net/mlx5: Add hash table for flow groups in flow table

When adding a flow table entry (fte) to a flow table (ft), we first
need to find its flow group (fg). Currently, this is done by
traversing a linear list of all flow groups in the flow table.
Furthermore, since multiple flow groups which correspond to the same
fte mask may exist in the same ft, we can't just stop at the first
match. Converting the linear list to rhltable in order to speed things
up.

The last four patches increases the steering rules update rate by a
factor of more than 7 (for insertion of 50K steering rules).

Signed-off-by: Matan Barak <[email protected]>
Reviewed-by: Maor Gottlieb <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

Author: matanb10

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

@@ -158,6 +158,15 @@ static const struct rhashtable_params rhash_fte = {
 	.min_size = 1,
 };
 
+static const struct rhashtable_params rhash_fg = {
+	.key_len = FIELD_SIZEOF(struct mlx5_flow_group, mask),
+	.key_offset = offsetof(struct mlx5_flow_group, mask),
+	.head_offset = offsetof(struct mlx5_flow_group, hash),
+	.automatic_shrinking = true,
+	.min_size = 1,
+
+};
+
 static void del_rule(struct fs_node *node);
 static void del_flow_table(struct fs_node *node);
 static void del_flow_group(struct fs_node *node);
@@ -318,12 +327,22 @@ static bool check_valid_mask(u8 match_criteria_enable, const u32 *match_criteria
 	return check_last_reserved(match_criteria);
 }
 
-static bool compare_match_criteria(u8 match_criteria_enable1,
-				   u8 match_criteria_enable2,
-				   void *mask1, void *mask2)
+static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 {
-	return match_criteria_enable1 == match_criteria_enable2 &&
-		!memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
+	int i;
+
+	if (!check_valid_mask(spec->match_criteria_enable, spec->match_criteria)) {
+		pr_warn("mlx5_core: Match criteria given mismatches match_criteria_enable\n");
+		return false;
+	}
+
+	for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
+		if (spec->match_value[i] & ~spec->match_criteria[i]) {
+			pr_warn("mlx5_core: match_value differs from match_criteria\n");
+			return false;
+		}
+
+	return check_last_reserved(spec->match_value);
 }
 
 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
@@ -365,6 +384,7 @@ static void del_flow_table(struct fs_node *node)
 	if (err)
 		mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 	ida_destroy(&ft->fte_allocator);
+	rhltable_destroy(&ft->fgs_hash);
 	fs_get_obj(prio, ft->node.parent);
 	prio->num_ft--;
 }
@@ -454,6 +474,7 @@ static void del_flow_group(struct fs_node *node)
 	struct mlx5_flow_group *fg;
 	struct mlx5_flow_table *ft;
 	struct mlx5_core_dev *dev;
+	int err;
 
 	fs_get_obj(fg, node);
 	fs_get_obj(ft, fg->node.parent);
@@ -463,6 +484,10 @@ static void del_flow_group(struct fs_node *node)
 		ft->autogroup.num_groups--;
 
 	rhashtable_destroy(&fg->ftes_hash);
+	err = rhltable_remove(&ft->fgs_hash,
+			      &fg->hash,
+			      rhash_fg);
+	WARN_ON(err);
 	if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 		mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 			       fg->id, ft->id);
@@ -525,10 +550,17 @@ static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_ft
 						u32 flags)
 {
 	struct mlx5_flow_table *ft;
+	int ret;
 
 	ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 	if (!ft)
-		return NULL;
+		return ERR_PTR(-ENOMEM);
+
+	ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
+	if (ret) {
+		kfree(ft);
+		return ERR_PTR(ret);
+	}
 
 	ft->level = level;
 	ft->node.type = FS_TYPE_FLOW_TABLE;
@@ -829,8 +861,8 @@ static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespa
 			      ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
 			      root->table_type,
 			      op_mod, ft_attr->flags);
-	if (!ft) {
-		err = -ENOMEM;
+	if (IS_ERR(ft)) {
+		err = PTR_ERR(ft);
 		goto unlock_root;
 	}
 
@@ -942,10 +974,14 @@ static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *
 	if (IS_ERR(fg))
 		return fg;
 
-	err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
+	err = rhltable_insert(&ft->fgs_hash, &fg->hash, rhash_fg);
 	if (err)
 		goto err_free_fg;
 
+	err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
+	if (err)
+		goto err_remove_fg;
+
 	if (ft->autogroup.active)
 		ft->autogroup.num_groups++;
 	/* Add node to tree */
@@ -956,6 +992,10 @@ static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *
 
 	return fg;
 
+err_remove_fg:
+	WARN_ON(rhltable_remove(&ft->fgs_hash,
+				&fg->hash,
+				rhash_fg));
 err_free_fg:
 	rhashtable_destroy(&fg->ftes_hash);
 	kfree(fg);
@@ -1291,18 +1331,13 @@ static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
 					    u32 *match_value,
 					    struct mlx5_flow_act *flow_act,
 					    struct mlx5_flow_destination *dest,
-					    int dest_num)
+					    int dest_num,
+					    struct fs_fte *fte)
 {
-	u32 masked_val[sizeof(fg->mask.match_criteria)];
 	struct mlx5_flow_handle *handle;
 	struct mlx5_flow_table *ft;
-	struct fs_fte *fte;
 	int i;
 
-	nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
-	for (i = 0; i < sizeof(masked_val); i++)
-		masked_val[i] = match_value[i] & fg->mask.match_criteria[i];
-	fte = rhashtable_lookup_fast(&fg->ftes_hash, masked_val, rhash_fte);
 	if (fte) {
 		int old_action;
 		int ret;
@@ -1324,23 +1359,20 @@ static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
 		} else {
 			goto add_rules;
 		}
-		unlock_ref_node(&fte->node);
 	}
 	fs_get_obj(ft, fg->node.parent);
 
 	fte = create_fte(fg, match_value, flow_act);
-	if (IS_ERR(fte)) {
-		handle = (void *)fte;
-		goto unlock_fg;
-	}
+	if (IS_ERR(fte))
+		return (void *)fte;
 	tree_init_node(&fte->node, 0, del_fte);
 	nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
 	handle = add_rule_fte(fte, fg, dest, dest_num, false);
 	if (IS_ERR(handle)) {
 		unlock_ref_node(&fte->node);
 		destroy_fte(fte, fg);
 		kfree(fte);
-		goto unlock_fg;
+		return handle;
 	}
 
 	tree_add_node(&fte->node, &fg->node);
@@ -1353,8 +1385,6 @@ static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
 	}
 unlock_fte:
 	unlock_ref_node(&fte->node);
-unlock_fg:
-	unlock_ref_node(&fg->node);
 	return handle;
 }
 
@@ -1402,6 +1432,96 @@ static bool dest_is_valid(struct mlx5_flow_destination *dest,
 	return true;
 }
 
+static struct mlx5_flow_handle *
+try_add_to_existing_fg(struct mlx5_flow_table *ft,
+		       struct mlx5_flow_spec *spec,
+		       struct mlx5_flow_act *flow_act,
+		       struct mlx5_flow_destination *dest,
+		       int dest_num)
+{
+	struct mlx5_flow_group *g;
+	struct mlx5_flow_handle *rule = ERR_PTR(-ENOENT);
+	struct rhlist_head *tmp, *list;
+	struct match_list {
+		struct list_head	list;
+		struct mlx5_flow_group *g;
+	} match_list, *iter;
+	LIST_HEAD(match_head);
+
+	rcu_read_lock();
+	/* Collect all fgs which has a matching match_criteria */
+	list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
+	rhl_for_each_entry_rcu(g, tmp, list, hash) {
+		struct match_list *curr_match;
+
+		if (likely(list_empty(&match_head))) {
+			match_list.g = g;
+			list_add_tail(&match_list.list, &match_head);
+			continue;
+		}
+		curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
+
+		if (!curr_match) {
+			rcu_read_unlock();
+			rule = ERR_PTR(-ENOMEM);
+			goto free_list;
+		}
+		curr_match->g = g;
+		list_add_tail(&curr_match->list, &match_head);
+	}
+	rcu_read_unlock();
+
+	/* Try to find a fg that already contains a matching fte */
+	list_for_each_entry(iter, &match_head, list) {
+		struct fs_fte *fte;
+
+		g = iter->g;
+		nested_lock_ref_node(&g->node, FS_MUTEX_PARENT);
+		fte = rhashtable_lookup_fast(&g->ftes_hash, spec->match_value,
+					     rhash_fte);
+		if (fte) {
+			rule = add_rule_fg(g, spec->match_value,
+					   flow_act, dest, dest_num, fte);
+			unlock_ref_node(&g->node);
+			goto free_list;
+		}
+		unlock_ref_node(&g->node);
+	}
+
+	/* No group with matching fte found. Try to add a new fte to any
+	 * matching fg.
+	 */
+	list_for_each_entry(iter, &match_head, list) {
+		g = iter->g;
+
+		nested_lock_ref_node(&g->node, FS_MUTEX_PARENT);
+		rule = add_rule_fg(g, spec->match_value,
+				   flow_act, dest, dest_num, NULL);
+		if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC) {
+			unlock_ref_node(&g->node);
+			goto free_list;
+		}
+		unlock_ref_node(&g->node);
+	}
+
+free_list:
+	if (!list_empty(&match_head)) {
+		struct match_list *match_tmp;
+
+		/* The most common case is having one FG. Since we want to
+		 * optimize this case, we save the first on the stack.
+		 * Therefore, no need to free it.
+		 */
+		list_del(&list_first_entry(&match_head, typeof(*iter), list)->list);
+		list_for_each_entry_safe(iter, match_tmp, &match_head, list) {
+			list_del(&iter->list);
+			kfree(iter);
+		}
+	}
+
+	return rule;
+}
+
 static struct mlx5_flow_handle *
 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
 		     struct mlx5_flow_spec *spec,
@@ -1414,8 +1534,7 @@ _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
 	struct mlx5_flow_handle *rule;
 	int i;
 
-	if (!check_valid_mask(spec->match_criteria_enable,
-			      spec->match_criteria))
+	if (!check_valid_spec(spec))
 		return ERR_PTR(-EINVAL);
 
 	for (i = 0; i < dest_num; i++) {
@@ -1424,16 +1543,9 @@ _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
 	}
 
 	nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
-	fs_for_each_fg(g, ft)
-		if (compare_match_criteria(g->mask.match_criteria_enable,
-					   spec->match_criteria_enable,
-					   g->mask.match_criteria,
-					   spec->match_criteria)) {
-			rule = add_rule_fg(g, spec->match_value,
-					   flow_act, dest, dest_num);
-			if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
-				goto unlock;
-		}
+	rule = try_add_to_existing_fg(ft, spec, flow_act, dest, dest_num);
+	if (!IS_ERR(rule))
+		goto unlock;
 
 	g = create_autogroup(ft, spec->match_criteria_enable,
 			     spec->match_criteria);
@@ -1442,7 +1554,8 @@ _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
 		goto unlock;
 	}
 
-	rule = add_rule_fg(g, spec->match_value, flow_act, dest, dest_num);
+	rule = add_rule_fg(g, spec->match_value, flow_act, dest,
+			   dest_num, NULL);
 	if (IS_ERR(rule)) {
 		/* Remove assumes refcount > 0 and autogroup creates a group
 		 * with a refcount = 0.

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

@@ -120,6 +120,7 @@ struct mlx5_flow_table {
 	struct list_head		fwd_rules;
 	u32				flags;
 	struct ida			fte_allocator;
+	struct rhltable			fgs_hash;
 };
 
 struct mlx5_fc_cache {
@@ -200,6 +201,7 @@ struct mlx5_flow_group {
 	u32				max_ftes;
 	u32				id;
 	struct rhashtable		ftes_hash;
+	struct rhlist_head		hash;
 };
 
 struct mlx5_flow_root_namespace {