2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/mutex.h>
34 #include <linux/mlx5/driver.h>
36 #include "mlx5_core.h"
40 #define INIT_TREE_NODE_ARRAY_SIZE(...) (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
41 sizeof(struct init_tree_node))
43 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
44 ...) {.type = FS_TYPE_PRIO,\
45 .min_ft_level = min_level_val,\
46 .num_levels = num_levels_val,\
47 .num_leaf_prios = num_prios_val,\
49 .children = (struct init_tree_node[]) {__VA_ARGS__},\
50 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
53 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
54 ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
57 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
58 .children = (struct init_tree_node[]) {__VA_ARGS__},\
59 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
62 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
65 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
67 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
68 .caps = (long[]) {__VA_ARGS__} }
70 #define FS_CHAINING_CAPS FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
71 FS_CAP(flow_table_properties_nic_receive.modify_root), \
72 FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
73 FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
75 #define LEFTOVERS_NUM_LEVELS 1
76 #define LEFTOVERS_NUM_PRIOS 1
78 #define BY_PASS_PRIO_NUM_LEVELS 1
79 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
82 #define ETHTOOL_PRIO_NUM_LEVELS 1
83 #define ETHTOOL_NUM_PRIOS 11
84 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
85 /* Vlan, mac, ttc, aRFS */
86 #define KERNEL_NIC_PRIO_NUM_LEVELS 4
87 #define KERNEL_NIC_NUM_PRIOS 1
88 /* One more level for tc */
89 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
91 #define ANCHOR_NUM_LEVELS 1
92 #define ANCHOR_NUM_PRIOS 1
93 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
95 #define OFFLOADS_MAX_FT 1
96 #define OFFLOADS_NUM_PRIOS 1
97 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
99 #define LAG_PRIO_NUM_LEVELS 1
100 #define LAG_NUM_PRIOS 1
101 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
107 static struct init_tree_node {
108 enum fs_node_type type;
109 struct init_tree_node *children;
111 struct node_caps caps;
117 .type = FS_TYPE_NAMESPACE,
119 .children = (struct init_tree_node[]) {
120 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
122 ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
123 BY_PASS_PRIO_NUM_LEVELS))),
124 ADD_PRIO(0, LAG_MIN_LEVEL, 0,
126 ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
127 LAG_PRIO_NUM_LEVELS))),
128 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
129 ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
130 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
132 ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
133 ETHTOOL_PRIO_NUM_LEVELS))),
134 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
135 ADD_NS(ADD_MULTIPLE_PRIO(1, 1),
136 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
137 KERNEL_NIC_PRIO_NUM_LEVELS))),
138 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
140 ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
141 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
142 ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
146 enum fs_i_mutex_lock_class {
147 FS_MUTEX_GRANDPARENT,
152 static void del_rule(struct fs_node *node);
153 static void del_flow_table(struct fs_node *node);
154 static void del_flow_group(struct fs_node *node);
155 static void del_fte(struct fs_node *node);
156 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
157 struct mlx5_flow_destination *d2);
158 static struct mlx5_flow_rule *
159 find_flow_rule(struct fs_fte *fte,
160 struct mlx5_flow_destination *dest);
162 static void tree_init_node(struct fs_node *node,
163 unsigned int refcount,
164 void (*remove_func)(struct fs_node *))
166 atomic_set(&node->refcount, refcount);
167 INIT_LIST_HEAD(&node->list);
168 INIT_LIST_HEAD(&node->children);
169 mutex_init(&node->lock);
170 node->remove_func = remove_func;
173 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
176 atomic_inc(&parent->refcount);
177 node->parent = parent;
179 /* Parent is the root */
183 node->root = parent->root;
186 static void tree_get_node(struct fs_node *node)
188 atomic_inc(&node->refcount);
191 static void nested_lock_ref_node(struct fs_node *node,
192 enum fs_i_mutex_lock_class class)
195 mutex_lock_nested(&node->lock, class);
196 atomic_inc(&node->refcount);
200 static void lock_ref_node(struct fs_node *node)
203 mutex_lock(&node->lock);
204 atomic_inc(&node->refcount);
208 static void unlock_ref_node(struct fs_node *node)
211 atomic_dec(&node->refcount);
212 mutex_unlock(&node->lock);
216 static void tree_put_node(struct fs_node *node)
218 struct fs_node *parent_node = node->parent;
220 lock_ref_node(parent_node);
221 if (atomic_dec_and_test(&node->refcount)) {
223 list_del_init(&node->list);
224 if (node->remove_func)
225 node->remove_func(node);
229 unlock_ref_node(parent_node);
230 if (!node && parent_node)
231 tree_put_node(parent_node);
234 static int tree_remove_node(struct fs_node *node)
236 if (atomic_read(&node->refcount) > 1) {
237 atomic_dec(&node->refcount);
244 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
247 struct fs_prio *iter_prio;
249 fs_for_each_prio(iter_prio, ns) {
250 if (iter_prio->prio == prio)
257 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
261 for (i = 0; i < size; i++, mask++, val1++, val2++)
262 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
263 ((*(u8 *)val2) & (*(u8 *)mask)))
269 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
270 void *fte_param1, void *fte_param2)
272 if (mask->match_criteria_enable &
273 1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
274 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
275 fte_param1, outer_headers);
276 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
277 fte_param2, outer_headers);
278 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
279 mask->match_criteria, outer_headers);
281 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
282 MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
286 if (mask->match_criteria_enable &
287 1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
288 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
289 fte_param1, misc_parameters);
290 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
291 fte_param2, misc_parameters);
292 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
293 mask->match_criteria, misc_parameters);
295 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
296 MLX5_ST_SZ_BYTES(fte_match_set_misc)))
300 if (mask->match_criteria_enable &
301 1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
302 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
303 fte_param1, inner_headers);
304 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
305 fte_param2, inner_headers);
306 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
307 mask->match_criteria, inner_headers);
309 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
310 MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
316 static bool compare_match_criteria(u8 match_criteria_enable1,
317 u8 match_criteria_enable2,
318 void *mask1, void *mask2)
320 return match_criteria_enable1 == match_criteria_enable2 &&
321 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
324 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
326 struct fs_node *root;
327 struct mlx5_flow_namespace *ns;
331 if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
332 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
336 ns = container_of(root, struct mlx5_flow_namespace, node);
337 return container_of(ns, struct mlx5_flow_root_namespace, ns);
340 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
342 struct mlx5_flow_root_namespace *root = find_root(node);
349 static void del_flow_table(struct fs_node *node)
351 struct mlx5_flow_table *ft;
352 struct mlx5_core_dev *dev;
353 struct fs_prio *prio;
356 fs_get_obj(ft, node);
357 dev = get_dev(&ft->node);
359 err = mlx5_cmd_destroy_flow_table(dev, ft);
361 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
362 fs_get_obj(prio, ft->node.parent);
366 static void del_rule(struct fs_node *node)
368 struct mlx5_flow_rule *rule;
369 struct mlx5_flow_table *ft;
370 struct mlx5_flow_group *fg;
374 struct mlx5_core_dev *dev = get_dev(node);
375 int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
377 bool update_fte = false;
379 match_value = mlx5_vzalloc(match_len);
381 mlx5_core_warn(dev, "failed to allocate inbox\n");
385 fs_get_obj(rule, node);
386 fs_get_obj(fte, rule->node.parent);
387 fs_get_obj(fg, fte->node.parent);
388 memcpy(match_value, fte->val, sizeof(fte->val));
389 fs_get_obj(ft, fg->node.parent);
390 list_del(&rule->node.list);
391 if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
392 mutex_lock(&rule->dest_attr.ft->lock);
393 list_del(&rule->next_ft);
394 mutex_unlock(&rule->dest_attr.ft->lock);
397 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER &&
399 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
400 fte->action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
405 if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
407 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST),
411 if (update_fte && fte->dests_size) {
412 err = mlx5_cmd_update_fte(dev, ft, fg->id, modify_mask, fte);
415 "%s can't del rule fg id=%d fte_index=%d\n",
416 __func__, fg->id, fte->index);
421 static void del_fte(struct fs_node *node)
423 struct mlx5_flow_table *ft;
424 struct mlx5_flow_group *fg;
425 struct mlx5_core_dev *dev;
429 fs_get_obj(fte, node);
430 fs_get_obj(fg, fte->node.parent);
431 fs_get_obj(ft, fg->node.parent);
433 dev = get_dev(&ft->node);
434 err = mlx5_cmd_delete_fte(dev, ft,
438 "flow steering can't delete fte in index %d of flow group id %d\n",
445 static void del_flow_group(struct fs_node *node)
447 struct mlx5_flow_group *fg;
448 struct mlx5_flow_table *ft;
449 struct mlx5_core_dev *dev;
451 fs_get_obj(fg, node);
452 fs_get_obj(ft, fg->node.parent);
453 dev = get_dev(&ft->node);
455 if (ft->autogroup.active)
456 ft->autogroup.num_groups--;
458 if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
459 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
463 static struct fs_fte *alloc_fte(struct mlx5_flow_act *flow_act,
469 fte = kzalloc(sizeof(*fte), GFP_KERNEL);
471 return ERR_PTR(-ENOMEM);
473 memcpy(fte->val, match_value, sizeof(fte->val));
474 fte->node.type = FS_TYPE_FLOW_ENTRY;
475 fte->flow_tag = flow_act->flow_tag;
477 fte->action = flow_act->action;
478 fte->encap_id = flow_act->encap_id;
479 fte->modify_id = flow_act->modify_id;
484 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
486 struct mlx5_flow_group *fg;
487 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
488 create_fg_in, match_criteria);
489 u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
491 match_criteria_enable);
492 fg = kzalloc(sizeof(*fg), GFP_KERNEL);
494 return ERR_PTR(-ENOMEM);
496 fg->mask.match_criteria_enable = match_criteria_enable;
497 memcpy(&fg->mask.match_criteria, match_criteria,
498 sizeof(fg->mask.match_criteria));
499 fg->node.type = FS_TYPE_FLOW_GROUP;
500 fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
502 fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
503 end_flow_index) - fg->start_index + 1;
507 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
508 enum fs_flow_table_type table_type,
509 enum fs_flow_table_op_mod op_mod,
512 struct mlx5_flow_table *ft;
514 ft = kzalloc(sizeof(*ft), GFP_KERNEL);
519 ft->node.type = FS_TYPE_FLOW_TABLE;
521 ft->type = table_type;
523 ft->max_fte = max_fte;
525 INIT_LIST_HEAD(&ft->fwd_rules);
526 mutex_init(&ft->lock);
531 /* If reverse is false, then we search for the first flow table in the
532 * root sub-tree from start(closest from right), else we search for the
533 * last flow table in the root sub-tree till start(closest from left).
535 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node *root,
536 struct list_head *start,
539 #define list_advance_entry(pos, reverse) \
540 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
542 #define list_for_each_advance_continue(pos, head, reverse) \
543 for (pos = list_advance_entry(pos, reverse); \
544 &pos->list != (head); \
545 pos = list_advance_entry(pos, reverse))
547 struct fs_node *iter = list_entry(start, struct fs_node, list);
548 struct mlx5_flow_table *ft = NULL;
553 list_for_each_advance_continue(iter, &root->children, reverse) {
554 if (iter->type == FS_TYPE_FLOW_TABLE) {
555 fs_get_obj(ft, iter);
558 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
566 /* If reverse if false then return the first flow table in next priority of
567 * prio in the tree, else return the last flow table in the previous priority
568 * of prio in the tree.
570 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
572 struct mlx5_flow_table *ft = NULL;
573 struct fs_node *curr_node;
574 struct fs_node *parent;
576 parent = prio->node.parent;
577 curr_node = &prio->node;
578 while (!ft && parent) {
579 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
581 parent = curr_node->parent;
586 /* Assuming all the tree is locked by mutex chain lock */
587 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
589 return find_closest_ft(prio, false);
592 /* Assuming all the tree is locked by mutex chain lock */
593 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
595 return find_closest_ft(prio, true);
598 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
599 struct fs_prio *prio,
600 struct mlx5_flow_table *ft)
602 struct mlx5_flow_table *iter;
606 fs_for_each_ft(iter, prio) {
608 err = mlx5_cmd_modify_flow_table(dev,
612 mlx5_core_warn(dev, "Failed to modify flow table %d\n",
614 /* The driver is out of sync with the FW */
623 /* Connect flow tables from previous priority of prio to ft */
624 static int connect_prev_fts(struct mlx5_core_dev *dev,
625 struct mlx5_flow_table *ft,
626 struct fs_prio *prio)
628 struct mlx5_flow_table *prev_ft;
630 prev_ft = find_prev_chained_ft(prio);
632 struct fs_prio *prev_prio;
634 fs_get_obj(prev_prio, prev_ft->node.parent);
635 return connect_fts_in_prio(dev, prev_prio, ft);
640 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
643 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
644 int min_level = INT_MAX;
648 min_level = root->root_ft->level;
650 if (ft->level >= min_level)
653 err = mlx5_cmd_update_root_ft(root->dev, ft);
655 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
663 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
664 struct mlx5_flow_destination *dest)
666 struct mlx5_flow_table *ft;
667 struct mlx5_flow_group *fg;
669 int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
672 fs_get_obj(fte, rule->node.parent);
673 if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
675 lock_ref_node(&fte->node);
676 fs_get_obj(fg, fte->node.parent);
677 fs_get_obj(ft, fg->node.parent);
679 memcpy(&rule->dest_attr, dest, sizeof(*dest));
680 err = mlx5_cmd_update_fte(get_dev(&ft->node),
684 unlock_ref_node(&fte->node);
689 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
690 struct mlx5_flow_destination *new_dest,
691 struct mlx5_flow_destination *old_dest)
696 if (handle->num_rules != 1)
698 return _mlx5_modify_rule_destination(handle->rule[0],
702 for (i = 0; i < handle->num_rules; i++) {
703 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
704 return _mlx5_modify_rule_destination(handle->rule[i],
711 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */
712 static int connect_fwd_rules(struct mlx5_core_dev *dev,
713 struct mlx5_flow_table *new_next_ft,
714 struct mlx5_flow_table *old_next_ft)
716 struct mlx5_flow_destination dest;
717 struct mlx5_flow_rule *iter;
720 /* new_next_ft and old_next_ft could be NULL only
721 * when we create/destroy the anchor flow table.
723 if (!new_next_ft || !old_next_ft)
726 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
727 dest.ft = new_next_ft;
729 mutex_lock(&old_next_ft->lock);
730 list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
731 mutex_unlock(&old_next_ft->lock);
732 list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
733 err = _mlx5_modify_rule_destination(iter, &dest);
735 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
741 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
742 struct fs_prio *prio)
744 struct mlx5_flow_table *next_ft;
747 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
749 if (list_empty(&prio->node.children)) {
750 err = connect_prev_fts(dev, ft, prio);
754 next_ft = find_next_chained_ft(prio);
755 err = connect_fwd_rules(dev, ft, next_ft);
760 if (MLX5_CAP_FLOWTABLE(dev,
761 flow_table_properties_nic_receive.modify_root))
762 err = update_root_ft_create(ft, prio);
766 static void list_add_flow_table(struct mlx5_flow_table *ft,
767 struct fs_prio *prio)
769 struct list_head *prev = &prio->node.children;
770 struct mlx5_flow_table *iter;
772 fs_for_each_ft(iter, prio) {
773 if (iter->level > ft->level)
775 prev = &iter->node.list;
777 list_add(&ft->node.list, prev);
780 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
781 struct mlx5_flow_table_attr *ft_attr,
782 enum fs_flow_table_op_mod op_mod,
785 struct mlx5_flow_root_namespace *root = find_root(&ns->node);
786 struct mlx5_flow_table *next_ft = NULL;
787 struct fs_prio *fs_prio = NULL;
788 struct mlx5_flow_table *ft;
793 pr_err("mlx5: flow steering failed to find root of namespace\n");
794 return ERR_PTR(-ENODEV);
797 mutex_lock(&root->chain_lock);
798 fs_prio = find_prio(ns, ft_attr->prio);
803 if (ft_attr->level >= fs_prio->num_levels) {
807 /* The level is related to the
808 * priority level range.
810 ft_attr->level += fs_prio->start_level;
811 ft = alloc_flow_table(ft_attr->level,
813 ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
815 op_mod, ft_attr->flags);
821 ft->underlay_qpn = ft_attr->underlay_qpn;
823 tree_init_node(&ft->node, 1, del_flow_table);
824 log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
825 next_ft = find_next_chained_ft(fs_prio);
826 err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->op_mod, ft->type,
827 ft->level, log_table_sz, next_ft, &ft->id,
832 err = connect_flow_table(root->dev, ft, fs_prio);
835 lock_ref_node(&fs_prio->node);
836 tree_add_node(&ft->node, &fs_prio->node);
837 list_add_flow_table(ft, fs_prio);
839 unlock_ref_node(&fs_prio->node);
840 mutex_unlock(&root->chain_lock);
843 mlx5_cmd_destroy_flow_table(root->dev, ft);
847 mutex_unlock(&root->chain_lock);
851 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
852 struct mlx5_flow_table_attr *ft_attr)
854 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
857 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
858 int prio, int max_fte,
859 u32 level, u16 vport)
861 struct mlx5_flow_table_attr ft_attr = {};
863 ft_attr.max_fte = max_fte;
864 ft_attr.level = level;
867 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, 0);
870 struct mlx5_flow_table*
871 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
874 struct mlx5_flow_table_attr ft_attr = {};
876 ft_attr.level = level;
878 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
880 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
882 struct mlx5_flow_table*
883 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
885 int num_flow_table_entries,
890 struct mlx5_flow_table_attr ft_attr = {};
891 struct mlx5_flow_table *ft;
893 if (max_num_groups > num_flow_table_entries)
894 return ERR_PTR(-EINVAL);
896 ft_attr.max_fte = num_flow_table_entries;
898 ft_attr.level = level;
899 ft_attr.flags = flags;
901 ft = mlx5_create_flow_table(ns, &ft_attr);
905 ft->autogroup.active = true;
906 ft->autogroup.required_groups = max_num_groups;
910 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
912 /* Flow table should be locked */
913 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
919 struct mlx5_flow_group *fg;
920 struct mlx5_core_dev *dev = get_dev(&ft->node);
924 return ERR_PTR(-ENODEV);
926 fg = alloc_flow_group(fg_in);
930 err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
936 if (ft->autogroup.active)
937 ft->autogroup.num_groups++;
938 /* Add node to tree */
939 tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
940 tree_add_node(&fg->node, &ft->node);
941 /* Add node to group list */
942 list_add(&fg->node.list, prev_fg);
947 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
950 struct mlx5_flow_group *fg;
952 if (ft->autogroup.active)
953 return ERR_PTR(-EPERM);
955 lock_ref_node(&ft->node);
956 fg = create_flow_group_common(ft, fg_in, ft->node.children.prev, false);
957 unlock_ref_node(&ft->node);
962 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
964 struct mlx5_flow_rule *rule;
966 rule = kzalloc(sizeof(*rule), GFP_KERNEL);
970 INIT_LIST_HEAD(&rule->next_ft);
971 rule->node.type = FS_TYPE_FLOW_DEST;
973 memcpy(&rule->dest_attr, dest, sizeof(*dest));
978 static struct mlx5_flow_handle *alloc_handle(int num_rules)
980 struct mlx5_flow_handle *handle;
982 handle = kzalloc(sizeof(*handle) + sizeof(handle->rule[0]) *
983 num_rules, GFP_KERNEL);
987 handle->num_rules = num_rules;
992 static void destroy_flow_handle(struct fs_fte *fte,
993 struct mlx5_flow_handle *handle,
994 struct mlx5_flow_destination *dest,
998 if (atomic_dec_and_test(&handle->rule[i]->node.refcount)) {
1000 list_del(&handle->rule[i]->node.list);
1001 kfree(handle->rule[i]);
1007 static struct mlx5_flow_handle *
1008 create_flow_handle(struct fs_fte *fte,
1009 struct mlx5_flow_destination *dest,
1014 struct mlx5_flow_handle *handle;
1015 struct mlx5_flow_rule *rule = NULL;
1016 static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1017 static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1021 handle = alloc_handle((dest_num) ? dest_num : 1);
1023 return ERR_PTR(-ENOMEM);
1027 rule = find_flow_rule(fte, dest + i);
1029 atomic_inc(&rule->node.refcount);
1035 rule = alloc_rule(dest + i);
1039 /* Add dest to dests list- we need flow tables to be in the
1040 * end of the list for forward to next prio rules.
1042 tree_init_node(&rule->node, 1, del_rule);
1044 dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1045 list_add(&rule->node.list, &fte->node.children);
1047 list_add_tail(&rule->node.list, &fte->node.children);
1051 type = dest[i].type ==
1052 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1053 *modify_mask |= type ? count : dst;
1056 handle->rule[i] = rule;
1057 } while (++i < dest_num);
1062 destroy_flow_handle(fte, handle, dest, i);
1063 return ERR_PTR(-ENOMEM);
1066 /* fte should not be deleted while calling this function */
1067 static struct mlx5_flow_handle *
1068 add_rule_fte(struct fs_fte *fte,
1069 struct mlx5_flow_group *fg,
1070 struct mlx5_flow_destination *dest,
1074 struct mlx5_flow_handle *handle;
1075 struct mlx5_flow_table *ft;
1076 int modify_mask = 0;
1078 bool new_rule = false;
1080 handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1082 if (IS_ERR(handle) || !new_rule)
1086 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1088 fs_get_obj(ft, fg->node.parent);
1089 if (!(fte->status & FS_FTE_STATUS_EXISTING))
1090 err = mlx5_cmd_create_fte(get_dev(&ft->node),
1093 err = mlx5_cmd_update_fte(get_dev(&ft->node),
1094 ft, fg->id, modify_mask, fte);
1098 fte->status |= FS_FTE_STATUS_EXISTING;
1104 destroy_flow_handle(fte, handle, dest, handle->num_rules);
1105 return ERR_PTR(err);
1108 /* Assumed fg is locked */
1109 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
1110 struct list_head **prev)
1113 unsigned int start = fg->start_index;
1116 *prev = &fg->node.children;
1118 /* assumed list is sorted by index */
1119 fs_for_each_fte(fte, fg) {
1120 if (fte->index != start)
1124 *prev = &fte->node.list;
1130 /* prev is output, prev->next = new_fte */
1131 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
1133 struct mlx5_flow_act *flow_act,
1134 struct list_head **prev)
1139 index = get_free_fte_index(fg, prev);
1140 fte = alloc_fte(flow_act, match_value, index);
1147 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
1148 u8 match_criteria_enable,
1149 u32 *match_criteria)
1151 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1152 struct list_head *prev = &ft->node.children;
1153 unsigned int candidate_index = 0;
1154 struct mlx5_flow_group *fg;
1155 void *match_criteria_addr;
1156 unsigned int group_size = 0;
1159 if (!ft->autogroup.active)
1160 return ERR_PTR(-ENOENT);
1162 in = mlx5_vzalloc(inlen);
1164 return ERR_PTR(-ENOMEM);
1166 if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1167 /* We save place for flow groups in addition to max types */
1168 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1170 /* ft->max_fte == ft->autogroup.max_types */
1171 if (group_size == 0)
1174 /* sorted by start_index */
1175 fs_for_each_fg(fg, ft) {
1176 if (candidate_index + group_size > fg->start_index)
1177 candidate_index = fg->start_index + fg->max_ftes;
1180 prev = &fg->node.list;
1183 if (candidate_index + group_size > ft->max_fte) {
1184 fg = ERR_PTR(-ENOSPC);
1188 MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1189 match_criteria_enable);
1190 MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1191 MLX5_SET(create_flow_group_in, in, end_flow_index, candidate_index +
1193 match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1194 in, match_criteria);
1195 memcpy(match_criteria_addr, match_criteria,
1196 MLX5_ST_SZ_BYTES(fte_match_param));
1198 fg = create_flow_group_common(ft, in, prev, true);
1204 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1205 struct mlx5_flow_destination *d2)
1207 if (d1->type == d2->type) {
1208 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1209 d1->vport_num == d2->vport_num) ||
1210 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1211 d1->ft == d2->ft) ||
1212 (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1213 d1->tir_num == d2->tir_num))
1220 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1221 struct mlx5_flow_destination *dest)
1223 struct mlx5_flow_rule *rule;
1225 list_for_each_entry(rule, &fte->node.children, node.list) {
1226 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1232 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1234 struct mlx5_flow_act *flow_act,
1235 struct mlx5_flow_destination *dest,
1238 struct mlx5_flow_handle *handle;
1239 struct mlx5_flow_table *ft;
1240 struct list_head *prev;
1244 nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1245 fs_for_each_fte(fte, fg) {
1246 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1247 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1248 (flow_act->action & fte->action)) {
1249 int old_action = fte->action;
1251 if (fte->flow_tag != flow_act->flow_tag) {
1252 mlx5_core_warn(get_dev(&fte->node),
1253 "FTE flow tag %u already exists with different flow tag %u\n",
1255 flow_act->flow_tag);
1256 handle = ERR_PTR(-EEXIST);
1260 fte->action |= flow_act->action;
1261 handle = add_rule_fte(fte, fg, dest, dest_num,
1262 old_action != flow_act->action);
1263 if (IS_ERR(handle)) {
1264 fte->action = old_action;
1270 unlock_ref_node(&fte->node);
1272 fs_get_obj(ft, fg->node.parent);
1273 if (fg->num_ftes >= fg->max_ftes) {
1274 handle = ERR_PTR(-ENOSPC);
1278 fte = create_fte(fg, match_value, flow_act, &prev);
1280 handle = (void *)fte;
1283 tree_init_node(&fte->node, 0, del_fte);
1284 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1285 handle = add_rule_fte(fte, fg, dest, dest_num, false);
1286 if (IS_ERR(handle)) {
1287 unlock_ref_node(&fte->node);
1294 tree_add_node(&fte->node, &fg->node);
1295 list_add(&fte->node.list, prev);
1297 for (i = 0; i < handle->num_rules; i++) {
1298 if (atomic_read(&handle->rule[i]->node.refcount) == 1)
1299 tree_add_node(&handle->rule[i]->node, &fte->node);
1302 unlock_ref_node(&fte->node);
1304 unlock_ref_node(&fg->node);
1308 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_handle *handle)
1310 struct mlx5_flow_rule *dst;
1313 fs_get_obj(fte, handle->rule[0]->node.parent);
1315 fs_for_each_dst(dst, fte) {
1316 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1317 return dst->dest_attr.counter;
1323 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1325 if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1331 return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1332 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1335 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1337 struct mlx5_flow_table *ft)
1339 if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1340 return counter_is_valid(dest->counter, action);
1342 if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1345 if (!dest || ((dest->type ==
1346 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1347 (dest->ft->level <= ft->level)))
1352 static struct mlx5_flow_handle *
1353 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1354 struct mlx5_flow_spec *spec,
1355 struct mlx5_flow_act *flow_act,
1356 struct mlx5_flow_destination *dest,
1360 struct mlx5_flow_group *g;
1361 struct mlx5_flow_handle *rule;
1364 for (i = 0; i < dest_num; i++) {
1365 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1366 return ERR_PTR(-EINVAL);
1369 nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1370 fs_for_each_fg(g, ft)
1371 if (compare_match_criteria(g->mask.match_criteria_enable,
1372 spec->match_criteria_enable,
1373 g->mask.match_criteria,
1374 spec->match_criteria)) {
1375 rule = add_rule_fg(g, spec->match_value,
1376 flow_act, dest, dest_num);
1377 if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1381 g = create_autogroup(ft, spec->match_criteria_enable,
1382 spec->match_criteria);
1388 rule = add_rule_fg(g, spec->match_value, flow_act, dest, dest_num);
1390 /* Remove assumes refcount > 0 and autogroup creates a group
1391 * with a refcount = 0.
1393 unlock_ref_node(&ft->node);
1394 tree_get_node(&g->node);
1395 tree_remove_node(&g->node);
1399 unlock_ref_node(&ft->node);
1403 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1405 return ((ft->type == FS_FT_NIC_RX) &&
1406 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1409 struct mlx5_flow_handle *
1410 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1411 struct mlx5_flow_spec *spec,
1412 struct mlx5_flow_act *flow_act,
1413 struct mlx5_flow_destination *dest,
1416 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1417 struct mlx5_flow_destination gen_dest;
1418 struct mlx5_flow_table *next_ft = NULL;
1419 struct mlx5_flow_handle *handle = NULL;
1420 u32 sw_action = flow_act->action;
1421 struct fs_prio *prio;
1423 fs_get_obj(prio, ft->node.parent);
1424 if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1425 if (!fwd_next_prio_supported(ft))
1426 return ERR_PTR(-EOPNOTSUPP);
1428 return ERR_PTR(-EINVAL);
1429 mutex_lock(&root->chain_lock);
1430 next_ft = find_next_chained_ft(prio);
1432 gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1433 gen_dest.ft = next_ft;
1436 flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1438 mutex_unlock(&root->chain_lock);
1439 return ERR_PTR(-EOPNOTSUPP);
1443 handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, dest_num);
1445 if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1446 if (!IS_ERR_OR_NULL(handle) &&
1447 (list_empty(&handle->rule[0]->next_ft))) {
1448 mutex_lock(&next_ft->lock);
1449 list_add(&handle->rule[0]->next_ft,
1450 &next_ft->fwd_rules);
1451 mutex_unlock(&next_ft->lock);
1452 handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1454 mutex_unlock(&root->chain_lock);
1458 EXPORT_SYMBOL(mlx5_add_flow_rules);
1460 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1464 for (i = handle->num_rules - 1; i >= 0; i--)
1465 tree_remove_node(&handle->rule[i]->node);
1468 EXPORT_SYMBOL(mlx5_del_flow_rules);
1470 /* Assuming prio->node.children(flow tables) is sorted by level */
1471 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1473 struct fs_prio *prio;
1475 fs_get_obj(prio, ft->node.parent);
1477 if (!list_is_last(&ft->node.list, &prio->node.children))
1478 return list_next_entry(ft, node.list);
1479 return find_next_chained_ft(prio);
1482 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1484 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1485 struct mlx5_flow_table *new_root_ft = NULL;
1487 if (root->root_ft != ft)
1490 new_root_ft = find_next_ft(ft);
1492 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1495 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1500 root->root_ft = new_root_ft;
1504 /* Connect flow table from previous priority to
1505 * the next flow table.
1507 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1509 struct mlx5_core_dev *dev = get_dev(&ft->node);
1510 struct mlx5_flow_table *next_ft;
1511 struct fs_prio *prio;
1514 err = update_root_ft_destroy(ft);
1518 fs_get_obj(prio, ft->node.parent);
1519 if (!(list_first_entry(&prio->node.children,
1520 struct mlx5_flow_table,
1524 next_ft = find_next_chained_ft(prio);
1525 err = connect_fwd_rules(dev, next_ft, ft);
1529 err = connect_prev_fts(dev, next_ft, prio);
1531 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1536 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1538 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1541 mutex_lock(&root->chain_lock);
1542 err = disconnect_flow_table(ft);
1544 mutex_unlock(&root->chain_lock);
1547 if (tree_remove_node(&ft->node))
1548 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1550 mutex_unlock(&root->chain_lock);
1554 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1556 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1558 if (tree_remove_node(&fg->node))
1559 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1563 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1564 enum mlx5_flow_namespace_type type)
1566 struct mlx5_flow_steering *steering = dev->priv.steering;
1567 struct mlx5_flow_root_namespace *root_ns;
1569 struct fs_prio *fs_prio;
1570 struct mlx5_flow_namespace *ns;
1576 case MLX5_FLOW_NAMESPACE_BYPASS:
1577 case MLX5_FLOW_NAMESPACE_LAG:
1578 case MLX5_FLOW_NAMESPACE_OFFLOADS:
1579 case MLX5_FLOW_NAMESPACE_ETHTOOL:
1580 case MLX5_FLOW_NAMESPACE_KERNEL:
1581 case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1582 case MLX5_FLOW_NAMESPACE_ANCHOR:
1585 case MLX5_FLOW_NAMESPACE_FDB:
1586 if (steering->fdb_root_ns)
1587 return &steering->fdb_root_ns->ns;
1590 case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1591 if (steering->esw_egress_root_ns)
1592 return &steering->esw_egress_root_ns->ns;
1595 case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1596 if (steering->esw_ingress_root_ns)
1597 return &steering->esw_ingress_root_ns->ns;
1600 case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
1601 if (steering->sniffer_rx_root_ns)
1602 return &steering->sniffer_rx_root_ns->ns;
1605 case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
1606 if (steering->sniffer_tx_root_ns)
1607 return &steering->sniffer_tx_root_ns->ns;
1614 root_ns = steering->root_ns;
1618 fs_prio = find_prio(&root_ns->ns, prio);
1622 ns = list_first_entry(&fs_prio->node.children,
1628 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1630 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1631 unsigned int prio, int num_levels)
1633 struct fs_prio *fs_prio;
1635 fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1637 return ERR_PTR(-ENOMEM);
1639 fs_prio->node.type = FS_TYPE_PRIO;
1640 tree_init_node(&fs_prio->node, 1, NULL);
1641 tree_add_node(&fs_prio->node, &ns->node);
1642 fs_prio->num_levels = num_levels;
1643 fs_prio->prio = prio;
1644 list_add_tail(&fs_prio->node.list, &ns->node.children);
1649 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1652 ns->node.type = FS_TYPE_NAMESPACE;
1657 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1659 struct mlx5_flow_namespace *ns;
1661 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1663 return ERR_PTR(-ENOMEM);
1665 fs_init_namespace(ns);
1666 tree_init_node(&ns->node, 1, NULL);
1667 tree_add_node(&ns->node, &prio->node);
1668 list_add_tail(&ns->node.list, &prio->node.children);
1673 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1674 struct init_tree_node *prio_metadata)
1676 struct fs_prio *fs_prio;
1679 for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1680 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1681 if (IS_ERR(fs_prio))
1682 return PTR_ERR(fs_prio);
1687 #define FLOW_TABLE_BIT_SZ 1
1688 #define GET_FLOW_TABLE_CAP(dev, offset) \
1689 ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) + \
1691 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1692 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1696 for (i = 0; i < caps->arr_sz; i++) {
1697 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1703 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
1704 struct init_tree_node *init_node,
1705 struct fs_node *fs_parent_node,
1706 struct init_tree_node *init_parent_node,
1709 int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
1710 flow_table_properties_nic_receive.
1712 struct mlx5_flow_namespace *fs_ns;
1713 struct fs_prio *fs_prio;
1714 struct fs_node *base;
1718 if (init_node->type == FS_TYPE_PRIO) {
1719 if ((init_node->min_ft_level > max_ft_level) ||
1720 !has_required_caps(steering->dev, &init_node->caps))
1723 fs_get_obj(fs_ns, fs_parent_node);
1724 if (init_node->num_leaf_prios)
1725 return create_leaf_prios(fs_ns, prio, init_node);
1726 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1727 if (IS_ERR(fs_prio))
1728 return PTR_ERR(fs_prio);
1729 base = &fs_prio->node;
1730 } else if (init_node->type == FS_TYPE_NAMESPACE) {
1731 fs_get_obj(fs_prio, fs_parent_node);
1732 fs_ns = fs_create_namespace(fs_prio);
1734 return PTR_ERR(fs_ns);
1735 base = &fs_ns->node;
1740 for (i = 0; i < init_node->ar_size; i++) {
1741 err = init_root_tree_recursive(steering, &init_node->children[i],
1742 base, init_node, prio);
1745 if (init_node->children[i].type == FS_TYPE_PRIO &&
1746 init_node->children[i].num_leaf_prios) {
1747 prio += init_node->children[i].num_leaf_prios;
1754 static int init_root_tree(struct mlx5_flow_steering *steering,
1755 struct init_tree_node *init_node,
1756 struct fs_node *fs_parent_node)
1759 struct mlx5_flow_namespace *fs_ns;
1762 fs_get_obj(fs_ns, fs_parent_node);
1763 for (i = 0; i < init_node->ar_size; i++) {
1764 err = init_root_tree_recursive(steering, &init_node->children[i],
1773 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
1774 enum fs_flow_table_type
1777 struct mlx5_flow_root_namespace *root_ns;
1778 struct mlx5_flow_namespace *ns;
1780 /* Create the root namespace */
1781 root_ns = mlx5_vzalloc(sizeof(*root_ns));
1785 root_ns->dev = steering->dev;
1786 root_ns->table_type = table_type;
1789 fs_init_namespace(ns);
1790 mutex_init(&root_ns->chain_lock);
1791 tree_init_node(&ns->node, 1, NULL);
1792 tree_add_node(&ns->node, NULL);
1797 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1799 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1801 struct fs_prio *prio;
1803 fs_for_each_prio(prio, ns) {
1804 /* This updates prio start_level and num_levels */
1805 set_prio_attrs_in_prio(prio, acc_level);
1806 acc_level += prio->num_levels;
1811 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1813 struct mlx5_flow_namespace *ns;
1814 int acc_level_ns = acc_level;
1816 prio->start_level = acc_level;
1817 fs_for_each_ns(ns, prio)
1818 /* This updates start_level and num_levels of ns's priority descendants */
1819 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1820 if (!prio->num_levels)
1821 prio->num_levels = acc_level_ns - prio->start_level;
1822 WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1825 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1827 struct mlx5_flow_namespace *ns = &root_ns->ns;
1828 struct fs_prio *prio;
1829 int start_level = 0;
1831 fs_for_each_prio(prio, ns) {
1832 set_prio_attrs_in_prio(prio, start_level);
1833 start_level += prio->num_levels;
1837 #define ANCHOR_PRIO 0
1838 #define ANCHOR_SIZE 1
1839 #define ANCHOR_LEVEL 0
1840 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
1842 struct mlx5_flow_namespace *ns = NULL;
1843 struct mlx5_flow_table_attr ft_attr = {};
1844 struct mlx5_flow_table *ft;
1846 ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1850 ft_attr.max_fte = ANCHOR_SIZE;
1851 ft_attr.level = ANCHOR_LEVEL;
1852 ft_attr.prio = ANCHOR_PRIO;
1854 ft = mlx5_create_flow_table(ns, &ft_attr);
1856 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
1862 static int init_root_ns(struct mlx5_flow_steering *steering)
1865 steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
1866 if (!steering->root_ns)
1869 if (init_root_tree(steering, &root_fs, &steering->root_ns->ns.node))
1872 set_prio_attrs(steering->root_ns);
1874 if (create_anchor_flow_table(steering))
1880 mlx5_cleanup_fs(steering->dev);
1884 static void clean_tree(struct fs_node *node)
1887 struct fs_node *iter;
1888 struct fs_node *temp;
1890 list_for_each_entry_safe(iter, temp, &node->children, list)
1892 tree_remove_node(node);
1896 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
1901 clean_tree(&root_ns->ns.node);
1904 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1906 struct mlx5_flow_steering *steering = dev->priv.steering;
1908 cleanup_root_ns(steering->root_ns);
1909 cleanup_root_ns(steering->esw_egress_root_ns);
1910 cleanup_root_ns(steering->esw_ingress_root_ns);
1911 cleanup_root_ns(steering->fdb_root_ns);
1912 cleanup_root_ns(steering->sniffer_rx_root_ns);
1913 cleanup_root_ns(steering->sniffer_tx_root_ns);
1914 mlx5_cleanup_fc_stats(dev);
1918 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
1920 struct fs_prio *prio;
1922 steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
1923 if (!steering->sniffer_tx_root_ns)
1926 /* Create single prio */
1927 prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
1929 cleanup_root_ns(steering->sniffer_tx_root_ns);
1930 return PTR_ERR(prio);
1935 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
1937 struct fs_prio *prio;
1939 steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
1940 if (!steering->sniffer_rx_root_ns)
1943 /* Create single prio */
1944 prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
1946 cleanup_root_ns(steering->sniffer_rx_root_ns);
1947 return PTR_ERR(prio);
1952 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
1954 struct fs_prio *prio;
1956 steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
1957 if (!steering->fdb_root_ns)
1960 prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
1964 prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
1968 set_prio_attrs(steering->fdb_root_ns);
1972 cleanup_root_ns(steering->fdb_root_ns);
1973 steering->fdb_root_ns = NULL;
1974 return PTR_ERR(prio);
1977 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
1979 struct fs_prio *prio;
1981 steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
1982 if (!steering->esw_egress_root_ns)
1986 prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
1987 MLX5_TOTAL_VPORTS(steering->dev));
1988 return PTR_ERR_OR_ZERO(prio);
1991 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
1993 struct fs_prio *prio;
1995 steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
1996 if (!steering->esw_ingress_root_ns)
2000 prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
2001 MLX5_TOTAL_VPORTS(steering->dev));
2002 return PTR_ERR_OR_ZERO(prio);
2005 int mlx5_init_fs(struct mlx5_core_dev *dev)
2007 struct mlx5_flow_steering *steering;
2010 err = mlx5_init_fc_stats(dev);
2014 steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2017 steering->dev = dev;
2018 dev->priv.steering = steering;
2020 if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2021 (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2022 ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2023 MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2024 MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2025 err = init_root_ns(steering);
2030 if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
2031 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2032 err = init_fdb_root_ns(steering);
2036 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2037 err = init_egress_acl_root_ns(steering);
2041 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2042 err = init_ingress_acl_root_ns(steering);
2048 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2049 err = init_sniffer_rx_root_ns(steering);
2054 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2055 err = init_sniffer_tx_root_ns(steering);
2062 mlx5_cleanup_fs(dev);