net/mlx5: DR, Allocate icm_chunks from their own slab allocator

SW steering allocates/frees lots of icm_chunk structs. To make this more
efficiently, create a separate kmem_cache and allocate these chunks from
this allocator.
By doing this we observe that the alloc/free "hiccups" frequency has
become much lower, which allows for a more steady rule insersion rate.
Signed-off-by: Yevgeny Kliteynik <kliteyn@nvidia.com>
Reviewed-by: Alex Vesker <valex@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

drivers/net/ethernet/mellanox/mlx5/core/steering/dr_domain.c

@@ -60,10 +60,19 @@ static int dr_domain_init_mem_resources(struct mlx5dr_domain *dmn)
 {
 	int ret;
 
+	dmn->chunks_kmem_cache = kmem_cache_create("mlx5_dr_chunks",
+						   sizeof(struct mlx5dr_icm_chunk), 0,
+						   SLAB_HWCACHE_ALIGN, NULL);
+	if (!dmn->chunks_kmem_cache) {
+		mlx5dr_err(dmn, "Couldn't create chunks kmem_cache\n");
+		return -ENOMEM;
+	}
+
 	dmn->ste_icm_pool = mlx5dr_icm_pool_create(dmn, DR_ICM_TYPE_STE);
 	if (!dmn->ste_icm_pool) {
 		mlx5dr_err(dmn, "Couldn't get icm memory\n");
-		return -ENOMEM;
+		ret = -ENOMEM;
+		goto free_chunks_kmem_cache;
 	}
 
 	dmn->action_icm_pool = mlx5dr_icm_pool_create(dmn, DR_ICM_TYPE_MODIFY_ACTION);
@@ -85,6 +94,9 @@ static int dr_domain_init_mem_resources(struct mlx5dr_domain *dmn)
 	mlx5dr_icm_pool_destroy(dmn->action_icm_pool);
 free_ste_icm_pool:
 	mlx5dr_icm_pool_destroy(dmn->ste_icm_pool);
+free_chunks_kmem_cache:
+	kmem_cache_destroy(dmn->chunks_kmem_cache);
+
 	return ret;
 }
 
@@ -93,6 +105,7 @@ static void dr_domain_uninit_mem_resources(struct mlx5dr_domain *dmn)
 	mlx5dr_send_info_pool_destroy(dmn);
 	mlx5dr_icm_pool_destroy(dmn->action_icm_pool);
 	mlx5dr_icm_pool_destroy(dmn->ste_icm_pool);
+	kmem_cache_destroy(dmn->chunks_kmem_cache);
 }
 
 static int dr_domain_init_resources(struct mlx5dr_domain *dmn)

drivers/net/ethernet/mellanox/mlx5/core/steering/dr_icm_pool.c

@@ -9,6 +9,8 @@ struct mlx5dr_icm_pool {
 	enum mlx5dr_icm_type icm_type;
 	enum mlx5dr_icm_chunk_size max_log_chunk_sz;
 	struct mlx5dr_domain *dmn;
+	struct kmem_cache *chunks_kmem_cache;
+
 	/* memory management */
 	struct mutex mutex; /* protect the ICM pool and ICM buddy */
 	struct list_head buddy_mem_list;
@@ -193,10 +195,13 @@ static void dr_icm_chunk_ste_init(struct mlx5dr_icm_chunk *chunk, int offset)
 
 static void dr_icm_chunk_destroy(struct mlx5dr_icm_chunk *chunk)
 {
+	struct kmem_cache *chunks_cache =
+		chunk->buddy_mem->pool->chunks_kmem_cache;
+
 	chunk->buddy_mem->used_memory -= mlx5dr_icm_pool_get_chunk_byte_size(chunk);
 	list_del(&chunk->chunk_list);
 
-	kvfree(chunk);
+	kmem_cache_free(chunks_cache, chunk);
 }
 
 static int dr_icm_buddy_init_ste_cache(struct mlx5dr_icm_buddy_mem *buddy)
@@ -302,10 +307,11 @@ dr_icm_chunk_create(struct mlx5dr_icm_pool *pool,
 		    struct mlx5dr_icm_buddy_mem *buddy_mem_pool,
 		    unsigned int seg)
 {
+	struct kmem_cache *chunks_cache = buddy_mem_pool->pool->chunks_kmem_cache;
 	struct mlx5dr_icm_chunk *chunk;
 	int offset;
 
-	chunk = kvzalloc(sizeof(*chunk), GFP_KERNEL);
+	chunk = kmem_cache_alloc(chunks_cache, GFP_KERNEL);
 	if (!chunk)
 		return NULL;
 
@@ -482,6 +488,7 @@ struct mlx5dr_icm_pool *mlx5dr_icm_pool_create(struct mlx5dr_domain *dmn,
 	pool->dmn = dmn;
 	pool->icm_type = icm_type;
 	pool->max_log_chunk_sz = max_log_chunk_sz;
+	pool->chunks_kmem_cache = dmn->chunks_kmem_cache;
 
 	INIT_LIST_HEAD(&pool->buddy_mem_list);
 

drivers/net/ethernet/mellanox/mlx5/core/steering/dr_types.h

@@ -915,6 +915,7 @@ struct mlx5dr_domain {
 	struct mlx5dr_icm_pool *action_icm_pool;
 	struct mlx5dr_send_info_pool *send_info_pool_rx;
 	struct mlx5dr_send_info_pool *send_info_pool_tx;
+	struct kmem_cache *chunks_kmem_cache;
 	struct mlx5dr_send_ring *send_ring;
 	struct mlx5dr_domain_info info;
 	struct xarray csum_fts_xa;