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;
483 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
485 struct mlx5_flow_group *fg;
486 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
487 create_fg_in, match_criteria);
488 u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
490 match_criteria_enable);
491 fg = kzalloc(sizeof(*fg), GFP_KERNEL);
493 return ERR_PTR(-ENOMEM);
495 fg->mask.match_criteria_enable = match_criteria_enable;
496 memcpy(&fg->mask.match_criteria, match_criteria,
497 sizeof(fg->mask.match_criteria));
498 fg->node.type = FS_TYPE_FLOW_GROUP;
499 fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
501 fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
502 end_flow_index) - fg->start_index + 1;
506 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
507 enum fs_flow_table_type table_type,
508 enum fs_flow_table_op_mod op_mod,
511 struct mlx5_flow_table *ft;
513 ft = kzalloc(sizeof(*ft), GFP_KERNEL);
518 ft->node.type = FS_TYPE_FLOW_TABLE;
520 ft->type = table_type;
522 ft->max_fte = max_fte;
524 INIT_LIST_HEAD(&ft->fwd_rules);
525 mutex_init(&ft->lock);
530 /* If reverse is false, then we search for the first flow table in the
531 * root sub-tree from start(closest from right), else we search for the
532 * last flow table in the root sub-tree till start(closest from left).
534 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node *root,
535 struct list_head *start,
538 #define list_advance_entry(pos, reverse) \
539 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
541 #define list_for_each_advance_continue(pos, head, reverse) \
542 for (pos = list_advance_entry(pos, reverse); \
543 &pos->list != (head); \
544 pos = list_advance_entry(pos, reverse))
546 struct fs_node *iter = list_entry(start, struct fs_node, list);
547 struct mlx5_flow_table *ft = NULL;
552 list_for_each_advance_continue(iter, &root->children, reverse) {
553 if (iter->type == FS_TYPE_FLOW_TABLE) {
554 fs_get_obj(ft, iter);
557 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
565 /* If reverse if false then return the first flow table in next priority of
566 * prio in the tree, else return the last flow table in the previous priority
567 * of prio in the tree.
569 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
571 struct mlx5_flow_table *ft = NULL;
572 struct fs_node *curr_node;
573 struct fs_node *parent;
575 parent = prio->node.parent;
576 curr_node = &prio->node;
577 while (!ft && parent) {
578 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
580 parent = curr_node->parent;
585 /* Assuming all the tree is locked by mutex chain lock */
586 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
588 return find_closest_ft(prio, false);
591 /* Assuming all the tree is locked by mutex chain lock */
592 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
594 return find_closest_ft(prio, true);
597 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
598 struct fs_prio *prio,
599 struct mlx5_flow_table *ft)
601 struct mlx5_flow_table *iter;
605 fs_for_each_ft(iter, prio) {
607 err = mlx5_cmd_modify_flow_table(dev,
611 mlx5_core_warn(dev, "Failed to modify flow table %d\n",
613 /* The driver is out of sync with the FW */
622 /* Connect flow tables from previous priority of prio to ft */
623 static int connect_prev_fts(struct mlx5_core_dev *dev,
624 struct mlx5_flow_table *ft,
625 struct fs_prio *prio)
627 struct mlx5_flow_table *prev_ft;
629 prev_ft = find_prev_chained_ft(prio);
631 struct fs_prio *prev_prio;
633 fs_get_obj(prev_prio, prev_ft->node.parent);
634 return connect_fts_in_prio(dev, prev_prio, ft);
639 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
642 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
643 int min_level = INT_MAX;
647 min_level = root->root_ft->level;
649 if (ft->level >= min_level)
652 err = mlx5_cmd_update_root_ft(root->dev, ft);
654 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
662 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
663 struct mlx5_flow_destination *dest)
665 struct mlx5_flow_table *ft;
666 struct mlx5_flow_group *fg;
668 int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
671 fs_get_obj(fte, rule->node.parent);
672 if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
674 lock_ref_node(&fte->node);
675 fs_get_obj(fg, fte->node.parent);
676 fs_get_obj(ft, fg->node.parent);
678 memcpy(&rule->dest_attr, dest, sizeof(*dest));
679 err = mlx5_cmd_update_fte(get_dev(&ft->node),
683 unlock_ref_node(&fte->node);
688 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
689 struct mlx5_flow_destination *new_dest,
690 struct mlx5_flow_destination *old_dest)
695 if (handle->num_rules != 1)
697 return _mlx5_modify_rule_destination(handle->rule[0],
701 for (i = 0; i < handle->num_rules; i++) {
702 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
703 return _mlx5_modify_rule_destination(handle->rule[i],
710 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */
711 static int connect_fwd_rules(struct mlx5_core_dev *dev,
712 struct mlx5_flow_table *new_next_ft,
713 struct mlx5_flow_table *old_next_ft)
715 struct mlx5_flow_destination dest;
716 struct mlx5_flow_rule *iter;
719 /* new_next_ft and old_next_ft could be NULL only
720 * when we create/destroy the anchor flow table.
722 if (!new_next_ft || !old_next_ft)
725 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
726 dest.ft = new_next_ft;
728 mutex_lock(&old_next_ft->lock);
729 list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
730 mutex_unlock(&old_next_ft->lock);
731 list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
732 err = _mlx5_modify_rule_destination(iter, &dest);
734 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
740 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
741 struct fs_prio *prio)
743 struct mlx5_flow_table *next_ft;
746 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
748 if (list_empty(&prio->node.children)) {
749 err = connect_prev_fts(dev, ft, prio);
753 next_ft = find_next_chained_ft(prio);
754 err = connect_fwd_rules(dev, ft, next_ft);
759 if (MLX5_CAP_FLOWTABLE(dev,
760 flow_table_properties_nic_receive.modify_root))
761 err = update_root_ft_create(ft, prio);
765 static void list_add_flow_table(struct mlx5_flow_table *ft,
766 struct fs_prio *prio)
768 struct list_head *prev = &prio->node.children;
769 struct mlx5_flow_table *iter;
771 fs_for_each_ft(iter, prio) {
772 if (iter->level > ft->level)
774 prev = &iter->node.list;
776 list_add(&ft->node.list, prev);
779 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
780 enum fs_flow_table_op_mod op_mod,
782 int max_fte, u32 level,
785 struct mlx5_flow_table *next_ft = NULL;
786 struct mlx5_flow_table *ft;
789 struct mlx5_flow_root_namespace *root =
790 find_root(&ns->node);
791 struct fs_prio *fs_prio = NULL;
794 pr_err("mlx5: flow steering failed to find root of namespace\n");
795 return ERR_PTR(-ENODEV);
798 mutex_lock(&root->chain_lock);
799 fs_prio = find_prio(ns, prio);
804 if (level >= fs_prio->num_levels) {
808 /* The level is related to the
809 * priority level range.
811 level += fs_prio->start_level;
812 ft = alloc_flow_table(level,
814 max_fte ? roundup_pow_of_two(max_fte) : 0,
822 tree_init_node(&ft->node, 1, del_flow_table);
823 log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
824 next_ft = find_next_chained_ft(fs_prio);
825 err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->op_mod, ft->type,
826 ft->level, log_table_sz, next_ft, &ft->id,
831 err = connect_flow_table(root->dev, ft, fs_prio);
834 lock_ref_node(&fs_prio->node);
835 tree_add_node(&ft->node, &fs_prio->node);
836 list_add_flow_table(ft, fs_prio);
838 unlock_ref_node(&fs_prio->node);
839 mutex_unlock(&root->chain_lock);
842 mlx5_cmd_destroy_flow_table(root->dev, ft);
846 mutex_unlock(&root->chain_lock);
850 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
851 int prio, int max_fte,
855 return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, 0, prio,
856 max_fte, level, flags);
859 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
860 int prio, int max_fte,
861 u32 level, u16 vport)
863 return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, vport, prio,
867 struct mlx5_flow_table *mlx5_create_lag_demux_flow_table(
868 struct mlx5_flow_namespace *ns,
871 return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_LAG_DEMUX, 0, prio, 0,
874 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
876 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
878 int num_flow_table_entries,
883 struct mlx5_flow_table *ft;
885 if (max_num_groups > num_flow_table_entries)
886 return ERR_PTR(-EINVAL);
888 ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries, level, flags);
892 ft->autogroup.active = true;
893 ft->autogroup.required_groups = max_num_groups;
897 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
899 /* Flow table should be locked */
900 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
906 struct mlx5_flow_group *fg;
907 struct mlx5_core_dev *dev = get_dev(&ft->node);
911 return ERR_PTR(-ENODEV);
913 fg = alloc_flow_group(fg_in);
917 err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
923 if (ft->autogroup.active)
924 ft->autogroup.num_groups++;
925 /* Add node to tree */
926 tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
927 tree_add_node(&fg->node, &ft->node);
928 /* Add node to group list */
929 list_add(&fg->node.list, prev_fg);
934 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
937 struct mlx5_flow_group *fg;
939 if (ft->autogroup.active)
940 return ERR_PTR(-EPERM);
942 lock_ref_node(&ft->node);
943 fg = create_flow_group_common(ft, fg_in, ft->node.children.prev, false);
944 unlock_ref_node(&ft->node);
949 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
951 struct mlx5_flow_rule *rule;
953 rule = kzalloc(sizeof(*rule), GFP_KERNEL);
957 INIT_LIST_HEAD(&rule->next_ft);
958 rule->node.type = FS_TYPE_FLOW_DEST;
960 memcpy(&rule->dest_attr, dest, sizeof(*dest));
965 static struct mlx5_flow_handle *alloc_handle(int num_rules)
967 struct mlx5_flow_handle *handle;
969 handle = kzalloc(sizeof(*handle) + sizeof(handle->rule[0]) *
970 num_rules, GFP_KERNEL);
974 handle->num_rules = num_rules;
979 static void destroy_flow_handle(struct fs_fte *fte,
980 struct mlx5_flow_handle *handle,
981 struct mlx5_flow_destination *dest,
985 if (atomic_dec_and_test(&handle->rule[i]->node.refcount)) {
987 list_del(&handle->rule[i]->node.list);
988 kfree(handle->rule[i]);
994 static struct mlx5_flow_handle *
995 create_flow_handle(struct fs_fte *fte,
996 struct mlx5_flow_destination *dest,
1001 struct mlx5_flow_handle *handle;
1002 struct mlx5_flow_rule *rule = NULL;
1003 static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1004 static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1008 handle = alloc_handle((dest_num) ? dest_num : 1);
1010 return ERR_PTR(-ENOMEM);
1014 rule = find_flow_rule(fte, dest + i);
1016 atomic_inc(&rule->node.refcount);
1022 rule = alloc_rule(dest + i);
1026 /* Add dest to dests list- we need flow tables to be in the
1027 * end of the list for forward to next prio rules.
1029 tree_init_node(&rule->node, 1, del_rule);
1031 dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1032 list_add(&rule->node.list, &fte->node.children);
1034 list_add_tail(&rule->node.list, &fte->node.children);
1038 type = dest[i].type ==
1039 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1040 *modify_mask |= type ? count : dst;
1043 handle->rule[i] = rule;
1044 } while (++i < dest_num);
1049 destroy_flow_handle(fte, handle, dest, i);
1050 return ERR_PTR(-ENOMEM);
1053 /* fte should not be deleted while calling this function */
1054 static struct mlx5_flow_handle *
1055 add_rule_fte(struct fs_fte *fte,
1056 struct mlx5_flow_group *fg,
1057 struct mlx5_flow_destination *dest,
1061 struct mlx5_flow_handle *handle;
1062 struct mlx5_flow_table *ft;
1063 int modify_mask = 0;
1065 bool new_rule = false;
1067 handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1069 if (IS_ERR(handle) || !new_rule)
1073 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1075 fs_get_obj(ft, fg->node.parent);
1076 if (!(fte->status & FS_FTE_STATUS_EXISTING))
1077 err = mlx5_cmd_create_fte(get_dev(&ft->node),
1080 err = mlx5_cmd_update_fte(get_dev(&ft->node),
1081 ft, fg->id, modify_mask, fte);
1085 fte->status |= FS_FTE_STATUS_EXISTING;
1091 destroy_flow_handle(fte, handle, dest, handle->num_rules);
1092 return ERR_PTR(err);
1095 /* Assumed fg is locked */
1096 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
1097 struct list_head **prev)
1100 unsigned int start = fg->start_index;
1103 *prev = &fg->node.children;
1105 /* assumed list is sorted by index */
1106 fs_for_each_fte(fte, fg) {
1107 if (fte->index != start)
1111 *prev = &fte->node.list;
1117 /* prev is output, prev->next = new_fte */
1118 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
1120 struct mlx5_flow_act *flow_act,
1121 struct list_head **prev)
1126 index = get_free_fte_index(fg, prev);
1127 fte = alloc_fte(flow_act, match_value, index);
1134 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
1135 u8 match_criteria_enable,
1136 u32 *match_criteria)
1138 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1139 struct list_head *prev = ft->node.children.prev;
1140 unsigned int candidate_index = 0;
1141 struct mlx5_flow_group *fg;
1142 void *match_criteria_addr;
1143 unsigned int group_size = 0;
1146 if (!ft->autogroup.active)
1147 return ERR_PTR(-ENOENT);
1149 in = mlx5_vzalloc(inlen);
1151 return ERR_PTR(-ENOMEM);
1153 if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1154 /* We save place for flow groups in addition to max types */
1155 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1157 /* ft->max_fte == ft->autogroup.max_types */
1158 if (group_size == 0)
1161 /* sorted by start_index */
1162 fs_for_each_fg(fg, ft) {
1163 if (candidate_index + group_size > fg->start_index)
1164 candidate_index = fg->start_index + fg->max_ftes;
1167 prev = &fg->node.list;
1170 if (candidate_index + group_size > ft->max_fte) {
1171 fg = ERR_PTR(-ENOSPC);
1175 MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1176 match_criteria_enable);
1177 MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1178 MLX5_SET(create_flow_group_in, in, end_flow_index, candidate_index +
1180 match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1181 in, match_criteria);
1182 memcpy(match_criteria_addr, match_criteria,
1183 MLX5_ST_SZ_BYTES(fte_match_param));
1185 fg = create_flow_group_common(ft, in, prev, true);
1191 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1192 struct mlx5_flow_destination *d2)
1194 if (d1->type == d2->type) {
1195 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1196 d1->vport_num == d2->vport_num) ||
1197 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1198 d1->ft == d2->ft) ||
1199 (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1200 d1->tir_num == d2->tir_num))
1207 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1208 struct mlx5_flow_destination *dest)
1210 struct mlx5_flow_rule *rule;
1212 list_for_each_entry(rule, &fte->node.children, node.list) {
1213 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1219 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1221 struct mlx5_flow_act *flow_act,
1222 struct mlx5_flow_destination *dest,
1225 struct mlx5_flow_handle *handle;
1226 struct mlx5_flow_table *ft;
1227 struct list_head *prev;
1231 nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1232 fs_for_each_fte(fte, fg) {
1233 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1234 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1235 (flow_act->action & fte->action) &&
1236 flow_act->flow_tag == fte->flow_tag) {
1237 int old_action = fte->action;
1239 fte->action |= flow_act->action;
1240 handle = add_rule_fte(fte, fg, dest, dest_num,
1241 old_action != flow_act->action);
1242 if (IS_ERR(handle)) {
1243 fte->action = old_action;
1249 unlock_ref_node(&fte->node);
1251 fs_get_obj(ft, fg->node.parent);
1252 if (fg->num_ftes >= fg->max_ftes) {
1253 handle = ERR_PTR(-ENOSPC);
1257 fte = create_fte(fg, match_value, flow_act, &prev);
1259 handle = (void *)fte;
1262 tree_init_node(&fte->node, 0, del_fte);
1263 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1264 handle = add_rule_fte(fte, fg, dest, dest_num, false);
1265 if (IS_ERR(handle)) {
1266 unlock_ref_node(&fte->node);
1273 tree_add_node(&fte->node, &fg->node);
1274 list_add(&fte->node.list, prev);
1276 for (i = 0; i < handle->num_rules; i++) {
1277 if (atomic_read(&handle->rule[i]->node.refcount) == 1)
1278 tree_add_node(&handle->rule[i]->node, &fte->node);
1281 unlock_ref_node(&fte->node);
1283 unlock_ref_node(&fg->node);
1287 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_handle *handle)
1289 struct mlx5_flow_rule *dst;
1292 fs_get_obj(fte, handle->rule[0]->node.parent);
1294 fs_for_each_dst(dst, fte) {
1295 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1296 return dst->dest_attr.counter;
1302 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1304 if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1310 return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1311 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1314 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1316 struct mlx5_flow_table *ft)
1318 if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1319 return counter_is_valid(dest->counter, action);
1321 if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1324 if (!dest || ((dest->type ==
1325 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1326 (dest->ft->level <= ft->level)))
1331 static struct mlx5_flow_handle *
1332 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1333 struct mlx5_flow_spec *spec,
1334 struct mlx5_flow_act *flow_act,
1335 struct mlx5_flow_destination *dest,
1339 struct mlx5_flow_group *g;
1340 struct mlx5_flow_handle *rule;
1343 for (i = 0; i < dest_num; i++) {
1344 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1345 return ERR_PTR(-EINVAL);
1348 nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1349 fs_for_each_fg(g, ft)
1350 if (compare_match_criteria(g->mask.match_criteria_enable,
1351 spec->match_criteria_enable,
1352 g->mask.match_criteria,
1353 spec->match_criteria)) {
1354 rule = add_rule_fg(g, spec->match_value,
1355 flow_act, dest, dest_num);
1356 if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1360 g = create_autogroup(ft, spec->match_criteria_enable,
1361 spec->match_criteria);
1367 rule = add_rule_fg(g, spec->match_value, flow_act, dest, dest_num);
1369 /* Remove assumes refcount > 0 and autogroup creates a group
1370 * with a refcount = 0.
1372 unlock_ref_node(&ft->node);
1373 tree_get_node(&g->node);
1374 tree_remove_node(&g->node);
1378 unlock_ref_node(&ft->node);
1382 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1384 return ((ft->type == FS_FT_NIC_RX) &&
1385 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1388 struct mlx5_flow_handle *
1389 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1390 struct mlx5_flow_spec *spec,
1391 struct mlx5_flow_act *flow_act,
1392 struct mlx5_flow_destination *dest,
1395 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1396 struct mlx5_flow_destination gen_dest;
1397 struct mlx5_flow_table *next_ft = NULL;
1398 struct mlx5_flow_handle *handle = NULL;
1399 u32 sw_action = flow_act->action;
1400 struct fs_prio *prio;
1402 fs_get_obj(prio, ft->node.parent);
1403 if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1404 if (!fwd_next_prio_supported(ft))
1405 return ERR_PTR(-EOPNOTSUPP);
1407 return ERR_PTR(-EINVAL);
1408 mutex_lock(&root->chain_lock);
1409 next_ft = find_next_chained_ft(prio);
1411 gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1412 gen_dest.ft = next_ft;
1415 flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1417 mutex_unlock(&root->chain_lock);
1418 return ERR_PTR(-EOPNOTSUPP);
1422 handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, dest_num);
1424 if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1425 if (!IS_ERR_OR_NULL(handle) &&
1426 (list_empty(&handle->rule[0]->next_ft))) {
1427 mutex_lock(&next_ft->lock);
1428 list_add(&handle->rule[0]->next_ft,
1429 &next_ft->fwd_rules);
1430 mutex_unlock(&next_ft->lock);
1431 handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1433 mutex_unlock(&root->chain_lock);
1437 EXPORT_SYMBOL(mlx5_add_flow_rules);
1439 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1443 for (i = handle->num_rules - 1; i >= 0; i--)
1444 tree_remove_node(&handle->rule[i]->node);
1447 EXPORT_SYMBOL(mlx5_del_flow_rules);
1449 /* Assuming prio->node.children(flow tables) is sorted by level */
1450 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1452 struct fs_prio *prio;
1454 fs_get_obj(prio, ft->node.parent);
1456 if (!list_is_last(&ft->node.list, &prio->node.children))
1457 return list_next_entry(ft, node.list);
1458 return find_next_chained_ft(prio);
1461 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1463 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1464 struct mlx5_flow_table *new_root_ft = NULL;
1466 if (root->root_ft != ft)
1469 new_root_ft = find_next_ft(ft);
1471 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1474 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1479 root->root_ft = new_root_ft;
1483 /* Connect flow table from previous priority to
1484 * the next flow table.
1486 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1488 struct mlx5_core_dev *dev = get_dev(&ft->node);
1489 struct mlx5_flow_table *next_ft;
1490 struct fs_prio *prio;
1493 err = update_root_ft_destroy(ft);
1497 fs_get_obj(prio, ft->node.parent);
1498 if (!(list_first_entry(&prio->node.children,
1499 struct mlx5_flow_table,
1503 next_ft = find_next_chained_ft(prio);
1504 err = connect_fwd_rules(dev, next_ft, ft);
1508 err = connect_prev_fts(dev, next_ft, prio);
1510 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1515 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1517 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1520 mutex_lock(&root->chain_lock);
1521 err = disconnect_flow_table(ft);
1523 mutex_unlock(&root->chain_lock);
1526 if (tree_remove_node(&ft->node))
1527 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1529 mutex_unlock(&root->chain_lock);
1533 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1535 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1537 if (tree_remove_node(&fg->node))
1538 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1542 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1543 enum mlx5_flow_namespace_type type)
1545 struct mlx5_flow_steering *steering = dev->priv.steering;
1546 struct mlx5_flow_root_namespace *root_ns;
1548 struct fs_prio *fs_prio;
1549 struct mlx5_flow_namespace *ns;
1555 case MLX5_FLOW_NAMESPACE_BYPASS:
1556 case MLX5_FLOW_NAMESPACE_LAG:
1557 case MLX5_FLOW_NAMESPACE_OFFLOADS:
1558 case MLX5_FLOW_NAMESPACE_ETHTOOL:
1559 case MLX5_FLOW_NAMESPACE_KERNEL:
1560 case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1561 case MLX5_FLOW_NAMESPACE_ANCHOR:
1564 case MLX5_FLOW_NAMESPACE_FDB:
1565 if (steering->fdb_root_ns)
1566 return &steering->fdb_root_ns->ns;
1569 case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1570 if (steering->esw_egress_root_ns)
1571 return &steering->esw_egress_root_ns->ns;
1574 case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1575 if (steering->esw_ingress_root_ns)
1576 return &steering->esw_ingress_root_ns->ns;
1579 case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
1580 if (steering->sniffer_rx_root_ns)
1581 return &steering->sniffer_rx_root_ns->ns;
1584 case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
1585 if (steering->sniffer_tx_root_ns)
1586 return &steering->sniffer_tx_root_ns->ns;
1593 root_ns = steering->root_ns;
1597 fs_prio = find_prio(&root_ns->ns, prio);
1601 ns = list_first_entry(&fs_prio->node.children,
1607 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1609 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1610 unsigned int prio, int num_levels)
1612 struct fs_prio *fs_prio;
1614 fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1616 return ERR_PTR(-ENOMEM);
1618 fs_prio->node.type = FS_TYPE_PRIO;
1619 tree_init_node(&fs_prio->node, 1, NULL);
1620 tree_add_node(&fs_prio->node, &ns->node);
1621 fs_prio->num_levels = num_levels;
1622 fs_prio->prio = prio;
1623 list_add_tail(&fs_prio->node.list, &ns->node.children);
1628 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1631 ns->node.type = FS_TYPE_NAMESPACE;
1636 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1638 struct mlx5_flow_namespace *ns;
1640 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1642 return ERR_PTR(-ENOMEM);
1644 fs_init_namespace(ns);
1645 tree_init_node(&ns->node, 1, NULL);
1646 tree_add_node(&ns->node, &prio->node);
1647 list_add_tail(&ns->node.list, &prio->node.children);
1652 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1653 struct init_tree_node *prio_metadata)
1655 struct fs_prio *fs_prio;
1658 for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1659 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1660 if (IS_ERR(fs_prio))
1661 return PTR_ERR(fs_prio);
1666 #define FLOW_TABLE_BIT_SZ 1
1667 #define GET_FLOW_TABLE_CAP(dev, offset) \
1668 ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) + \
1670 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1671 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1675 for (i = 0; i < caps->arr_sz; i++) {
1676 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1682 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
1683 struct init_tree_node *init_node,
1684 struct fs_node *fs_parent_node,
1685 struct init_tree_node *init_parent_node,
1688 int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
1689 flow_table_properties_nic_receive.
1691 struct mlx5_flow_namespace *fs_ns;
1692 struct fs_prio *fs_prio;
1693 struct fs_node *base;
1697 if (init_node->type == FS_TYPE_PRIO) {
1698 if ((init_node->min_ft_level > max_ft_level) ||
1699 !has_required_caps(steering->dev, &init_node->caps))
1702 fs_get_obj(fs_ns, fs_parent_node);
1703 if (init_node->num_leaf_prios)
1704 return create_leaf_prios(fs_ns, prio, init_node);
1705 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1706 if (IS_ERR(fs_prio))
1707 return PTR_ERR(fs_prio);
1708 base = &fs_prio->node;
1709 } else if (init_node->type == FS_TYPE_NAMESPACE) {
1710 fs_get_obj(fs_prio, fs_parent_node);
1711 fs_ns = fs_create_namespace(fs_prio);
1713 return PTR_ERR(fs_ns);
1714 base = &fs_ns->node;
1719 for (i = 0; i < init_node->ar_size; i++) {
1720 err = init_root_tree_recursive(steering, &init_node->children[i],
1721 base, init_node, prio);
1724 if (init_node->children[i].type == FS_TYPE_PRIO &&
1725 init_node->children[i].num_leaf_prios) {
1726 prio += init_node->children[i].num_leaf_prios;
1733 static int init_root_tree(struct mlx5_flow_steering *steering,
1734 struct init_tree_node *init_node,
1735 struct fs_node *fs_parent_node)
1738 struct mlx5_flow_namespace *fs_ns;
1741 fs_get_obj(fs_ns, fs_parent_node);
1742 for (i = 0; i < init_node->ar_size; i++) {
1743 err = init_root_tree_recursive(steering, &init_node->children[i],
1752 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
1753 enum fs_flow_table_type
1756 struct mlx5_flow_root_namespace *root_ns;
1757 struct mlx5_flow_namespace *ns;
1759 /* Create the root namespace */
1760 root_ns = mlx5_vzalloc(sizeof(*root_ns));
1764 root_ns->dev = steering->dev;
1765 root_ns->table_type = table_type;
1768 fs_init_namespace(ns);
1769 mutex_init(&root_ns->chain_lock);
1770 tree_init_node(&ns->node, 1, NULL);
1771 tree_add_node(&ns->node, NULL);
1776 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1778 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1780 struct fs_prio *prio;
1782 fs_for_each_prio(prio, ns) {
1783 /* This updates prio start_level and num_levels */
1784 set_prio_attrs_in_prio(prio, acc_level);
1785 acc_level += prio->num_levels;
1790 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1792 struct mlx5_flow_namespace *ns;
1793 int acc_level_ns = acc_level;
1795 prio->start_level = acc_level;
1796 fs_for_each_ns(ns, prio)
1797 /* This updates start_level and num_levels of ns's priority descendants */
1798 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1799 if (!prio->num_levels)
1800 prio->num_levels = acc_level_ns - prio->start_level;
1801 WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1804 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1806 struct mlx5_flow_namespace *ns = &root_ns->ns;
1807 struct fs_prio *prio;
1808 int start_level = 0;
1810 fs_for_each_prio(prio, ns) {
1811 set_prio_attrs_in_prio(prio, start_level);
1812 start_level += prio->num_levels;
1816 #define ANCHOR_PRIO 0
1817 #define ANCHOR_SIZE 1
1818 #define ANCHOR_LEVEL 0
1819 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
1821 struct mlx5_flow_namespace *ns = NULL;
1822 struct mlx5_flow_table *ft;
1824 ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1827 ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL, 0);
1829 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
1835 static int init_root_ns(struct mlx5_flow_steering *steering)
1838 steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
1839 if (!steering->root_ns)
1842 if (init_root_tree(steering, &root_fs, &steering->root_ns->ns.node))
1845 set_prio_attrs(steering->root_ns);
1847 if (create_anchor_flow_table(steering))
1853 mlx5_cleanup_fs(steering->dev);
1857 static void clean_tree(struct fs_node *node)
1860 struct fs_node *iter;
1861 struct fs_node *temp;
1863 list_for_each_entry_safe(iter, temp, &node->children, list)
1865 tree_remove_node(node);
1869 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
1874 clean_tree(&root_ns->ns.node);
1877 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1879 struct mlx5_flow_steering *steering = dev->priv.steering;
1881 if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1884 cleanup_root_ns(steering->root_ns);
1885 cleanup_root_ns(steering->esw_egress_root_ns);
1886 cleanup_root_ns(steering->esw_ingress_root_ns);
1887 cleanup_root_ns(steering->fdb_root_ns);
1888 cleanup_root_ns(steering->sniffer_rx_root_ns);
1889 cleanup_root_ns(steering->sniffer_tx_root_ns);
1890 mlx5_cleanup_fc_stats(dev);
1894 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
1896 struct fs_prio *prio;
1898 steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
1899 if (!steering->sniffer_tx_root_ns)
1902 /* Create single prio */
1903 prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
1905 cleanup_root_ns(steering->sniffer_tx_root_ns);
1906 return PTR_ERR(prio);
1911 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
1913 struct fs_prio *prio;
1915 steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
1916 if (!steering->sniffer_rx_root_ns)
1919 /* Create single prio */
1920 prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
1922 cleanup_root_ns(steering->sniffer_rx_root_ns);
1923 return PTR_ERR(prio);
1928 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
1930 struct fs_prio *prio;
1932 steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
1933 if (!steering->fdb_root_ns)
1936 prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
1940 prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
1944 set_prio_attrs(steering->fdb_root_ns);
1948 cleanup_root_ns(steering->fdb_root_ns);
1949 steering->fdb_root_ns = NULL;
1950 return PTR_ERR(prio);
1953 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
1955 struct fs_prio *prio;
1957 steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
1958 if (!steering->esw_egress_root_ns)
1962 prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
1963 MLX5_TOTAL_VPORTS(steering->dev));
1964 return PTR_ERR_OR_ZERO(prio);
1967 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
1969 struct fs_prio *prio;
1971 steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
1972 if (!steering->esw_ingress_root_ns)
1976 prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
1977 MLX5_TOTAL_VPORTS(steering->dev));
1978 return PTR_ERR_OR_ZERO(prio);
1981 int mlx5_init_fs(struct mlx5_core_dev *dev)
1983 struct mlx5_flow_steering *steering;
1986 if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1989 err = mlx5_init_fc_stats(dev);
1993 steering = kzalloc(sizeof(*steering), GFP_KERNEL);
1996 steering->dev = dev;
1997 dev->priv.steering = steering;
1999 if (MLX5_CAP_GEN(dev, nic_flow_table) &&
2000 MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2001 err = init_root_ns(steering);
2006 if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
2007 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2008 err = init_fdb_root_ns(steering);
2012 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2013 err = init_egress_acl_root_ns(steering);
2017 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2018 err = init_ingress_acl_root_ns(steering);
2024 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2025 err = init_sniffer_rx_root_ns(steering);
2030 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2031 err = init_sniffer_tx_root_ns(steering);
2038 mlx5_cleanup_fs(dev);