Merge branch 'net-mlx5-frag-buffer-improvements'

Tariq Toukan says:

====================
net/mlx5: frag buffer improvements

This series adds observability for mlx5 fragment buffer DMA pools and
improves the default NUMA placement policy for fragment buffer
allocations.

Patch 1 adds a debugfs interface exposing per-node DMA pool usage
statistics for mlx5_frag_buf allocations, helping with debugging and
visibility into pool utilization.

Patch 2 improves locality of default fragment buffer allocations by
using numa_mem_id() when no explicit NUMA node is requested, allowing
allocations to prefer the current CPU's local memory node.

Together, these changes improve both introspection and memory locality
behavior of mlx5 fragment buffer allocations.
====================

Link: https://patch.msgid.link/20260514104925.337570-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/mellanox/mlx5/core/alloc.c

@@ -38,6 +38,8 @@
 #include <linux/dma-mapping.h>
 #include <linux/vmalloc.h>
 #include <linux/nodemask.h>
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
 #include <linux/mlx5/driver.h>
 
 #include "mlx5_core.h"
@@ -74,6 +76,13 @@ struct mlx5_frag_buf_node_pools {
 	struct mlx5_dma_pool *pools[MLX5_FRAG_BUF_POOLS_NUM];
 };
 
+struct mlx5_dma_pool_stats {
+	int node;
+	size_t block_size;
+	size_t used_blocks;
+	size_t allocated_blocks;
+};
+
 /* Handling for queue buffers -- we allocate a bunch of memory and
  * register it in a memory region at HCA virtual address 0.
  */
@@ -225,6 +234,43 @@ static void mlx5_dma_pool_free(struct mlx5_dma_pool *pool,
 	mutex_unlock(&pool->lock);
 }
 
+static void mlx5_dma_pool_debugfs_get_stats(struct mlx5_dma_pool *pool,
+					    struct mlx5_dma_pool_stats *stats)
+{
+	int blocks_per_page = BIT(PAGE_SHIFT - pool->block_shift);
+	struct mlx5_dma_pool_page *page;
+	size_t free_blocks = 0;
+	size_t pages = 0;
+
+	mutex_lock(&pool->lock);
+	list_for_each_entry(page, &pool->page_list, pool_link) {
+		pages++;
+		free_blocks += bitmap_weight(page->bitmap, blocks_per_page);
+	}
+	mutex_unlock(&pool->lock);
+
+	stats->node = pool->node;
+	stats->block_size = BIT(pool->block_shift);
+	stats->allocated_blocks = pages * blocks_per_page;
+	stats->used_blocks = stats->allocated_blocks - free_blocks;
+}
+
+static void mlx5_dma_pool_debugfs_stats_print(struct seq_file *file,
+					      struct mlx5_dma_pool *pool)
+{
+	struct mlx5_dma_pool_stats stats = {};
+
+	mlx5_dma_pool_debugfs_get_stats(pool, &stats);
+	seq_printf(file, "%4d       %5zu      %7zu           %7zu\n",
+		   stats.node, stats.block_size, stats.used_blocks,
+		   stats.allocated_blocks);
+}
+
+static void mlx5_dma_pools_debugfs_print_header(struct seq_file *file)
+{
+	seq_puts(file, "node  block_size  used_blocks  allocated_blocks\n");
+}
+
 static void
 mlx5_frag_buf_node_pools_destroy(struct mlx5_frag_buf_node_pools *node_pools)
 {
@@ -257,11 +303,46 @@ mlx5_frag_buf_node_pools_create(struct mlx5_core_dev *dev, int node)
 	return node_pools;
 }
 
+static int
+mlx5_frag_buf_dma_pools_debugfs_show(struct seq_file *file, void *priv)
+{
+	struct mlx5_core_dev *dev = file->private;
+	int node;
+
+	mlx5_dma_pools_debugfs_print_header(file);
+
+	if (!dev->priv.frag_buf_node_pools)
+		return 0;
+
+	for_each_node_state(node, N_POSSIBLE) {
+		struct mlx5_frag_buf_node_pools *node_pools;
+
+		node_pools = dev->priv.frag_buf_node_pools[node];
+		if (!node_pools)
+			continue;
+
+		for (int i = 0; i < MLX5_FRAG_BUF_POOLS_NUM; i++) {
+			struct mlx5_dma_pool *pool = node_pools->pools[i];
+
+			if (!pool)
+				continue;
+
+			mlx5_dma_pool_debugfs_stats_print(file, pool);
+		}
+	}
+
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(mlx5_frag_buf_dma_pools_debugfs);
+
 void mlx5_frag_buf_pools_cleanup(struct mlx5_core_dev *dev)
 {
 	struct mlx5_priv *priv = &dev->priv;
 	int node;
 
+	debugfs_remove(priv->dbg.frag_buf_dma_pools_debugfs);
+	priv->dbg.frag_buf_dma_pools_debugfs = NULL;
+
 	for_each_node_state(node, N_POSSIBLE) {
 		struct mlx5_frag_buf_node_pools *node_pools;
 
@@ -296,6 +377,11 @@ int mlx5_frag_buf_pools_init(struct mlx5_core_dev *dev)
 		priv->frag_buf_node_pools[node] = node_pools;
 	}
 
+	priv->dbg.frag_buf_dma_pools_debugfs =
+		debugfs_create_file("frag_buf_dma_pools", 0444,
+				    priv->dbg.dbg_root, dev,
+				    &mlx5_frag_buf_dma_pools_debugfs_fops);
+
 	return 0;
 }
 
@@ -305,7 +391,7 @@ int mlx5_frag_buf_alloc_node(struct mlx5_core_dev *dev, int size,
 	struct mlx5_dma_pool *pool;
 	int pool_idx;
 
-	node = node == NUMA_NO_NODE ? first_online_node : node;
+	node = node == NUMA_NO_NODE ? numa_mem_id() : node;
 
 	buf->size = size;
 	buf->npages = DIV_ROUND_UP(size, PAGE_SIZE);

include/linux/mlx5/driver.h

@@ -547,6 +547,7 @@ struct mlx5_debugfs_entries {
 	struct dentry *eq_debugfs;
 	struct dentry *cq_debugfs;
 	struct dentry *cmdif_debugfs;
+	struct dentry *frag_buf_dma_pools_debugfs;
 	struct dentry *pages_debugfs;
 	struct dentry *lag_debugfs;
 };