2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/mutex.h>
34 #include <linux/mlx5/driver.h>
36 #include "mlx5_core.h"
40 #define INIT_TREE_NODE_ARRAY_SIZE(...) (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
41 sizeof(struct init_tree_node))
43 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
44 ...) {.type = FS_TYPE_PRIO,\
45 .min_ft_level = min_level_val,\
46 .num_levels = num_levels_val,\
47 .num_leaf_prios = num_prios_val,\
49 .children = (struct init_tree_node[]) {__VA_ARGS__},\
50 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
53 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
54 ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
57 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
58 .children = (struct init_tree_node[]) {__VA_ARGS__},\
59 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
62 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
65 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
67 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
68 .caps = (long[]) {__VA_ARGS__} }
70 #define FS_CHAINING_CAPS FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
71 FS_CAP(flow_table_properties_nic_receive.modify_root), \
72 FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
73 FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
75 #define LEFTOVERS_NUM_LEVELS 1
76 #define LEFTOVERS_NUM_PRIOS 1
78 #define BY_PASS_PRIO_NUM_LEVELS 1
79 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
82 #define ETHTOOL_PRIO_NUM_LEVELS 1
83 #define ETHTOOL_NUM_PRIOS 11
84 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
85 /* Vlan, mac, ttc, aRFS */
86 #define KERNEL_NIC_PRIO_NUM_LEVELS 4
87 #define KERNEL_NIC_NUM_PRIOS 1
88 /* One more level for tc */
89 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
91 #define ANCHOR_NUM_LEVELS 1
92 #define ANCHOR_NUM_PRIOS 1
93 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
95 #define OFFLOADS_MAX_FT 1
96 #define OFFLOADS_NUM_PRIOS 1
97 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
99 #define LAG_PRIO_NUM_LEVELS 1
100 #define LAG_NUM_PRIOS 1
101 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
107 static struct init_tree_node
{
108 enum fs_node_type type
;
109 struct init_tree_node
*children
;
111 struct node_caps caps
;
117 .type
= FS_TYPE_NAMESPACE
,
119 .children
= (struct init_tree_node
[]) {
120 ADD_PRIO(0, BY_PASS_MIN_LEVEL
, 0,
122 ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS
,
123 BY_PASS_PRIO_NUM_LEVELS
))),
124 ADD_PRIO(0, LAG_MIN_LEVEL
, 0,
126 ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS
,
127 LAG_PRIO_NUM_LEVELS
))),
128 ADD_PRIO(0, OFFLOADS_MIN_LEVEL
, 0, {},
129 ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS
, OFFLOADS_MAX_FT
))),
130 ADD_PRIO(0, ETHTOOL_MIN_LEVEL
, 0,
132 ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS
,
133 ETHTOOL_PRIO_NUM_LEVELS
))),
134 ADD_PRIO(0, KERNEL_MIN_LEVEL
, 0, {},
135 ADD_NS(ADD_MULTIPLE_PRIO(1, 1),
136 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS
,
137 KERNEL_NIC_PRIO_NUM_LEVELS
))),
138 ADD_PRIO(0, BY_PASS_MIN_LEVEL
, 0,
140 ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS
, LEFTOVERS_NUM_LEVELS
))),
141 ADD_PRIO(0, ANCHOR_MIN_LEVEL
, 0, {},
142 ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS
, ANCHOR_NUM_LEVELS
))),
146 enum fs_i_mutex_lock_class
{
147 FS_MUTEX_GRANDPARENT
,
152 static void del_rule(struct fs_node
*node
);
153 static void del_flow_table(struct fs_node
*node
);
154 static void del_flow_group(struct fs_node
*node
);
155 static void del_fte(struct fs_node
*node
);
157 static void tree_init_node(struct fs_node
*node
,
158 unsigned int refcount
,
159 void (*remove_func
)(struct fs_node
*))
161 atomic_set(&node
->refcount
, refcount
);
162 INIT_LIST_HEAD(&node
->list
);
163 INIT_LIST_HEAD(&node
->children
);
164 mutex_init(&node
->lock
);
165 node
->remove_func
= remove_func
;
168 static void tree_add_node(struct fs_node
*node
, struct fs_node
*parent
)
171 atomic_inc(&parent
->refcount
);
172 node
->parent
= parent
;
174 /* Parent is the root */
178 node
->root
= parent
->root
;
181 static void tree_get_node(struct fs_node
*node
)
183 atomic_inc(&node
->refcount
);
186 static void nested_lock_ref_node(struct fs_node
*node
,
187 enum fs_i_mutex_lock_class
class)
190 mutex_lock_nested(&node
->lock
, class);
191 atomic_inc(&node
->refcount
);
195 static void lock_ref_node(struct fs_node
*node
)
198 mutex_lock(&node
->lock
);
199 atomic_inc(&node
->refcount
);
203 static void unlock_ref_node(struct fs_node
*node
)
206 atomic_dec(&node
->refcount
);
207 mutex_unlock(&node
->lock
);
211 static void tree_put_node(struct fs_node
*node
)
213 struct fs_node
*parent_node
= node
->parent
;
215 lock_ref_node(parent_node
);
216 if (atomic_dec_and_test(&node
->refcount
)) {
218 list_del_init(&node
->list
);
219 if (node
->remove_func
)
220 node
->remove_func(node
);
224 unlock_ref_node(parent_node
);
225 if (!node
&& parent_node
)
226 tree_put_node(parent_node
);
229 static int tree_remove_node(struct fs_node
*node
)
231 if (atomic_read(&node
->refcount
) > 1) {
232 atomic_dec(&node
->refcount
);
239 static struct fs_prio
*find_prio(struct mlx5_flow_namespace
*ns
,
242 struct fs_prio
*iter_prio
;
244 fs_for_each_prio(iter_prio
, ns
) {
245 if (iter_prio
->prio
== prio
)
252 static bool masked_memcmp(void *mask
, void *val1
, void *val2
, size_t size
)
256 for (i
= 0; i
< size
; i
++, mask
++, val1
++, val2
++)
257 if ((*((u8
*)val1
) & (*(u8
*)mask
)) !=
258 ((*(u8
*)val2
) & (*(u8
*)mask
)))
264 static bool compare_match_value(struct mlx5_flow_group_mask
*mask
,
265 void *fte_param1
, void *fte_param2
)
267 if (mask
->match_criteria_enable
&
268 1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS
) {
269 void *fte_match1
= MLX5_ADDR_OF(fte_match_param
,
270 fte_param1
, outer_headers
);
271 void *fte_match2
= MLX5_ADDR_OF(fte_match_param
,
272 fte_param2
, outer_headers
);
273 void *fte_mask
= MLX5_ADDR_OF(fte_match_param
,
274 mask
->match_criteria
, outer_headers
);
276 if (!masked_memcmp(fte_mask
, fte_match1
, fte_match2
,
277 MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4
)))
281 if (mask
->match_criteria_enable
&
282 1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS
) {
283 void *fte_match1
= MLX5_ADDR_OF(fte_match_param
,
284 fte_param1
, misc_parameters
);
285 void *fte_match2
= MLX5_ADDR_OF(fte_match_param
,
286 fte_param2
, misc_parameters
);
287 void *fte_mask
= MLX5_ADDR_OF(fte_match_param
,
288 mask
->match_criteria
, misc_parameters
);
290 if (!masked_memcmp(fte_mask
, fte_match1
, fte_match2
,
291 MLX5_ST_SZ_BYTES(fte_match_set_misc
)))
295 if (mask
->match_criteria_enable
&
296 1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS
) {
297 void *fte_match1
= MLX5_ADDR_OF(fte_match_param
,
298 fte_param1
, inner_headers
);
299 void *fte_match2
= MLX5_ADDR_OF(fte_match_param
,
300 fte_param2
, inner_headers
);
301 void *fte_mask
= MLX5_ADDR_OF(fte_match_param
,
302 mask
->match_criteria
, inner_headers
);
304 if (!masked_memcmp(fte_mask
, fte_match1
, fte_match2
,
305 MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4
)))
311 static bool compare_match_criteria(u8 match_criteria_enable1
,
312 u8 match_criteria_enable2
,
313 void *mask1
, void *mask2
)
315 return match_criteria_enable1
== match_criteria_enable2
&&
316 !memcmp(mask1
, mask2
, MLX5_ST_SZ_BYTES(fte_match_param
));
319 static struct mlx5_flow_root_namespace
*find_root(struct fs_node
*node
)
321 struct fs_node
*root
;
322 struct mlx5_flow_namespace
*ns
;
326 if (WARN_ON(root
->type
!= FS_TYPE_NAMESPACE
)) {
327 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
331 ns
= container_of(root
, struct mlx5_flow_namespace
, node
);
332 return container_of(ns
, struct mlx5_flow_root_namespace
, ns
);
335 static inline struct mlx5_core_dev
*get_dev(struct fs_node
*node
)
337 struct mlx5_flow_root_namespace
*root
= find_root(node
);
344 static void del_flow_table(struct fs_node
*node
)
346 struct mlx5_flow_table
*ft
;
347 struct mlx5_core_dev
*dev
;
348 struct fs_prio
*prio
;
351 fs_get_obj(ft
, node
);
352 dev
= get_dev(&ft
->node
);
354 err
= mlx5_cmd_destroy_flow_table(dev
, ft
);
356 mlx5_core_warn(dev
, "flow steering can't destroy ft\n");
357 fs_get_obj(prio
, ft
->node
.parent
);
361 static void del_rule(struct fs_node
*node
)
363 struct mlx5_flow_rule
*rule
;
364 struct mlx5_flow_table
*ft
;
365 struct mlx5_flow_group
*fg
;
369 struct mlx5_core_dev
*dev
= get_dev(node
);
370 int match_len
= MLX5_ST_SZ_BYTES(fte_match_param
);
373 match_value
= mlx5_vzalloc(match_len
);
375 mlx5_core_warn(dev
, "failed to allocate inbox\n");
379 fs_get_obj(rule
, node
);
380 fs_get_obj(fte
, rule
->node
.parent
);
381 fs_get_obj(fg
, fte
->node
.parent
);
382 memcpy(match_value
, fte
->val
, sizeof(fte
->val
));
383 fs_get_obj(ft
, fg
->node
.parent
);
384 list_del(&rule
->node
.list
);
385 if (rule
->sw_action
== MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO
) {
386 mutex_lock(&rule
->dest_attr
.ft
->lock
);
387 list_del(&rule
->next_ft
);
388 mutex_unlock(&rule
->dest_attr
.ft
->lock
);
390 if ((fte
->action
& MLX5_FLOW_CONTEXT_ACTION_FWD_DEST
) &&
392 modify_mask
= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST
),
393 err
= mlx5_cmd_update_fte(dev
, ft
,
399 "%s can't del rule fg id=%d fte_index=%d\n",
400 __func__
, fg
->id
, fte
->index
);
405 static void del_fte(struct fs_node
*node
)
407 struct mlx5_flow_table
*ft
;
408 struct mlx5_flow_group
*fg
;
409 struct mlx5_core_dev
*dev
;
413 fs_get_obj(fte
, node
);
414 fs_get_obj(fg
, fte
->node
.parent
);
415 fs_get_obj(ft
, fg
->node
.parent
);
417 dev
= get_dev(&ft
->node
);
418 err
= mlx5_cmd_delete_fte(dev
, ft
,
422 "flow steering can't delete fte in index %d of flow group id %d\n",
429 static void del_flow_group(struct fs_node
*node
)
431 struct mlx5_flow_group
*fg
;
432 struct mlx5_flow_table
*ft
;
433 struct mlx5_core_dev
*dev
;
435 fs_get_obj(fg
, node
);
436 fs_get_obj(ft
, fg
->node
.parent
);
437 dev
= get_dev(&ft
->node
);
439 if (mlx5_cmd_destroy_flow_group(dev
, ft
, fg
->id
))
440 mlx5_core_warn(dev
, "flow steering can't destroy fg %d of ft %d\n",
444 static struct fs_fte
*alloc_fte(u8 action
,
451 fte
= kzalloc(sizeof(*fte
), GFP_KERNEL
);
453 return ERR_PTR(-ENOMEM
);
455 memcpy(fte
->val
, match_value
, sizeof(fte
->val
));
456 fte
->node
.type
= FS_TYPE_FLOW_ENTRY
;
457 fte
->flow_tag
= flow_tag
;
459 fte
->action
= action
;
464 static struct mlx5_flow_group
*alloc_flow_group(u32
*create_fg_in
)
466 struct mlx5_flow_group
*fg
;
467 void *match_criteria
= MLX5_ADDR_OF(create_flow_group_in
,
468 create_fg_in
, match_criteria
);
469 u8 match_criteria_enable
= MLX5_GET(create_flow_group_in
,
471 match_criteria_enable
);
472 fg
= kzalloc(sizeof(*fg
), GFP_KERNEL
);
474 return ERR_PTR(-ENOMEM
);
476 fg
->mask
.match_criteria_enable
= match_criteria_enable
;
477 memcpy(&fg
->mask
.match_criteria
, match_criteria
,
478 sizeof(fg
->mask
.match_criteria
));
479 fg
->node
.type
= FS_TYPE_FLOW_GROUP
;
480 fg
->start_index
= MLX5_GET(create_flow_group_in
, create_fg_in
,
482 fg
->max_ftes
= MLX5_GET(create_flow_group_in
, create_fg_in
,
483 end_flow_index
) - fg
->start_index
+ 1;
487 static struct mlx5_flow_table
*alloc_flow_table(int level
, u16 vport
, int max_fte
,
488 enum fs_flow_table_type table_type
,
489 enum fs_flow_table_op_mod op_mod
)
491 struct mlx5_flow_table
*ft
;
493 ft
= kzalloc(sizeof(*ft
), GFP_KERNEL
);
498 ft
->node
.type
= FS_TYPE_FLOW_TABLE
;
500 ft
->type
= table_type
;
502 ft
->max_fte
= max_fte
;
503 INIT_LIST_HEAD(&ft
->fwd_rules
);
504 mutex_init(&ft
->lock
);
509 /* If reverse is false, then we search for the first flow table in the
510 * root sub-tree from start(closest from right), else we search for the
511 * last flow table in the root sub-tree till start(closest from left).
513 static struct mlx5_flow_table
*find_closest_ft_recursive(struct fs_node
*root
,
514 struct list_head
*start
,
517 #define list_advance_entry(pos, reverse) \
518 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
520 #define list_for_each_advance_continue(pos, head, reverse) \
521 for (pos = list_advance_entry(pos, reverse); \
522 &pos->list != (head); \
523 pos = list_advance_entry(pos, reverse))
525 struct fs_node
*iter
= list_entry(start
, struct fs_node
, list
);
526 struct mlx5_flow_table
*ft
= NULL
;
531 list_for_each_advance_continue(iter
, &root
->children
, reverse
) {
532 if (iter
->type
== FS_TYPE_FLOW_TABLE
) {
533 fs_get_obj(ft
, iter
);
536 ft
= find_closest_ft_recursive(iter
, &iter
->children
, reverse
);
544 /* If reverse if false then return the first flow table in next priority of
545 * prio in the tree, else return the last flow table in the previous priority
546 * of prio in the tree.
548 static struct mlx5_flow_table
*find_closest_ft(struct fs_prio
*prio
, bool reverse
)
550 struct mlx5_flow_table
*ft
= NULL
;
551 struct fs_node
*curr_node
;
552 struct fs_node
*parent
;
554 parent
= prio
->node
.parent
;
555 curr_node
= &prio
->node
;
556 while (!ft
&& parent
) {
557 ft
= find_closest_ft_recursive(parent
, &curr_node
->list
, reverse
);
559 parent
= curr_node
->parent
;
564 /* Assuming all the tree is locked by mutex chain lock */
565 static struct mlx5_flow_table
*find_next_chained_ft(struct fs_prio
*prio
)
567 return find_closest_ft(prio
, false);
570 /* Assuming all the tree is locked by mutex chain lock */
571 static struct mlx5_flow_table
*find_prev_chained_ft(struct fs_prio
*prio
)
573 return find_closest_ft(prio
, true);
576 static int connect_fts_in_prio(struct mlx5_core_dev
*dev
,
577 struct fs_prio
*prio
,
578 struct mlx5_flow_table
*ft
)
580 struct mlx5_flow_table
*iter
;
584 fs_for_each_ft(iter
, prio
) {
586 err
= mlx5_cmd_modify_flow_table(dev
,
590 mlx5_core_warn(dev
, "Failed to modify flow table %d\n",
592 /* The driver is out of sync with the FW */
601 /* Connect flow tables from previous priority of prio to ft */
602 static int connect_prev_fts(struct mlx5_core_dev
*dev
,
603 struct mlx5_flow_table
*ft
,
604 struct fs_prio
*prio
)
606 struct mlx5_flow_table
*prev_ft
;
608 prev_ft
= find_prev_chained_ft(prio
);
610 struct fs_prio
*prev_prio
;
612 fs_get_obj(prev_prio
, prev_ft
->node
.parent
);
613 return connect_fts_in_prio(dev
, prev_prio
, ft
);
618 static int update_root_ft_create(struct mlx5_flow_table
*ft
, struct fs_prio
621 struct mlx5_flow_root_namespace
*root
= find_root(&prio
->node
);
622 int min_level
= INT_MAX
;
626 min_level
= root
->root_ft
->level
;
628 if (ft
->level
>= min_level
)
631 err
= mlx5_cmd_update_root_ft(root
->dev
, ft
);
633 mlx5_core_warn(root
->dev
, "Update root flow table of id=%u failed\n",
641 int mlx5_modify_rule_destination(struct mlx5_flow_rule
*rule
,
642 struct mlx5_flow_destination
*dest
)
644 struct mlx5_flow_table
*ft
;
645 struct mlx5_flow_group
*fg
;
647 int modify_mask
= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST
);
650 fs_get_obj(fte
, rule
->node
.parent
);
651 if (!(fte
->action
& MLX5_FLOW_CONTEXT_ACTION_FWD_DEST
))
653 lock_ref_node(&fte
->node
);
654 fs_get_obj(fg
, fte
->node
.parent
);
655 fs_get_obj(ft
, fg
->node
.parent
);
657 memcpy(&rule
->dest_attr
, dest
, sizeof(*dest
));
658 err
= mlx5_cmd_update_fte(get_dev(&ft
->node
),
662 unlock_ref_node(&fte
->node
);
667 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft */
668 static int connect_fwd_rules(struct mlx5_core_dev
*dev
,
669 struct mlx5_flow_table
*new_next_ft
,
670 struct mlx5_flow_table
*old_next_ft
)
672 struct mlx5_flow_destination dest
;
673 struct mlx5_flow_rule
*iter
;
676 /* new_next_ft and old_next_ft could be NULL only
677 * when we create/destroy the anchor flow table.
679 if (!new_next_ft
|| !old_next_ft
)
682 dest
.type
= MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE
;
683 dest
.ft
= new_next_ft
;
685 mutex_lock(&old_next_ft
->lock
);
686 list_splice_init(&old_next_ft
->fwd_rules
, &new_next_ft
->fwd_rules
);
687 mutex_unlock(&old_next_ft
->lock
);
688 list_for_each_entry(iter
, &new_next_ft
->fwd_rules
, next_ft
) {
689 err
= mlx5_modify_rule_destination(iter
, &dest
);
691 pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
697 static int connect_flow_table(struct mlx5_core_dev
*dev
, struct mlx5_flow_table
*ft
,
698 struct fs_prio
*prio
)
700 struct mlx5_flow_table
*next_ft
;
703 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
705 if (list_empty(&prio
->node
.children
)) {
706 err
= connect_prev_fts(dev
, ft
, prio
);
710 next_ft
= find_next_chained_ft(prio
);
711 err
= connect_fwd_rules(dev
, ft
, next_ft
);
716 if (MLX5_CAP_FLOWTABLE(dev
,
717 flow_table_properties_nic_receive
.modify_root
))
718 err
= update_root_ft_create(ft
, prio
);
722 static void list_add_flow_table(struct mlx5_flow_table
*ft
,
723 struct fs_prio
*prio
)
725 struct list_head
*prev
= &prio
->node
.children
;
726 struct mlx5_flow_table
*iter
;
728 fs_for_each_ft(iter
, prio
) {
729 if (iter
->level
> ft
->level
)
731 prev
= &iter
->node
.list
;
733 list_add(&ft
->node
.list
, prev
);
736 static struct mlx5_flow_table
*__mlx5_create_flow_table(struct mlx5_flow_namespace
*ns
,
737 enum fs_flow_table_op_mod op_mod
,
739 int max_fte
, u32 level
)
741 struct mlx5_flow_table
*next_ft
= NULL
;
742 struct mlx5_flow_table
*ft
;
745 struct mlx5_flow_root_namespace
*root
=
746 find_root(&ns
->node
);
747 struct fs_prio
*fs_prio
= NULL
;
750 pr_err("mlx5: flow steering failed to find root of namespace\n");
751 return ERR_PTR(-ENODEV
);
754 mutex_lock(&root
->chain_lock
);
755 fs_prio
= find_prio(ns
, prio
);
760 if (level
>= fs_prio
->num_levels
) {
764 /* The level is related to the
765 * priority level range.
767 level
+= fs_prio
->start_level
;
768 ft
= alloc_flow_table(level
,
770 max_fte
? roundup_pow_of_two(max_fte
) : 0,
778 tree_init_node(&ft
->node
, 1, del_flow_table
);
779 log_table_sz
= ft
->max_fte
? ilog2(ft
->max_fte
) : 0;
780 next_ft
= find_next_chained_ft(fs_prio
);
781 err
= mlx5_cmd_create_flow_table(root
->dev
, ft
->vport
, ft
->op_mod
, ft
->type
,
782 ft
->level
, log_table_sz
, next_ft
, &ft
->id
);
786 err
= connect_flow_table(root
->dev
, ft
, fs_prio
);
789 lock_ref_node(&fs_prio
->node
);
790 tree_add_node(&ft
->node
, &fs_prio
->node
);
791 list_add_flow_table(ft
, fs_prio
);
793 unlock_ref_node(&fs_prio
->node
);
794 mutex_unlock(&root
->chain_lock
);
797 mlx5_cmd_destroy_flow_table(root
->dev
, ft
);
801 mutex_unlock(&root
->chain_lock
);
805 struct mlx5_flow_table
*mlx5_create_flow_table(struct mlx5_flow_namespace
*ns
,
806 int prio
, int max_fte
,
809 return __mlx5_create_flow_table(ns
, FS_FT_OP_MOD_NORMAL
, 0, prio
,
813 struct mlx5_flow_table
*mlx5_create_vport_flow_table(struct mlx5_flow_namespace
*ns
,
814 int prio
, int max_fte
,
815 u32 level
, u16 vport
)
817 return __mlx5_create_flow_table(ns
, FS_FT_OP_MOD_NORMAL
, vport
, prio
,
821 struct mlx5_flow_table
*mlx5_create_lag_demux_flow_table(
822 struct mlx5_flow_namespace
*ns
,
825 return __mlx5_create_flow_table(ns
, FS_FT_OP_MOD_LAG_DEMUX
, 0, prio
, 0,
828 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table
);
830 struct mlx5_flow_table
*mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace
*ns
,
832 int num_flow_table_entries
,
836 struct mlx5_flow_table
*ft
;
838 if (max_num_groups
> num_flow_table_entries
)
839 return ERR_PTR(-EINVAL
);
841 ft
= mlx5_create_flow_table(ns
, prio
, num_flow_table_entries
, level
);
845 ft
->autogroup
.active
= true;
846 ft
->autogroup
.required_groups
= max_num_groups
;
850 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table
);
852 /* Flow table should be locked */
853 static struct mlx5_flow_group
*create_flow_group_common(struct mlx5_flow_table
*ft
,
859 struct mlx5_flow_group
*fg
;
860 struct mlx5_core_dev
*dev
= get_dev(&ft
->node
);
864 return ERR_PTR(-ENODEV
);
866 fg
= alloc_flow_group(fg_in
);
870 err
= mlx5_cmd_create_flow_group(dev
, ft
, fg_in
, &fg
->id
);
876 if (ft
->autogroup
.active
)
877 ft
->autogroup
.num_groups
++;
878 /* Add node to tree */
879 tree_init_node(&fg
->node
, !is_auto_fg
, del_flow_group
);
880 tree_add_node(&fg
->node
, &ft
->node
);
881 /* Add node to group list */
882 list_add(&fg
->node
.list
, ft
->node
.children
.prev
);
887 struct mlx5_flow_group
*mlx5_create_flow_group(struct mlx5_flow_table
*ft
,
890 struct mlx5_flow_group
*fg
;
892 if (ft
->autogroup
.active
)
893 return ERR_PTR(-EPERM
);
895 lock_ref_node(&ft
->node
);
896 fg
= create_flow_group_common(ft
, fg_in
, &ft
->node
.children
, false);
897 unlock_ref_node(&ft
->node
);
902 static struct mlx5_flow_rule
*alloc_rule(struct mlx5_flow_destination
*dest
)
904 struct mlx5_flow_rule
*rule
;
906 rule
= kzalloc(sizeof(*rule
), GFP_KERNEL
);
910 INIT_LIST_HEAD(&rule
->next_ft
);
911 rule
->node
.type
= FS_TYPE_FLOW_DEST
;
913 memcpy(&rule
->dest_attr
, dest
, sizeof(*dest
));
918 /* fte should not be deleted while calling this function */
919 static struct mlx5_flow_rule
*add_rule_fte(struct fs_fte
*fte
,
920 struct mlx5_flow_group
*fg
,
921 struct mlx5_flow_destination
*dest
)
923 struct mlx5_flow_table
*ft
;
924 struct mlx5_flow_rule
*rule
;
928 rule
= alloc_rule(dest
);
930 return ERR_PTR(-ENOMEM
);
932 fs_get_obj(ft
, fg
->node
.parent
);
933 /* Add dest to dests list- we need flow tables to be in the
934 * end of the list for forward to next prio rules.
936 tree_init_node(&rule
->node
, 1, del_rule
);
937 if (dest
&& dest
->type
!= MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE
)
938 list_add(&rule
->node
.list
, &fte
->node
.children
);
940 list_add_tail(&rule
->node
.list
, &fte
->node
.children
);
944 modify_mask
|= dest
->type
== MLX5_FLOW_DESTINATION_TYPE_COUNTER
?
945 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS
) :
946 BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST
);
949 if (fte
->dests_size
== 1 || !dest
)
950 err
= mlx5_cmd_create_fte(get_dev(&ft
->node
),
953 err
= mlx5_cmd_update_fte(get_dev(&ft
->node
),
954 ft
, fg
->id
, modify_mask
, fte
);
958 fte
->status
|= FS_FTE_STATUS_EXISTING
;
963 list_del(&rule
->node
.list
);
970 /* Assumed fg is locked */
971 static unsigned int get_free_fte_index(struct mlx5_flow_group
*fg
,
972 struct list_head
**prev
)
975 unsigned int start
= fg
->start_index
;
978 *prev
= &fg
->node
.children
;
980 /* assumed list is sorted by index */
981 fs_for_each_fte(fte
, fg
) {
982 if (fte
->index
!= start
)
986 *prev
= &fte
->node
.list
;
992 /* prev is output, prev->next = new_fte */
993 static struct fs_fte
*create_fte(struct mlx5_flow_group
*fg
,
997 struct list_head
**prev
)
1002 index
= get_free_fte_index(fg
, prev
);
1003 fte
= alloc_fte(action
, flow_tag
, match_value
, index
);
1010 static struct mlx5_flow_group
*create_autogroup(struct mlx5_flow_table
*ft
,
1011 u8 match_criteria_enable
,
1012 u32
*match_criteria
)
1014 int inlen
= MLX5_ST_SZ_BYTES(create_flow_group_in
);
1015 struct list_head
*prev
= &ft
->node
.children
;
1016 unsigned int candidate_index
= 0;
1017 struct mlx5_flow_group
*fg
;
1018 void *match_criteria_addr
;
1019 unsigned int group_size
= 0;
1022 if (!ft
->autogroup
.active
)
1023 return ERR_PTR(-ENOENT
);
1025 in
= mlx5_vzalloc(inlen
);
1027 return ERR_PTR(-ENOMEM
);
1029 if (ft
->autogroup
.num_groups
< ft
->autogroup
.required_groups
)
1030 /* We save place for flow groups in addition to max types */
1031 group_size
= ft
->max_fte
/ (ft
->autogroup
.required_groups
+ 1);
1033 /* ft->max_fte == ft->autogroup.max_types */
1034 if (group_size
== 0)
1037 /* sorted by start_index */
1038 fs_for_each_fg(fg
, ft
) {
1039 if (candidate_index
+ group_size
> fg
->start_index
)
1040 candidate_index
= fg
->start_index
+ fg
->max_ftes
;
1043 prev
= &fg
->node
.list
;
1046 if (candidate_index
+ group_size
> ft
->max_fte
) {
1047 fg
= ERR_PTR(-ENOSPC
);
1051 MLX5_SET(create_flow_group_in
, in
, match_criteria_enable
,
1052 match_criteria_enable
);
1053 MLX5_SET(create_flow_group_in
, in
, start_flow_index
, candidate_index
);
1054 MLX5_SET(create_flow_group_in
, in
, end_flow_index
, candidate_index
+
1056 match_criteria_addr
= MLX5_ADDR_OF(create_flow_group_in
,
1057 in
, match_criteria
);
1058 memcpy(match_criteria_addr
, match_criteria
,
1059 MLX5_ST_SZ_BYTES(fte_match_param
));
1061 fg
= create_flow_group_common(ft
, in
, prev
, true);
1067 static struct mlx5_flow_rule
*find_flow_rule(struct fs_fte
*fte
,
1068 struct mlx5_flow_destination
*dest
)
1070 struct mlx5_flow_rule
*rule
;
1072 list_for_each_entry(rule
, &fte
->node
.children
, node
.list
) {
1073 if (rule
->dest_attr
.type
== dest
->type
) {
1074 if ((dest
->type
== MLX5_FLOW_DESTINATION_TYPE_VPORT
&&
1075 dest
->vport_num
== rule
->dest_attr
.vport_num
) ||
1076 (dest
->type
== MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE
&&
1077 dest
->ft
== rule
->dest_attr
.ft
) ||
1078 (dest
->type
== MLX5_FLOW_DESTINATION_TYPE_TIR
&&
1079 dest
->tir_num
== rule
->dest_attr
.tir_num
))
1086 static struct mlx5_flow_rule
*add_rule_fg(struct mlx5_flow_group
*fg
,
1090 struct mlx5_flow_destination
*dest
)
1093 struct mlx5_flow_rule
*rule
;
1094 struct mlx5_flow_table
*ft
;
1095 struct list_head
*prev
;
1097 nested_lock_ref_node(&fg
->node
, FS_MUTEX_PARENT
);
1098 fs_for_each_fte(fte
, fg
) {
1099 nested_lock_ref_node(&fte
->node
, FS_MUTEX_CHILD
);
1100 if (compare_match_value(&fg
->mask
, match_value
, &fte
->val
) &&
1101 action
== fte
->action
&& flow_tag
== fte
->flow_tag
) {
1102 rule
= find_flow_rule(fte
, dest
);
1104 atomic_inc(&rule
->node
.refcount
);
1105 unlock_ref_node(&fte
->node
);
1106 unlock_ref_node(&fg
->node
);
1109 rule
= add_rule_fte(fte
, fg
, dest
);
1110 unlock_ref_node(&fte
->node
);
1116 unlock_ref_node(&fte
->node
);
1118 fs_get_obj(ft
, fg
->node
.parent
);
1119 if (fg
->num_ftes
>= fg
->max_ftes
) {
1120 rule
= ERR_PTR(-ENOSPC
);
1124 fte
= create_fte(fg
, match_value
, action
, flow_tag
, &prev
);
1129 tree_init_node(&fte
->node
, 0, del_fte
);
1130 rule
= add_rule_fte(fte
, fg
, dest
);
1138 tree_add_node(&fte
->node
, &fg
->node
);
1139 list_add(&fte
->node
.list
, prev
);
1141 tree_add_node(&rule
->node
, &fte
->node
);
1143 unlock_ref_node(&fg
->node
);
1147 struct mlx5_fc
*mlx5_flow_rule_counter(struct mlx5_flow_rule
*rule
)
1149 struct mlx5_flow_rule
*dst
;
1152 fs_get_obj(fte
, rule
->node
.parent
);
1154 fs_for_each_dst(dst
, fte
) {
1155 if (dst
->dest_attr
.type
== MLX5_FLOW_DESTINATION_TYPE_COUNTER
)
1156 return dst
->dest_attr
.counter
;
1162 static bool counter_is_valid(struct mlx5_fc
*counter
, u32 action
)
1164 if (!(action
& MLX5_FLOW_CONTEXT_ACTION_COUNT
))
1170 /* Hardware support counter for a drop action only */
1171 return action
== (MLX5_FLOW_CONTEXT_ACTION_DROP
| MLX5_FLOW_CONTEXT_ACTION_COUNT
);
1174 static bool dest_is_valid(struct mlx5_flow_destination
*dest
,
1176 struct mlx5_flow_table
*ft
)
1178 if (dest
&& (dest
->type
== MLX5_FLOW_DESTINATION_TYPE_COUNTER
))
1179 return counter_is_valid(dest
->counter
, action
);
1181 if (!(action
& MLX5_FLOW_CONTEXT_ACTION_FWD_DEST
))
1184 if (!dest
|| ((dest
->type
==
1185 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE
) &&
1186 (dest
->ft
->level
<= ft
->level
)))
1191 static struct mlx5_flow_rule
*
1192 _mlx5_add_flow_rule(struct mlx5_flow_table
*ft
,
1193 struct mlx5_flow_spec
*spec
,
1196 struct mlx5_flow_destination
*dest
)
1198 struct mlx5_flow_group
*g
;
1199 struct mlx5_flow_rule
*rule
;
1201 if (!dest_is_valid(dest
, action
, ft
))
1202 return ERR_PTR(-EINVAL
);
1204 nested_lock_ref_node(&ft
->node
, FS_MUTEX_GRANDPARENT
);
1205 fs_for_each_fg(g
, ft
)
1206 if (compare_match_criteria(g
->mask
.match_criteria_enable
,
1207 spec
->match_criteria_enable
,
1208 g
->mask
.match_criteria
,
1209 spec
->match_criteria
)) {
1210 rule
= add_rule_fg(g
, spec
->match_value
,
1211 action
, flow_tag
, dest
);
1212 if (!IS_ERR(rule
) || PTR_ERR(rule
) != -ENOSPC
)
1216 g
= create_autogroup(ft
, spec
->match_criteria_enable
,
1217 spec
->match_criteria
);
1223 rule
= add_rule_fg(g
, spec
->match_value
,
1224 action
, flow_tag
, dest
);
1226 /* Remove assumes refcount > 0 and autogroup creates a group
1227 * with a refcount = 0.
1229 unlock_ref_node(&ft
->node
);
1230 tree_get_node(&g
->node
);
1231 tree_remove_node(&g
->node
);
1235 unlock_ref_node(&ft
->node
);
1239 static bool fwd_next_prio_supported(struct mlx5_flow_table
*ft
)
1241 return ((ft
->type
== FS_FT_NIC_RX
) &&
1242 (MLX5_CAP_FLOWTABLE(get_dev(&ft
->node
), nic_rx_multi_path_tirs
)));
1245 struct mlx5_flow_rule
*
1246 mlx5_add_flow_rule(struct mlx5_flow_table
*ft
,
1247 struct mlx5_flow_spec
*spec
,
1250 struct mlx5_flow_destination
*dest
)
1252 struct mlx5_flow_root_namespace
*root
= find_root(&ft
->node
);
1253 struct mlx5_flow_destination gen_dest
;
1254 struct mlx5_flow_table
*next_ft
= NULL
;
1255 struct mlx5_flow_rule
*rule
= NULL
;
1256 u32 sw_action
= action
;
1257 struct fs_prio
*prio
;
1259 fs_get_obj(prio
, ft
->node
.parent
);
1260 if (action
== MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO
) {
1261 if (!fwd_next_prio_supported(ft
))
1262 return ERR_PTR(-EOPNOTSUPP
);
1264 return ERR_PTR(-EINVAL
);
1265 mutex_lock(&root
->chain_lock
);
1266 next_ft
= find_next_chained_ft(prio
);
1268 gen_dest
.type
= MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE
;
1269 gen_dest
.ft
= next_ft
;
1271 action
= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST
;
1273 mutex_unlock(&root
->chain_lock
);
1274 return ERR_PTR(-EOPNOTSUPP
);
1278 rule
= _mlx5_add_flow_rule(ft
, spec
, action
, flow_tag
, dest
);
1280 if (sw_action
== MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO
) {
1281 if (!IS_ERR_OR_NULL(rule
) &&
1282 (list_empty(&rule
->next_ft
))) {
1283 mutex_lock(&next_ft
->lock
);
1284 list_add(&rule
->next_ft
, &next_ft
->fwd_rules
);
1285 mutex_unlock(&next_ft
->lock
);
1286 rule
->sw_action
= MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO
;
1288 mutex_unlock(&root
->chain_lock
);
1292 EXPORT_SYMBOL(mlx5_add_flow_rule
);
1294 void mlx5_del_flow_rule(struct mlx5_flow_rule
*rule
)
1296 tree_remove_node(&rule
->node
);
1298 EXPORT_SYMBOL(mlx5_del_flow_rule
);
1300 /* Assuming prio->node.children(flow tables) is sorted by level */
1301 static struct mlx5_flow_table
*find_next_ft(struct mlx5_flow_table
*ft
)
1303 struct fs_prio
*prio
;
1305 fs_get_obj(prio
, ft
->node
.parent
);
1307 if (!list_is_last(&ft
->node
.list
, &prio
->node
.children
))
1308 return list_next_entry(ft
, node
.list
);
1309 return find_next_chained_ft(prio
);
1312 static int update_root_ft_destroy(struct mlx5_flow_table
*ft
)
1314 struct mlx5_flow_root_namespace
*root
= find_root(&ft
->node
);
1315 struct mlx5_flow_table
*new_root_ft
= NULL
;
1317 if (root
->root_ft
!= ft
)
1320 new_root_ft
= find_next_ft(ft
);
1322 int err
= mlx5_cmd_update_root_ft(root
->dev
, new_root_ft
);
1325 mlx5_core_warn(root
->dev
, "Update root flow table of id=%u failed\n",
1330 root
->root_ft
= new_root_ft
;
1334 /* Connect flow table from previous priority to
1335 * the next flow table.
1337 static int disconnect_flow_table(struct mlx5_flow_table
*ft
)
1339 struct mlx5_core_dev
*dev
= get_dev(&ft
->node
);
1340 struct mlx5_flow_table
*next_ft
;
1341 struct fs_prio
*prio
;
1344 err
= update_root_ft_destroy(ft
);
1348 fs_get_obj(prio
, ft
->node
.parent
);
1349 if (!(list_first_entry(&prio
->node
.children
,
1350 struct mlx5_flow_table
,
1354 next_ft
= find_next_chained_ft(prio
);
1355 err
= connect_fwd_rules(dev
, next_ft
, ft
);
1359 err
= connect_prev_fts(dev
, next_ft
, prio
);
1361 mlx5_core_warn(dev
, "Failed to disconnect flow table %d\n",
1366 int mlx5_destroy_flow_table(struct mlx5_flow_table
*ft
)
1368 struct mlx5_flow_root_namespace
*root
= find_root(&ft
->node
);
1371 mutex_lock(&root
->chain_lock
);
1372 err
= disconnect_flow_table(ft
);
1374 mutex_unlock(&root
->chain_lock
);
1377 if (tree_remove_node(&ft
->node
))
1378 mlx5_core_warn(get_dev(&ft
->node
), "Flow table %d wasn't destroyed, refcount > 1\n",
1380 mutex_unlock(&root
->chain_lock
);
1384 EXPORT_SYMBOL(mlx5_destroy_flow_table
);
1386 void mlx5_destroy_flow_group(struct mlx5_flow_group
*fg
)
1388 if (tree_remove_node(&fg
->node
))
1389 mlx5_core_warn(get_dev(&fg
->node
), "Flow group %d wasn't destroyed, refcount > 1\n",
1393 struct mlx5_flow_namespace
*mlx5_get_flow_namespace(struct mlx5_core_dev
*dev
,
1394 enum mlx5_flow_namespace_type type
)
1396 struct mlx5_flow_steering
*steering
= dev
->priv
.steering
;
1397 struct mlx5_flow_root_namespace
*root_ns
;
1399 struct fs_prio
*fs_prio
;
1400 struct mlx5_flow_namespace
*ns
;
1406 case MLX5_FLOW_NAMESPACE_BYPASS
:
1407 case MLX5_FLOW_NAMESPACE_LAG
:
1408 case MLX5_FLOW_NAMESPACE_OFFLOADS
:
1409 case MLX5_FLOW_NAMESPACE_ETHTOOL
:
1410 case MLX5_FLOW_NAMESPACE_KERNEL
:
1411 case MLX5_FLOW_NAMESPACE_LEFTOVERS
:
1412 case MLX5_FLOW_NAMESPACE_ANCHOR
:
1415 case MLX5_FLOW_NAMESPACE_FDB
:
1416 if (steering
->fdb_root_ns
)
1417 return &steering
->fdb_root_ns
->ns
;
1420 case MLX5_FLOW_NAMESPACE_ESW_EGRESS
:
1421 if (steering
->esw_egress_root_ns
)
1422 return &steering
->esw_egress_root_ns
->ns
;
1425 case MLX5_FLOW_NAMESPACE_ESW_INGRESS
:
1426 if (steering
->esw_ingress_root_ns
)
1427 return &steering
->esw_ingress_root_ns
->ns
;
1430 case MLX5_FLOW_NAMESPACE_SNIFFER_RX
:
1431 if (steering
->sniffer_rx_root_ns
)
1432 return &steering
->sniffer_rx_root_ns
->ns
;
1435 case MLX5_FLOW_NAMESPACE_SNIFFER_TX
:
1436 if (steering
->sniffer_tx_root_ns
)
1437 return &steering
->sniffer_tx_root_ns
->ns
;
1444 root_ns
= steering
->root_ns
;
1448 fs_prio
= find_prio(&root_ns
->ns
, prio
);
1452 ns
= list_first_entry(&fs_prio
->node
.children
,
1458 EXPORT_SYMBOL(mlx5_get_flow_namespace
);
1460 static struct fs_prio
*fs_create_prio(struct mlx5_flow_namespace
*ns
,
1461 unsigned int prio
, int num_levels
)
1463 struct fs_prio
*fs_prio
;
1465 fs_prio
= kzalloc(sizeof(*fs_prio
), GFP_KERNEL
);
1467 return ERR_PTR(-ENOMEM
);
1469 fs_prio
->node
.type
= FS_TYPE_PRIO
;
1470 tree_init_node(&fs_prio
->node
, 1, NULL
);
1471 tree_add_node(&fs_prio
->node
, &ns
->node
);
1472 fs_prio
->num_levels
= num_levels
;
1473 fs_prio
->prio
= prio
;
1474 list_add_tail(&fs_prio
->node
.list
, &ns
->node
.children
);
1479 static struct mlx5_flow_namespace
*fs_init_namespace(struct mlx5_flow_namespace
1482 ns
->node
.type
= FS_TYPE_NAMESPACE
;
1487 static struct mlx5_flow_namespace
*fs_create_namespace(struct fs_prio
*prio
)
1489 struct mlx5_flow_namespace
*ns
;
1491 ns
= kzalloc(sizeof(*ns
), GFP_KERNEL
);
1493 return ERR_PTR(-ENOMEM
);
1495 fs_init_namespace(ns
);
1496 tree_init_node(&ns
->node
, 1, NULL
);
1497 tree_add_node(&ns
->node
, &prio
->node
);
1498 list_add_tail(&ns
->node
.list
, &prio
->node
.children
);
1503 static int create_leaf_prios(struct mlx5_flow_namespace
*ns
, int prio
,
1504 struct init_tree_node
*prio_metadata
)
1506 struct fs_prio
*fs_prio
;
1509 for (i
= 0; i
< prio_metadata
->num_leaf_prios
; i
++) {
1510 fs_prio
= fs_create_prio(ns
, prio
++, prio_metadata
->num_levels
);
1511 if (IS_ERR(fs_prio
))
1512 return PTR_ERR(fs_prio
);
1517 #define FLOW_TABLE_BIT_SZ 1
1518 #define GET_FLOW_TABLE_CAP(dev, offset) \
1519 ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) + \
1521 (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1522 static bool has_required_caps(struct mlx5_core_dev
*dev
, struct node_caps
*caps
)
1526 for (i
= 0; i
< caps
->arr_sz
; i
++) {
1527 if (!GET_FLOW_TABLE_CAP(dev
, caps
->caps
[i
]))
1533 static int init_root_tree_recursive(struct mlx5_flow_steering
*steering
,
1534 struct init_tree_node
*init_node
,
1535 struct fs_node
*fs_parent_node
,
1536 struct init_tree_node
*init_parent_node
,
1539 int max_ft_level
= MLX5_CAP_FLOWTABLE(steering
->dev
,
1540 flow_table_properties_nic_receive
.
1542 struct mlx5_flow_namespace
*fs_ns
;
1543 struct fs_prio
*fs_prio
;
1544 struct fs_node
*base
;
1548 if (init_node
->type
== FS_TYPE_PRIO
) {
1549 if ((init_node
->min_ft_level
> max_ft_level
) ||
1550 !has_required_caps(steering
->dev
, &init_node
->caps
))
1553 fs_get_obj(fs_ns
, fs_parent_node
);
1554 if (init_node
->num_leaf_prios
)
1555 return create_leaf_prios(fs_ns
, prio
, init_node
);
1556 fs_prio
= fs_create_prio(fs_ns
, prio
, init_node
->num_levels
);
1557 if (IS_ERR(fs_prio
))
1558 return PTR_ERR(fs_prio
);
1559 base
= &fs_prio
->node
;
1560 } else if (init_node
->type
== FS_TYPE_NAMESPACE
) {
1561 fs_get_obj(fs_prio
, fs_parent_node
);
1562 fs_ns
= fs_create_namespace(fs_prio
);
1564 return PTR_ERR(fs_ns
);
1565 base
= &fs_ns
->node
;
1570 for (i
= 0; i
< init_node
->ar_size
; i
++) {
1571 err
= init_root_tree_recursive(steering
, &init_node
->children
[i
],
1572 base
, init_node
, prio
);
1575 if (init_node
->children
[i
].type
== FS_TYPE_PRIO
&&
1576 init_node
->children
[i
].num_leaf_prios
) {
1577 prio
+= init_node
->children
[i
].num_leaf_prios
;
1584 static int init_root_tree(struct mlx5_flow_steering
*steering
,
1585 struct init_tree_node
*init_node
,
1586 struct fs_node
*fs_parent_node
)
1589 struct mlx5_flow_namespace
*fs_ns
;
1592 fs_get_obj(fs_ns
, fs_parent_node
);
1593 for (i
= 0; i
< init_node
->ar_size
; i
++) {
1594 err
= init_root_tree_recursive(steering
, &init_node
->children
[i
],
1603 static struct mlx5_flow_root_namespace
*create_root_ns(struct mlx5_flow_steering
*steering
,
1604 enum fs_flow_table_type
1607 struct mlx5_flow_root_namespace
*root_ns
;
1608 struct mlx5_flow_namespace
*ns
;
1610 /* Create the root namespace */
1611 root_ns
= mlx5_vzalloc(sizeof(*root_ns
));
1615 root_ns
->dev
= steering
->dev
;
1616 root_ns
->table_type
= table_type
;
1619 fs_init_namespace(ns
);
1620 mutex_init(&root_ns
->chain_lock
);
1621 tree_init_node(&ns
->node
, 1, NULL
);
1622 tree_add_node(&ns
->node
, NULL
);
1627 static void set_prio_attrs_in_prio(struct fs_prio
*prio
, int acc_level
);
1629 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace
*ns
, int acc_level
)
1631 struct fs_prio
*prio
;
1633 fs_for_each_prio(prio
, ns
) {
1634 /* This updates prio start_level and num_levels */
1635 set_prio_attrs_in_prio(prio
, acc_level
);
1636 acc_level
+= prio
->num_levels
;
1641 static void set_prio_attrs_in_prio(struct fs_prio
*prio
, int acc_level
)
1643 struct mlx5_flow_namespace
*ns
;
1644 int acc_level_ns
= acc_level
;
1646 prio
->start_level
= acc_level
;
1647 fs_for_each_ns(ns
, prio
)
1648 /* This updates start_level and num_levels of ns's priority descendants */
1649 acc_level_ns
= set_prio_attrs_in_ns(ns
, acc_level
);
1650 if (!prio
->num_levels
)
1651 prio
->num_levels
= acc_level_ns
- prio
->start_level
;
1652 WARN_ON(prio
->num_levels
< acc_level_ns
- prio
->start_level
);
1655 static void set_prio_attrs(struct mlx5_flow_root_namespace
*root_ns
)
1657 struct mlx5_flow_namespace
*ns
= &root_ns
->ns
;
1658 struct fs_prio
*prio
;
1659 int start_level
= 0;
1661 fs_for_each_prio(prio
, ns
) {
1662 set_prio_attrs_in_prio(prio
, start_level
);
1663 start_level
+= prio
->num_levels
;
1667 #define ANCHOR_PRIO 0
1668 #define ANCHOR_SIZE 1
1669 #define ANCHOR_LEVEL 0
1670 static int create_anchor_flow_table(struct mlx5_flow_steering
*steering
)
1672 struct mlx5_flow_namespace
*ns
= NULL
;
1673 struct mlx5_flow_table
*ft
;
1675 ns
= mlx5_get_flow_namespace(steering
->dev
, MLX5_FLOW_NAMESPACE_ANCHOR
);
1678 ft
= mlx5_create_flow_table(ns
, ANCHOR_PRIO
, ANCHOR_SIZE
, ANCHOR_LEVEL
);
1680 mlx5_core_err(steering
->dev
, "Failed to create last anchor flow table");
1686 static int init_root_ns(struct mlx5_flow_steering
*steering
)
1689 steering
->root_ns
= create_root_ns(steering
, FS_FT_NIC_RX
);
1690 if (IS_ERR_OR_NULL(steering
->root_ns
))
1693 if (init_root_tree(steering
, &root_fs
, &steering
->root_ns
->ns
.node
))
1696 set_prio_attrs(steering
->root_ns
);
1698 if (create_anchor_flow_table(steering
))
1704 mlx5_cleanup_fs(steering
->dev
);
1708 static void clean_tree(struct fs_node
*node
)
1711 struct fs_node
*iter
;
1712 struct fs_node
*temp
;
1714 list_for_each_entry_safe(iter
, temp
, &node
->children
, list
)
1716 tree_remove_node(node
);
1720 static void cleanup_root_ns(struct mlx5_flow_root_namespace
*root_ns
)
1725 clean_tree(&root_ns
->ns
.node
);
1728 void mlx5_cleanup_fs(struct mlx5_core_dev
*dev
)
1730 struct mlx5_flow_steering
*steering
= dev
->priv
.steering
;
1732 if (MLX5_CAP_GEN(dev
, port_type
) != MLX5_CAP_PORT_TYPE_ETH
)
1735 cleanup_root_ns(steering
->root_ns
);
1736 cleanup_root_ns(steering
->esw_egress_root_ns
);
1737 cleanup_root_ns(steering
->esw_ingress_root_ns
);
1738 cleanup_root_ns(steering
->fdb_root_ns
);
1739 cleanup_root_ns(steering
->sniffer_rx_root_ns
);
1740 cleanup_root_ns(steering
->sniffer_tx_root_ns
);
1741 mlx5_cleanup_fc_stats(dev
);
1745 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering
*steering
)
1747 struct fs_prio
*prio
;
1749 steering
->sniffer_tx_root_ns
= create_root_ns(steering
, FS_FT_SNIFFER_TX
);
1750 if (!steering
->sniffer_tx_root_ns
)
1753 /* Create single prio */
1754 prio
= fs_create_prio(&steering
->sniffer_tx_root_ns
->ns
, 0, 1);
1756 cleanup_root_ns(steering
->sniffer_tx_root_ns
);
1757 return PTR_ERR(prio
);
1762 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering
*steering
)
1764 struct fs_prio
*prio
;
1766 steering
->sniffer_rx_root_ns
= create_root_ns(steering
, FS_FT_SNIFFER_RX
);
1767 if (!steering
->sniffer_rx_root_ns
)
1770 /* Create single prio */
1771 prio
= fs_create_prio(&steering
->sniffer_rx_root_ns
->ns
, 0, 1);
1773 cleanup_root_ns(steering
->sniffer_rx_root_ns
);
1774 return PTR_ERR(prio
);
1779 static int init_fdb_root_ns(struct mlx5_flow_steering
*steering
)
1781 struct fs_prio
*prio
;
1783 steering
->fdb_root_ns
= create_root_ns(steering
, FS_FT_FDB
);
1784 if (!steering
->fdb_root_ns
)
1787 prio
= fs_create_prio(&steering
->fdb_root_ns
->ns
, 0, 1);
1791 prio
= fs_create_prio(&steering
->fdb_root_ns
->ns
, 1, 1);
1795 set_prio_attrs(steering
->fdb_root_ns
);
1799 cleanup_root_ns(steering
->fdb_root_ns
);
1800 steering
->fdb_root_ns
= NULL
;
1801 return PTR_ERR(prio
);
1804 static int init_ingress_acl_root_ns(struct mlx5_flow_steering
*steering
)
1806 struct fs_prio
*prio
;
1808 steering
->esw_egress_root_ns
= create_root_ns(steering
, FS_FT_ESW_EGRESS_ACL
);
1809 if (!steering
->esw_egress_root_ns
)
1813 prio
= fs_create_prio(&steering
->esw_egress_root_ns
->ns
, 0,
1814 MLX5_TOTAL_VPORTS(steering
->dev
));
1815 return PTR_ERR_OR_ZERO(prio
);
1818 static int init_egress_acl_root_ns(struct mlx5_flow_steering
*steering
)
1820 struct fs_prio
*prio
;
1822 steering
->esw_ingress_root_ns
= create_root_ns(steering
, FS_FT_ESW_INGRESS_ACL
);
1823 if (!steering
->esw_ingress_root_ns
)
1827 prio
= fs_create_prio(&steering
->esw_ingress_root_ns
->ns
, 0,
1828 MLX5_TOTAL_VPORTS(steering
->dev
));
1829 return PTR_ERR_OR_ZERO(prio
);
1832 int mlx5_init_fs(struct mlx5_core_dev
*dev
)
1834 struct mlx5_flow_steering
*steering
;
1837 if (MLX5_CAP_GEN(dev
, port_type
) != MLX5_CAP_PORT_TYPE_ETH
)
1840 err
= mlx5_init_fc_stats(dev
);
1844 steering
= kzalloc(sizeof(*steering
), GFP_KERNEL
);
1847 steering
->dev
= dev
;
1848 dev
->priv
.steering
= steering
;
1850 if (MLX5_CAP_GEN(dev
, nic_flow_table
) &&
1851 MLX5_CAP_FLOWTABLE_NIC_RX(dev
, ft_support
)) {
1852 err
= init_root_ns(steering
);
1857 if (MLX5_CAP_GEN(dev
, eswitch_flow_table
)) {
1858 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev
, ft_support
)) {
1859 err
= init_fdb_root_ns(steering
);
1863 if (MLX5_CAP_ESW_EGRESS_ACL(dev
, ft_support
)) {
1864 err
= init_egress_acl_root_ns(steering
);
1868 if (MLX5_CAP_ESW_INGRESS_ACL(dev
, ft_support
)) {
1869 err
= init_ingress_acl_root_ns(steering
);
1875 if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev
, ft_support
)) {
1876 err
= init_sniffer_rx_root_ns(steering
);
1881 if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev
, ft_support
)) {
1882 err
= init_sniffer_tx_root_ns(steering
);
1889 mlx5_cleanup_fs(dev
);