net/mlx5: fs_core: Introduce unmanaged flow tables

Currently, Most of the steering tree is statically declared ahead of time,
with steering prios instances allocated for each fdb chain to assign max
number of levels for each of them. This allows fs_core to manage the
connections and  levels of the flow tables hierarcy to prevent loops, but
restricts us with the number of supported chains and priorities.

Introduce unmananged flow tables, allowing the user to manage the flow
table connections. A unamanged table is detached from the fs_core flow
table hierarcy, and is only connected back to the hierarchy by explicit
FTEs forward actions.

This will be used together with firmware that supports ignoring the flow
table levels to increase the number of supported chains and prios.

Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Reviewed-by: Oz Shlomo <ozsh@mellanox.com>
Reviewed-by: Mark Bloch <markb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

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

@@ -1006,7 +1006,8 @@ static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespa
 							u16 vport)
 {
 	struct mlx5_flow_root_namespace *root = find_root(&ns->node);
-	struct mlx5_flow_table *next_ft = NULL;
+	bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED;
+	struct mlx5_flow_table *next_ft;
 	struct fs_prio *fs_prio = NULL;
 	struct mlx5_flow_table *ft;
 	int log_table_sz;
@@ -1023,14 +1024,21 @@ static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespa
 		err = -EINVAL;
 		goto unlock_root;
 	}
-	if (ft_attr->level >= fs_prio->num_levels) {
-		err = -ENOSPC;
-		goto unlock_root;
+	if (!unmanaged) {
+		/* The level is related to the
+		 * priority level range.
+		 */
+		if (ft_attr->level >= fs_prio->num_levels) {
+			err = -ENOSPC;
+			goto unlock_root;
+		}
+
+		ft_attr->level += fs_prio->start_level;
 	}
+
 	/* The level is related to the
 	 * priority level range.
 	 */
-	ft_attr->level += fs_prio->start_level;
 	ft = alloc_flow_table(ft_attr->level,
 			      vport,
 			      ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
@@ -1043,19 +1051,27 @@ static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespa
 
 	tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
 	log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
-	next_ft = find_next_chained_ft(fs_prio);
+	next_ft = unmanaged ? ft_attr->next_ft :
+			      find_next_chained_ft(fs_prio);
 	ft->def_miss_action = ns->def_miss_action;
 	err = root->cmds->create_flow_table(root, ft, log_table_sz, next_ft);
 	if (err)
 		goto free_ft;
 
-	err = connect_flow_table(root->dev, ft, fs_prio);
-	if (err)
-		goto destroy_ft;
+	if (!unmanaged) {
+		err = connect_flow_table(root->dev, ft, fs_prio);
+		if (err)
+			goto destroy_ft;
+	}
+
 	ft->node.active = true;
 	down_write_ref_node(&fs_prio->node, false);
-	tree_add_node(&ft->node, &fs_prio->node);
-	list_add_flow_table(ft, fs_prio);
+	if (!unmanaged) {
+		tree_add_node(&ft->node, &fs_prio->node);
+		list_add_flow_table(ft, fs_prio);
+	} else {
+		ft->node.root = fs_prio->node.root;
+	}
 	fs_prio->num_ft++;
 	up_write_ref_node(&fs_prio->node, false);
 	mutex_unlock(&root->chain_lock);
@@ -2024,7 +2040,8 @@ int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
 	int err = 0;
 
 	mutex_lock(&root->chain_lock);
-	err = disconnect_flow_table(ft);
+	if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
+		err = disconnect_flow_table(ft);
 	if (err) {
 		mutex_unlock(&root->chain_lock);
 		return err;

include/linux/mlx5/fs.h

@@ -48,6 +48,7 @@ enum {
 	MLX5_FLOW_TABLE_TUNNEL_EN_REFORMAT = BIT(0),
 	MLX5_FLOW_TABLE_TUNNEL_EN_DECAP = BIT(1),
 	MLX5_FLOW_TABLE_TERMINATION = BIT(2),
+	MLX5_FLOW_TABLE_UNMANAGED = BIT(3),
 };
 
 #define LEFTOVERS_RULE_NUM	 2
@@ -150,6 +151,7 @@ struct mlx5_flow_table_attr {
 	int max_fte;
 	u32 level;
 	u32 flags;
+	struct mlx5_flow_table *next_ft;
 
 	struct {
 		int max_num_groups;