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/vport.h>
36 #include <linux/mlx5/eswitch.h>
38 #include "mlx5_core.h"
41 #include "diag/fs_tracepoint.h"
42 #include "accel/ipsec.h"
43 #include "fpga/ipsec.h"
46 #define INIT_TREE_NODE_ARRAY_SIZE(...) (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
47 sizeof(struct init_tree_node))
49 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
50 ...) {.type = FS_TYPE_PRIO,\
51 .min_ft_level = min_level_val,\
52 .num_levels = num_levels_val,\
53 .num_leaf_prios = num_prios_val,\
55 .children = (struct init_tree_node[]) {__VA_ARGS__},\
56 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
59 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
60 ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
63 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
64 .children = (struct init_tree_node[]) {__VA_ARGS__},\
65 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
68 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
71 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
73 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
74 .caps = (long[]) {__VA_ARGS__} }
76 #define FS_CHAINING_CAPS FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
77 FS_CAP(flow_table_properties_nic_receive.modify_root), \
78 FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
79 FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
81 #define FS_CHAINING_CAPS_EGRESS \
83 FS_CAP(flow_table_properties_nic_transmit.flow_modify_en), \
84 FS_CAP(flow_table_properties_nic_transmit.modify_root), \
85 FS_CAP(flow_table_properties_nic_transmit \
86 .identified_miss_table_mode), \
87 FS_CAP(flow_table_properties_nic_transmit.flow_table_modify))
89 #define LEFTOVERS_NUM_LEVELS 1
90 #define LEFTOVERS_NUM_PRIOS 1
92 #define BY_PASS_PRIO_NUM_LEVELS 1
93 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
96 #define ETHTOOL_PRIO_NUM_LEVELS 1
97 #define ETHTOOL_NUM_PRIOS 11
98 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
99 /* Vlan, mac, ttc, inner ttc, aRFS */
100 #define KERNEL_NIC_PRIO_NUM_LEVELS 5
101 #define KERNEL_NIC_NUM_PRIOS 1
102 /* One more level for tc */
103 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
105 #define KERNEL_NIC_TC_NUM_PRIOS 1
106 #define KERNEL_NIC_TC_NUM_LEVELS 2
108 #define ANCHOR_NUM_LEVELS 1
109 #define ANCHOR_NUM_PRIOS 1
110 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
112 #define OFFLOADS_MAX_FT 1
113 #define OFFLOADS_NUM_PRIOS 1
114 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
116 #define LAG_PRIO_NUM_LEVELS 1
117 #define LAG_NUM_PRIOS 1
118 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
125 static struct init_tree_node {
126 enum fs_node_type type;
127 struct init_tree_node *children;
129 struct node_caps caps;
135 .type = FS_TYPE_NAMESPACE,
137 .children = (struct init_tree_node[]) {
138 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
140 ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
141 BY_PASS_PRIO_NUM_LEVELS))),
142 ADD_PRIO(0, LAG_MIN_LEVEL, 0,
144 ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
145 LAG_PRIO_NUM_LEVELS))),
146 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
147 ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
148 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
150 ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
151 ETHTOOL_PRIO_NUM_LEVELS))),
152 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
153 ADD_NS(ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS, KERNEL_NIC_TC_NUM_LEVELS),
154 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
155 KERNEL_NIC_PRIO_NUM_LEVELS))),
156 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
158 ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
159 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
160 ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
164 static struct init_tree_node egress_root_fs = {
165 .type = FS_TYPE_NAMESPACE,
167 .children = (struct init_tree_node[]) {
168 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
169 FS_CHAINING_CAPS_EGRESS,
170 ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
171 BY_PASS_PRIO_NUM_LEVELS))),
175 enum fs_i_lock_class {
181 static const struct rhashtable_params rhash_fte = {
182 .key_len = FIELD_SIZEOF(struct fs_fte, val),
183 .key_offset = offsetof(struct fs_fte, val),
184 .head_offset = offsetof(struct fs_fte, hash),
185 .automatic_shrinking = true,
189 static const struct rhashtable_params rhash_fg = {
190 .key_len = FIELD_SIZEOF(struct mlx5_flow_group, mask),
191 .key_offset = offsetof(struct mlx5_flow_group, mask),
192 .head_offset = offsetof(struct mlx5_flow_group, hash),
193 .automatic_shrinking = true,
198 static void del_hw_flow_table(struct fs_node *node);
199 static void del_hw_flow_group(struct fs_node *node);
200 static void del_hw_fte(struct fs_node *node);
201 static void del_sw_flow_table(struct fs_node *node);
202 static void del_sw_flow_group(struct fs_node *node);
203 static void del_sw_fte(struct fs_node *node);
204 static void del_sw_prio(struct fs_node *node);
205 static void del_sw_ns(struct fs_node *node);
206 /* Delete rule (destination) is special case that
207 * requires to lock the FTE for all the deletion process.
209 static void del_sw_hw_rule(struct fs_node *node);
210 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
211 struct mlx5_flow_destination *d2);
212 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
213 static struct mlx5_flow_rule *
214 find_flow_rule(struct fs_fte *fte,
215 struct mlx5_flow_destination *dest);
217 static void tree_init_node(struct fs_node *node,
218 void (*del_hw_func)(struct fs_node *),
219 void (*del_sw_func)(struct fs_node *))
221 refcount_set(&node->refcount, 1);
222 INIT_LIST_HEAD(&node->list);
223 INIT_LIST_HEAD(&node->children);
224 init_rwsem(&node->lock);
225 node->del_hw_func = del_hw_func;
226 node->del_sw_func = del_sw_func;
227 node->active = false;
230 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
233 refcount_inc(&parent->refcount);
234 node->parent = parent;
236 /* Parent is the root */
240 node->root = parent->root;
243 static int tree_get_node(struct fs_node *node)
245 return refcount_inc_not_zero(&node->refcount);
248 static void nested_down_read_ref_node(struct fs_node *node,
249 enum fs_i_lock_class class)
252 down_read_nested(&node->lock, class);
253 refcount_inc(&node->refcount);
257 static void nested_down_write_ref_node(struct fs_node *node,
258 enum fs_i_lock_class class)
261 down_write_nested(&node->lock, class);
262 refcount_inc(&node->refcount);
266 static void down_write_ref_node(struct fs_node *node)
269 down_write(&node->lock);
270 refcount_inc(&node->refcount);
274 static void up_read_ref_node(struct fs_node *node)
276 refcount_dec(&node->refcount);
277 up_read(&node->lock);
280 static void up_write_ref_node(struct fs_node *node)
282 refcount_dec(&node->refcount);
283 up_write(&node->lock);
286 static void tree_put_node(struct fs_node *node)
288 struct fs_node *parent_node = node->parent;
290 if (refcount_dec_and_test(&node->refcount)) {
291 if (node->del_hw_func)
292 node->del_hw_func(node);
294 /* Only root namespace doesn't have parent and we just
295 * need to free its node.
297 down_write_ref_node(parent_node);
298 list_del_init(&node->list);
299 if (node->del_sw_func)
300 node->del_sw_func(node);
301 up_write_ref_node(parent_node);
307 if (!node && parent_node)
308 tree_put_node(parent_node);
311 static int tree_remove_node(struct fs_node *node)
313 if (refcount_read(&node->refcount) > 1) {
314 refcount_dec(&node->refcount);
321 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
324 struct fs_prio *iter_prio;
326 fs_for_each_prio(iter_prio, ns) {
327 if (iter_prio->prio == prio)
334 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
338 for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
339 if (spec->match_value[i] & ~spec->match_criteria[i]) {
340 pr_warn("mlx5_core: match_value differs from match_criteria\n");
347 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
349 struct fs_node *root;
350 struct mlx5_flow_namespace *ns;
354 if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
355 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
359 ns = container_of(root, struct mlx5_flow_namespace, node);
360 return container_of(ns, struct mlx5_flow_root_namespace, ns);
363 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
365 struct mlx5_flow_root_namespace *root = find_root(node);
368 return root->dev->priv.steering;
372 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
374 struct mlx5_flow_root_namespace *root = find_root(node);
381 static void del_sw_ns(struct fs_node *node)
386 static void del_sw_prio(struct fs_node *node)
391 static void del_hw_flow_table(struct fs_node *node)
393 struct mlx5_flow_root_namespace *root;
394 struct mlx5_flow_table *ft;
395 struct mlx5_core_dev *dev;
398 fs_get_obj(ft, node);
399 dev = get_dev(&ft->node);
400 root = find_root(&ft->node);
403 err = root->cmds->destroy_flow_table(dev, ft);
405 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
409 static void del_sw_flow_table(struct fs_node *node)
411 struct mlx5_flow_table *ft;
412 struct fs_prio *prio;
414 fs_get_obj(ft, node);
416 rhltable_destroy(&ft->fgs_hash);
417 fs_get_obj(prio, ft->node.parent);
422 static void del_sw_hw_rule(struct fs_node *node)
424 struct mlx5_flow_root_namespace *root;
425 struct mlx5_flow_rule *rule;
426 struct mlx5_flow_table *ft;
427 struct mlx5_flow_group *fg;
430 struct mlx5_core_dev *dev = get_dev(node);
432 bool update_fte = false;
434 fs_get_obj(rule, node);
435 fs_get_obj(fte, rule->node.parent);
436 fs_get_obj(fg, fte->node.parent);
437 fs_get_obj(ft, fg->node.parent);
438 trace_mlx5_fs_del_rule(rule);
439 if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
440 mutex_lock(&rule->dest_attr.ft->lock);
441 list_del(&rule->next_ft);
442 mutex_unlock(&rule->dest_attr.ft->lock);
445 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER &&
447 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
448 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
449 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
454 if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
456 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
460 root = find_root(&ft->node);
461 if (update_fte && fte->dests_size) {
462 err = root->cmds->update_fte(dev, ft, fg->id, modify_mask, fte);
465 "%s can't del rule fg id=%d fte_index=%d\n",
466 __func__, fg->id, fte->index);
471 static void del_hw_fte(struct fs_node *node)
473 struct mlx5_flow_root_namespace *root;
474 struct mlx5_flow_table *ft;
475 struct mlx5_flow_group *fg;
476 struct mlx5_core_dev *dev;
480 fs_get_obj(fte, node);
481 fs_get_obj(fg, fte->node.parent);
482 fs_get_obj(ft, fg->node.parent);
484 trace_mlx5_fs_del_fte(fte);
485 dev = get_dev(&ft->node);
486 root = find_root(&ft->node);
488 err = root->cmds->delete_fte(dev, ft, fte);
491 "flow steering can't delete fte in index %d of flow group id %d\n",
496 static void del_sw_fte(struct fs_node *node)
498 struct mlx5_flow_steering *steering = get_steering(node);
499 struct mlx5_flow_group *fg;
503 fs_get_obj(fte, node);
504 fs_get_obj(fg, fte->node.parent);
506 err = rhashtable_remove_fast(&fg->ftes_hash,
510 ida_simple_remove(&fg->fte_allocator, fte->index - fg->start_index);
511 kmem_cache_free(steering->ftes_cache, fte);
514 static void del_hw_flow_group(struct fs_node *node)
516 struct mlx5_flow_root_namespace *root;
517 struct mlx5_flow_group *fg;
518 struct mlx5_flow_table *ft;
519 struct mlx5_core_dev *dev;
521 fs_get_obj(fg, node);
522 fs_get_obj(ft, fg->node.parent);
523 dev = get_dev(&ft->node);
524 trace_mlx5_fs_del_fg(fg);
526 root = find_root(&ft->node);
527 if (fg->node.active && root->cmds->destroy_flow_group(dev, ft, fg->id))
528 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
532 static void del_sw_flow_group(struct fs_node *node)
534 struct mlx5_flow_steering *steering = get_steering(node);
535 struct mlx5_flow_group *fg;
536 struct mlx5_flow_table *ft;
539 fs_get_obj(fg, node);
540 fs_get_obj(ft, fg->node.parent);
542 rhashtable_destroy(&fg->ftes_hash);
543 ida_destroy(&fg->fte_allocator);
544 if (ft->autogroup.active)
545 ft->autogroup.num_groups--;
546 err = rhltable_remove(&ft->fgs_hash,
550 kmem_cache_free(steering->fgs_cache, fg);
553 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
558 index = ida_simple_get(&fg->fte_allocator, 0, fg->max_ftes, GFP_KERNEL);
562 fte->index = index + fg->start_index;
563 ret = rhashtable_insert_fast(&fg->ftes_hash,
569 tree_add_node(&fte->node, &fg->node);
570 list_add_tail(&fte->node.list, &fg->node.children);
574 ida_simple_remove(&fg->fte_allocator, index);
578 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
580 struct mlx5_flow_act *flow_act)
582 struct mlx5_flow_steering *steering = get_steering(&ft->node);
585 fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
587 return ERR_PTR(-ENOMEM);
589 memcpy(fte->val, match_value, sizeof(fte->val));
590 fte->node.type = FS_TYPE_FLOW_ENTRY;
591 fte->action = *flow_act;
593 tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
598 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
599 struct mlx5_flow_group *fg)
601 rhashtable_destroy(&fg->ftes_hash);
602 kmem_cache_free(steering->fgs_cache, fg);
605 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
606 u8 match_criteria_enable,
607 void *match_criteria,
611 struct mlx5_flow_group *fg;
614 fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
616 return ERR_PTR(-ENOMEM);
618 ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
620 kmem_cache_free(steering->fgs_cache, fg);
623 ida_init(&fg->fte_allocator);
624 fg->mask.match_criteria_enable = match_criteria_enable;
625 memcpy(&fg->mask.match_criteria, match_criteria,
626 sizeof(fg->mask.match_criteria));
627 fg->node.type = FS_TYPE_FLOW_GROUP;
628 fg->start_index = start_index;
629 fg->max_ftes = end_index - start_index + 1;
634 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
635 u8 match_criteria_enable,
636 void *match_criteria,
639 struct list_head *prev)
641 struct mlx5_flow_steering *steering = get_steering(&ft->node);
642 struct mlx5_flow_group *fg;
645 fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
646 start_index, end_index);
650 /* initialize refcnt, add to parent list */
651 ret = rhltable_insert(&ft->fgs_hash,
655 dealloc_flow_group(steering, fg);
659 tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
660 tree_add_node(&fg->node, &ft->node);
661 /* Add node to group list */
662 list_add(&fg->node.list, prev);
663 atomic_inc(&ft->node.version);
668 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
669 enum fs_flow_table_type table_type,
670 enum fs_flow_table_op_mod op_mod,
673 struct mlx5_flow_table *ft;
676 ft = kzalloc(sizeof(*ft), GFP_KERNEL);
678 return ERR_PTR(-ENOMEM);
680 ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
687 ft->node.type = FS_TYPE_FLOW_TABLE;
689 ft->type = table_type;
691 ft->max_fte = max_fte;
693 INIT_LIST_HEAD(&ft->fwd_rules);
694 mutex_init(&ft->lock);
699 /* If reverse is false, then we search for the first flow table in the
700 * root sub-tree from start(closest from right), else we search for the
701 * last flow table in the root sub-tree till start(closest from left).
703 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node *root,
704 struct list_head *start,
707 #define list_advance_entry(pos, reverse) \
708 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
710 #define list_for_each_advance_continue(pos, head, reverse) \
711 for (pos = list_advance_entry(pos, reverse); \
712 &pos->list != (head); \
713 pos = list_advance_entry(pos, reverse))
715 struct fs_node *iter = list_entry(start, struct fs_node, list);
716 struct mlx5_flow_table *ft = NULL;
718 if (!root || root->type == FS_TYPE_PRIO_CHAINS)
721 list_for_each_advance_continue(iter, &root->children, reverse) {
722 if (iter->type == FS_TYPE_FLOW_TABLE) {
723 fs_get_obj(ft, iter);
726 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
734 /* If reverse if false then return the first flow table in next priority of
735 * prio in the tree, else return the last flow table in the previous priority
736 * of prio in the tree.
738 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
740 struct mlx5_flow_table *ft = NULL;
741 struct fs_node *curr_node;
742 struct fs_node *parent;
744 parent = prio->node.parent;
745 curr_node = &prio->node;
746 while (!ft && parent) {
747 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
749 parent = curr_node->parent;
754 /* Assuming all the tree is locked by mutex chain lock */
755 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
757 return find_closest_ft(prio, false);
760 /* Assuming all the tree is locked by mutex chain lock */
761 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
763 return find_closest_ft(prio, true);
766 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
767 struct fs_prio *prio,
768 struct mlx5_flow_table *ft)
770 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
771 struct mlx5_flow_table *iter;
775 fs_for_each_ft(iter, prio) {
777 err = root->cmds->modify_flow_table(dev, iter, ft);
779 mlx5_core_warn(dev, "Failed to modify flow table %d\n",
781 /* The driver is out of sync with the FW */
790 /* Connect flow tables from previous priority of prio to ft */
791 static int connect_prev_fts(struct mlx5_core_dev *dev,
792 struct mlx5_flow_table *ft,
793 struct fs_prio *prio)
795 struct mlx5_flow_table *prev_ft;
797 prev_ft = find_prev_chained_ft(prio);
799 struct fs_prio *prev_prio;
801 fs_get_obj(prev_prio, prev_ft->node.parent);
802 return connect_fts_in_prio(dev, prev_prio, ft);
807 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
810 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
811 struct mlx5_ft_underlay_qp *uqp;
812 int min_level = INT_MAX;
817 min_level = root->root_ft->level;
819 if (ft->level >= min_level)
822 if (list_empty(&root->underlay_qpns)) {
823 /* Don't set any QPN (zero) in case QPN list is empty */
825 err = root->cmds->update_root_ft(root->dev, ft, qpn, false);
827 list_for_each_entry(uqp, &root->underlay_qpns, list) {
829 err = root->cmds->update_root_ft(root->dev, ft,
837 mlx5_core_warn(root->dev,
838 "Update root flow table of id(%u) qpn(%d) failed\n",
846 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
847 struct mlx5_flow_destination *dest)
849 struct mlx5_flow_root_namespace *root;
850 struct mlx5_flow_table *ft;
851 struct mlx5_flow_group *fg;
853 int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
856 fs_get_obj(fte, rule->node.parent);
857 if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
859 down_write_ref_node(&fte->node);
860 fs_get_obj(fg, fte->node.parent);
861 fs_get_obj(ft, fg->node.parent);
863 memcpy(&rule->dest_attr, dest, sizeof(*dest));
864 root = find_root(&ft->node);
865 err = root->cmds->update_fte(get_dev(&ft->node), ft, fg->id,
867 up_write_ref_node(&fte->node);
872 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
873 struct mlx5_flow_destination *new_dest,
874 struct mlx5_flow_destination *old_dest)
879 if (handle->num_rules != 1)
881 return _mlx5_modify_rule_destination(handle->rule[0],
885 for (i = 0; i < handle->num_rules; i++) {
886 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
887 return _mlx5_modify_rule_destination(handle->rule[i],
894 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */
895 static int connect_fwd_rules(struct mlx5_core_dev *dev,
896 struct mlx5_flow_table *new_next_ft,
897 struct mlx5_flow_table *old_next_ft)
899 struct mlx5_flow_destination dest = {};
900 struct mlx5_flow_rule *iter;
903 /* new_next_ft and old_next_ft could be NULL only
904 * when we create/destroy the anchor flow table.
906 if (!new_next_ft || !old_next_ft)
909 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
910 dest.ft = new_next_ft;
912 mutex_lock(&old_next_ft->lock);
913 list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
914 mutex_unlock(&old_next_ft->lock);
915 list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
916 err = _mlx5_modify_rule_destination(iter, &dest);
918 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
924 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
925 struct fs_prio *prio)
927 struct mlx5_flow_table *next_ft;
930 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
932 if (list_empty(&prio->node.children)) {
933 err = connect_prev_fts(dev, ft, prio);
937 next_ft = find_next_chained_ft(prio);
938 err = connect_fwd_rules(dev, ft, next_ft);
943 if (MLX5_CAP_FLOWTABLE(dev,
944 flow_table_properties_nic_receive.modify_root))
945 err = update_root_ft_create(ft, prio);
949 static void list_add_flow_table(struct mlx5_flow_table *ft,
950 struct fs_prio *prio)
952 struct list_head *prev = &prio->node.children;
953 struct mlx5_flow_table *iter;
955 fs_for_each_ft(iter, prio) {
956 if (iter->level > ft->level)
958 prev = &iter->node.list;
960 list_add(&ft->node.list, prev);
963 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
964 struct mlx5_flow_table_attr *ft_attr,
965 enum fs_flow_table_op_mod op_mod,
968 struct mlx5_flow_root_namespace *root = find_root(&ns->node);
969 struct mlx5_flow_table *next_ft = NULL;
970 struct fs_prio *fs_prio = NULL;
971 struct mlx5_flow_table *ft;
976 pr_err("mlx5: flow steering failed to find root of namespace\n");
977 return ERR_PTR(-ENODEV);
980 mutex_lock(&root->chain_lock);
981 fs_prio = find_prio(ns, ft_attr->prio);
986 if (ft_attr->level >= fs_prio->num_levels) {
990 /* The level is related to the
991 * priority level range.
993 ft_attr->level += fs_prio->start_level;
994 ft = alloc_flow_table(ft_attr->level,
996 ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
998 op_mod, ft_attr->flags);
1004 tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1005 log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
1006 next_ft = find_next_chained_ft(fs_prio);
1007 err = root->cmds->create_flow_table(root->dev, ft->vport, ft->op_mod,
1008 ft->type, ft->level, log_table_sz,
1009 next_ft, &ft->id, ft->flags);
1013 err = connect_flow_table(root->dev, ft, fs_prio);
1016 ft->node.active = true;
1017 down_write_ref_node(&fs_prio->node);
1018 tree_add_node(&ft->node, &fs_prio->node);
1019 list_add_flow_table(ft, fs_prio);
1021 up_write_ref_node(&fs_prio->node);
1022 mutex_unlock(&root->chain_lock);
1025 root->cmds->destroy_flow_table(root->dev, ft);
1029 mutex_unlock(&root->chain_lock);
1030 return ERR_PTR(err);
1033 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1034 struct mlx5_flow_table_attr *ft_attr)
1036 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1039 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1040 int prio, int max_fte,
1041 u32 level, u16 vport)
1043 struct mlx5_flow_table_attr ft_attr = {};
1045 ft_attr.max_fte = max_fte;
1046 ft_attr.level = level;
1047 ft_attr.prio = prio;
1049 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1052 struct mlx5_flow_table*
1053 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1054 int prio, u32 level)
1056 struct mlx5_flow_table_attr ft_attr = {};
1058 ft_attr.level = level;
1059 ft_attr.prio = prio;
1060 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1062 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1064 struct mlx5_flow_table*
1065 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1067 int num_flow_table_entries,
1072 struct mlx5_flow_table_attr ft_attr = {};
1073 struct mlx5_flow_table *ft;
1075 if (max_num_groups > num_flow_table_entries)
1076 return ERR_PTR(-EINVAL);
1078 ft_attr.max_fte = num_flow_table_entries;
1079 ft_attr.prio = prio;
1080 ft_attr.level = level;
1081 ft_attr.flags = flags;
1083 ft = mlx5_create_flow_table(ns, &ft_attr);
1087 ft->autogroup.active = true;
1088 ft->autogroup.required_groups = max_num_groups;
1092 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1094 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1097 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1098 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1099 fg_in, match_criteria);
1100 u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1102 match_criteria_enable);
1103 int start_index = MLX5_GET(create_flow_group_in, fg_in,
1105 int end_index = MLX5_GET(create_flow_group_in, fg_in,
1107 struct mlx5_core_dev *dev = get_dev(&ft->node);
1108 struct mlx5_flow_group *fg;
1111 if (ft->autogroup.active)
1112 return ERR_PTR(-EPERM);
1114 down_write_ref_node(&ft->node);
1115 fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1116 start_index, end_index,
1117 ft->node.children.prev);
1118 up_write_ref_node(&ft->node);
1122 err = root->cmds->create_flow_group(dev, ft, fg_in, &fg->id);
1124 tree_put_node(&fg->node);
1125 return ERR_PTR(err);
1127 trace_mlx5_fs_add_fg(fg);
1128 fg->node.active = true;
1133 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1135 struct mlx5_flow_rule *rule;
1137 rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1141 INIT_LIST_HEAD(&rule->next_ft);
1142 rule->node.type = FS_TYPE_FLOW_DEST;
1144 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1149 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1151 struct mlx5_flow_handle *handle;
1153 handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1157 handle->num_rules = num_rules;
1162 static void destroy_flow_handle(struct fs_fte *fte,
1163 struct mlx5_flow_handle *handle,
1164 struct mlx5_flow_destination *dest,
1168 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1170 list_del(&handle->rule[i]->node.list);
1171 kfree(handle->rule[i]);
1177 static struct mlx5_flow_handle *
1178 create_flow_handle(struct fs_fte *fte,
1179 struct mlx5_flow_destination *dest,
1184 struct mlx5_flow_handle *handle;
1185 struct mlx5_flow_rule *rule = NULL;
1186 static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1187 static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1191 handle = alloc_handle((dest_num) ? dest_num : 1);
1193 return ERR_PTR(-ENOMEM);
1197 rule = find_flow_rule(fte, dest + i);
1199 refcount_inc(&rule->node.refcount);
1205 rule = alloc_rule(dest + i);
1209 /* Add dest to dests list- we need flow tables to be in the
1210 * end of the list for forward to next prio rules.
1212 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1214 dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1215 list_add(&rule->node.list, &fte->node.children);
1217 list_add_tail(&rule->node.list, &fte->node.children);
1221 type = dest[i].type ==
1222 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1223 *modify_mask |= type ? count : dst;
1226 handle->rule[i] = rule;
1227 } while (++i < dest_num);
1232 destroy_flow_handle(fte, handle, dest, i);
1233 return ERR_PTR(-ENOMEM);
1236 /* fte should not be deleted while calling this function */
1237 static struct mlx5_flow_handle *
1238 add_rule_fte(struct fs_fte *fte,
1239 struct mlx5_flow_group *fg,
1240 struct mlx5_flow_destination *dest,
1244 struct mlx5_flow_root_namespace *root;
1245 struct mlx5_flow_handle *handle;
1246 struct mlx5_flow_table *ft;
1247 int modify_mask = 0;
1249 bool new_rule = false;
1251 handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1253 if (IS_ERR(handle) || !new_rule)
1257 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1259 fs_get_obj(ft, fg->node.parent);
1260 root = find_root(&fg->node);
1261 if (!(fte->status & FS_FTE_STATUS_EXISTING))
1262 err = root->cmds->create_fte(get_dev(&ft->node),
1265 err = root->cmds->update_fte(get_dev(&ft->node), ft, fg->id,
1270 fte->node.active = true;
1271 fte->status |= FS_FTE_STATUS_EXISTING;
1272 atomic_inc(&fte->node.version);
1278 destroy_flow_handle(fte, handle, dest, handle->num_rules);
1279 return ERR_PTR(err);
1282 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table *ft,
1283 struct mlx5_flow_spec *spec)
1285 struct list_head *prev = &ft->node.children;
1286 struct mlx5_flow_group *fg;
1287 unsigned int candidate_index = 0;
1288 unsigned int group_size = 0;
1290 if (!ft->autogroup.active)
1291 return ERR_PTR(-ENOENT);
1293 if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1294 /* We save place for flow groups in addition to max types */
1295 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1297 /* ft->max_fte == ft->autogroup.max_types */
1298 if (group_size == 0)
1301 /* sorted by start_index */
1302 fs_for_each_fg(fg, ft) {
1303 if (candidate_index + group_size > fg->start_index)
1304 candidate_index = fg->start_index + fg->max_ftes;
1307 prev = &fg->node.list;
1310 if (candidate_index + group_size > ft->max_fte)
1311 return ERR_PTR(-ENOSPC);
1313 fg = alloc_insert_flow_group(ft,
1314 spec->match_criteria_enable,
1315 spec->match_criteria,
1317 candidate_index + group_size - 1,
1322 ft->autogroup.num_groups++;
1328 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1329 struct mlx5_flow_group *fg)
1331 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1332 struct mlx5_core_dev *dev = get_dev(&ft->node);
1333 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1334 void *match_criteria_addr;
1335 u8 src_esw_owner_mask_on;
1340 in = kvzalloc(inlen, GFP_KERNEL);
1344 MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1345 fg->mask.match_criteria_enable);
1346 MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1347 MLX5_SET(create_flow_group_in, in, end_flow_index, fg->start_index +
1350 misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1352 src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1353 source_eswitch_owner_vhca_id);
1354 MLX5_SET(create_flow_group_in, in,
1355 source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1357 match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1358 in, match_criteria);
1359 memcpy(match_criteria_addr, fg->mask.match_criteria,
1360 sizeof(fg->mask.match_criteria));
1362 err = root->cmds->create_flow_group(dev, ft, in, &fg->id);
1364 fg->node.active = true;
1365 trace_mlx5_fs_add_fg(fg);
1372 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1373 struct mlx5_flow_destination *d2)
1375 if (d1->type == d2->type) {
1376 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1377 d1->vport.num == d2->vport.num &&
1378 d1->vport.flags == d2->vport.flags &&
1379 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1380 (d1->vport.reformat_id == d2->vport.reformat_id) : true)) ||
1381 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1382 d1->ft == d2->ft) ||
1383 (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1384 d1->tir_num == d2->tir_num) ||
1385 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1386 d1->ft_num == d2->ft_num))
1393 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1394 struct mlx5_flow_destination *dest)
1396 struct mlx5_flow_rule *rule;
1398 list_for_each_entry(rule, &fte->node.children, node.list) {
1399 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1405 static bool check_conflicting_actions(u32 action1, u32 action2)
1407 u32 xored_actions = action1 ^ action2;
1409 /* if one rule only wants to count, it's ok */
1410 if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1411 action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1414 if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1415 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1416 MLX5_FLOW_CONTEXT_ACTION_DECAP |
1417 MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
1418 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1419 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1420 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1421 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1427 static int check_conflicting_ftes(struct fs_fte *fte, const struct mlx5_flow_act *flow_act)
1429 if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1430 mlx5_core_warn(get_dev(&fte->node),
1431 "Found two FTEs with conflicting actions\n");
1435 if ((flow_act->flags & FLOW_ACT_HAS_TAG) &&
1436 fte->action.flow_tag != flow_act->flow_tag) {
1437 mlx5_core_warn(get_dev(&fte->node),
1438 "FTE flow tag %u already exists with different flow tag %u\n",
1439 fte->action.flow_tag,
1440 flow_act->flow_tag);
1447 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1449 struct mlx5_flow_act *flow_act,
1450 struct mlx5_flow_destination *dest,
1454 struct mlx5_flow_handle *handle;
1459 ret = check_conflicting_ftes(fte, flow_act);
1461 return ERR_PTR(ret);
1463 old_action = fte->action.action;
1464 fte->action.action |= flow_act->action;
1465 handle = add_rule_fte(fte, fg, dest, dest_num,
1466 old_action != flow_act->action);
1467 if (IS_ERR(handle)) {
1468 fte->action.action = old_action;
1471 trace_mlx5_fs_set_fte(fte, false);
1473 for (i = 0; i < handle->num_rules; i++) {
1474 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1475 tree_add_node(&handle->rule[i]->node, &fte->node);
1476 trace_mlx5_fs_add_rule(handle->rule[i]);
1482 static bool counter_is_valid(u32 action)
1484 return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1485 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1488 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1490 struct mlx5_flow_table *ft)
1492 if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1493 return counter_is_valid(action);
1495 if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1498 if (!dest || ((dest->type ==
1499 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1500 (dest->ft->level <= ft->level)))
1506 struct list_head list;
1507 struct mlx5_flow_group *g;
1510 struct match_list_head {
1511 struct list_head list;
1512 struct match_list first;
1515 static void free_match_list(struct match_list_head *head)
1517 if (!list_empty(&head->list)) {
1518 struct match_list *iter, *match_tmp;
1520 list_del(&head->first.list);
1521 tree_put_node(&head->first.g->node);
1522 list_for_each_entry_safe(iter, match_tmp, &head->list,
1524 tree_put_node(&iter->g->node);
1525 list_del(&iter->list);
1531 static int build_match_list(struct match_list_head *match_head,
1532 struct mlx5_flow_table *ft,
1533 struct mlx5_flow_spec *spec)
1535 struct rhlist_head *tmp, *list;
1536 struct mlx5_flow_group *g;
1540 INIT_LIST_HEAD(&match_head->list);
1541 /* Collect all fgs which has a matching match_criteria */
1542 list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1543 /* RCU is atomic, we can't execute FW commands here */
1544 rhl_for_each_entry_rcu(g, tmp, list, hash) {
1545 struct match_list *curr_match;
1547 if (likely(list_empty(&match_head->list))) {
1548 if (!tree_get_node(&g->node))
1550 match_head->first.g = g;
1551 list_add_tail(&match_head->first.list,
1556 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1558 free_match_list(match_head);
1562 if (!tree_get_node(&g->node)) {
1567 list_add_tail(&curr_match->list, &match_head->list);
1574 static u64 matched_fgs_get_version(struct list_head *match_head)
1576 struct match_list *iter;
1579 list_for_each_entry(iter, match_head, list)
1580 version += (u64)atomic_read(&iter->g->node.version);
1584 static struct fs_fte *
1585 lookup_fte_locked(struct mlx5_flow_group *g,
1589 struct fs_fte *fte_tmp;
1592 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1594 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1595 fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1597 if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1602 nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1605 up_write_ref_node(&g->node);
1607 up_read_ref_node(&g->node);
1611 static struct mlx5_flow_handle *
1612 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1613 struct list_head *match_head,
1614 struct mlx5_flow_spec *spec,
1615 struct mlx5_flow_act *flow_act,
1616 struct mlx5_flow_destination *dest,
1620 struct mlx5_flow_steering *steering = get_steering(&ft->node);
1621 struct mlx5_flow_group *g;
1622 struct mlx5_flow_handle *rule;
1623 struct match_list *iter;
1624 bool take_write = false;
1629 fte = alloc_fte(ft, spec->match_value, flow_act);
1631 return ERR_PTR(-ENOMEM);
1633 search_again_locked:
1634 version = matched_fgs_get_version(match_head);
1635 if (flow_act->flags & FLOW_ACT_NO_APPEND)
1637 /* Try to find a fg that already contains a matching fte */
1638 list_for_each_entry(iter, match_head, list) {
1639 struct fs_fte *fte_tmp;
1642 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1645 rule = add_rule_fg(g, spec->match_value,
1646 flow_act, dest, dest_num, fte_tmp);
1647 up_write_ref_node(&fte_tmp->node);
1648 tree_put_node(&fte_tmp->node);
1649 kmem_cache_free(steering->ftes_cache, fte);
1654 /* No group with matching fte found, or we skipped the search.
1655 * Try to add a new fte to any matching fg.
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_chained_ft(prio);
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_fdb_sub_ns(struct mlx5_core_dev *dev,
1991 struct mlx5_flow_steering *steering = dev->priv.steering;
1993 if (!steering || !steering->fdb_sub_ns)
1996 return steering->fdb_sub_ns[n];
1998 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2000 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2001 enum mlx5_flow_namespace_type type)
2003 struct mlx5_flow_steering *steering = dev->priv.steering;
2004 struct mlx5_flow_root_namespace *root_ns;
2006 struct fs_prio *fs_prio;
2007 struct mlx5_flow_namespace *ns;
2013 case MLX5_FLOW_NAMESPACE_FDB:
2014 if (steering->fdb_root_ns)
2015 return &steering->fdb_root_ns->ns;
2017 case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2018 if (steering->sniffer_rx_root_ns)
2019 return &steering->sniffer_rx_root_ns->ns;
2021 case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2022 if (steering->sniffer_tx_root_ns)
2023 return &steering->sniffer_tx_root_ns->ns;
2029 if (type == MLX5_FLOW_NAMESPACE_EGRESS) {
2030 root_ns = steering->egress_root_ns;
2031 } else { /* Must be NIC RX */
2032 root_ns = steering->root_ns;
2039 fs_prio = find_prio(&root_ns->ns, prio);
2043 ns = list_first_entry(&fs_prio->node.children,
2049 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2051 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2052 enum mlx5_flow_namespace_type type,
2055 struct mlx5_flow_steering *steering = dev->priv.steering;
2057 if (!steering || vport >= MLX5_TOTAL_VPORTS(dev))
2061 case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2062 if (steering->esw_egress_root_ns &&
2063 steering->esw_egress_root_ns[vport])
2064 return &steering->esw_egress_root_ns[vport]->ns;
2067 case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2068 if (steering->esw_ingress_root_ns &&
2069 steering->esw_ingress_root_ns[vport])
2070 return &steering->esw_ingress_root_ns[vport]->ns;
2078 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2081 enum fs_node_type type)
2083 struct fs_prio *fs_prio;
2085 fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2087 return ERR_PTR(-ENOMEM);
2089 fs_prio->node.type = type;
2090 tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2091 tree_add_node(&fs_prio->node, &ns->node);
2092 fs_prio->num_levels = num_levels;
2093 fs_prio->prio = prio;
2094 list_add_tail(&fs_prio->node.list, &ns->node.children);
2099 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2103 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2106 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2107 unsigned int prio, int num_levels)
2109 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2112 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2115 ns->node.type = FS_TYPE_NAMESPACE;
2120 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
2122 struct mlx5_flow_namespace *ns;
2124 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2126 return ERR_PTR(-ENOMEM);
2128 fs_init_namespace(ns);
2129 tree_init_node(&ns->node, NULL, del_sw_ns);
2130 tree_add_node(&ns->node, &prio->node);
2131 list_add_tail(&ns->node.list, &prio->node.children);
2136 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2137 struct init_tree_node *prio_metadata)
2139 struct fs_prio *fs_prio;
2142 for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2143 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2144 if (IS_ERR(fs_prio))
2145 return PTR_ERR(fs_prio);
2150 #define FLOW_TABLE_BIT_SZ 1
2151 #define GET_FLOW_TABLE_CAP(dev, offset) \
2152 ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) + \
2154 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2155 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2159 for (i = 0; i < caps->arr_sz; i++) {
2160 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2166 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2167 struct init_tree_node *init_node,
2168 struct fs_node *fs_parent_node,
2169 struct init_tree_node *init_parent_node,
2172 int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2173 flow_table_properties_nic_receive.
2175 struct mlx5_flow_namespace *fs_ns;
2176 struct fs_prio *fs_prio;
2177 struct fs_node *base;
2181 if (init_node->type == FS_TYPE_PRIO) {
2182 if ((init_node->min_ft_level > max_ft_level) ||
2183 !has_required_caps(steering->dev, &init_node->caps))
2186 fs_get_obj(fs_ns, fs_parent_node);
2187 if (init_node->num_leaf_prios)
2188 return create_leaf_prios(fs_ns, prio, init_node);
2189 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2190 if (IS_ERR(fs_prio))
2191 return PTR_ERR(fs_prio);
2192 base = &fs_prio->node;
2193 } else if (init_node->type == FS_TYPE_NAMESPACE) {
2194 fs_get_obj(fs_prio, fs_parent_node);
2195 fs_ns = fs_create_namespace(fs_prio);
2197 return PTR_ERR(fs_ns);
2198 base = &fs_ns->node;
2203 for (i = 0; i < init_node->ar_size; i++) {
2204 err = init_root_tree_recursive(steering, &init_node->children[i],
2205 base, init_node, prio);
2208 if (init_node->children[i].type == FS_TYPE_PRIO &&
2209 init_node->children[i].num_leaf_prios) {
2210 prio += init_node->children[i].num_leaf_prios;
2217 static int init_root_tree(struct mlx5_flow_steering *steering,
2218 struct init_tree_node *init_node,
2219 struct fs_node *fs_parent_node)
2222 struct mlx5_flow_namespace *fs_ns;
2225 fs_get_obj(fs_ns, fs_parent_node);
2226 for (i = 0; i < init_node->ar_size; i++) {
2227 err = init_root_tree_recursive(steering, &init_node->children[i],
2236 static struct mlx5_flow_root_namespace
2237 *create_root_ns(struct mlx5_flow_steering *steering,
2238 enum fs_flow_table_type table_type)
2240 const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2241 struct mlx5_flow_root_namespace *root_ns;
2242 struct mlx5_flow_namespace *ns;
2244 if (mlx5_accel_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2245 (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2246 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2248 /* Create the root namespace */
2249 root_ns = kvzalloc(sizeof(*root_ns), GFP_KERNEL);
2253 root_ns->dev = steering->dev;
2254 root_ns->table_type = table_type;
2255 root_ns->cmds = cmds;
2257 INIT_LIST_HEAD(&root_ns->underlay_qpns);
2260 fs_init_namespace(ns);
2261 mutex_init(&root_ns->chain_lock);
2262 tree_init_node(&ns->node, NULL, NULL);
2263 tree_add_node(&ns->node, NULL);
2268 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2270 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2272 struct fs_prio *prio;
2274 fs_for_each_prio(prio, ns) {
2275 /* This updates prio start_level and num_levels */
2276 set_prio_attrs_in_prio(prio, acc_level);
2277 acc_level += prio->num_levels;
2282 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2284 struct mlx5_flow_namespace *ns;
2285 int acc_level_ns = acc_level;
2287 prio->start_level = acc_level;
2288 fs_for_each_ns(ns, prio)
2289 /* This updates start_level and num_levels of ns's priority descendants */
2290 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2291 if (!prio->num_levels)
2292 prio->num_levels = acc_level_ns - prio->start_level;
2293 WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2296 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2298 struct mlx5_flow_namespace *ns = &root_ns->ns;
2299 struct fs_prio *prio;
2300 int start_level = 0;
2302 fs_for_each_prio(prio, ns) {
2303 set_prio_attrs_in_prio(prio, start_level);
2304 start_level += prio->num_levels;
2308 #define ANCHOR_PRIO 0
2309 #define ANCHOR_SIZE 1
2310 #define ANCHOR_LEVEL 0
2311 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2313 struct mlx5_flow_namespace *ns = NULL;
2314 struct mlx5_flow_table_attr ft_attr = {};
2315 struct mlx5_flow_table *ft;
2317 ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2321 ft_attr.max_fte = ANCHOR_SIZE;
2322 ft_attr.level = ANCHOR_LEVEL;
2323 ft_attr.prio = ANCHOR_PRIO;
2325 ft = mlx5_create_flow_table(ns, &ft_attr);
2327 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2333 static int init_root_ns(struct mlx5_flow_steering *steering)
2337 steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2338 if (!steering->root_ns)
2341 err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2345 set_prio_attrs(steering->root_ns);
2346 err = create_anchor_flow_table(steering);
2353 cleanup_root_ns(steering->root_ns);
2354 steering->root_ns = NULL;
2358 static void clean_tree(struct fs_node *node)
2361 struct fs_node *iter;
2362 struct fs_node *temp;
2364 tree_get_node(node);
2365 list_for_each_entry_safe(iter, temp, &node->children, list)
2367 tree_put_node(node);
2368 tree_remove_node(node);
2372 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2377 clean_tree(&root_ns->ns.node);
2380 static void cleanup_egress_acls_root_ns(struct mlx5_core_dev *dev)
2382 struct mlx5_flow_steering *steering = dev->priv.steering;
2385 if (!steering->esw_egress_root_ns)
2388 for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2389 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2391 kfree(steering->esw_egress_root_ns);
2394 static void cleanup_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2396 struct mlx5_flow_steering *steering = dev->priv.steering;
2399 if (!steering->esw_ingress_root_ns)
2402 for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2403 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2405 kfree(steering->esw_ingress_root_ns);
2408 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2410 struct mlx5_flow_steering *steering = dev->priv.steering;
2412 cleanup_root_ns(steering->root_ns);
2413 cleanup_egress_acls_root_ns(dev);
2414 cleanup_ingress_acls_root_ns(dev);
2415 cleanup_root_ns(steering->fdb_root_ns);
2416 steering->fdb_root_ns = NULL;
2417 kfree(steering->fdb_sub_ns);
2418 steering->fdb_sub_ns = NULL;
2419 cleanup_root_ns(steering->sniffer_rx_root_ns);
2420 cleanup_root_ns(steering->sniffer_tx_root_ns);
2421 cleanup_root_ns(steering->egress_root_ns);
2422 mlx5_cleanup_fc_stats(dev);
2423 kmem_cache_destroy(steering->ftes_cache);
2424 kmem_cache_destroy(steering->fgs_cache);
2428 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2430 struct fs_prio *prio;
2432 steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2433 if (!steering->sniffer_tx_root_ns)
2436 /* Create single prio */
2437 prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2439 cleanup_root_ns(steering->sniffer_tx_root_ns);
2440 return PTR_ERR(prio);
2445 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2447 struct fs_prio *prio;
2449 steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2450 if (!steering->sniffer_rx_root_ns)
2453 /* Create single prio */
2454 prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2456 cleanup_root_ns(steering->sniffer_rx_root_ns);
2457 return PTR_ERR(prio);
2462 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2464 struct mlx5_flow_namespace *ns;
2465 struct fs_prio *maj_prio;
2466 struct fs_prio *min_prio;
2472 steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2473 if (!steering->fdb_root_ns)
2476 steering->fdb_sub_ns = kzalloc(sizeof(steering->fdb_sub_ns) *
2477 (FDB_MAX_CHAIN + 1), GFP_KERNEL);
2478 if (!steering->fdb_sub_ns)
2481 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2483 if (IS_ERR(maj_prio)) {
2484 err = PTR_ERR(maj_prio);
2488 levels = 2 * FDB_MAX_PRIO * (FDB_MAX_CHAIN + 1);
2489 maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2492 if (IS_ERR(maj_prio)) {
2493 err = PTR_ERR(maj_prio);
2497 for (chain = 0; chain <= FDB_MAX_CHAIN; chain++) {
2498 ns = fs_create_namespace(maj_prio);
2504 for (prio = 0; prio < FDB_MAX_PRIO * (chain + 1); prio++) {
2505 min_prio = fs_create_prio(ns, prio, 2);
2506 if (IS_ERR(min_prio)) {
2507 err = PTR_ERR(min_prio);
2512 steering->fdb_sub_ns[chain] = ns;
2515 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2516 if (IS_ERR(maj_prio)) {
2517 err = PTR_ERR(maj_prio);
2521 set_prio_attrs(steering->fdb_root_ns);
2525 cleanup_root_ns(steering->fdb_root_ns);
2526 kfree(steering->fdb_sub_ns);
2527 steering->fdb_sub_ns = NULL;
2528 steering->fdb_root_ns = NULL;
2532 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2534 struct fs_prio *prio;
2536 steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2537 if (!steering->esw_egress_root_ns[vport])
2541 prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2542 return PTR_ERR_OR_ZERO(prio);
2545 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2547 struct fs_prio *prio;
2549 steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2550 if (!steering->esw_ingress_root_ns[vport])
2554 prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2555 return PTR_ERR_OR_ZERO(prio);
2558 static int init_egress_acls_root_ns(struct mlx5_core_dev *dev)
2560 struct mlx5_flow_steering *steering = dev->priv.steering;
2564 steering->esw_egress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2565 sizeof(*steering->esw_egress_root_ns),
2567 if (!steering->esw_egress_root_ns)
2570 for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2571 err = init_egress_acl_root_ns(steering, i);
2573 goto cleanup_root_ns;
2579 for (i--; i >= 0; i--)
2580 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2581 kfree(steering->esw_egress_root_ns);
2585 static int init_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2587 struct mlx5_flow_steering *steering = dev->priv.steering;
2591 steering->esw_ingress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2592 sizeof(*steering->esw_ingress_root_ns),
2594 if (!steering->esw_ingress_root_ns)
2597 for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2598 err = init_ingress_acl_root_ns(steering, i);
2600 goto cleanup_root_ns;
2606 for (i--; i >= 0; i--)
2607 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2608 kfree(steering->esw_ingress_root_ns);
2612 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2616 steering->egress_root_ns = create_root_ns(steering,
2618 if (!steering->egress_root_ns)
2621 err = init_root_tree(steering, &egress_root_fs,
2622 &steering->egress_root_ns->ns.node);
2625 set_prio_attrs(steering->egress_root_ns);
2628 cleanup_root_ns(steering->egress_root_ns);
2629 steering->egress_root_ns = NULL;
2633 int mlx5_init_fs(struct mlx5_core_dev *dev)
2635 struct mlx5_flow_steering *steering;
2638 err = mlx5_init_fc_stats(dev);
2642 steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2645 steering->dev = dev;
2646 dev->priv.steering = steering;
2648 steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2649 sizeof(struct mlx5_flow_group), 0,
2651 steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2653 if (!steering->ftes_cache || !steering->fgs_cache) {
2658 if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2659 (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2660 ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2661 MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2662 MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2663 err = init_root_ns(steering);
2668 if (MLX5_ESWITCH_MANAGER(dev)) {
2669 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2670 err = init_fdb_root_ns(steering);
2674 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2675 err = init_egress_acls_root_ns(dev);
2679 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2680 err = init_ingress_acls_root_ns(dev);
2686 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2687 err = init_sniffer_rx_root_ns(steering);
2692 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2693 err = init_sniffer_tx_root_ns(steering);
2698 if (MLX5_IPSEC_DEV(dev) || MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
2699 err = init_egress_root_ns(steering);
2706 mlx5_cleanup_fs(dev);
2710 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2712 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2713 struct mlx5_ft_underlay_qp *new_uqp;
2716 new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
2720 mutex_lock(&root->chain_lock);
2722 if (!root->root_ft) {
2724 goto update_ft_fail;
2727 err = root->cmds->update_root_ft(dev, root->root_ft, underlay_qpn,
2730 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
2732 goto update_ft_fail;
2735 new_uqp->qpn = underlay_qpn;
2736 list_add_tail(&new_uqp->list, &root->underlay_qpns);
2738 mutex_unlock(&root->chain_lock);
2743 mutex_unlock(&root->chain_lock);
2747 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
2749 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2751 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2752 struct mlx5_ft_underlay_qp *uqp;
2756 mutex_lock(&root->chain_lock);
2757 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2758 if (uqp->qpn == underlay_qpn) {
2765 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
2771 err = root->cmds->update_root_ft(dev, root->root_ft, underlay_qpn,
2774 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
2777 list_del(&uqp->list);
2778 mutex_unlock(&root->chain_lock);
2784 mutex_unlock(&root->chain_lock);
2787 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);