_en_deactivate_cq(priv, cq);
+			mlx4_en_free_affinity_hint(priv, i);
 			goto cq_err;
 		}
 		mlx4_en_arm_cq(priv, cq);
@@ -1715,8 +1744,10 @@ rss_err:
 mac_err:
 	mlx4_en_put_qp(priv);
 cq_err:
-	while (rx_index--)
+	while (rx_index--) {
 		mlx4_en_deactivate_cq(priv, priv->rx_cq[rx_index]);
+		mlx4_en_free_affinity_hint(priv, i);
+	}
 	for (i = 0; i < priv->rx_ring_num; i++)
 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
 
@@ -1847,6 +1878,8 @@ void mlx4_en_stop_port(struct net_device *dev, int detach)
 			msleep(1);
 		mlx4_en_deactivate_rx_ring(priv, priv->rx_ring[i]);
 		mlx4_en_deactivate_cq(priv, cq);
+
+		mlx4_en_free_affinity_hint(priv, i);
 	}
 }
 

drivers/net/ethernet/mellanox/mlx4/mlx4_en.h

@@ -313,6 +313,7 @@ struct mlx4_en_rx_ring {
 	unsigned long csum_ok;
 	unsigned long csum_none;
 	int hwtstamp_rx_filter;
+	cpumask_var_t affinity_mask;
 };
 
 struct mlx4_en_cq {