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>
35 #include <linux/mlx5/eswitch.h>
37 #include "mlx5_core.h"
40 #include "diag/fs_tracepoint.h"
41 #include "accel/ipsec.h"
42 #include "fpga/ipsec.h"
44 #define INIT_TREE_NODE_ARRAY_SIZE(...) (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
45 sizeof(struct init_tree_node))
47 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
48 ...) {.type = FS_TYPE_PRIO,\
49 .min_ft_level = min_level_val,\
50 .num_levels = num_levels_val,\
51 .num_leaf_prios = num_prios_val,\
53 .children = (struct init_tree_node[]) {__VA_ARGS__},\
54 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
57 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
58 ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
61 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
62 .children = (struct init_tree_node[]) {__VA_ARGS__},\
63 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
66 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
69 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
71 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
72 .caps = (long[]) {__VA_ARGS__} }
74 #define FS_CHAINING_CAPS FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
75 FS_CAP(flow_table_properties_nic_receive.modify_root), \
76 FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
77 FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
79 #define LEFTOVERS_NUM_LEVELS 1
80 #define LEFTOVERS_NUM_PRIOS 1
82 #define BY_PASS_PRIO_NUM_LEVELS 1
83 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
86 #define ETHTOOL_PRIO_NUM_LEVELS 1
87 #define ETHTOOL_NUM_PRIOS 11
88 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
89 /* Vlan, mac, ttc, inner ttc, aRFS */
90 #define KERNEL_NIC_PRIO_NUM_LEVELS 5
91 #define KERNEL_NIC_NUM_PRIOS 1
92 /* One more level for tc */
93 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
95 #define KERNEL_NIC_TC_NUM_PRIOS 1
96 #define KERNEL_NIC_TC_NUM_LEVELS 2
98 #define ANCHOR_NUM_LEVELS 1
99 #define ANCHOR_NUM_PRIOS 1
100 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
102 #define OFFLOADS_MAX_FT 1
103 #define OFFLOADS_NUM_PRIOS 1
104 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
106 #define LAG_PRIO_NUM_LEVELS 1
107 #define LAG_NUM_PRIOS 1
108 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
115 static struct init_tree_node {
116 enum fs_node_type type;
117 struct init_tree_node *children;
119 struct node_caps caps;
125 .type = FS_TYPE_NAMESPACE,
127 .children = (struct init_tree_node[]) {
128 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
130 ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
131 BY_PASS_PRIO_NUM_LEVELS))),
132 ADD_PRIO(0, LAG_MIN_LEVEL, 0,
134 ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
135 LAG_PRIO_NUM_LEVELS))),
136 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
137 ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
138 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
140 ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
141 ETHTOOL_PRIO_NUM_LEVELS))),
142 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
143 ADD_NS(ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS, KERNEL_NIC_TC_NUM_LEVELS),
144 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
145 KERNEL_NIC_PRIO_NUM_LEVELS))),
146 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
148 ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
149 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
150 ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
154 enum fs_i_lock_class {
160 static const struct rhashtable_params rhash_fte = {
161 .key_len = FIELD_SIZEOF(struct fs_fte, val),
162 .key_offset = offsetof(struct fs_fte, val),
163 .head_offset = offsetof(struct fs_fte, hash),
164 .automatic_shrinking = true,
168 static const struct rhashtable_params rhash_fg = {
169 .key_len = FIELD_SIZEOF(struct mlx5_flow_group, mask),
170 .key_offset = offsetof(struct mlx5_flow_group, mask),
171 .head_offset = offsetof(struct mlx5_flow_group, hash),
172 .automatic_shrinking = true,
177 static void del_hw_flow_table(struct fs_node *node);
178 static void del_hw_flow_group(struct fs_node *node);
179 static void del_hw_fte(struct fs_node *node);
180 static void del_sw_flow_table(struct fs_node *node);
181 static void del_sw_flow_group(struct fs_node *node);
182 static void del_sw_fte(struct fs_node *node);
183 static void del_sw_prio(struct fs_node *node);
184 static void del_sw_ns(struct fs_node *node);
185 /* Delete rule (destination) is special case that
186 * requires to lock the FTE for all the deletion process.
188 static void del_sw_hw_rule(struct fs_node *node);
189 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
190 struct mlx5_flow_destination *d2);
191 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
192 static struct mlx5_flow_rule *
193 find_flow_rule(struct fs_fte *fte,
194 struct mlx5_flow_destination *dest);
196 static void tree_init_node(struct fs_node *node,
197 void (*del_hw_func)(struct fs_node *),
198 void (*del_sw_func)(struct fs_node *))
200 refcount_set(&node->refcount, 1);
201 INIT_LIST_HEAD(&node->list);
202 INIT_LIST_HEAD(&node->children);
203 init_rwsem(&node->lock);
204 node->del_hw_func = del_hw_func;
205 node->del_sw_func = del_sw_func;
206 node->active = false;
209 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
212 refcount_inc(&parent->refcount);
213 node->parent = parent;
215 /* Parent is the root */
219 node->root = parent->root;
222 static int tree_get_node(struct fs_node *node)
224 return refcount_inc_not_zero(&node->refcount);
227 static void nested_down_read_ref_node(struct fs_node *node,
228 enum fs_i_lock_class class)
231 down_read_nested(&node->lock, class);
232 refcount_inc(&node->refcount);
236 static void nested_down_write_ref_node(struct fs_node *node,
237 enum fs_i_lock_class class)
240 down_write_nested(&node->lock, class);
241 refcount_inc(&node->refcount);
245 static void down_write_ref_node(struct fs_node *node)
248 down_write(&node->lock);
249 refcount_inc(&node->refcount);
253 static void up_read_ref_node(struct fs_node *node)
255 refcount_dec(&node->refcount);
256 up_read(&node->lock);
259 static void up_write_ref_node(struct fs_node *node)
261 refcount_dec(&node->refcount);
262 up_write(&node->lock);
265 static void tree_put_node(struct fs_node *node)
267 struct fs_node *parent_node = node->parent;
269 if (refcount_dec_and_test(&node->refcount)) {
270 if (node->del_hw_func)
271 node->del_hw_func(node);
273 /* Only root namespace doesn't have parent and we just
274 * need to free its node.
276 down_write_ref_node(parent_node);
277 list_del_init(&node->list);
278 if (node->del_sw_func)
279 node->del_sw_func(node);
280 up_write_ref_node(parent_node);
286 if (!node && parent_node)
287 tree_put_node(parent_node);
290 static int tree_remove_node(struct fs_node *node)
292 if (refcount_read(&node->refcount) > 1) {
293 refcount_dec(&node->refcount);
300 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
303 struct fs_prio *iter_prio;
305 fs_for_each_prio(iter_prio, ns) {
306 if (iter_prio->prio == prio)
313 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
317 for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
318 if (spec->match_value[i] & ~spec->match_criteria[i]) {
319 pr_warn("mlx5_core: match_value differs from match_criteria\n");
326 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
328 struct fs_node *root;
329 struct mlx5_flow_namespace *ns;
333 if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
334 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
338 ns = container_of(root, struct mlx5_flow_namespace, node);
339 return container_of(ns, struct mlx5_flow_root_namespace, ns);
342 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
344 struct mlx5_flow_root_namespace *root = find_root(node);
347 return root->dev->priv.steering;
351 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
353 struct mlx5_flow_root_namespace *root = find_root(node);
360 static void del_sw_ns(struct fs_node *node)
365 static void del_sw_prio(struct fs_node *node)
370 static void del_hw_flow_table(struct fs_node *node)
372 struct mlx5_flow_root_namespace *root;
373 struct mlx5_flow_table *ft;
374 struct mlx5_core_dev *dev;
377 fs_get_obj(ft, node);
378 dev = get_dev(&ft->node);
379 root = find_root(&ft->node);
382 err = root->cmds->destroy_flow_table(dev, ft);
384 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
388 static void del_sw_flow_table(struct fs_node *node)
390 struct mlx5_flow_table *ft;
391 struct fs_prio *prio;
393 fs_get_obj(ft, node);
395 rhltable_destroy(&ft->fgs_hash);
396 fs_get_obj(prio, ft->node.parent);
401 static void del_sw_hw_rule(struct fs_node *node)
403 struct mlx5_flow_root_namespace *root;
404 struct mlx5_flow_rule *rule;
405 struct mlx5_flow_table *ft;
406 struct mlx5_flow_group *fg;
409 struct mlx5_core_dev *dev = get_dev(node);
411 bool update_fte = false;
413 fs_get_obj(rule, node);
414 fs_get_obj(fte, rule->node.parent);
415 fs_get_obj(fg, fte->node.parent);
416 fs_get_obj(ft, fg->node.parent);
417 trace_mlx5_fs_del_rule(rule);
418 if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
419 mutex_lock(&rule->dest_attr.ft->lock);
420 list_del(&rule->next_ft);
421 mutex_unlock(&rule->dest_attr.ft->lock);
424 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER &&
426 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
427 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
428 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
433 if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
435 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
439 root = find_root(&ft->node);
440 if (update_fte && fte->dests_size) {
441 err = root->cmds->update_fte(dev, ft, fg->id, modify_mask, fte);
444 "%s can't del rule fg id=%d fte_index=%d\n",
445 __func__, fg->id, fte->index);
450 static void del_hw_fte(struct fs_node *node)
452 struct mlx5_flow_root_namespace *root;
453 struct mlx5_flow_table *ft;
454 struct mlx5_flow_group *fg;
455 struct mlx5_core_dev *dev;
459 fs_get_obj(fte, node);
460 fs_get_obj(fg, fte->node.parent);
461 fs_get_obj(ft, fg->node.parent);
463 trace_mlx5_fs_del_fte(fte);
464 dev = get_dev(&ft->node);
465 root = find_root(&ft->node);
467 err = root->cmds->delete_fte(dev, ft, fte);
470 "flow steering can't delete fte in index %d of flow group id %d\n",
476 static void del_sw_fte(struct fs_node *node)
478 struct mlx5_flow_steering *steering = get_steering(node);
479 struct mlx5_flow_group *fg;
483 fs_get_obj(fte, node);
484 fs_get_obj(fg, fte->node.parent);
486 err = rhashtable_remove_fast(&fg->ftes_hash,
490 ida_simple_remove(&fg->fte_allocator, fte->index - fg->start_index);
491 kmem_cache_free(steering->ftes_cache, fte);
494 static void del_hw_flow_group(struct fs_node *node)
496 struct mlx5_flow_root_namespace *root;
497 struct mlx5_flow_group *fg;
498 struct mlx5_flow_table *ft;
499 struct mlx5_core_dev *dev;
501 fs_get_obj(fg, node);
502 fs_get_obj(ft, fg->node.parent);
503 dev = get_dev(&ft->node);
504 trace_mlx5_fs_del_fg(fg);
506 root = find_root(&ft->node);
507 if (fg->node.active && root->cmds->destroy_flow_group(dev, ft, fg->id))
508 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
512 static void del_sw_flow_group(struct fs_node *node)
514 struct mlx5_flow_steering *steering = get_steering(node);
515 struct mlx5_flow_group *fg;
516 struct mlx5_flow_table *ft;
519 fs_get_obj(fg, node);
520 fs_get_obj(ft, fg->node.parent);
522 rhashtable_destroy(&fg->ftes_hash);
523 ida_destroy(&fg->fte_allocator);
524 if (ft->autogroup.active && fg->max_ftes == ft->autogroup.group_size)
525 ft->autogroup.num_groups--;
526 err = rhltable_remove(&ft->fgs_hash,
530 kmem_cache_free(steering->fgs_cache, fg);
533 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
538 index = ida_simple_get(&fg->fte_allocator, 0, fg->max_ftes, GFP_KERNEL);
542 fte->index = index + fg->start_index;
543 ret = rhashtable_insert_fast(&fg->ftes_hash,
549 tree_add_node(&fte->node, &fg->node);
550 list_add_tail(&fte->node.list, &fg->node.children);
554 ida_simple_remove(&fg->fte_allocator, index);
558 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
560 struct mlx5_flow_act *flow_act)
562 struct mlx5_flow_steering *steering = get_steering(&ft->node);
565 fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
567 return ERR_PTR(-ENOMEM);
569 memcpy(fte->val, match_value, sizeof(fte->val));
570 fte->node.type = FS_TYPE_FLOW_ENTRY;
571 fte->action = *flow_act;
573 tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
578 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
579 struct mlx5_flow_group *fg)
581 rhashtable_destroy(&fg->ftes_hash);
582 kmem_cache_free(steering->fgs_cache, fg);
585 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
586 u8 match_criteria_enable,
587 void *match_criteria,
591 struct mlx5_flow_group *fg;
594 fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
596 return ERR_PTR(-ENOMEM);
598 ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
600 kmem_cache_free(steering->fgs_cache, fg);
603 ida_init(&fg->fte_allocator);
604 fg->mask.match_criteria_enable = match_criteria_enable;
605 memcpy(&fg->mask.match_criteria, match_criteria,
606 sizeof(fg->mask.match_criteria));
607 fg->node.type = FS_TYPE_FLOW_GROUP;
608 fg->start_index = start_index;
609 fg->max_ftes = end_index - start_index + 1;
614 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
615 u8 match_criteria_enable,
616 void *match_criteria,
619 struct list_head *prev)
621 struct mlx5_flow_steering *steering = get_steering(&ft->node);
622 struct mlx5_flow_group *fg;
625 fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
626 start_index, end_index);
630 /* initialize refcnt, add to parent list */
631 ret = rhltable_insert(&ft->fgs_hash,
635 dealloc_flow_group(steering, fg);
639 tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
640 tree_add_node(&fg->node, &ft->node);
641 /* Add node to group list */
642 list_add(&fg->node.list, prev);
643 atomic_inc(&ft->node.version);
648 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
649 enum fs_flow_table_type table_type,
650 enum fs_flow_table_op_mod op_mod,
653 struct mlx5_flow_table *ft;
656 ft = kzalloc(sizeof(*ft), GFP_KERNEL);
658 return ERR_PTR(-ENOMEM);
660 ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
667 ft->node.type = FS_TYPE_FLOW_TABLE;
669 ft->type = table_type;
671 ft->max_fte = max_fte;
673 INIT_LIST_HEAD(&ft->fwd_rules);
674 mutex_init(&ft->lock);
679 /* If reverse is false, then we search for the first flow table in the
680 * root sub-tree from start(closest from right), else we search for the
681 * last flow table in the root sub-tree till start(closest from left).
683 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node *root,
684 struct list_head *start,
687 #define list_advance_entry(pos, reverse) \
688 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
690 #define list_for_each_advance_continue(pos, head, reverse) \
691 for (pos = list_advance_entry(pos, reverse); \
692 &pos->list != (head); \
693 pos = list_advance_entry(pos, reverse))
695 struct fs_node *iter = list_entry(start, struct fs_node, list);
696 struct mlx5_flow_table *ft = NULL;
701 list_for_each_advance_continue(iter, &root->children, reverse) {
702 if (iter->type == FS_TYPE_FLOW_TABLE) {
703 fs_get_obj(ft, iter);
706 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
714 /* If reverse if false then return the first flow table in next priority of
715 * prio in the tree, else return the last flow table in the previous priority
716 * of prio in the tree.
718 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
720 struct mlx5_flow_table *ft = NULL;
721 struct fs_node *curr_node;
722 struct fs_node *parent;
724 parent = prio->node.parent;
725 curr_node = &prio->node;
726 while (!ft && parent) {
727 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
729 parent = curr_node->parent;
734 /* Assuming all the tree is locked by mutex chain lock */
735 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
737 return find_closest_ft(prio, false);
740 /* Assuming all the tree is locked by mutex chain lock */
741 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
743 return find_closest_ft(prio, true);
746 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
747 struct fs_prio *prio,
748 struct mlx5_flow_table *ft)
750 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
751 struct mlx5_flow_table *iter;
755 fs_for_each_ft(iter, prio) {
757 err = root->cmds->modify_flow_table(dev, iter, ft);
759 mlx5_core_warn(dev, "Failed to modify flow table %d\n",
761 /* The driver is out of sync with the FW */
770 /* Connect flow tables from previous priority of prio to ft */
771 static int connect_prev_fts(struct mlx5_core_dev *dev,
772 struct mlx5_flow_table *ft,
773 struct fs_prio *prio)
775 struct mlx5_flow_table *prev_ft;
777 prev_ft = find_prev_chained_ft(prio);
779 struct fs_prio *prev_prio;
781 fs_get_obj(prev_prio, prev_ft->node.parent);
782 return connect_fts_in_prio(dev, prev_prio, ft);
787 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
790 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
791 struct mlx5_ft_underlay_qp *uqp;
792 int min_level = INT_MAX;
797 min_level = root->root_ft->level;
799 if (ft->level >= min_level)
802 if (list_empty(&root->underlay_qpns)) {
803 /* Don't set any QPN (zero) in case QPN list is empty */
805 err = root->cmds->update_root_ft(root->dev, ft, qpn, false);
807 list_for_each_entry(uqp, &root->underlay_qpns, list) {
809 err = root->cmds->update_root_ft(root->dev, ft,
817 mlx5_core_warn(root->dev,
818 "Update root flow table of id(%u) qpn(%d) failed\n",
826 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
827 struct mlx5_flow_destination *dest)
829 struct mlx5_flow_root_namespace *root;
830 struct mlx5_flow_table *ft;
831 struct mlx5_flow_group *fg;
833 int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
836 fs_get_obj(fte, rule->node.parent);
837 if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
839 down_write_ref_node(&fte->node);
840 fs_get_obj(fg, fte->node.parent);
841 fs_get_obj(ft, fg->node.parent);
843 memcpy(&rule->dest_attr, dest, sizeof(*dest));
844 root = find_root(&ft->node);
845 err = root->cmds->update_fte(get_dev(&ft->node), ft, fg->id,
847 up_write_ref_node(&fte->node);
852 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
853 struct mlx5_flow_destination *new_dest,
854 struct mlx5_flow_destination *old_dest)
859 if (handle->num_rules != 1)
861 return _mlx5_modify_rule_destination(handle->rule[0],
865 for (i = 0; i < handle->num_rules; i++) {
866 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
867 return _mlx5_modify_rule_destination(handle->rule[i],
874 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */
875 static int connect_fwd_rules(struct mlx5_core_dev *dev,
876 struct mlx5_flow_table *new_next_ft,
877 struct mlx5_flow_table *old_next_ft)
879 struct mlx5_flow_destination dest = {};
880 struct mlx5_flow_rule *iter;
883 /* new_next_ft and old_next_ft could be NULL only
884 * when we create/destroy the anchor flow table.
886 if (!new_next_ft || !old_next_ft)
889 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
890 dest.ft = new_next_ft;
892 mutex_lock(&old_next_ft->lock);
893 list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
894 mutex_unlock(&old_next_ft->lock);
895 list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
896 err = _mlx5_modify_rule_destination(iter, &dest);
898 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
904 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
905 struct fs_prio *prio)
907 struct mlx5_flow_table *next_ft, *first_ft;
910 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
912 first_ft = list_first_entry_or_null(&prio->node.children,
913 struct mlx5_flow_table, node.list);
914 if (!first_ft || first_ft->level > ft->level) {
915 err = connect_prev_fts(dev, ft, prio);
919 next_ft = first_ft ? first_ft : find_next_chained_ft(prio);
920 err = connect_fwd_rules(dev, ft, next_ft);
925 if (MLX5_CAP_FLOWTABLE(dev,
926 flow_table_properties_nic_receive.modify_root))
927 err = update_root_ft_create(ft, prio);
931 static void list_add_flow_table(struct mlx5_flow_table *ft,
932 struct fs_prio *prio)
934 struct list_head *prev = &prio->node.children;
935 struct mlx5_flow_table *iter;
937 fs_for_each_ft(iter, prio) {
938 if (iter->level > ft->level)
940 prev = &iter->node.list;
942 list_add(&ft->node.list, prev);
945 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
946 struct mlx5_flow_table_attr *ft_attr,
947 enum fs_flow_table_op_mod op_mod,
950 struct mlx5_flow_root_namespace *root = find_root(&ns->node);
951 struct mlx5_flow_table *next_ft = NULL;
952 struct fs_prio *fs_prio = NULL;
953 struct mlx5_flow_table *ft;
958 pr_err("mlx5: flow steering failed to find root of namespace\n");
959 return ERR_PTR(-ENODEV);
962 mutex_lock(&root->chain_lock);
963 fs_prio = find_prio(ns, ft_attr->prio);
968 if (ft_attr->level >= fs_prio->num_levels) {
972 /* The level is related to the
973 * priority level range.
975 ft_attr->level += fs_prio->start_level;
976 ft = alloc_flow_table(ft_attr->level,
978 ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
980 op_mod, ft_attr->flags);
986 tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
987 log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
988 next_ft = find_next_chained_ft(fs_prio);
989 err = root->cmds->create_flow_table(root->dev, ft->vport, ft->op_mod,
990 ft->type, ft->level, log_table_sz,
991 next_ft, &ft->id, ft->flags);
995 err = connect_flow_table(root->dev, ft, fs_prio);
998 ft->node.active = true;
999 down_write_ref_node(&fs_prio->node);
1000 tree_add_node(&ft->node, &fs_prio->node);
1001 list_add_flow_table(ft, fs_prio);
1003 up_write_ref_node(&fs_prio->node);
1004 mutex_unlock(&root->chain_lock);
1007 root->cmds->destroy_flow_table(root->dev, ft);
1009 rhltable_destroy(&ft->fgs_hash);
1012 mutex_unlock(&root->chain_lock);
1013 return ERR_PTR(err);
1016 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1017 struct mlx5_flow_table_attr *ft_attr)
1019 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1022 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1023 int prio, int max_fte,
1024 u32 level, u16 vport)
1026 struct mlx5_flow_table_attr ft_attr = {};
1028 ft_attr.max_fte = max_fte;
1029 ft_attr.level = level;
1030 ft_attr.prio = prio;
1032 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1035 struct mlx5_flow_table*
1036 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1037 int prio, u32 level)
1039 struct mlx5_flow_table_attr ft_attr = {};
1041 ft_attr.level = level;
1042 ft_attr.prio = prio;
1043 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1045 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1047 struct mlx5_flow_table*
1048 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1050 int num_flow_table_entries,
1055 struct mlx5_flow_table_attr ft_attr = {};
1056 struct mlx5_flow_table *ft;
1058 if (max_num_groups > num_flow_table_entries)
1059 return ERR_PTR(-EINVAL);
1061 ft_attr.max_fte = num_flow_table_entries;
1062 ft_attr.prio = prio;
1063 ft_attr.level = level;
1064 ft_attr.flags = flags;
1066 ft = mlx5_create_flow_table(ns, &ft_attr);
1070 ft->autogroup.active = true;
1071 ft->autogroup.required_groups = max_num_groups;
1072 /* We save place for flow groups in addition to max types */
1073 ft->autogroup.group_size = ft->max_fte / (max_num_groups + 1);
1077 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1079 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1082 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1083 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1084 fg_in, match_criteria);
1085 u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1087 match_criteria_enable);
1088 int start_index = MLX5_GET(create_flow_group_in, fg_in,
1090 int end_index = MLX5_GET(create_flow_group_in, fg_in,
1092 struct mlx5_core_dev *dev = get_dev(&ft->node);
1093 struct mlx5_flow_group *fg;
1096 if (ft->autogroup.active)
1097 return ERR_PTR(-EPERM);
1099 down_write_ref_node(&ft->node);
1100 fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1101 start_index, end_index,
1102 ft->node.children.prev);
1103 up_write_ref_node(&ft->node);
1107 err = root->cmds->create_flow_group(dev, ft, fg_in, &fg->id);
1109 tree_put_node(&fg->node);
1110 return ERR_PTR(err);
1112 trace_mlx5_fs_add_fg(fg);
1113 fg->node.active = true;
1118 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1120 struct mlx5_flow_rule *rule;
1122 rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1126 INIT_LIST_HEAD(&rule->next_ft);
1127 rule->node.type = FS_TYPE_FLOW_DEST;
1129 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1134 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1136 struct mlx5_flow_handle *handle;
1138 handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1142 handle->num_rules = num_rules;
1147 static void destroy_flow_handle(struct fs_fte *fte,
1148 struct mlx5_flow_handle *handle,
1149 struct mlx5_flow_destination *dest,
1153 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1155 list_del(&handle->rule[i]->node.list);
1156 kfree(handle->rule[i]);
1162 static struct mlx5_flow_handle *
1163 create_flow_handle(struct fs_fte *fte,
1164 struct mlx5_flow_destination *dest,
1169 struct mlx5_flow_handle *handle;
1170 struct mlx5_flow_rule *rule = NULL;
1171 static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1172 static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1176 handle = alloc_handle((dest_num) ? dest_num : 1);
1178 return ERR_PTR(-ENOMEM);
1182 rule = find_flow_rule(fte, dest + i);
1184 refcount_inc(&rule->node.refcount);
1190 rule = alloc_rule(dest + i);
1194 /* Add dest to dests list- we need flow tables to be in the
1195 * end of the list for forward to next prio rules.
1197 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1199 dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1200 list_add(&rule->node.list, &fte->node.children);
1202 list_add_tail(&rule->node.list, &fte->node.children);
1206 type = dest[i].type ==
1207 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1208 *modify_mask |= type ? count : dst;
1211 handle->rule[i] = rule;
1212 } while (++i < dest_num);
1217 destroy_flow_handle(fte, handle, dest, i);
1218 return ERR_PTR(-ENOMEM);
1221 /* fte should not be deleted while calling this function */
1222 static struct mlx5_flow_handle *
1223 add_rule_fte(struct fs_fte *fte,
1224 struct mlx5_flow_group *fg,
1225 struct mlx5_flow_destination *dest,
1229 struct mlx5_flow_root_namespace *root;
1230 struct mlx5_flow_handle *handle;
1231 struct mlx5_flow_table *ft;
1232 int modify_mask = 0;
1234 bool new_rule = false;
1236 handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1238 if (IS_ERR(handle) || !new_rule)
1242 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1244 fs_get_obj(ft, fg->node.parent);
1245 root = find_root(&fg->node);
1246 if (!(fte->status & FS_FTE_STATUS_EXISTING))
1247 err = root->cmds->create_fte(get_dev(&ft->node),
1250 err = root->cmds->update_fte(get_dev(&ft->node), ft, fg->id,
1255 fte->node.active = true;
1256 fte->status |= FS_FTE_STATUS_EXISTING;
1257 atomic_inc(&fte->node.version);
1263 destroy_flow_handle(fte, handle, dest, handle->num_rules);
1264 return ERR_PTR(err);
1267 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table *ft,
1268 struct mlx5_flow_spec *spec)
1270 struct list_head *prev = &ft->node.children;
1271 struct mlx5_flow_group *fg;
1272 unsigned int candidate_index = 0;
1273 unsigned int group_size = 0;
1275 if (!ft->autogroup.active)
1276 return ERR_PTR(-ENOENT);
1278 if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1279 group_size = ft->autogroup.group_size;
1281 /* ft->max_fte == ft->autogroup.max_types */
1282 if (group_size == 0)
1285 /* sorted by start_index */
1286 fs_for_each_fg(fg, ft) {
1287 if (candidate_index + group_size > fg->start_index)
1288 candidate_index = fg->start_index + fg->max_ftes;
1291 prev = &fg->node.list;
1294 if (candidate_index + group_size > ft->max_fte)
1295 return ERR_PTR(-ENOSPC);
1297 fg = alloc_insert_flow_group(ft,
1298 spec->match_criteria_enable,
1299 spec->match_criteria,
1301 candidate_index + group_size - 1,
1306 if (group_size == ft->autogroup.group_size)
1307 ft->autogroup.num_groups++;
1313 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1314 struct mlx5_flow_group *fg)
1316 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1317 struct mlx5_core_dev *dev = get_dev(&ft->node);
1318 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1319 void *match_criteria_addr;
1320 u8 src_esw_owner_mask_on;
1325 in = kvzalloc(inlen, GFP_KERNEL);
1329 MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1330 fg->mask.match_criteria_enable);
1331 MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1332 MLX5_SET(create_flow_group_in, in, end_flow_index, fg->start_index +
1335 misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1337 src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1338 source_eswitch_owner_vhca_id);
1339 MLX5_SET(create_flow_group_in, in,
1340 source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1342 match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1343 in, match_criteria);
1344 memcpy(match_criteria_addr, fg->mask.match_criteria,
1345 sizeof(fg->mask.match_criteria));
1347 err = root->cmds->create_flow_group(dev, ft, in, &fg->id);
1349 fg->node.active = true;
1350 trace_mlx5_fs_add_fg(fg);
1357 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1358 struct mlx5_flow_destination *d2)
1360 if (d1->type == d2->type) {
1361 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1362 d1->vport.num == d2->vport.num) ||
1363 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1364 d1->ft == d2->ft) ||
1365 (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1366 d1->tir_num == d2->tir_num) ||
1367 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1368 d1->ft_num == d2->ft_num))
1375 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1376 struct mlx5_flow_destination *dest)
1378 struct mlx5_flow_rule *rule;
1380 list_for_each_entry(rule, &fte->node.children, node.list) {
1381 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1387 static bool check_conflicting_actions(u32 action1, u32 action2)
1389 u32 xored_actions = action1 ^ action2;
1391 /* if one rule only wants to count, it's ok */
1392 if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1393 action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1396 if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1397 MLX5_FLOW_CONTEXT_ACTION_ENCAP |
1398 MLX5_FLOW_CONTEXT_ACTION_DECAP |
1399 MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
1400 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1401 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1402 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1403 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1409 static int check_conflicting_ftes(struct fs_fte *fte, const struct mlx5_flow_act *flow_act)
1411 if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1412 mlx5_core_warn(get_dev(&fte->node),
1413 "Found two FTEs with conflicting actions\n");
1417 if (flow_act->has_flow_tag &&
1418 fte->action.flow_tag != flow_act->flow_tag) {
1419 mlx5_core_warn(get_dev(&fte->node),
1420 "FTE flow tag %u already exists with different flow tag %u\n",
1421 fte->action.flow_tag,
1422 flow_act->flow_tag);
1429 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1431 struct mlx5_flow_act *flow_act,
1432 struct mlx5_flow_destination *dest,
1436 struct mlx5_flow_handle *handle;
1441 ret = check_conflicting_ftes(fte, flow_act);
1443 return ERR_PTR(ret);
1445 old_action = fte->action.action;
1446 fte->action.action |= flow_act->action;
1447 handle = add_rule_fte(fte, fg, dest, dest_num,
1448 old_action != flow_act->action);
1449 if (IS_ERR(handle)) {
1450 fte->action.action = old_action;
1453 trace_mlx5_fs_set_fte(fte, false);
1455 for (i = 0; i < handle->num_rules; i++) {
1456 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1457 tree_add_node(&handle->rule[i]->node, &fte->node);
1458 trace_mlx5_fs_add_rule(handle->rule[i]);
1464 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_handle *handle)
1466 struct mlx5_flow_rule *dst;
1469 fs_get_obj(fte, handle->rule[0]->node.parent);
1471 fs_for_each_dst(dst, fte) {
1472 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1473 return dst->dest_attr.counter;
1479 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1481 if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1487 return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1488 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1491 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1493 struct mlx5_flow_table *ft)
1495 if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1496 return counter_is_valid(dest->counter, action);
1498 if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1501 if (!dest || ((dest->type ==
1502 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1503 (dest->ft->level <= ft->level)))
1509 struct list_head list;
1510 struct mlx5_flow_group *g;
1513 struct match_list_head {
1514 struct list_head list;
1515 struct match_list first;
1518 static void free_match_list(struct match_list_head *head)
1520 if (!list_empty(&head->list)) {
1521 struct match_list *iter, *match_tmp;
1523 list_del(&head->first.list);
1524 tree_put_node(&head->first.g->node);
1525 list_for_each_entry_safe(iter, match_tmp, &head->list,
1527 tree_put_node(&iter->g->node);
1528 list_del(&iter->list);
1534 static int build_match_list(struct match_list_head *match_head,
1535 struct mlx5_flow_table *ft,
1536 struct mlx5_flow_spec *spec)
1538 struct rhlist_head *tmp, *list;
1539 struct mlx5_flow_group *g;
1543 INIT_LIST_HEAD(&match_head->list);
1544 /* Collect all fgs which has a matching match_criteria */
1545 list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1546 /* RCU is atomic, we can't execute FW commands here */
1547 rhl_for_each_entry_rcu(g, tmp, list, hash) {
1548 struct match_list *curr_match;
1550 if (likely(list_empty(&match_head->list))) {
1551 if (!tree_get_node(&g->node))
1553 match_head->first.g = g;
1554 list_add_tail(&match_head->first.list,
1559 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1562 free_match_list(match_head);
1565 if (!tree_get_node(&g->node)) {
1570 list_add_tail(&curr_match->list, &match_head->list);
1576 static u64 matched_fgs_get_version(struct list_head *match_head)
1578 struct match_list *iter;
1581 list_for_each_entry(iter, match_head, list)
1582 version += (u64)atomic_read(&iter->g->node.version);
1586 static struct fs_fte *
1587 lookup_fte_locked(struct mlx5_flow_group *g,
1591 struct fs_fte *fte_tmp;
1594 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1596 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1597 fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1599 if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1603 if (!fte_tmp->node.active) {
1604 tree_put_node(&fte_tmp->node);
1609 nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1612 up_write_ref_node(&g->node);
1614 up_read_ref_node(&g->node);
1618 static struct mlx5_flow_handle *
1619 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1620 struct list_head *match_head,
1621 struct mlx5_flow_spec *spec,
1622 struct mlx5_flow_act *flow_act,
1623 struct mlx5_flow_destination *dest,
1627 struct mlx5_flow_steering *steering = get_steering(&ft->node);
1628 struct mlx5_flow_group *g;
1629 struct mlx5_flow_handle *rule;
1630 struct match_list *iter;
1631 bool take_write = false;
1636 fte = alloc_fte(ft, spec->match_value, flow_act);
1638 return ERR_PTR(-ENOMEM);
1640 search_again_locked:
1641 version = matched_fgs_get_version(match_head);
1642 /* Try to find a fg that already contains a matching fte */
1643 list_for_each_entry(iter, match_head, list) {
1644 struct fs_fte *fte_tmp;
1647 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1650 rule = add_rule_fg(g, spec->match_value,
1651 flow_act, dest, dest_num, fte_tmp);
1652 up_write_ref_node(&fte_tmp->node);
1653 tree_put_node(&fte_tmp->node);
1654 kmem_cache_free(steering->ftes_cache, fte);
1658 /* Check the ft version, for case that new flow group
1659 * was added while the fgs weren't locked
1661 if (atomic_read(&ft->node.version) != ft_version) {
1662 rule = ERR_PTR(-EAGAIN);
1666 /* Check the fgs version, for case the new FTE with the
1667 * same values was added while the fgs weren't locked
1669 if (version != matched_fgs_get_version(match_head)) {
1671 goto search_again_locked;
1674 list_for_each_entry(iter, match_head, list) {
1677 if (!g->node.active)
1680 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1682 err = insert_fte(g, fte);
1684 up_write_ref_node(&g->node);
1687 kmem_cache_free(steering->ftes_cache, fte);
1688 return ERR_PTR(err);
1691 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1692 up_write_ref_node(&g->node);
1693 rule = add_rule_fg(g, spec->match_value,
1694 flow_act, dest, dest_num, fte);
1695 up_write_ref_node(&fte->node);
1696 tree_put_node(&fte->node);
1699 rule = ERR_PTR(-ENOENT);
1701 kmem_cache_free(steering->ftes_cache, fte);
1705 static struct mlx5_flow_handle *
1706 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1707 struct mlx5_flow_spec *spec,
1708 struct mlx5_flow_act *flow_act,
1709 struct mlx5_flow_destination *dest,
1713 struct mlx5_flow_steering *steering = get_steering(&ft->node);
1714 struct mlx5_flow_group *g;
1715 struct mlx5_flow_handle *rule;
1716 struct match_list_head match_head;
1717 bool take_write = false;
1723 if (!check_valid_spec(spec))
1724 return ERR_PTR(-EINVAL);
1726 for (i = 0; i < dest_num; i++) {
1727 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1728 return ERR_PTR(-EINVAL);
1730 nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1731 search_again_locked:
1732 version = atomic_read(&ft->node.version);
1734 /* Collect all fgs which has a matching match_criteria */
1735 err = build_match_list(&match_head, ft, spec);
1738 up_write_ref_node(&ft->node);
1740 up_read_ref_node(&ft->node);
1741 return ERR_PTR(err);
1745 up_read_ref_node(&ft->node);
1747 rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1749 free_match_list(&match_head);
1750 if (!IS_ERR(rule) ||
1751 (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1753 up_write_ref_node(&ft->node);
1758 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1762 if (PTR_ERR(rule) == -EAGAIN ||
1763 version != atomic_read(&ft->node.version))
1764 goto search_again_locked;
1766 g = alloc_auto_flow_group(ft, spec);
1769 up_write_ref_node(&ft->node);
1773 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1774 up_write_ref_node(&ft->node);
1776 err = create_auto_flow_group(ft, g);
1778 goto err_release_fg;
1780 fte = alloc_fte(ft, spec->match_value, flow_act);
1783 goto err_release_fg;
1786 err = insert_fte(g, fte);
1788 kmem_cache_free(steering->ftes_cache, fte);
1789 goto err_release_fg;
1792 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1793 up_write_ref_node(&g->node);
1794 rule = add_rule_fg(g, spec->match_value, flow_act, dest,
1796 up_write_ref_node(&fte->node);
1797 tree_put_node(&fte->node);
1798 tree_put_node(&g->node);
1802 up_write_ref_node(&g->node);
1803 tree_put_node(&g->node);
1804 return ERR_PTR(err);
1807 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1809 return ((ft->type == FS_FT_NIC_RX) &&
1810 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1813 struct mlx5_flow_handle *
1814 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1815 struct mlx5_flow_spec *spec,
1816 struct mlx5_flow_act *flow_act,
1817 struct mlx5_flow_destination *dest,
1820 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1821 struct mlx5_flow_destination gen_dest = {};
1822 struct mlx5_flow_table *next_ft = NULL;
1823 struct mlx5_flow_handle *handle = NULL;
1824 u32 sw_action = flow_act->action;
1825 struct fs_prio *prio;
1827 fs_get_obj(prio, ft->node.parent);
1828 if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1829 if (!fwd_next_prio_supported(ft))
1830 return ERR_PTR(-EOPNOTSUPP);
1832 return ERR_PTR(-EINVAL);
1833 mutex_lock(&root->chain_lock);
1834 next_ft = find_next_chained_ft(prio);
1836 gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1837 gen_dest.ft = next_ft;
1840 flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1842 mutex_unlock(&root->chain_lock);
1843 return ERR_PTR(-EOPNOTSUPP);
1847 handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1849 if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1850 if (!IS_ERR_OR_NULL(handle) &&
1851 (list_empty(&handle->rule[0]->next_ft))) {
1852 mutex_lock(&next_ft->lock);
1853 list_add(&handle->rule[0]->next_ft,
1854 &next_ft->fwd_rules);
1855 mutex_unlock(&next_ft->lock);
1856 handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1858 mutex_unlock(&root->chain_lock);
1862 EXPORT_SYMBOL(mlx5_add_flow_rules);
1864 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1868 for (i = handle->num_rules - 1; i >= 0; i--)
1869 tree_remove_node(&handle->rule[i]->node);
1872 EXPORT_SYMBOL(mlx5_del_flow_rules);
1874 /* Assuming prio->node.children(flow tables) is sorted by level */
1875 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1877 struct fs_prio *prio;
1879 fs_get_obj(prio, ft->node.parent);
1881 if (!list_is_last(&ft->node.list, &prio->node.children))
1882 return list_next_entry(ft, node.list);
1883 return find_next_chained_ft(prio);
1886 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1888 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1889 struct mlx5_ft_underlay_qp *uqp;
1890 struct mlx5_flow_table *new_root_ft = NULL;
1894 if (root->root_ft != ft)
1897 new_root_ft = find_next_ft(ft);
1899 root->root_ft = NULL;
1903 if (list_empty(&root->underlay_qpns)) {
1904 /* Don't set any QPN (zero) in case QPN list is empty */
1906 err = root->cmds->update_root_ft(root->dev, new_root_ft,
1909 list_for_each_entry(uqp, &root->underlay_qpns, list) {
1911 err = root->cmds->update_root_ft(root->dev,
1920 mlx5_core_warn(root->dev,
1921 "Update root flow table of id(%u) qpn(%d) failed\n",
1924 root->root_ft = new_root_ft;
1929 /* Connect flow table from previous priority to
1930 * the next flow table.
1932 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1934 struct mlx5_core_dev *dev = get_dev(&ft->node);
1935 struct mlx5_flow_table *next_ft;
1936 struct fs_prio *prio;
1939 err = update_root_ft_destroy(ft);
1943 fs_get_obj(prio, ft->node.parent);
1944 if (!(list_first_entry(&prio->node.children,
1945 struct mlx5_flow_table,
1949 next_ft = find_next_ft(ft);
1950 err = connect_fwd_rules(dev, next_ft, ft);
1954 err = connect_prev_fts(dev, next_ft, prio);
1956 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1961 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1963 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1966 mutex_lock(&root->chain_lock);
1967 err = disconnect_flow_table(ft);
1969 mutex_unlock(&root->chain_lock);
1972 if (tree_remove_node(&ft->node))
1973 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1975 mutex_unlock(&root->chain_lock);
1979 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1981 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1983 if (tree_remove_node(&fg->node))
1984 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1988 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1989 enum mlx5_flow_namespace_type type)
1991 struct mlx5_flow_steering *steering = dev->priv.steering;
1992 struct mlx5_flow_root_namespace *root_ns;
1994 struct fs_prio *fs_prio;
1995 struct mlx5_flow_namespace *ns;
2001 case MLX5_FLOW_NAMESPACE_BYPASS:
2002 case MLX5_FLOW_NAMESPACE_LAG:
2003 case MLX5_FLOW_NAMESPACE_OFFLOADS:
2004 case MLX5_FLOW_NAMESPACE_ETHTOOL:
2005 case MLX5_FLOW_NAMESPACE_KERNEL:
2006 case MLX5_FLOW_NAMESPACE_LEFTOVERS:
2007 case MLX5_FLOW_NAMESPACE_ANCHOR:
2010 case MLX5_FLOW_NAMESPACE_FDB:
2011 if (steering->fdb_root_ns)
2012 return &steering->fdb_root_ns->ns;
2015 case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2016 if (steering->sniffer_rx_root_ns)
2017 return &steering->sniffer_rx_root_ns->ns;
2020 case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2021 if (steering->sniffer_tx_root_ns)
2022 return &steering->sniffer_tx_root_ns->ns;
2025 case MLX5_FLOW_NAMESPACE_EGRESS:
2026 if (steering->egress_root_ns)
2027 return &steering->egress_root_ns->ns;
2034 root_ns = steering->root_ns;
2038 fs_prio = find_prio(&root_ns->ns, prio);
2042 ns = list_first_entry(&fs_prio->node.children,
2048 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2050 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2051 enum mlx5_flow_namespace_type type,
2054 struct mlx5_flow_steering *steering = dev->priv.steering;
2056 if (!steering || vport >= MLX5_TOTAL_VPORTS(dev))
2060 case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2061 if (steering->esw_egress_root_ns &&
2062 steering->esw_egress_root_ns[vport])
2063 return &steering->esw_egress_root_ns[vport]->ns;
2066 case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2067 if (steering->esw_ingress_root_ns &&
2068 steering->esw_ingress_root_ns[vport])
2069 return &steering->esw_ingress_root_ns[vport]->ns;
2077 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2078 unsigned int prio, int num_levels)
2080 struct fs_prio *fs_prio;
2082 fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2084 return ERR_PTR(-ENOMEM);
2086 fs_prio->node.type = FS_TYPE_PRIO;
2087 tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2088 tree_add_node(&fs_prio->node, &ns->node);
2089 fs_prio->num_levels = num_levels;
2090 fs_prio->prio = prio;
2091 list_add_tail(&fs_prio->node.list, &ns->node.children);
2096 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2099 ns->node.type = FS_TYPE_NAMESPACE;
2104 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
2106 struct mlx5_flow_namespace *ns;
2108 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2110 return ERR_PTR(-ENOMEM);
2112 fs_init_namespace(ns);
2113 tree_init_node(&ns->node, NULL, del_sw_ns);
2114 tree_add_node(&ns->node, &prio->node);
2115 list_add_tail(&ns->node.list, &prio->node.children);
2120 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2121 struct init_tree_node *prio_metadata)
2123 struct fs_prio *fs_prio;
2126 for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2127 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2128 if (IS_ERR(fs_prio))
2129 return PTR_ERR(fs_prio);
2134 #define FLOW_TABLE_BIT_SZ 1
2135 #define GET_FLOW_TABLE_CAP(dev, offset) \
2136 ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) + \
2138 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2139 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2143 for (i = 0; i < caps->arr_sz; i++) {
2144 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2150 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2151 struct init_tree_node *init_node,
2152 struct fs_node *fs_parent_node,
2153 struct init_tree_node *init_parent_node,
2156 int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2157 flow_table_properties_nic_receive.
2159 struct mlx5_flow_namespace *fs_ns;
2160 struct fs_prio *fs_prio;
2161 struct fs_node *base;
2165 if (init_node->type == FS_TYPE_PRIO) {
2166 if ((init_node->min_ft_level > max_ft_level) ||
2167 !has_required_caps(steering->dev, &init_node->caps))
2170 fs_get_obj(fs_ns, fs_parent_node);
2171 if (init_node->num_leaf_prios)
2172 return create_leaf_prios(fs_ns, prio, init_node);
2173 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2174 if (IS_ERR(fs_prio))
2175 return PTR_ERR(fs_prio);
2176 base = &fs_prio->node;
2177 } else if (init_node->type == FS_TYPE_NAMESPACE) {
2178 fs_get_obj(fs_prio, fs_parent_node);
2179 fs_ns = fs_create_namespace(fs_prio);
2181 return PTR_ERR(fs_ns);
2182 base = &fs_ns->node;
2187 for (i = 0; i < init_node->ar_size; i++) {
2188 err = init_root_tree_recursive(steering, &init_node->children[i],
2189 base, init_node, prio);
2192 if (init_node->children[i].type == FS_TYPE_PRIO &&
2193 init_node->children[i].num_leaf_prios) {
2194 prio += init_node->children[i].num_leaf_prios;
2201 static int init_root_tree(struct mlx5_flow_steering *steering,
2202 struct init_tree_node *init_node,
2203 struct fs_node *fs_parent_node)
2206 struct mlx5_flow_namespace *fs_ns;
2209 fs_get_obj(fs_ns, fs_parent_node);
2210 for (i = 0; i < init_node->ar_size; i++) {
2211 err = init_root_tree_recursive(steering, &init_node->children[i],
2220 static struct mlx5_flow_root_namespace
2221 *create_root_ns(struct mlx5_flow_steering *steering,
2222 enum fs_flow_table_type table_type)
2224 const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2225 struct mlx5_flow_root_namespace *root_ns;
2226 struct mlx5_flow_namespace *ns;
2228 if (mlx5_accel_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2229 (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2230 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2232 /* Create the root namespace */
2233 root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2237 root_ns->dev = steering->dev;
2238 root_ns->table_type = table_type;
2239 root_ns->cmds = cmds;
2241 INIT_LIST_HEAD(&root_ns->underlay_qpns);
2244 fs_init_namespace(ns);
2245 mutex_init(&root_ns->chain_lock);
2246 tree_init_node(&ns->node, NULL, NULL);
2247 tree_add_node(&ns->node, NULL);
2252 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2254 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2256 struct fs_prio *prio;
2258 fs_for_each_prio(prio, ns) {
2259 /* This updates prio start_level and num_levels */
2260 set_prio_attrs_in_prio(prio, acc_level);
2261 acc_level += prio->num_levels;
2266 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2268 struct mlx5_flow_namespace *ns;
2269 int acc_level_ns = acc_level;
2271 prio->start_level = acc_level;
2272 fs_for_each_ns(ns, prio)
2273 /* This updates start_level and num_levels of ns's priority descendants */
2274 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2275 if (!prio->num_levels)
2276 prio->num_levels = acc_level_ns - prio->start_level;
2277 WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2280 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2282 struct mlx5_flow_namespace *ns = &root_ns->ns;
2283 struct fs_prio *prio;
2284 int start_level = 0;
2286 fs_for_each_prio(prio, ns) {
2287 set_prio_attrs_in_prio(prio, start_level);
2288 start_level += prio->num_levels;
2292 #define ANCHOR_PRIO 0
2293 #define ANCHOR_SIZE 1
2294 #define ANCHOR_LEVEL 0
2295 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2297 struct mlx5_flow_namespace *ns = NULL;
2298 struct mlx5_flow_table_attr ft_attr = {};
2299 struct mlx5_flow_table *ft;
2301 ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2305 ft_attr.max_fte = ANCHOR_SIZE;
2306 ft_attr.level = ANCHOR_LEVEL;
2307 ft_attr.prio = ANCHOR_PRIO;
2309 ft = mlx5_create_flow_table(ns, &ft_attr);
2311 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2317 static int init_root_ns(struct mlx5_flow_steering *steering)
2321 steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2322 if (!steering->root_ns)
2325 err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2329 set_prio_attrs(steering->root_ns);
2330 err = create_anchor_flow_table(steering);
2337 cleanup_root_ns(steering->root_ns);
2338 steering->root_ns = NULL;
2342 static void clean_tree(struct fs_node *node)
2345 struct fs_node *iter;
2346 struct fs_node *temp;
2348 tree_get_node(node);
2349 list_for_each_entry_safe(iter, temp, &node->children, list)
2351 tree_put_node(node);
2352 tree_remove_node(node);
2356 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2361 clean_tree(&root_ns->ns.node);
2364 static void cleanup_egress_acls_root_ns(struct mlx5_core_dev *dev)
2366 struct mlx5_flow_steering *steering = dev->priv.steering;
2369 if (!steering->esw_egress_root_ns)
2372 for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2373 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2375 kfree(steering->esw_egress_root_ns);
2376 steering->esw_egress_root_ns = NULL;
2379 static void cleanup_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2381 struct mlx5_flow_steering *steering = dev->priv.steering;
2384 if (!steering->esw_ingress_root_ns)
2387 for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2388 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2390 kfree(steering->esw_ingress_root_ns);
2391 steering->esw_ingress_root_ns = NULL;
2394 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2396 struct mlx5_flow_steering *steering = dev->priv.steering;
2398 cleanup_root_ns(steering->root_ns);
2399 cleanup_egress_acls_root_ns(dev);
2400 cleanup_ingress_acls_root_ns(dev);
2401 cleanup_root_ns(steering->fdb_root_ns);
2402 cleanup_root_ns(steering->sniffer_rx_root_ns);
2403 cleanup_root_ns(steering->sniffer_tx_root_ns);
2404 cleanup_root_ns(steering->egress_root_ns);
2405 mlx5_cleanup_fc_stats(dev);
2406 kmem_cache_destroy(steering->ftes_cache);
2407 kmem_cache_destroy(steering->fgs_cache);
2411 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2413 struct fs_prio *prio;
2415 steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2416 if (!steering->sniffer_tx_root_ns)
2419 /* Create single prio */
2420 prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2422 cleanup_root_ns(steering->sniffer_tx_root_ns);
2423 return PTR_ERR(prio);
2428 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2430 struct fs_prio *prio;
2432 steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2433 if (!steering->sniffer_rx_root_ns)
2436 /* Create single prio */
2437 prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2439 cleanup_root_ns(steering->sniffer_rx_root_ns);
2440 return PTR_ERR(prio);
2445 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2447 struct fs_prio *prio;
2449 steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2450 if (!steering->fdb_root_ns)
2453 prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 2);
2457 prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
2461 set_prio_attrs(steering->fdb_root_ns);
2465 cleanup_root_ns(steering->fdb_root_ns);
2466 steering->fdb_root_ns = NULL;
2467 return PTR_ERR(prio);
2470 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2472 struct fs_prio *prio;
2474 steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2475 if (!steering->esw_egress_root_ns[vport])
2479 prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2480 return PTR_ERR_OR_ZERO(prio);
2483 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2485 struct fs_prio *prio;
2487 steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2488 if (!steering->esw_ingress_root_ns[vport])
2492 prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2493 return PTR_ERR_OR_ZERO(prio);
2496 static int init_egress_acls_root_ns(struct mlx5_core_dev *dev)
2498 struct mlx5_flow_steering *steering = dev->priv.steering;
2502 steering->esw_egress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2503 sizeof(*steering->esw_egress_root_ns),
2505 if (!steering->esw_egress_root_ns)
2508 for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2509 err = init_egress_acl_root_ns(steering, i);
2511 goto cleanup_root_ns;
2517 for (i--; i >= 0; i--)
2518 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2519 kfree(steering->esw_egress_root_ns);
2520 steering->esw_egress_root_ns = NULL;
2524 static int init_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2526 struct mlx5_flow_steering *steering = dev->priv.steering;
2530 steering->esw_ingress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2531 sizeof(*steering->esw_ingress_root_ns),
2533 if (!steering->esw_ingress_root_ns)
2536 for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2537 err = init_ingress_acl_root_ns(steering, i);
2539 goto cleanup_root_ns;
2545 for (i--; i >= 0; i--)
2546 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2547 kfree(steering->esw_ingress_root_ns);
2548 steering->esw_ingress_root_ns = NULL;
2552 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2554 struct fs_prio *prio;
2556 steering->egress_root_ns = create_root_ns(steering,
2558 if (!steering->egress_root_ns)
2562 prio = fs_create_prio(&steering->egress_root_ns->ns, 0, 1);
2563 return PTR_ERR_OR_ZERO(prio);
2566 int mlx5_init_fs(struct mlx5_core_dev *dev)
2568 struct mlx5_flow_steering *steering;
2571 err = mlx5_init_fc_stats(dev);
2575 steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2578 steering->dev = dev;
2579 dev->priv.steering = steering;
2581 steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2582 sizeof(struct mlx5_flow_group), 0,
2584 steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2586 if (!steering->ftes_cache || !steering->fgs_cache) {
2591 if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2592 (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2593 ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2594 MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2595 MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2596 err = init_root_ns(steering);
2601 if (MLX5_ESWITCH_MANAGER(dev)) {
2602 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2603 err = init_fdb_root_ns(steering);
2607 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2608 err = init_egress_acls_root_ns(dev);
2612 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2613 err = init_ingress_acls_root_ns(dev);
2619 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2620 err = init_sniffer_rx_root_ns(steering);
2625 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2626 err = init_sniffer_tx_root_ns(steering);
2631 if (MLX5_IPSEC_DEV(dev)) {
2632 err = init_egress_root_ns(steering);
2639 mlx5_cleanup_fs(dev);
2643 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2645 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2646 struct mlx5_ft_underlay_qp *new_uqp;
2649 new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
2653 mutex_lock(&root->chain_lock);
2655 if (!root->root_ft) {
2657 goto update_ft_fail;
2660 err = root->cmds->update_root_ft(dev, root->root_ft, underlay_qpn,
2663 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
2665 goto update_ft_fail;
2668 new_uqp->qpn = underlay_qpn;
2669 list_add_tail(&new_uqp->list, &root->underlay_qpns);
2671 mutex_unlock(&root->chain_lock);
2676 mutex_unlock(&root->chain_lock);
2680 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
2682 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2684 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2685 struct mlx5_ft_underlay_qp *uqp;
2689 mutex_lock(&root->chain_lock);
2690 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2691 if (uqp->qpn == underlay_qpn) {
2698 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
2704 err = root->cmds->update_root_ft(dev, root->root_ft, underlay_qpn,
2707 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
2710 list_del(&uqp->list);
2711 mutex_unlock(&root->chain_lock);
2717 mutex_unlock(&root->chain_lock);
2720 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);