* [PATCH net-next V2 1/3] net/mlx5: initialize doorbell dma pools
2026-08-03 13:25 [PATCH net-next V2 0/3] net/mlx5: allocate doorbells from dma pools Tariq Toukan
@ 2026-08-03 13:25 ` Tariq Toukan
2026-08-03 13:25 ` [PATCH net-next V2 2/3] net/mlx5: allocate doorbells from " Tariq Toukan
2026-08-03 13:25 ` [PATCH net-next V2 3/3] net/mlx5: add debugfs stats for doorbell " Tariq Toukan
2 siblings, 0 replies; 4+ messages in thread
From: Tariq Toukan @ 2026-08-03 13:25 UTC (permalink / raw)
To: Andrew Lunn, David S. Miller, Eric Dumazet, Jakub Kicinski,
netdev, Paolo Abeni
Cc: Dragos Tatulea, Gal Pressman, Leon Romanovsky, linux-kernel,
linux-rdma, Mark Bloch, Nimrod Oren, Saeed Mahameed, Tariq Toukan
From: Nimrod Oren <noren@nvidia.com>
Add per-node doorbell dma pool creation and cleanup to mdev lifecycle.
Signed-off-by: Nimrod Oren <noren@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
.../net/ethernet/mellanox/mlx5/core/alloc.c | 37 +++++++++++++++++++
.../net/ethernet/mellanox/mlx5/core/main.c | 7 ++++
.../ethernet/mellanox/mlx5/core/mlx5_core.h | 2 +
include/linux/mlx5/driver.h | 2 +
4 files changed, 48 insertions(+)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/alloc.c b/drivers/net/ethernet/mellanox/mlx5/core/alloc.c
index 4fe9d7d4f143..3c9938068c56 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/alloc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/alloc.c
@@ -444,6 +444,43 @@ void mlx5_frag_buf_free(struct mlx5_core_dev *dev, struct mlx5_frag_buf *buf)
}
EXPORT_SYMBOL_GPL(mlx5_frag_buf_free);
+void mlx5_db_pools_cleanup(struct mlx5_core_dev *dev)
+{
+ struct mlx5_priv *priv = &dev->priv;
+ int node;
+
+ for_each_node_state(node, N_POSSIBLE)
+ if (priv->db_node_pools[node])
+ mlx5_dma_pool_destroy(priv->db_node_pools[node]);
+
+ kfree(priv->db_node_pools);
+ priv->db_node_pools = NULL;
+}
+
+int mlx5_db_pools_init(struct mlx5_core_dev *dev)
+{
+ struct mlx5_priv *priv = &dev->priv;
+ int node;
+
+ priv->db_node_pools = kzalloc_objs(*priv->db_node_pools, nr_node_ids);
+ if (!priv->db_node_pools)
+ return -ENOMEM;
+
+ for_each_node_state(node, N_POSSIBLE) {
+ struct mlx5_dma_pool *pool;
+
+ pool = mlx5_dma_pool_create(dev, node,
+ order_base_2(cache_line_size()));
+ if (!pool) {
+ mlx5_db_pools_cleanup(dev);
+ return -ENOMEM;
+ }
+ priv->db_node_pools[node] = pool;
+ }
+
+ return 0;
+}
+
static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev,
int node)
{
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index 643b4aac2033..b3cb090b5677 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -1830,6 +1830,10 @@ int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
if (err)
goto err_frag_buf_pools_init;
+ err = mlx5_db_pools_init(dev);
+ if (err)
+ goto err_db_pools_init;
+
INIT_LIST_HEAD(&priv->traps);
err = mlx5_cmd_init(dev);
@@ -1891,6 +1895,8 @@ int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
err_timeout_init:
mlx5_cmd_cleanup(dev);
err_cmd_init:
+ mlx5_db_pools_cleanup(dev);
+err_db_pools_init:
mlx5_frag_buf_pools_cleanup(dev);
err_frag_buf_pools_init:
debugfs_remove(dev->priv.dbg.dbg_root);
@@ -1917,6 +1923,7 @@ void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
mlx5_health_cleanup(dev);
mlx5_tout_cleanup(dev);
mlx5_cmd_cleanup(dev);
+ mlx5_db_pools_cleanup(dev);
mlx5_frag_buf_pools_cleanup(dev);
debugfs_remove_recursive(dev->priv.dbg.dbg_root);
mutex_destroy(&priv->pgdir_mutex);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h b/drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h
index 09e669f83dba..d6713a2ce676 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h
@@ -438,6 +438,8 @@ int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx);
void mlx5_mdev_uninit(struct mlx5_core_dev *dev);
int mlx5_frag_buf_pools_init(struct mlx5_core_dev *dev);
void mlx5_frag_buf_pools_cleanup(struct mlx5_core_dev *dev);
+int mlx5_db_pools_init(struct mlx5_core_dev *dev);
+void mlx5_db_pools_cleanup(struct mlx5_core_dev *dev);
int mlx5_init_one(struct mlx5_core_dev *dev);
int mlx5_init_one_devl_locked(struct mlx5_core_dev *dev);
void mlx5_uninit_one(struct mlx5_core_dev *dev);
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index b1871c0821d0..4246d6d904ba 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -569,6 +569,7 @@ enum mlx5_page_mgt_mode {
};
struct mlx5_frag_buf_node_pools;
+struct mlx5_dma_pool;
struct mlx5_ft_pool;
struct mlx5_priv {
/* IRQ table valid only for real pci devices PF or VF */
@@ -602,6 +603,7 @@ struct mlx5_priv {
struct list_head pgdir_list;
struct mlx5_frag_buf_node_pools **frag_buf_node_pools;
+ struct mlx5_dma_pool **db_node_pools;
/* end: alloc stuff */
struct mlx5_adev **adev;
--
2.44.0
^ permalink raw reply related [flat|nested] 4+ messages in thread* [PATCH net-next V2 2/3] net/mlx5: allocate doorbells from dma pools
2026-08-03 13:25 [PATCH net-next V2 0/3] net/mlx5: allocate doorbells from dma pools Tariq Toukan
2026-08-03 13:25 ` [PATCH net-next V2 1/3] net/mlx5: initialize doorbell " Tariq Toukan
@ 2026-08-03 13:25 ` Tariq Toukan
2026-08-03 13:25 ` [PATCH net-next V2 3/3] net/mlx5: add debugfs stats for doorbell " Tariq Toukan
2 siblings, 0 replies; 4+ messages in thread
From: Tariq Toukan @ 2026-08-03 13:25 UTC (permalink / raw)
To: Andrew Lunn, David S. Miller, Eric Dumazet, Jakub Kicinski,
netdev, Paolo Abeni
Cc: Dragos Tatulea, Gal Pressman, Leon Romanovsky, linux-kernel,
linux-rdma, Mark Bloch, Nimrod Oren, Saeed Mahameed, Tariq Toukan
From: Nimrod Oren <noren@nvidia.com>
Allocate doorbells from dma pools instead of the pgdir allocator.
Doorbell records remain cache-line sized coherent DMA allocations, but
their sub-allocation is now handled by the common mlx5 DMA pool
infrastructure.
This also makes doorbell allocation honor the requested NUMA node when
reusing existing backing pages. The old pgdir allocator used the
requested node only when allocating a new pgdir page; later
allocations scanned one global pgdir list and could take any pgdir with
a free entry, even if that page had been allocated for a different
NUMA node. Selecting the per-node DMA pool before sub-allocation keeps
reused doorbell records on pages allocated for the requested node.
Signed-off-by: Nimrod Oren <noren@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
.../net/ethernet/mellanox/mlx5/core/alloc.c | 111 +++---------------
.../net/ethernet/mellanox/mlx5/core/main.c | 4 -
include/linux/mlx5/driver.h | 5 +-
3 files changed, 18 insertions(+), 102 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/alloc.c b/drivers/net/ethernet/mellanox/mlx5/core/alloc.c
index 3c9938068c56..3975c726d35c 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/alloc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/alloc.c
@@ -48,13 +48,6 @@
#define MLX5_FRAG_BUF_POOLS_NUM \
(PAGE_SHIFT - MLX5_FRAG_BUF_POOL_MIN_BLOCK_SHIFT + 1)
-struct mlx5_db_pgdir {
- struct list_head list;
- unsigned long *bitmap;
- __be32 *db_page;
- dma_addr_t db_dma;
-};
-
struct mlx5_dma_pool {
/* Protects page_list and per-page allocation bitmaps. */
struct mutex lock;
@@ -481,106 +474,36 @@ int mlx5_db_pools_init(struct mlx5_core_dev *dev)
return 0;
}
-static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev,
- int node)
-{
- u32 db_per_page = PAGE_SIZE / cache_line_size();
- struct mlx5_db_pgdir *pgdir;
-
- pgdir = kzalloc_node(sizeof(*pgdir), GFP_KERNEL, node);
- if (!pgdir)
- return NULL;
-
- pgdir->bitmap = bitmap_zalloc_node(db_per_page, GFP_KERNEL, node);
- if (!pgdir->bitmap) {
- kfree(pgdir);
- return NULL;
- }
-
- bitmap_fill(pgdir->bitmap, db_per_page);
-
- pgdir->db_page = mlx5_dma_zalloc_coherent_node(dev, PAGE_SIZE,
- &pgdir->db_dma, node);
- if (!pgdir->db_page) {
- bitmap_free(pgdir->bitmap);
- kfree(pgdir);
- return NULL;
- }
-
- return pgdir;
-}
-
-static int mlx5_alloc_db_from_pgdir(struct mlx5_db_pgdir *pgdir,
- struct mlx5_db *db)
+int mlx5_db_alloc_node(struct mlx5_core_dev *dev, struct mlx5_db *db, int node)
{
- u32 db_per_page = PAGE_SIZE / cache_line_size();
+ struct mlx5_dma_pool_page *page;
+ struct mlx5_dma_pool *pool;
+ unsigned long idx;
int offset;
- int i;
-
- i = find_first_bit(pgdir->bitmap, db_per_page);
- if (i >= db_per_page)
- return -ENOMEM;
- __clear_bit(i, pgdir->bitmap);
+ node = node == NUMA_NO_NODE ? numa_mem_id() : node;
- db->u.pgdir = pgdir;
- db->index = i;
- offset = db->index * cache_line_size();
- db->db = pgdir->db_page + offset / sizeof(*pgdir->db_page);
- db->dma = pgdir->db_dma + offset;
+ pool = dev->priv.db_node_pools[node];
+ page = mlx5_dma_pool_alloc(pool, &idx);
+ if (!page)
+ return -ENOMEM;
- db->db[0] = 0;
- db->db[1] = 0;
+ offset = idx << pool->block_shift;
+ db->u.pool_page = page;
+ db->index = idx;
+ db->db = (__be32 *)((u8 *)page->buf + offset);
+ db->dma = page->dma + offset;
return 0;
}
-
-int mlx5_db_alloc_node(struct mlx5_core_dev *dev, struct mlx5_db *db, int node)
-{
- struct mlx5_db_pgdir *pgdir;
- int ret = 0;
-
- mutex_lock(&dev->priv.pgdir_mutex);
-
- list_for_each_entry(pgdir, &dev->priv.pgdir_list, list)
- if (!mlx5_alloc_db_from_pgdir(pgdir, db))
- goto out;
-
- pgdir = mlx5_alloc_db_pgdir(dev, node);
- if (!pgdir) {
- ret = -ENOMEM;
- goto out;
- }
-
- list_add(&pgdir->list, &dev->priv.pgdir_list);
-
- /* This should never fail -- we just allocated an empty page: */
- WARN_ON(mlx5_alloc_db_from_pgdir(pgdir, db));
-
-out:
- mutex_unlock(&dev->priv.pgdir_mutex);
-
- return ret;
-}
EXPORT_SYMBOL_GPL(mlx5_db_alloc_node);
void mlx5_db_free(struct mlx5_core_dev *dev, struct mlx5_db *db)
{
- u32 db_per_page = PAGE_SIZE / cache_line_size();
-
- mutex_lock(&dev->priv.pgdir_mutex);
-
- __set_bit(db->index, db->u.pgdir->bitmap);
-
- if (bitmap_full(db->u.pgdir->bitmap, db_per_page)) {
- dma_free_coherent(mlx5_core_dma_dev(dev), PAGE_SIZE,
- db->u.pgdir->db_page, db->u.pgdir->db_dma);
- list_del(&db->u.pgdir->list);
- bitmap_free(db->u.pgdir->bitmap);
- kfree(db->u.pgdir);
- }
+ struct mlx5_dma_pool_page *page = db->u.pool_page;
+ struct mlx5_dma_pool *pool = page->pool;
- mutex_unlock(&dev->priv.pgdir_mutex);
+ mlx5_dma_pool_free(pool, page, db->index);
}
EXPORT_SYMBOL_GPL(mlx5_db_free);
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index b3cb090b5677..5f28d906c35b 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -1819,8 +1819,6 @@ int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
mutex_init(&priv->alloc_mutex);
- mutex_init(&priv->pgdir_mutex);
- INIT_LIST_HEAD(&priv->pgdir_list);
priv->numa_node = dev_to_node(mlx5_core_dma_dev(dev));
priv->dbg.dbg_root = debugfs_create_dir(dev_name(dev->device),
@@ -1900,7 +1898,6 @@ int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
mlx5_frag_buf_pools_cleanup(dev);
err_frag_buf_pools_init:
debugfs_remove(dev->priv.dbg.dbg_root);
- mutex_destroy(&priv->pgdir_mutex);
mutex_destroy(&priv->alloc_mutex);
mutex_destroy(&priv->bfregs.wc_head.lock);
mutex_destroy(&priv->bfregs.reg_head.lock);
@@ -1926,7 +1923,6 @@ void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
mlx5_db_pools_cleanup(dev);
mlx5_frag_buf_pools_cleanup(dev);
debugfs_remove_recursive(dev->priv.dbg.dbg_root);
- mutex_destroy(&priv->pgdir_mutex);
mutex_destroy(&priv->alloc_mutex);
mutex_destroy(&priv->bfregs.wc_head.lock);
mutex_destroy(&priv->bfregs.reg_head.lock);
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index 4246d6d904ba..2c56bb1d676d 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -599,9 +599,6 @@ struct mlx5_priv {
struct mutex alloc_mutex;
int numa_node;
- struct mutex pgdir_mutex;
- struct list_head pgdir_list;
-
struct mlx5_frag_buf_node_pools **frag_buf_node_pools;
struct mlx5_dma_pool **db_node_pools;
/* end: alloc stuff */
@@ -808,7 +805,7 @@ struct mlx5_core_dev {
struct mlx5_db {
__be32 *db;
union {
- struct mlx5_db_pgdir *pgdir;
+ struct mlx5_dma_pool_page *pool_page;
struct mlx5_ib_user_db_page *user_page;
} u;
dma_addr_t dma;
--
2.44.0
^ permalink raw reply related [flat|nested] 4+ messages in thread* [PATCH net-next V2 3/3] net/mlx5: add debugfs stats for doorbell dma pools
2026-08-03 13:25 [PATCH net-next V2 0/3] net/mlx5: allocate doorbells from dma pools Tariq Toukan
2026-08-03 13:25 ` [PATCH net-next V2 1/3] net/mlx5: initialize doorbell " Tariq Toukan
2026-08-03 13:25 ` [PATCH net-next V2 2/3] net/mlx5: allocate doorbells from " Tariq Toukan
@ 2026-08-03 13:25 ` Tariq Toukan
2 siblings, 0 replies; 4+ messages in thread
From: Tariq Toukan @ 2026-08-03 13:25 UTC (permalink / raw)
To: Andrew Lunn, David S. Miller, Eric Dumazet, Jakub Kicinski,
netdev, Paolo Abeni
Cc: Dragos Tatulea, Gal Pressman, Leon Romanovsky, linux-kernel,
linux-rdma, Mark Bloch, Nimrod Oren, Saeed Mahameed, Tariq Toukan
From: Nimrod Oren <noren@nvidia.com>
Add a debugfs file exposing per-node DMA pool usage for doorbell
allocations.
# cat /sys/kernel/debug/mlx5/<dev>/db_dma_pools
node block_size used_blocks allocated_blocks
0 64 0 0
1 64 0 0
Signed-off-by: Nimrod Oren <noren@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
---
.../net/ethernet/mellanox/mlx5/core/alloc.c | 27 +++++++++++++++++++
include/linux/mlx5/driver.h | 1 +
2 files changed, 28 insertions(+)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/alloc.c b/drivers/net/ethernet/mellanox/mlx5/core/alloc.c
index 3975c726d35c..a92cf545bdaf 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/alloc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/alloc.c
@@ -437,11 +437,34 @@ void mlx5_frag_buf_free(struct mlx5_core_dev *dev, struct mlx5_frag_buf *buf)
}
EXPORT_SYMBOL_GPL(mlx5_frag_buf_free);
+static int mlx5_db_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);
+
+ for_each_node_state(node, N_POSSIBLE) {
+ struct mlx5_dma_pool *pool = dev->priv.db_node_pools[node];
+
+ if (!pool)
+ continue;
+
+ mlx5_dma_pool_debugfs_stats_print(file, pool);
+ }
+
+ return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(mlx5_db_dma_pools_debugfs);
+
void mlx5_db_pools_cleanup(struct mlx5_core_dev *dev)
{
struct mlx5_priv *priv = &dev->priv;
int node;
+ debugfs_remove(priv->dbg.db_dma_pools_debugfs);
+ priv->dbg.db_dma_pools_debugfs = NULL;
+
for_each_node_state(node, N_POSSIBLE)
if (priv->db_node_pools[node])
mlx5_dma_pool_destroy(priv->db_node_pools[node]);
@@ -471,6 +494,10 @@ int mlx5_db_pools_init(struct mlx5_core_dev *dev)
priv->db_node_pools[node] = pool;
}
+ priv->dbg.db_dma_pools_debugfs =
+ debugfs_create_file("db_dma_pools", 0444, priv->dbg.dbg_root,
+ dev, &mlx5_db_dma_pools_debugfs_fops);
+
return 0;
}
diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h
index 2c56bb1d676d..83d0a83bbfbc 100644
--- a/include/linux/mlx5/driver.h
+++ b/include/linux/mlx5/driver.h
@@ -548,6 +548,7 @@ struct mlx5_debugfs_entries {
struct dentry *cq_debugfs;
struct dentry *cmdif_debugfs;
struct dentry *frag_buf_dma_pools_debugfs;
+ struct dentry *db_dma_pools_debugfs;
struct dentry *pages_debugfs;
struct dentry *lag_debugfs;
};
--
2.44.0
^ permalink raw reply related [flat|nested] 4+ messages in thread