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, bool locked)
270 down_write(&node->lock);
271 refcount_inc(&node->refcount);
275 static void up_read_ref_node(struct fs_node *node)
277 refcount_dec(&node->refcount);
278 up_read(&node->lock);
281 static void up_write_ref_node(struct fs_node *node, bool locked)
283 refcount_dec(&node->refcount);
285 up_write(&node->lock);
288 static void tree_put_node(struct fs_node *node, bool locked)
290 struct fs_node *parent_node = node->parent;
292 if (refcount_dec_and_test(&node->refcount)) {
293 if (node->del_hw_func)
294 node->del_hw_func(node);
296 /* Only root namespace doesn't have parent and we just
297 * need to free its node.
299 down_write_ref_node(parent_node, locked);
300 list_del_init(&node->list);
301 if (node->del_sw_func)
302 node->del_sw_func(node);
303 up_write_ref_node(parent_node, locked);
309 if (!node && parent_node)
310 tree_put_node(parent_node, locked);
313 static int tree_remove_node(struct fs_node *node, bool locked)
315 if (refcount_read(&node->refcount) > 1) {
316 refcount_dec(&node->refcount);
319 tree_put_node(node, locked);
323 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
326 struct fs_prio *iter_prio;
328 fs_for_each_prio(iter_prio, ns) {
329 if (iter_prio->prio == prio)
336 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
340 for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
341 if (spec->match_value[i] & ~spec->match_criteria[i]) {
342 pr_warn("mlx5_core: match_value differs from match_criteria\n");
349 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
351 struct fs_node *root;
352 struct mlx5_flow_namespace *ns;
356 if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
357 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
361 ns = container_of(root, struct mlx5_flow_namespace, node);
362 return container_of(ns, struct mlx5_flow_root_namespace, ns);
365 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
367 struct mlx5_flow_root_namespace *root = find_root(node);
370 return root->dev->priv.steering;
374 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
376 struct mlx5_flow_root_namespace *root = find_root(node);
383 static void del_sw_ns(struct fs_node *node)
388 static void del_sw_prio(struct fs_node *node)
393 static void del_hw_flow_table(struct fs_node *node)
395 struct mlx5_flow_root_namespace *root;
396 struct mlx5_flow_table *ft;
397 struct mlx5_core_dev *dev;
400 fs_get_obj(ft, node);
401 dev = get_dev(&ft->node);
402 root = find_root(&ft->node);
403 trace_mlx5_fs_del_ft(ft);
406 err = root->cmds->destroy_flow_table(root, ft);
408 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
412 static void del_sw_flow_table(struct fs_node *node)
414 struct mlx5_flow_table *ft;
415 struct fs_prio *prio;
417 fs_get_obj(ft, node);
419 rhltable_destroy(&ft->fgs_hash);
420 fs_get_obj(prio, ft->node.parent);
425 static void modify_fte(struct fs_fte *fte)
427 struct mlx5_flow_root_namespace *root;
428 struct mlx5_flow_table *ft;
429 struct mlx5_flow_group *fg;
430 struct mlx5_core_dev *dev;
433 fs_get_obj(fg, fte->node.parent);
434 fs_get_obj(ft, fg->node.parent);
435 dev = get_dev(&fte->node);
437 root = find_root(&ft->node);
438 err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte);
441 "%s can't del rule fg id=%d fte_index=%d\n",
442 __func__, fg->id, fte->index);
443 fte->modify_mask = 0;
446 static void del_sw_hw_rule(struct fs_node *node)
448 struct mlx5_flow_rule *rule;
451 fs_get_obj(rule, node);
452 fs_get_obj(fte, rule->node.parent);
453 trace_mlx5_fs_del_rule(rule);
454 if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
455 mutex_lock(&rule->dest_attr.ft->lock);
456 list_del(&rule->next_ft);
457 mutex_unlock(&rule->dest_attr.ft->lock);
460 if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER &&
463 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
464 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
465 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
469 if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
472 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
478 static void del_hw_fte(struct fs_node *node)
480 struct mlx5_flow_root_namespace *root;
481 struct mlx5_flow_table *ft;
482 struct mlx5_flow_group *fg;
483 struct mlx5_core_dev *dev;
487 fs_get_obj(fte, node);
488 fs_get_obj(fg, fte->node.parent);
489 fs_get_obj(ft, fg->node.parent);
491 trace_mlx5_fs_del_fte(fte);
492 dev = get_dev(&ft->node);
493 root = find_root(&ft->node);
495 err = root->cmds->delete_fte(root, ft, fte);
498 "flow steering can't delete fte in index %d of flow group id %d\n",
504 static void del_sw_fte(struct fs_node *node)
506 struct mlx5_flow_steering *steering = get_steering(node);
507 struct mlx5_flow_group *fg;
511 fs_get_obj(fte, node);
512 fs_get_obj(fg, fte->node.parent);
514 err = rhashtable_remove_fast(&fg->ftes_hash,
518 ida_simple_remove(&fg->fte_allocator, fte->index - fg->start_index);
519 kmem_cache_free(steering->ftes_cache, fte);
522 static void del_hw_flow_group(struct fs_node *node)
524 struct mlx5_flow_root_namespace *root;
525 struct mlx5_flow_group *fg;
526 struct mlx5_flow_table *ft;
527 struct mlx5_core_dev *dev;
529 fs_get_obj(fg, node);
530 fs_get_obj(ft, fg->node.parent);
531 dev = get_dev(&ft->node);
532 trace_mlx5_fs_del_fg(fg);
534 root = find_root(&ft->node);
535 if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
536 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
540 static void del_sw_flow_group(struct fs_node *node)
542 struct mlx5_flow_steering *steering = get_steering(node);
543 struct mlx5_flow_group *fg;
544 struct mlx5_flow_table *ft;
547 fs_get_obj(fg, node);
548 fs_get_obj(ft, fg->node.parent);
550 rhashtable_destroy(&fg->ftes_hash);
551 ida_destroy(&fg->fte_allocator);
552 if (ft->autogroup.active)
553 ft->autogroup.num_groups--;
554 err = rhltable_remove(&ft->fgs_hash,
558 kmem_cache_free(steering->fgs_cache, fg);
561 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
566 index = ida_simple_get(&fg->fte_allocator, 0, fg->max_ftes, GFP_KERNEL);
570 fte->index = index + fg->start_index;
571 ret = rhashtable_insert_fast(&fg->ftes_hash,
577 tree_add_node(&fte->node, &fg->node);
578 list_add_tail(&fte->node.list, &fg->node.children);
582 ida_simple_remove(&fg->fte_allocator, index);
586 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
587 struct mlx5_flow_spec *spec,
588 struct mlx5_flow_act *flow_act)
590 struct mlx5_flow_steering *steering = get_steering(&ft->node);
593 fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
595 return ERR_PTR(-ENOMEM);
597 memcpy(fte->val, &spec->match_value, sizeof(fte->val));
598 fte->node.type = FS_TYPE_FLOW_ENTRY;
599 fte->action = *flow_act;
600 fte->flow_context = spec->flow_context;
602 tree_init_node(&fte->node, NULL, del_sw_fte);
607 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
608 struct mlx5_flow_group *fg)
610 rhashtable_destroy(&fg->ftes_hash);
611 kmem_cache_free(steering->fgs_cache, fg);
614 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
615 u8 match_criteria_enable,
616 void *match_criteria,
620 struct mlx5_flow_group *fg;
623 fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
625 return ERR_PTR(-ENOMEM);
627 ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
629 kmem_cache_free(steering->fgs_cache, fg);
633 ida_init(&fg->fte_allocator);
634 fg->mask.match_criteria_enable = match_criteria_enable;
635 memcpy(&fg->mask.match_criteria, match_criteria,
636 sizeof(fg->mask.match_criteria));
637 fg->node.type = FS_TYPE_FLOW_GROUP;
638 fg->start_index = start_index;
639 fg->max_ftes = end_index - start_index + 1;
644 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
645 u8 match_criteria_enable,
646 void *match_criteria,
649 struct list_head *prev)
651 struct mlx5_flow_steering *steering = get_steering(&ft->node);
652 struct mlx5_flow_group *fg;
655 fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
656 start_index, end_index);
660 /* initialize refcnt, add to parent list */
661 ret = rhltable_insert(&ft->fgs_hash,
665 dealloc_flow_group(steering, fg);
669 tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
670 tree_add_node(&fg->node, &ft->node);
671 /* Add node to group list */
672 list_add(&fg->node.list, prev);
673 atomic_inc(&ft->node.version);
678 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
679 enum fs_flow_table_type table_type,
680 enum fs_flow_table_op_mod op_mod,
683 struct mlx5_flow_table *ft;
686 ft = kzalloc(sizeof(*ft), GFP_KERNEL);
688 return ERR_PTR(-ENOMEM);
690 ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
697 ft->node.type = FS_TYPE_FLOW_TABLE;
699 ft->type = table_type;
701 ft->max_fte = max_fte;
703 INIT_LIST_HEAD(&ft->fwd_rules);
704 mutex_init(&ft->lock);
709 /* If reverse is false, then we search for the first flow table in the
710 * root sub-tree from start(closest from right), else we search for the
711 * last flow table in the root sub-tree till start(closest from left).
713 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node *root,
714 struct list_head *start,
717 #define list_advance_entry(pos, reverse) \
718 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
720 #define list_for_each_advance_continue(pos, head, reverse) \
721 for (pos = list_advance_entry(pos, reverse); \
722 &pos->list != (head); \
723 pos = list_advance_entry(pos, reverse))
725 struct fs_node *iter = list_entry(start, struct fs_node, list);
726 struct mlx5_flow_table *ft = NULL;
728 if (!root || root->type == FS_TYPE_PRIO_CHAINS)
731 list_for_each_advance_continue(iter, &root->children, reverse) {
732 if (iter->type == FS_TYPE_FLOW_TABLE) {
733 fs_get_obj(ft, iter);
736 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
744 /* If reverse if false then return the first flow table in next priority of
745 * prio in the tree, else return the last flow table in the previous priority
746 * of prio in the tree.
748 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
750 struct mlx5_flow_table *ft = NULL;
751 struct fs_node *curr_node;
752 struct fs_node *parent;
754 parent = prio->node.parent;
755 curr_node = &prio->node;
756 while (!ft && parent) {
757 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
759 parent = curr_node->parent;
764 /* Assuming all the tree is locked by mutex chain lock */
765 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
767 return find_closest_ft(prio, false);
770 /* Assuming all the tree is locked by mutex chain lock */
771 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
773 return find_closest_ft(prio, true);
776 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
777 struct fs_prio *prio,
778 struct mlx5_flow_table *ft)
780 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
781 struct mlx5_flow_table *iter;
785 fs_for_each_ft(iter, prio) {
787 err = root->cmds->modify_flow_table(root, iter, ft);
789 mlx5_core_warn(dev, "Failed to modify flow table %d\n",
791 /* The driver is out of sync with the FW */
800 /* Connect flow tables from previous priority of prio to ft */
801 static int connect_prev_fts(struct mlx5_core_dev *dev,
802 struct mlx5_flow_table *ft,
803 struct fs_prio *prio)
805 struct mlx5_flow_table *prev_ft;
807 prev_ft = find_prev_chained_ft(prio);
809 struct fs_prio *prev_prio;
811 fs_get_obj(prev_prio, prev_ft->node.parent);
812 return connect_fts_in_prio(dev, prev_prio, ft);
817 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
820 struct mlx5_flow_root_namespace *root = find_root(&prio->node);
821 struct mlx5_ft_underlay_qp *uqp;
822 int min_level = INT_MAX;
827 min_level = root->root_ft->level;
829 if (ft->level >= min_level)
832 if (list_empty(&root->underlay_qpns)) {
833 /* Don't set any QPN (zero) in case QPN list is empty */
835 err = root->cmds->update_root_ft(root, ft, qpn, false);
837 list_for_each_entry(uqp, &root->underlay_qpns, list) {
839 err = root->cmds->update_root_ft(root, ft,
847 mlx5_core_warn(root->dev,
848 "Update root flow table of id(%u) qpn(%d) failed\n",
856 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
857 struct mlx5_flow_destination *dest)
859 struct mlx5_flow_root_namespace *root;
860 struct mlx5_flow_table *ft;
861 struct mlx5_flow_group *fg;
863 int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
866 fs_get_obj(fte, rule->node.parent);
867 if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
869 down_write_ref_node(&fte->node, false);
870 fs_get_obj(fg, fte->node.parent);
871 fs_get_obj(ft, fg->node.parent);
873 memcpy(&rule->dest_attr, dest, sizeof(*dest));
874 root = find_root(&ft->node);
875 err = root->cmds->update_fte(root, ft, fg,
877 up_write_ref_node(&fte->node, false);
882 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
883 struct mlx5_flow_destination *new_dest,
884 struct mlx5_flow_destination *old_dest)
889 if (handle->num_rules != 1)
891 return _mlx5_modify_rule_destination(handle->rule[0],
895 for (i = 0; i < handle->num_rules; i++) {
896 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
897 return _mlx5_modify_rule_destination(handle->rule[i],
904 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */
905 static int connect_fwd_rules(struct mlx5_core_dev *dev,
906 struct mlx5_flow_table *new_next_ft,
907 struct mlx5_flow_table *old_next_ft)
909 struct mlx5_flow_destination dest = {};
910 struct mlx5_flow_rule *iter;
913 /* new_next_ft and old_next_ft could be NULL only
914 * when we create/destroy the anchor flow table.
916 if (!new_next_ft || !old_next_ft)
919 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
920 dest.ft = new_next_ft;
922 mutex_lock(&old_next_ft->lock);
923 list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
924 mutex_unlock(&old_next_ft->lock);
925 list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
926 err = _mlx5_modify_rule_destination(iter, &dest);
928 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
934 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
935 struct fs_prio *prio)
937 struct mlx5_flow_table *next_ft;
940 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
942 if (list_empty(&prio->node.children)) {
943 err = connect_prev_fts(dev, ft, prio);
947 next_ft = find_next_chained_ft(prio);
948 err = connect_fwd_rules(dev, ft, next_ft);
953 if (MLX5_CAP_FLOWTABLE(dev,
954 flow_table_properties_nic_receive.modify_root))
955 err = update_root_ft_create(ft, prio);
959 static void list_add_flow_table(struct mlx5_flow_table *ft,
960 struct fs_prio *prio)
962 struct list_head *prev = &prio->node.children;
963 struct mlx5_flow_table *iter;
965 fs_for_each_ft(iter, prio) {
966 if (iter->level > ft->level)
968 prev = &iter->node.list;
970 list_add(&ft->node.list, prev);
973 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
974 struct mlx5_flow_table_attr *ft_attr,
975 enum fs_flow_table_op_mod op_mod,
978 struct mlx5_flow_root_namespace *root = find_root(&ns->node);
979 struct mlx5_flow_table *next_ft = NULL;
980 struct fs_prio *fs_prio = NULL;
981 struct mlx5_flow_table *ft;
986 pr_err("mlx5: flow steering failed to find root of namespace\n");
987 return ERR_PTR(-ENODEV);
990 mutex_lock(&root->chain_lock);
991 fs_prio = find_prio(ns, ft_attr->prio);
996 if (ft_attr->level >= fs_prio->num_levels) {
1000 /* The level is related to the
1001 * priority level range.
1003 ft_attr->level += fs_prio->start_level;
1004 ft = alloc_flow_table(ft_attr->level,
1006 ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
1008 op_mod, ft_attr->flags);
1014 tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1015 log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
1016 next_ft = find_next_chained_ft(fs_prio);
1017 err = root->cmds->create_flow_table(root, ft, log_table_sz, next_ft);
1021 err = connect_flow_table(root->dev, ft, fs_prio);
1024 ft->node.active = true;
1025 down_write_ref_node(&fs_prio->node, false);
1026 tree_add_node(&ft->node, &fs_prio->node);
1027 list_add_flow_table(ft, fs_prio);
1029 up_write_ref_node(&fs_prio->node, false);
1030 mutex_unlock(&root->chain_lock);
1031 trace_mlx5_fs_add_ft(ft);
1034 root->cmds->destroy_flow_table(root, ft);
1038 mutex_unlock(&root->chain_lock);
1039 return ERR_PTR(err);
1042 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1043 struct mlx5_flow_table_attr *ft_attr)
1045 return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1048 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1049 int prio, int max_fte,
1050 u32 level, u16 vport)
1052 struct mlx5_flow_table_attr ft_attr = {};
1054 ft_attr.max_fte = max_fte;
1055 ft_attr.level = level;
1056 ft_attr.prio = prio;
1058 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1061 struct mlx5_flow_table*
1062 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1063 int prio, u32 level)
1065 struct mlx5_flow_table_attr ft_attr = {};
1067 ft_attr.level = level;
1068 ft_attr.prio = prio;
1069 return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1071 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1073 struct mlx5_flow_table*
1074 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1076 int num_flow_table_entries,
1081 struct mlx5_flow_table_attr ft_attr = {};
1082 struct mlx5_flow_table *ft;
1084 if (max_num_groups > num_flow_table_entries)
1085 return ERR_PTR(-EINVAL);
1087 ft_attr.max_fte = num_flow_table_entries;
1088 ft_attr.prio = prio;
1089 ft_attr.level = level;
1090 ft_attr.flags = flags;
1092 ft = mlx5_create_flow_table(ns, &ft_attr);
1096 ft->autogroup.active = true;
1097 ft->autogroup.required_groups = max_num_groups;
1101 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1103 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1106 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1107 void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1108 fg_in, match_criteria);
1109 u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1111 match_criteria_enable);
1112 int start_index = MLX5_GET(create_flow_group_in, fg_in,
1114 int end_index = MLX5_GET(create_flow_group_in, fg_in,
1116 struct mlx5_flow_group *fg;
1119 if (ft->autogroup.active)
1120 return ERR_PTR(-EPERM);
1122 down_write_ref_node(&ft->node, false);
1123 fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1124 start_index, end_index,
1125 ft->node.children.prev);
1126 up_write_ref_node(&ft->node, false);
1130 err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1132 tree_put_node(&fg->node, false);
1133 return ERR_PTR(err);
1135 trace_mlx5_fs_add_fg(fg);
1136 fg->node.active = true;
1141 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1143 struct mlx5_flow_rule *rule;
1145 rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1149 INIT_LIST_HEAD(&rule->next_ft);
1150 rule->node.type = FS_TYPE_FLOW_DEST;
1152 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1157 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1159 struct mlx5_flow_handle *handle;
1161 handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1165 handle->num_rules = num_rules;
1170 static void destroy_flow_handle(struct fs_fte *fte,
1171 struct mlx5_flow_handle *handle,
1172 struct mlx5_flow_destination *dest,
1176 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1178 list_del(&handle->rule[i]->node.list);
1179 kfree(handle->rule[i]);
1185 static struct mlx5_flow_handle *
1186 create_flow_handle(struct fs_fte *fte,
1187 struct mlx5_flow_destination *dest,
1192 struct mlx5_flow_handle *handle;
1193 struct mlx5_flow_rule *rule = NULL;
1194 static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1195 static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1199 handle = alloc_handle((dest_num) ? dest_num : 1);
1201 return ERR_PTR(-ENOMEM);
1205 rule = find_flow_rule(fte, dest + i);
1207 refcount_inc(&rule->node.refcount);
1213 rule = alloc_rule(dest + i);
1217 /* Add dest to dests list- we need flow tables to be in the
1218 * end of the list for forward to next prio rules.
1220 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1222 dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1223 list_add(&rule->node.list, &fte->node.children);
1225 list_add_tail(&rule->node.list, &fte->node.children);
1229 type = dest[i].type ==
1230 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1231 *modify_mask |= type ? count : dst;
1234 handle->rule[i] = rule;
1235 } while (++i < dest_num);
1240 destroy_flow_handle(fte, handle, dest, i);
1241 return ERR_PTR(-ENOMEM);
1244 /* fte should not be deleted while calling this function */
1245 static struct mlx5_flow_handle *
1246 add_rule_fte(struct fs_fte *fte,
1247 struct mlx5_flow_group *fg,
1248 struct mlx5_flow_destination *dest,
1252 struct mlx5_flow_root_namespace *root;
1253 struct mlx5_flow_handle *handle;
1254 struct mlx5_flow_table *ft;
1255 int modify_mask = 0;
1257 bool new_rule = false;
1259 handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1261 if (IS_ERR(handle) || !new_rule)
1265 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1267 fs_get_obj(ft, fg->node.parent);
1268 root = find_root(&fg->node);
1269 if (!(fte->status & FS_FTE_STATUS_EXISTING))
1270 err = root->cmds->create_fte(root, ft, fg, fte);
1272 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1276 fte->node.active = true;
1277 fte->status |= FS_FTE_STATUS_EXISTING;
1278 atomic_inc(&fte->node.version);
1284 destroy_flow_handle(fte, handle, dest, handle->num_rules);
1285 return ERR_PTR(err);
1288 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table *ft,
1289 struct mlx5_flow_spec *spec)
1291 struct list_head *prev = &ft->node.children;
1292 struct mlx5_flow_group *fg;
1293 unsigned int candidate_index = 0;
1294 unsigned int group_size = 0;
1296 if (!ft->autogroup.active)
1297 return ERR_PTR(-ENOENT);
1299 if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1300 /* We save place for flow groups in addition to max types */
1301 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1303 /* ft->max_fte == ft->autogroup.max_types */
1304 if (group_size == 0)
1307 /* sorted by start_index */
1308 fs_for_each_fg(fg, ft) {
1309 if (candidate_index + group_size > fg->start_index)
1310 candidate_index = fg->start_index + fg->max_ftes;
1313 prev = &fg->node.list;
1316 if (candidate_index + group_size > ft->max_fte)
1317 return ERR_PTR(-ENOSPC);
1319 fg = alloc_insert_flow_group(ft,
1320 spec->match_criteria_enable,
1321 spec->match_criteria,
1323 candidate_index + group_size - 1,
1328 ft->autogroup.num_groups++;
1334 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1335 struct mlx5_flow_group *fg)
1337 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1338 int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1339 void *match_criteria_addr;
1340 u8 src_esw_owner_mask_on;
1345 in = kvzalloc(inlen, GFP_KERNEL);
1349 MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1350 fg->mask.match_criteria_enable);
1351 MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1352 MLX5_SET(create_flow_group_in, in, end_flow_index, fg->start_index +
1355 misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1357 src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1358 source_eswitch_owner_vhca_id);
1359 MLX5_SET(create_flow_group_in, in,
1360 source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1362 match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1363 in, match_criteria);
1364 memcpy(match_criteria_addr, fg->mask.match_criteria,
1365 sizeof(fg->mask.match_criteria));
1367 err = root->cmds->create_flow_group(root, ft, in, fg);
1369 fg->node.active = true;
1370 trace_mlx5_fs_add_fg(fg);
1377 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1378 struct mlx5_flow_destination *d2)
1380 if (d1->type == d2->type) {
1381 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1382 d1->vport.num == d2->vport.num &&
1383 d1->vport.flags == d2->vport.flags &&
1384 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1385 (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1386 ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1387 (d1->vport.reformat_id == d2->vport.reformat_id) : true)) ||
1388 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1389 d1->ft == d2->ft) ||
1390 (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1391 d1->tir_num == d2->tir_num) ||
1392 (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1393 d1->ft_num == d2->ft_num))
1400 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1401 struct mlx5_flow_destination *dest)
1403 struct mlx5_flow_rule *rule;
1405 list_for_each_entry(rule, &fte->node.children, node.list) {
1406 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1412 static bool check_conflicting_actions(u32 action1, u32 action2)
1414 u32 xored_actions = action1 ^ action2;
1416 /* if one rule only wants to count, it's ok */
1417 if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1418 action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1421 if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1422 MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1423 MLX5_FLOW_CONTEXT_ACTION_DECAP |
1424 MLX5_FLOW_CONTEXT_ACTION_MOD_HDR |
1425 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1426 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1427 MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1428 MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1434 static int check_conflicting_ftes(struct fs_fte *fte,
1435 const struct mlx5_flow_context *flow_context,
1436 const struct mlx5_flow_act *flow_act)
1438 if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1439 mlx5_core_warn(get_dev(&fte->node),
1440 "Found two FTEs with conflicting actions\n");
1444 if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1445 fte->flow_context.flow_tag != flow_context->flow_tag) {
1446 mlx5_core_warn(get_dev(&fte->node),
1447 "FTE flow tag %u already exists with different flow tag %u\n",
1448 fte->flow_context.flow_tag,
1449 flow_context->flow_tag);
1456 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1457 struct mlx5_flow_spec *spec,
1458 struct mlx5_flow_act *flow_act,
1459 struct mlx5_flow_destination *dest,
1463 struct mlx5_flow_handle *handle;
1468 ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1470 return ERR_PTR(ret);
1472 old_action = fte->action.action;
1473 fte->action.action |= flow_act->action;
1474 handle = add_rule_fte(fte, fg, dest, dest_num,
1475 old_action != flow_act->action);
1476 if (IS_ERR(handle)) {
1477 fte->action.action = old_action;
1480 trace_mlx5_fs_set_fte(fte, false);
1482 for (i = 0; i < handle->num_rules; i++) {
1483 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1484 tree_add_node(&handle->rule[i]->node, &fte->node);
1485 trace_mlx5_fs_add_rule(handle->rule[i]);
1491 static bool counter_is_valid(u32 action)
1493 return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1494 MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1497 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1499 struct mlx5_flow_table *ft)
1501 if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1502 return counter_is_valid(action);
1504 if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1507 if (!dest || ((dest->type ==
1508 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1509 (dest->ft->level <= ft->level)))
1515 struct list_head list;
1516 struct mlx5_flow_group *g;
1519 struct match_list_head {
1520 struct list_head list;
1521 struct match_list first;
1524 static void free_match_list(struct match_list_head *head)
1526 if (!list_empty(&head->list)) {
1527 struct match_list *iter, *match_tmp;
1529 list_del(&head->first.list);
1530 tree_put_node(&head->first.g->node, false);
1531 list_for_each_entry_safe(iter, match_tmp, &head->list,
1533 tree_put_node(&iter->g->node, false);
1534 list_del(&iter->list);
1540 static int build_match_list(struct match_list_head *match_head,
1541 struct mlx5_flow_table *ft,
1542 struct mlx5_flow_spec *spec)
1544 struct rhlist_head *tmp, *list;
1545 struct mlx5_flow_group *g;
1549 INIT_LIST_HEAD(&match_head->list);
1550 /* Collect all fgs which has a matching match_criteria */
1551 list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1552 /* RCU is atomic, we can't execute FW commands here */
1553 rhl_for_each_entry_rcu(g, tmp, list, hash) {
1554 struct match_list *curr_match;
1556 if (likely(list_empty(&match_head->list))) {
1557 if (!tree_get_node(&g->node))
1559 match_head->first.g = g;
1560 list_add_tail(&match_head->first.list,
1565 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1567 free_match_list(match_head);
1571 if (!tree_get_node(&g->node)) {
1576 list_add_tail(&curr_match->list, &match_head->list);
1583 static u64 matched_fgs_get_version(struct list_head *match_head)
1585 struct match_list *iter;
1588 list_for_each_entry(iter, match_head, list)
1589 version += (u64)atomic_read(&iter->g->node.version);
1593 static struct fs_fte *
1594 lookup_fte_locked(struct mlx5_flow_group *g,
1598 struct fs_fte *fte_tmp;
1601 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1603 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1604 fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1606 if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1610 if (!fte_tmp->node.active) {
1611 tree_put_node(&fte_tmp->node, false);
1616 nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1619 up_write_ref_node(&g->node, false);
1621 up_read_ref_node(&g->node);
1625 static struct mlx5_flow_handle *
1626 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1627 struct list_head *match_head,
1628 struct mlx5_flow_spec *spec,
1629 struct mlx5_flow_act *flow_act,
1630 struct mlx5_flow_destination *dest,
1634 struct mlx5_flow_steering *steering = get_steering(&ft->node);
1635 struct mlx5_flow_group *g;
1636 struct mlx5_flow_handle *rule;
1637 struct match_list *iter;
1638 bool take_write = false;
1643 fte = alloc_fte(ft, spec, flow_act);
1645 return ERR_PTR(-ENOMEM);
1647 search_again_locked:
1648 version = matched_fgs_get_version(match_head);
1649 if (flow_act->flags & FLOW_ACT_NO_APPEND)
1651 /* Try to find a fg that already contains a matching fte */
1652 list_for_each_entry(iter, match_head, list) {
1653 struct fs_fte *fte_tmp;
1656 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1659 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1660 up_write_ref_node(&fte_tmp->node, false);
1661 tree_put_node(&fte_tmp->node, false);
1662 kmem_cache_free(steering->ftes_cache, fte);
1667 /* No group with matching fte found, or we skipped the search.
1668 * Try to add a new fte to any matching fg.
1671 /* Check the ft version, for case that new flow group
1672 * was added while the fgs weren't locked
1674 if (atomic_read(&ft->node.version) != ft_version) {
1675 rule = ERR_PTR(-EAGAIN);
1679 /* Check the fgs version, for case the new FTE with the
1680 * same values was added while the fgs weren't locked
1682 if (version != matched_fgs_get_version(match_head)) {
1684 goto search_again_locked;
1687 list_for_each_entry(iter, match_head, list) {
1690 if (!g->node.active)
1693 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1695 err = insert_fte(g, fte);
1697 up_write_ref_node(&g->node, false);
1700 kmem_cache_free(steering->ftes_cache, fte);
1701 return ERR_PTR(err);
1704 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1705 up_write_ref_node(&g->node, false);
1706 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1707 up_write_ref_node(&fte->node, false);
1708 tree_put_node(&fte->node, false);
1711 rule = ERR_PTR(-ENOENT);
1713 kmem_cache_free(steering->ftes_cache, fte);
1717 static struct mlx5_flow_handle *
1718 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1719 struct mlx5_flow_spec *spec,
1720 struct mlx5_flow_act *flow_act,
1721 struct mlx5_flow_destination *dest,
1725 struct mlx5_flow_steering *steering = get_steering(&ft->node);
1726 struct mlx5_flow_group *g;
1727 struct mlx5_flow_handle *rule;
1728 struct match_list_head match_head;
1729 bool take_write = false;
1735 if (!check_valid_spec(spec))
1736 return ERR_PTR(-EINVAL);
1738 for (i = 0; i < dest_num; i++) {
1739 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1740 return ERR_PTR(-EINVAL);
1742 nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1743 search_again_locked:
1744 version = atomic_read(&ft->node.version);
1746 /* Collect all fgs which has a matching match_criteria */
1747 err = build_match_list(&match_head, ft, spec);
1750 up_write_ref_node(&ft->node, false);
1752 up_read_ref_node(&ft->node);
1753 return ERR_PTR(err);
1757 up_read_ref_node(&ft->node);
1759 rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1761 free_match_list(&match_head);
1762 if (!IS_ERR(rule) ||
1763 (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1765 up_write_ref_node(&ft->node, false);
1770 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1774 if (PTR_ERR(rule) == -EAGAIN ||
1775 version != atomic_read(&ft->node.version))
1776 goto search_again_locked;
1778 g = alloc_auto_flow_group(ft, spec);
1781 up_write_ref_node(&ft->node, false);
1785 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1786 up_write_ref_node(&ft->node, false);
1788 err = create_auto_flow_group(ft, g);
1790 goto err_release_fg;
1792 fte = alloc_fte(ft, spec, flow_act);
1795 goto err_release_fg;
1798 err = insert_fte(g, fte);
1800 kmem_cache_free(steering->ftes_cache, fte);
1801 goto err_release_fg;
1804 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1805 up_write_ref_node(&g->node, false);
1806 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1807 up_write_ref_node(&fte->node, false);
1808 tree_put_node(&fte->node, false);
1809 tree_put_node(&g->node, false);
1813 up_write_ref_node(&g->node, false);
1814 tree_put_node(&g->node, false);
1815 return ERR_PTR(err);
1818 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1820 return ((ft->type == FS_FT_NIC_RX) &&
1821 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1824 struct mlx5_flow_handle *
1825 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1826 struct mlx5_flow_spec *spec,
1827 struct mlx5_flow_act *flow_act,
1828 struct mlx5_flow_destination *dest,
1831 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1832 struct mlx5_flow_destination gen_dest = {};
1833 struct mlx5_flow_table *next_ft = NULL;
1834 struct mlx5_flow_handle *handle = NULL;
1835 u32 sw_action = flow_act->action;
1836 struct fs_prio *prio;
1838 fs_get_obj(prio, ft->node.parent);
1839 if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1840 if (!fwd_next_prio_supported(ft))
1841 return ERR_PTR(-EOPNOTSUPP);
1843 return ERR_PTR(-EINVAL);
1844 mutex_lock(&root->chain_lock);
1845 next_ft = find_next_chained_ft(prio);
1847 gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1848 gen_dest.ft = next_ft;
1851 flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1853 mutex_unlock(&root->chain_lock);
1854 return ERR_PTR(-EOPNOTSUPP);
1858 handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1860 if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1861 if (!IS_ERR_OR_NULL(handle) &&
1862 (list_empty(&handle->rule[0]->next_ft))) {
1863 mutex_lock(&next_ft->lock);
1864 list_add(&handle->rule[0]->next_ft,
1865 &next_ft->fwd_rules);
1866 mutex_unlock(&next_ft->lock);
1867 handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1869 mutex_unlock(&root->chain_lock);
1873 EXPORT_SYMBOL(mlx5_add_flow_rules);
1875 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1880 /* In order to consolidate the HW changes we lock the FTE for other
1881 * changes, and increase its refcount, in order not to perform the
1882 * "del" functions of the FTE. Will handle them here.
1883 * The removal of the rules is done under locked FTE.
1884 * After removing all the handle's rules, if there are remaining
1885 * rules, it means we just need to modify the FTE in FW, and
1886 * unlock/decrease the refcount we increased before.
1887 * Otherwise, it means the FTE should be deleted. First delete the
1888 * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
1889 * the FTE, which will handle the last decrease of the refcount, as
1890 * well as required handling of its parent.
1892 fs_get_obj(fte, handle->rule[0]->node.parent);
1893 down_write_ref_node(&fte->node, false);
1894 for (i = handle->num_rules - 1; i >= 0; i--)
1895 tree_remove_node(&handle->rule[i]->node, true);
1896 if (fte->modify_mask && fte->dests_size) {
1898 up_write_ref_node(&fte->node, false);
1900 del_hw_fte(&fte->node);
1901 up_write(&fte->node.lock);
1902 tree_put_node(&fte->node, false);
1906 EXPORT_SYMBOL(mlx5_del_flow_rules);
1908 /* Assuming prio->node.children(flow tables) is sorted by level */
1909 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1911 struct fs_prio *prio;
1913 fs_get_obj(prio, ft->node.parent);
1915 if (!list_is_last(&ft->node.list, &prio->node.children))
1916 return list_next_entry(ft, node.list);
1917 return find_next_chained_ft(prio);
1920 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1922 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1923 struct mlx5_ft_underlay_qp *uqp;
1924 struct mlx5_flow_table *new_root_ft = NULL;
1928 if (root->root_ft != ft)
1931 new_root_ft = find_next_ft(ft);
1933 root->root_ft = NULL;
1937 if (list_empty(&root->underlay_qpns)) {
1938 /* Don't set any QPN (zero) in case QPN list is empty */
1940 err = root->cmds->update_root_ft(root, new_root_ft,
1943 list_for_each_entry(uqp, &root->underlay_qpns, list) {
1945 err = root->cmds->update_root_ft(root,
1954 mlx5_core_warn(root->dev,
1955 "Update root flow table of id(%u) qpn(%d) failed\n",
1958 root->root_ft = new_root_ft;
1963 /* Connect flow table from previous priority to
1964 * the next flow table.
1966 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1968 struct mlx5_core_dev *dev = get_dev(&ft->node);
1969 struct mlx5_flow_table *next_ft;
1970 struct fs_prio *prio;
1973 err = update_root_ft_destroy(ft);
1977 fs_get_obj(prio, ft->node.parent);
1978 if (!(list_first_entry(&prio->node.children,
1979 struct mlx5_flow_table,
1983 next_ft = find_next_chained_ft(prio);
1984 err = connect_fwd_rules(dev, next_ft, ft);
1988 err = connect_prev_fts(dev, next_ft, prio);
1990 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1995 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1997 struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2000 mutex_lock(&root->chain_lock);
2001 err = disconnect_flow_table(ft);
2003 mutex_unlock(&root->chain_lock);
2006 if (tree_remove_node(&ft->node, false))
2007 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2009 mutex_unlock(&root->chain_lock);
2013 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2015 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2017 if (tree_remove_node(&fg->node, false))
2018 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2022 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2025 struct mlx5_flow_steering *steering = dev->priv.steering;
2027 if (!steering || !steering->fdb_sub_ns)
2030 return steering->fdb_sub_ns[n];
2032 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2034 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2035 enum mlx5_flow_namespace_type type)
2037 struct mlx5_flow_steering *steering = dev->priv.steering;
2038 struct mlx5_flow_root_namespace *root_ns;
2040 struct fs_prio *fs_prio;
2041 struct mlx5_flow_namespace *ns;
2047 case MLX5_FLOW_NAMESPACE_FDB:
2048 if (steering->fdb_root_ns)
2049 return &steering->fdb_root_ns->ns;
2051 case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2052 if (steering->sniffer_rx_root_ns)
2053 return &steering->sniffer_rx_root_ns->ns;
2055 case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2056 if (steering->sniffer_tx_root_ns)
2057 return &steering->sniffer_tx_root_ns->ns;
2059 case MLX5_FLOW_NAMESPACE_RDMA_RX:
2060 if (steering->rdma_rx_root_ns)
2061 return &steering->rdma_rx_root_ns->ns;
2067 if (type == MLX5_FLOW_NAMESPACE_EGRESS) {
2068 root_ns = steering->egress_root_ns;
2069 } else { /* Must be NIC RX */
2070 root_ns = steering->root_ns;
2077 fs_prio = find_prio(&root_ns->ns, prio);
2081 ns = list_first_entry(&fs_prio->node.children,
2087 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2089 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2090 enum mlx5_flow_namespace_type type,
2093 struct mlx5_flow_steering *steering = dev->priv.steering;
2095 if (!steering || vport >= mlx5_eswitch_get_total_vports(dev))
2099 case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2100 if (steering->esw_egress_root_ns &&
2101 steering->esw_egress_root_ns[vport])
2102 return &steering->esw_egress_root_ns[vport]->ns;
2105 case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2106 if (steering->esw_ingress_root_ns &&
2107 steering->esw_ingress_root_ns[vport])
2108 return &steering->esw_ingress_root_ns[vport]->ns;
2116 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2119 enum fs_node_type type)
2121 struct fs_prio *fs_prio;
2123 fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2125 return ERR_PTR(-ENOMEM);
2127 fs_prio->node.type = type;
2128 tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2129 tree_add_node(&fs_prio->node, &ns->node);
2130 fs_prio->num_levels = num_levels;
2131 fs_prio->prio = prio;
2132 list_add_tail(&fs_prio->node.list, &ns->node.children);
2137 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2141 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2144 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2145 unsigned int prio, int num_levels)
2147 return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2150 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2153 ns->node.type = FS_TYPE_NAMESPACE;
2158 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
2160 struct mlx5_flow_namespace *ns;
2162 ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2164 return ERR_PTR(-ENOMEM);
2166 fs_init_namespace(ns);
2167 tree_init_node(&ns->node, NULL, del_sw_ns);
2168 tree_add_node(&ns->node, &prio->node);
2169 list_add_tail(&ns->node.list, &prio->node.children);
2174 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2175 struct init_tree_node *prio_metadata)
2177 struct fs_prio *fs_prio;
2180 for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2181 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2182 if (IS_ERR(fs_prio))
2183 return PTR_ERR(fs_prio);
2188 #define FLOW_TABLE_BIT_SZ 1
2189 #define GET_FLOW_TABLE_CAP(dev, offset) \
2190 ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) + \
2192 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2193 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2197 for (i = 0; i < caps->arr_sz; i++) {
2198 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2204 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2205 struct init_tree_node *init_node,
2206 struct fs_node *fs_parent_node,
2207 struct init_tree_node *init_parent_node,
2210 int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2211 flow_table_properties_nic_receive.
2213 struct mlx5_flow_namespace *fs_ns;
2214 struct fs_prio *fs_prio;
2215 struct fs_node *base;
2219 if (init_node->type == FS_TYPE_PRIO) {
2220 if ((init_node->min_ft_level > max_ft_level) ||
2221 !has_required_caps(steering->dev, &init_node->caps))
2224 fs_get_obj(fs_ns, fs_parent_node);
2225 if (init_node->num_leaf_prios)
2226 return create_leaf_prios(fs_ns, prio, init_node);
2227 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2228 if (IS_ERR(fs_prio))
2229 return PTR_ERR(fs_prio);
2230 base = &fs_prio->node;
2231 } else if (init_node->type == FS_TYPE_NAMESPACE) {
2232 fs_get_obj(fs_prio, fs_parent_node);
2233 fs_ns = fs_create_namespace(fs_prio);
2235 return PTR_ERR(fs_ns);
2236 base = &fs_ns->node;
2241 for (i = 0; i < init_node->ar_size; i++) {
2242 err = init_root_tree_recursive(steering, &init_node->children[i],
2243 base, init_node, prio);
2246 if (init_node->children[i].type == FS_TYPE_PRIO &&
2247 init_node->children[i].num_leaf_prios) {
2248 prio += init_node->children[i].num_leaf_prios;
2255 static int init_root_tree(struct mlx5_flow_steering *steering,
2256 struct init_tree_node *init_node,
2257 struct fs_node *fs_parent_node)
2260 struct mlx5_flow_namespace *fs_ns;
2263 fs_get_obj(fs_ns, fs_parent_node);
2264 for (i = 0; i < init_node->ar_size; i++) {
2265 err = init_root_tree_recursive(steering, &init_node->children[i],
2274 static struct mlx5_flow_root_namespace
2275 *create_root_ns(struct mlx5_flow_steering *steering,
2276 enum fs_flow_table_type table_type)
2278 const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2279 struct mlx5_flow_root_namespace *root_ns;
2280 struct mlx5_flow_namespace *ns;
2282 if (mlx5_accel_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2283 (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2284 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2286 /* Create the root namespace */
2287 root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2291 root_ns->dev = steering->dev;
2292 root_ns->table_type = table_type;
2293 root_ns->cmds = cmds;
2295 INIT_LIST_HEAD(&root_ns->underlay_qpns);
2298 fs_init_namespace(ns);
2299 mutex_init(&root_ns->chain_lock);
2300 tree_init_node(&ns->node, NULL, NULL);
2301 tree_add_node(&ns->node, NULL);
2306 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2308 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2310 struct fs_prio *prio;
2312 fs_for_each_prio(prio, ns) {
2313 /* This updates prio start_level and num_levels */
2314 set_prio_attrs_in_prio(prio, acc_level);
2315 acc_level += prio->num_levels;
2320 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2322 struct mlx5_flow_namespace *ns;
2323 int acc_level_ns = acc_level;
2325 prio->start_level = acc_level;
2326 fs_for_each_ns(ns, prio)
2327 /* This updates start_level and num_levels of ns's priority descendants */
2328 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2329 if (!prio->num_levels)
2330 prio->num_levels = acc_level_ns - prio->start_level;
2331 WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2334 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2336 struct mlx5_flow_namespace *ns = &root_ns->ns;
2337 struct fs_prio *prio;
2338 int start_level = 0;
2340 fs_for_each_prio(prio, ns) {
2341 set_prio_attrs_in_prio(prio, start_level);
2342 start_level += prio->num_levels;
2346 #define ANCHOR_PRIO 0
2347 #define ANCHOR_SIZE 1
2348 #define ANCHOR_LEVEL 0
2349 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2351 struct mlx5_flow_namespace *ns = NULL;
2352 struct mlx5_flow_table_attr ft_attr = {};
2353 struct mlx5_flow_table *ft;
2355 ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2359 ft_attr.max_fte = ANCHOR_SIZE;
2360 ft_attr.level = ANCHOR_LEVEL;
2361 ft_attr.prio = ANCHOR_PRIO;
2363 ft = mlx5_create_flow_table(ns, &ft_attr);
2365 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2371 static int init_root_ns(struct mlx5_flow_steering *steering)
2375 steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2376 if (!steering->root_ns)
2379 err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2383 set_prio_attrs(steering->root_ns);
2384 err = create_anchor_flow_table(steering);
2391 cleanup_root_ns(steering->root_ns);
2392 steering->root_ns = NULL;
2396 static void clean_tree(struct fs_node *node)
2399 struct fs_node *iter;
2400 struct fs_node *temp;
2402 tree_get_node(node);
2403 list_for_each_entry_safe(iter, temp, &node->children, list)
2405 tree_put_node(node, false);
2406 tree_remove_node(node, false);
2410 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2415 clean_tree(&root_ns->ns.node);
2418 static void cleanup_egress_acls_root_ns(struct mlx5_core_dev *dev)
2420 struct mlx5_flow_steering *steering = dev->priv.steering;
2423 if (!steering->esw_egress_root_ns)
2426 for (i = 0; i < mlx5_eswitch_get_total_vports(dev); i++)
2427 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2429 kfree(steering->esw_egress_root_ns);
2430 steering->esw_egress_root_ns = NULL;
2433 static void cleanup_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2435 struct mlx5_flow_steering *steering = dev->priv.steering;
2438 if (!steering->esw_ingress_root_ns)
2441 for (i = 0; i < mlx5_eswitch_get_total_vports(dev); i++)
2442 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2444 kfree(steering->esw_ingress_root_ns);
2445 steering->esw_ingress_root_ns = NULL;
2448 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2450 struct mlx5_flow_steering *steering = dev->priv.steering;
2452 cleanup_root_ns(steering->root_ns);
2453 cleanup_egress_acls_root_ns(dev);
2454 cleanup_ingress_acls_root_ns(dev);
2455 cleanup_root_ns(steering->fdb_root_ns);
2456 steering->fdb_root_ns = NULL;
2457 kfree(steering->fdb_sub_ns);
2458 steering->fdb_sub_ns = NULL;
2459 cleanup_root_ns(steering->sniffer_rx_root_ns);
2460 cleanup_root_ns(steering->sniffer_tx_root_ns);
2461 cleanup_root_ns(steering->rdma_rx_root_ns);
2462 cleanup_root_ns(steering->egress_root_ns);
2463 mlx5_cleanup_fc_stats(dev);
2464 kmem_cache_destroy(steering->ftes_cache);
2465 kmem_cache_destroy(steering->fgs_cache);
2469 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2471 struct fs_prio *prio;
2473 steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2474 if (!steering->sniffer_tx_root_ns)
2477 /* Create single prio */
2478 prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2479 return PTR_ERR_OR_ZERO(prio);
2482 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2484 struct fs_prio *prio;
2486 steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2487 if (!steering->sniffer_rx_root_ns)
2490 /* Create single prio */
2491 prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2492 return PTR_ERR_OR_ZERO(prio);
2495 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2497 struct fs_prio *prio;
2499 steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2500 if (!steering->rdma_rx_root_ns)
2503 steering->rdma_rx_root_ns->def_miss_action =
2504 MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN;
2506 /* Create single prio */
2507 prio = fs_create_prio(&steering->rdma_rx_root_ns->ns, 0, 1);
2508 return PTR_ERR_OR_ZERO(prio);
2510 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2512 struct mlx5_flow_namespace *ns;
2513 struct fs_prio *maj_prio;
2514 struct fs_prio *min_prio;
2520 steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2521 if (!steering->fdb_root_ns)
2524 steering->fdb_sub_ns = kzalloc(sizeof(steering->fdb_sub_ns) *
2525 (FDB_MAX_CHAIN + 1), GFP_KERNEL);
2526 if (!steering->fdb_sub_ns)
2529 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2531 if (IS_ERR(maj_prio)) {
2532 err = PTR_ERR(maj_prio);
2536 levels = 2 * FDB_MAX_PRIO * (FDB_MAX_CHAIN + 1);
2537 maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2540 if (IS_ERR(maj_prio)) {
2541 err = PTR_ERR(maj_prio);
2545 for (chain = 0; chain <= FDB_MAX_CHAIN; chain++) {
2546 ns = fs_create_namespace(maj_prio);
2552 for (prio = 0; prio < FDB_MAX_PRIO * (chain + 1); prio++) {
2553 min_prio = fs_create_prio(ns, prio, 2);
2554 if (IS_ERR(min_prio)) {
2555 err = PTR_ERR(min_prio);
2560 steering->fdb_sub_ns[chain] = ns;
2563 maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2564 if (IS_ERR(maj_prio)) {
2565 err = PTR_ERR(maj_prio);
2569 set_prio_attrs(steering->fdb_root_ns);
2573 cleanup_root_ns(steering->fdb_root_ns);
2574 kfree(steering->fdb_sub_ns);
2575 steering->fdb_sub_ns = NULL;
2576 steering->fdb_root_ns = NULL;
2580 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2582 struct fs_prio *prio;
2584 steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2585 if (!steering->esw_egress_root_ns[vport])
2589 prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2590 return PTR_ERR_OR_ZERO(prio);
2593 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2595 struct fs_prio *prio;
2597 steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2598 if (!steering->esw_ingress_root_ns[vport])
2602 prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2603 return PTR_ERR_OR_ZERO(prio);
2606 static int init_egress_acls_root_ns(struct mlx5_core_dev *dev)
2608 struct mlx5_flow_steering *steering = dev->priv.steering;
2609 int total_vports = mlx5_eswitch_get_total_vports(dev);
2613 steering->esw_egress_root_ns =
2614 kcalloc(total_vports,
2615 sizeof(*steering->esw_egress_root_ns),
2617 if (!steering->esw_egress_root_ns)
2620 for (i = 0; i < total_vports; i++) {
2621 err = init_egress_acl_root_ns(steering, i);
2623 goto cleanup_root_ns;
2629 for (i--; i >= 0; i--)
2630 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2631 kfree(steering->esw_egress_root_ns);
2632 steering->esw_egress_root_ns = NULL;
2636 static int init_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2638 struct mlx5_flow_steering *steering = dev->priv.steering;
2639 int total_vports = mlx5_eswitch_get_total_vports(dev);
2643 steering->esw_ingress_root_ns =
2644 kcalloc(total_vports,
2645 sizeof(*steering->esw_ingress_root_ns),
2647 if (!steering->esw_ingress_root_ns)
2650 for (i = 0; i < total_vports; i++) {
2651 err = init_ingress_acl_root_ns(steering, i);
2653 goto cleanup_root_ns;
2659 for (i--; i >= 0; i--)
2660 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2661 kfree(steering->esw_ingress_root_ns);
2662 steering->esw_ingress_root_ns = NULL;
2666 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2670 steering->egress_root_ns = create_root_ns(steering,
2672 if (!steering->egress_root_ns)
2675 err = init_root_tree(steering, &egress_root_fs,
2676 &steering->egress_root_ns->ns.node);
2679 set_prio_attrs(steering->egress_root_ns);
2682 cleanup_root_ns(steering->egress_root_ns);
2683 steering->egress_root_ns = NULL;
2687 int mlx5_init_fs(struct mlx5_core_dev *dev)
2689 struct mlx5_flow_steering *steering;
2692 err = mlx5_init_fc_stats(dev);
2696 steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2699 steering->dev = dev;
2700 dev->priv.steering = steering;
2702 steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2703 sizeof(struct mlx5_flow_group), 0,
2705 steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2707 if (!steering->ftes_cache || !steering->fgs_cache) {
2712 if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2713 (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2714 ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2715 MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2716 MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2717 err = init_root_ns(steering);
2722 if (MLX5_ESWITCH_MANAGER(dev)) {
2723 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2724 err = init_fdb_root_ns(steering);
2728 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2729 err = init_egress_acls_root_ns(dev);
2733 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2734 err = init_ingress_acls_root_ns(dev);
2740 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2741 err = init_sniffer_rx_root_ns(steering);
2746 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2747 err = init_sniffer_tx_root_ns(steering);
2752 if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
2753 MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
2754 err = init_rdma_rx_root_ns(steering);
2759 if (MLX5_IPSEC_DEV(dev) || MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
2760 err = init_egress_root_ns(steering);
2767 mlx5_cleanup_fs(dev);
2771 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2773 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2774 struct mlx5_ft_underlay_qp *new_uqp;
2777 new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
2781 mutex_lock(&root->chain_lock);
2783 if (!root->root_ft) {
2785 goto update_ft_fail;
2788 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
2791 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
2793 goto update_ft_fail;
2796 new_uqp->qpn = underlay_qpn;
2797 list_add_tail(&new_uqp->list, &root->underlay_qpns);
2799 mutex_unlock(&root->chain_lock);
2804 mutex_unlock(&root->chain_lock);
2808 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
2810 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2812 struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2813 struct mlx5_ft_underlay_qp *uqp;
2817 mutex_lock(&root->chain_lock);
2818 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2819 if (uqp->qpn == underlay_qpn) {
2826 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
2832 err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
2835 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
2838 list_del(&uqp->list);
2839 mutex_unlock(&root->chain_lock);
2845 mutex_unlock(&root->chain_lock);
2848 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);