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 INIT_PRIO(min_level_val, max_ft_val,\
44 start_level_val, ...) {.type = FS_TYPE_PRIO,\
45 .min_ft_level = min_level_val,\
46 .start_level = start_level_val,\
47 .max_ft = max_ft_val,\
48 .children = (struct init_tree_node[]) {__VA_ARGS__},\
49 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
52 #define ADD_PRIO(min_level_val, max_ft_val, start_level_val, ...)\
53 INIT_PRIO(min_level_val, max_ft_val, start_level_val,\
56 #define ADD_FT_PRIO(max_ft_val, start_level_val, ...)\
57 INIT_PRIO(0, max_ft_val, start_level_val,\
60 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
61 .children = (struct init_tree_node[]) {__VA_ARGS__},\
62 .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
65 #define KERNEL_START_LEVEL 0
66 #define KERNEL_P0_START_LEVEL KERNEL_START_LEVEL
67 #define KERNEL_MAX_FT 2
68 #define KENREL_MIN_LEVEL 2
69 static struct init_tree_node
{
70 enum fs_node_type type
;
71 struct init_tree_node
*children
;
78 .type
= FS_TYPE_NAMESPACE
,
80 .children
= (struct init_tree_node
[]) {
81 ADD_PRIO(KENREL_MIN_LEVEL
, KERNEL_MAX_FT
,
83 ADD_NS(ADD_FT_PRIO(KERNEL_MAX_FT
,
84 KERNEL_P0_START_LEVEL
))),
88 enum fs_i_mutex_lock_class
{
94 static void del_rule(struct fs_node
*node
);
95 static void del_flow_table(struct fs_node
*node
);
96 static void del_flow_group(struct fs_node
*node
);
97 static void del_fte(struct fs_node
*node
);
99 static void tree_init_node(struct fs_node
*node
,
100 unsigned int refcount
,
101 void (*remove_func
)(struct fs_node
*))
103 atomic_set(&node
->refcount
, refcount
);
104 INIT_LIST_HEAD(&node
->list
);
105 INIT_LIST_HEAD(&node
->children
);
106 mutex_init(&node
->lock
);
107 node
->remove_func
= remove_func
;
110 static void tree_add_node(struct fs_node
*node
, struct fs_node
*parent
)
113 atomic_inc(&parent
->refcount
);
114 node
->parent
= parent
;
116 /* Parent is the root */
120 node
->root
= parent
->root
;
123 static void tree_get_node(struct fs_node
*node
)
125 atomic_inc(&node
->refcount
);
128 static void nested_lock_ref_node(struct fs_node
*node
,
129 enum fs_i_mutex_lock_class
class)
132 mutex_lock_nested(&node
->lock
, class);
133 atomic_inc(&node
->refcount
);
137 static void lock_ref_node(struct fs_node
*node
)
140 mutex_lock(&node
->lock
);
141 atomic_inc(&node
->refcount
);
145 static void unlock_ref_node(struct fs_node
*node
)
148 atomic_dec(&node
->refcount
);
149 mutex_unlock(&node
->lock
);
153 static void tree_put_node(struct fs_node
*node
)
155 struct fs_node
*parent_node
= node
->parent
;
157 lock_ref_node(parent_node
);
158 if (atomic_dec_and_test(&node
->refcount
)) {
160 list_del_init(&node
->list
);
161 if (node
->remove_func
)
162 node
->remove_func(node
);
166 unlock_ref_node(parent_node
);
167 if (!node
&& parent_node
)
168 tree_put_node(parent_node
);
171 static int tree_remove_node(struct fs_node
*node
)
173 if (atomic_read(&node
->refcount
) > 1)
179 static struct fs_prio
*find_prio(struct mlx5_flow_namespace
*ns
,
182 struct fs_prio
*iter_prio
;
184 fs_for_each_prio(iter_prio
, ns
) {
185 if (iter_prio
->prio
== prio
)
192 static unsigned int find_next_free_level(struct fs_prio
*prio
)
194 if (!list_empty(&prio
->node
.children
)) {
195 struct mlx5_flow_table
*ft
;
197 ft
= list_last_entry(&prio
->node
.children
,
198 struct mlx5_flow_table
,
200 return ft
->level
+ 1;
202 return prio
->start_level
;
205 static bool masked_memcmp(void *mask
, void *val1
, void *val2
, size_t size
)
209 for (i
= 0; i
< size
; i
++, mask
++, val1
++, val2
++)
210 if ((*((u8
*)val1
) & (*(u8
*)mask
)) !=
211 ((*(u8
*)val2
) & (*(u8
*)mask
)))
217 static bool compare_match_value(struct mlx5_flow_group_mask
*mask
,
218 void *fte_param1
, void *fte_param2
)
220 if (mask
->match_criteria_enable
&
221 1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS
) {
222 void *fte_match1
= MLX5_ADDR_OF(fte_match_param
,
223 fte_param1
, outer_headers
);
224 void *fte_match2
= MLX5_ADDR_OF(fte_match_param
,
225 fte_param2
, outer_headers
);
226 void *fte_mask
= MLX5_ADDR_OF(fte_match_param
,
227 mask
->match_criteria
, outer_headers
);
229 if (!masked_memcmp(fte_mask
, fte_match1
, fte_match2
,
230 MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4
)))
234 if (mask
->match_criteria_enable
&
235 1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS
) {
236 void *fte_match1
= MLX5_ADDR_OF(fte_match_param
,
237 fte_param1
, misc_parameters
);
238 void *fte_match2
= MLX5_ADDR_OF(fte_match_param
,
239 fte_param2
, misc_parameters
);
240 void *fte_mask
= MLX5_ADDR_OF(fte_match_param
,
241 mask
->match_criteria
, misc_parameters
);
243 if (!masked_memcmp(fte_mask
, fte_match1
, fte_match2
,
244 MLX5_ST_SZ_BYTES(fte_match_set_misc
)))
248 if (mask
->match_criteria_enable
&
249 1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS
) {
250 void *fte_match1
= MLX5_ADDR_OF(fte_match_param
,
251 fte_param1
, inner_headers
);
252 void *fte_match2
= MLX5_ADDR_OF(fte_match_param
,
253 fte_param2
, inner_headers
);
254 void *fte_mask
= MLX5_ADDR_OF(fte_match_param
,
255 mask
->match_criteria
, inner_headers
);
257 if (!masked_memcmp(fte_mask
, fte_match1
, fte_match2
,
258 MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4
)))
264 static bool compare_match_criteria(u8 match_criteria_enable1
,
265 u8 match_criteria_enable2
,
266 void *mask1
, void *mask2
)
268 return match_criteria_enable1
== match_criteria_enable2
&&
269 !memcmp(mask1
, mask2
, MLX5_ST_SZ_BYTES(fte_match_param
));
272 static struct mlx5_flow_root_namespace
*find_root(struct fs_node
*node
)
274 struct fs_node
*root
;
275 struct mlx5_flow_namespace
*ns
;
279 if (WARN_ON(root
->type
!= FS_TYPE_NAMESPACE
)) {
280 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
284 ns
= container_of(root
, struct mlx5_flow_namespace
, node
);
285 return container_of(ns
, struct mlx5_flow_root_namespace
, ns
);
288 static inline struct mlx5_core_dev
*get_dev(struct fs_node
*node
)
290 struct mlx5_flow_root_namespace
*root
= find_root(node
);
297 static void del_flow_table(struct fs_node
*node
)
299 struct mlx5_flow_table
*ft
;
300 struct mlx5_core_dev
*dev
;
301 struct fs_prio
*prio
;
304 fs_get_obj(ft
, node
);
305 dev
= get_dev(&ft
->node
);
307 err
= mlx5_cmd_destroy_flow_table(dev
, ft
);
309 pr_warn("flow steering can't destroy ft\n");
310 fs_get_obj(prio
, ft
->node
.parent
);
314 static void del_rule(struct fs_node
*node
)
316 struct mlx5_flow_rule
*rule
;
317 struct mlx5_flow_table
*ft
;
318 struct mlx5_flow_group
*fg
;
321 struct mlx5_core_dev
*dev
= get_dev(node
);
322 int match_len
= MLX5_ST_SZ_BYTES(fte_match_param
);
325 match_value
= mlx5_vzalloc(match_len
);
327 pr_warn("failed to allocate inbox\n");
331 fs_get_obj(rule
, node
);
332 fs_get_obj(fte
, rule
->node
.parent
);
333 fs_get_obj(fg
, fte
->node
.parent
);
334 memcpy(match_value
, fte
->val
, sizeof(fte
->val
));
335 fs_get_obj(ft
, fg
->node
.parent
);
336 list_del(&rule
->node
.list
);
338 if (fte
->dests_size
) {
339 err
= mlx5_cmd_update_fte(dev
, ft
,
342 pr_warn("%s can't del rule fg id=%d fte_index=%d\n",
343 __func__
, fg
->id
, fte
->index
);
348 static void del_fte(struct fs_node
*node
)
350 struct mlx5_flow_table
*ft
;
351 struct mlx5_flow_group
*fg
;
352 struct mlx5_core_dev
*dev
;
356 fs_get_obj(fte
, node
);
357 fs_get_obj(fg
, fte
->node
.parent
);
358 fs_get_obj(ft
, fg
->node
.parent
);
360 dev
= get_dev(&ft
->node
);
361 err
= mlx5_cmd_delete_fte(dev
, ft
,
364 pr_warn("flow steering can't delete fte in index %d of flow group id %d\n",
371 static void del_flow_group(struct fs_node
*node
)
373 struct mlx5_flow_group
*fg
;
374 struct mlx5_flow_table
*ft
;
375 struct mlx5_core_dev
*dev
;
377 fs_get_obj(fg
, node
);
378 fs_get_obj(ft
, fg
->node
.parent
);
379 dev
= get_dev(&ft
->node
);
381 if (mlx5_cmd_destroy_flow_group(dev
, ft
, fg
->id
))
382 pr_warn("flow steering can't destroy fg %d of ft %d\n",
386 static struct fs_fte
*alloc_fte(u8 action
,
393 fte
= kzalloc(sizeof(*fte
), GFP_KERNEL
);
395 return ERR_PTR(-ENOMEM
);
397 memcpy(fte
->val
, match_value
, sizeof(fte
->val
));
398 fte
->node
.type
= FS_TYPE_FLOW_ENTRY
;
399 fte
->flow_tag
= flow_tag
;
401 fte
->action
= action
;
406 static struct mlx5_flow_group
*alloc_flow_group(u32
*create_fg_in
)
408 struct mlx5_flow_group
*fg
;
409 void *match_criteria
= MLX5_ADDR_OF(create_flow_group_in
,
410 create_fg_in
, match_criteria
);
411 u8 match_criteria_enable
= MLX5_GET(create_flow_group_in
,
413 match_criteria_enable
);
414 fg
= kzalloc(sizeof(*fg
), GFP_KERNEL
);
416 return ERR_PTR(-ENOMEM
);
418 fg
->mask
.match_criteria_enable
= match_criteria_enable
;
419 memcpy(&fg
->mask
.match_criteria
, match_criteria
,
420 sizeof(fg
->mask
.match_criteria
));
421 fg
->node
.type
= FS_TYPE_FLOW_GROUP
;
422 fg
->start_index
= MLX5_GET(create_flow_group_in
, create_fg_in
,
424 fg
->max_ftes
= MLX5_GET(create_flow_group_in
, create_fg_in
,
425 end_flow_index
) - fg
->start_index
+ 1;
429 static struct mlx5_flow_table
*alloc_flow_table(int level
, int max_fte
,
430 enum fs_flow_table_type table_type
)
432 struct mlx5_flow_table
*ft
;
434 ft
= kzalloc(sizeof(*ft
), GFP_KERNEL
);
439 ft
->node
.type
= FS_TYPE_FLOW_TABLE
;
440 ft
->type
= table_type
;
441 ft
->max_fte
= max_fte
;
446 /* If reverse is false, then we search for the first flow table in the
447 * root sub-tree from start(closest from right), else we search for the
448 * last flow table in the root sub-tree till start(closest from left).
450 static struct mlx5_flow_table
*find_closest_ft_recursive(struct fs_node
*root
,
451 struct list_head
*start
,
454 #define list_advance_entry(pos, reverse) \
455 ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
457 #define list_for_each_advance_continue(pos, head, reverse) \
458 for (pos = list_advance_entry(pos, reverse); \
459 &pos->list != (head); \
460 pos = list_advance_entry(pos, reverse))
462 struct fs_node
*iter
= list_entry(start
, struct fs_node
, list
);
463 struct mlx5_flow_table
*ft
= NULL
;
468 list_for_each_advance_continue(iter
, &root
->children
, reverse
) {
469 if (iter
->type
== FS_TYPE_FLOW_TABLE
) {
470 fs_get_obj(ft
, iter
);
473 ft
= find_closest_ft_recursive(iter
, &iter
->children
, reverse
);
481 /* If reverse if false then return the first flow table in next priority of
482 * prio in the tree, else return the last flow table in the previous priority
483 * of prio in the tree.
485 static struct mlx5_flow_table
*find_closest_ft(struct fs_prio
*prio
, bool reverse
)
487 struct mlx5_flow_table
*ft
= NULL
;
488 struct fs_node
*curr_node
;
489 struct fs_node
*parent
;
491 parent
= prio
->node
.parent
;
492 curr_node
= &prio
->node
;
493 while (!ft
&& parent
) {
494 ft
= find_closest_ft_recursive(parent
, &curr_node
->list
, reverse
);
496 parent
= curr_node
->parent
;
501 /* Assuming all the tree is locked by mutex chain lock */
502 static struct mlx5_flow_table
*find_next_chained_ft(struct fs_prio
*prio
)
504 return find_closest_ft(prio
, false);
507 /* Assuming all the tree is locked by mutex chain lock */
508 static struct mlx5_flow_table
*find_prev_chained_ft(struct fs_prio
*prio
)
510 return find_closest_ft(prio
, true);
513 static int connect_fts_in_prio(struct mlx5_core_dev
*dev
,
514 struct fs_prio
*prio
,
515 struct mlx5_flow_table
*ft
)
517 struct mlx5_flow_table
*iter
;
521 fs_for_each_ft(iter
, prio
) {
523 err
= mlx5_cmd_modify_flow_table(dev
,
527 mlx5_core_warn(dev
, "Failed to modify flow table %d\n",
529 /* The driver is out of sync with the FW */
538 /* Connect flow tables from previous priority of prio to ft */
539 static int connect_prev_fts(struct mlx5_core_dev
*dev
,
540 struct mlx5_flow_table
*ft
,
541 struct fs_prio
*prio
)
543 struct mlx5_flow_table
*prev_ft
;
545 prev_ft
= find_prev_chained_ft(prio
);
547 struct fs_prio
*prev_prio
;
549 fs_get_obj(prev_prio
, prev_ft
->node
.parent
);
550 return connect_fts_in_prio(dev
, prev_prio
, ft
);
555 static int update_root_ft_create(struct mlx5_flow_table
*ft
, struct fs_prio
558 struct mlx5_flow_root_namespace
*root
= find_root(&prio
->node
);
559 int min_level
= INT_MAX
;
563 min_level
= root
->root_ft
->level
;
565 if (ft
->level
>= min_level
)
568 err
= mlx5_cmd_update_root_ft(root
->dev
, ft
);
570 mlx5_core_warn(root
->dev
, "Update root flow table of id=%u failed\n",
578 static int connect_flow_table(struct mlx5_core_dev
*dev
, struct mlx5_flow_table
*ft
,
579 struct fs_prio
*prio
)
583 /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
585 if (list_empty(&prio
->node
.children
)) {
586 err
= connect_prev_fts(dev
, ft
, prio
);
591 if (MLX5_CAP_FLOWTABLE(dev
,
592 flow_table_properties_nic_receive
.modify_root
))
593 err
= update_root_ft_create(ft
, prio
);
597 struct mlx5_flow_table
*mlx5_create_flow_table(struct mlx5_flow_namespace
*ns
,
601 struct mlx5_flow_table
*next_ft
= NULL
;
602 struct mlx5_flow_table
*ft
;
605 struct mlx5_flow_root_namespace
*root
=
606 find_root(&ns
->node
);
607 struct fs_prio
*fs_prio
= NULL
;
610 pr_err("mlx5: flow steering failed to find root of namespace\n");
611 return ERR_PTR(-ENODEV
);
614 mutex_lock(&root
->chain_lock
);
615 fs_prio
= find_prio(ns
, prio
);
620 if (fs_prio
->num_ft
== fs_prio
->max_ft
) {
625 ft
= alloc_flow_table(find_next_free_level(fs_prio
),
626 roundup_pow_of_two(max_fte
),
633 tree_init_node(&ft
->node
, 1, del_flow_table
);
634 log_table_sz
= ilog2(ft
->max_fte
);
635 next_ft
= find_next_chained_ft(fs_prio
);
636 err
= mlx5_cmd_create_flow_table(root
->dev
, ft
->type
, ft
->level
,
637 log_table_sz
, next_ft
, &ft
->id
);
641 err
= connect_flow_table(root
->dev
, ft
, fs_prio
);
644 lock_ref_node(&fs_prio
->node
);
645 tree_add_node(&ft
->node
, &fs_prio
->node
);
646 list_add_tail(&ft
->node
.list
, &fs_prio
->node
.children
);
648 unlock_ref_node(&fs_prio
->node
);
649 mutex_unlock(&root
->chain_lock
);
652 mlx5_cmd_destroy_flow_table(root
->dev
, ft
);
656 mutex_unlock(&root
->chain_lock
);
660 struct mlx5_flow_table
*mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace
*ns
,
662 int num_flow_table_entries
,
665 struct mlx5_flow_table
*ft
;
667 if (max_num_groups
> num_flow_table_entries
)
668 return ERR_PTR(-EINVAL
);
670 ft
= mlx5_create_flow_table(ns
, prio
, num_flow_table_entries
);
674 ft
->autogroup
.active
= true;
675 ft
->autogroup
.required_groups
= max_num_groups
;
680 /* Flow table should be locked */
681 static struct mlx5_flow_group
*create_flow_group_common(struct mlx5_flow_table
*ft
,
687 struct mlx5_flow_group
*fg
;
688 struct mlx5_core_dev
*dev
= get_dev(&ft
->node
);
692 return ERR_PTR(-ENODEV
);
694 fg
= alloc_flow_group(fg_in
);
698 err
= mlx5_cmd_create_flow_group(dev
, ft
, fg_in
, &fg
->id
);
704 if (ft
->autogroup
.active
)
705 ft
->autogroup
.num_groups
++;
706 /* Add node to tree */
707 tree_init_node(&fg
->node
, !is_auto_fg
, del_flow_group
);
708 tree_add_node(&fg
->node
, &ft
->node
);
709 /* Add node to group list */
710 list_add(&fg
->node
.list
, ft
->node
.children
.prev
);
715 struct mlx5_flow_group
*mlx5_create_flow_group(struct mlx5_flow_table
*ft
,
718 struct mlx5_flow_group
*fg
;
720 if (ft
->autogroup
.active
)
721 return ERR_PTR(-EPERM
);
723 lock_ref_node(&ft
->node
);
724 fg
= create_flow_group_common(ft
, fg_in
, &ft
->node
.children
, false);
725 unlock_ref_node(&ft
->node
);
730 static struct mlx5_flow_rule
*alloc_rule(struct mlx5_flow_destination
*dest
)
732 struct mlx5_flow_rule
*rule
;
734 rule
= kzalloc(sizeof(*rule
), GFP_KERNEL
);
738 rule
->node
.type
= FS_TYPE_FLOW_DEST
;
739 memcpy(&rule
->dest_attr
, dest
, sizeof(*dest
));
744 /* fte should not be deleted while calling this function */
745 static struct mlx5_flow_rule
*add_rule_fte(struct fs_fte
*fte
,
746 struct mlx5_flow_group
*fg
,
747 struct mlx5_flow_destination
*dest
)
749 struct mlx5_flow_table
*ft
;
750 struct mlx5_flow_rule
*rule
;
753 rule
= alloc_rule(dest
);
755 return ERR_PTR(-ENOMEM
);
757 fs_get_obj(ft
, fg
->node
.parent
);
758 /* Add dest to dests list- added as first element after the head */
759 tree_init_node(&rule
->node
, 1, del_rule
);
760 list_add_tail(&rule
->node
.list
, &fte
->node
.children
);
762 if (fte
->dests_size
== 1)
763 err
= mlx5_cmd_create_fte(get_dev(&ft
->node
),
766 err
= mlx5_cmd_update_fte(get_dev(&ft
->node
),
771 fte
->status
|= FS_FTE_STATUS_EXISTING
;
776 list_del(&rule
->node
.list
);
782 /* Assumed fg is locked */
783 static unsigned int get_free_fte_index(struct mlx5_flow_group
*fg
,
784 struct list_head
**prev
)
787 unsigned int start
= fg
->start_index
;
790 *prev
= &fg
->node
.children
;
792 /* assumed list is sorted by index */
793 fs_for_each_fte(fte
, fg
) {
794 if (fte
->index
!= start
)
798 *prev
= &fte
->node
.list
;
804 /* prev is output, prev->next = new_fte */
805 static struct fs_fte
*create_fte(struct mlx5_flow_group
*fg
,
809 struct list_head
**prev
)
814 index
= get_free_fte_index(fg
, prev
);
815 fte
= alloc_fte(action
, flow_tag
, match_value
, index
);
822 static struct mlx5_flow_group
*create_autogroup(struct mlx5_flow_table
*ft
,
823 u8 match_criteria_enable
,
826 int inlen
= MLX5_ST_SZ_BYTES(create_flow_group_in
);
827 struct list_head
*prev
= &ft
->node
.children
;
828 unsigned int candidate_index
= 0;
829 struct mlx5_flow_group
*fg
;
830 void *match_criteria_addr
;
831 unsigned int group_size
= 0;
834 if (!ft
->autogroup
.active
)
835 return ERR_PTR(-ENOENT
);
837 in
= mlx5_vzalloc(inlen
);
839 return ERR_PTR(-ENOMEM
);
841 if (ft
->autogroup
.num_groups
< ft
->autogroup
.required_groups
)
842 /* We save place for flow groups in addition to max types */
843 group_size
= ft
->max_fte
/ (ft
->autogroup
.required_groups
+ 1);
845 /* ft->max_fte == ft->autogroup.max_types */
849 /* sorted by start_index */
850 fs_for_each_fg(fg
, ft
) {
851 if (candidate_index
+ group_size
> fg
->start_index
)
852 candidate_index
= fg
->start_index
+ fg
->max_ftes
;
855 prev
= &fg
->node
.list
;
858 if (candidate_index
+ group_size
> ft
->max_fte
) {
859 fg
= ERR_PTR(-ENOSPC
);
863 MLX5_SET(create_flow_group_in
, in
, match_criteria_enable
,
864 match_criteria_enable
);
865 MLX5_SET(create_flow_group_in
, in
, start_flow_index
, candidate_index
);
866 MLX5_SET(create_flow_group_in
, in
, end_flow_index
, candidate_index
+
868 match_criteria_addr
= MLX5_ADDR_OF(create_flow_group_in
,
870 memcpy(match_criteria_addr
, match_criteria
,
871 MLX5_ST_SZ_BYTES(fte_match_param
));
873 fg
= create_flow_group_common(ft
, in
, prev
, true);
879 static struct mlx5_flow_rule
*add_rule_fg(struct mlx5_flow_group
*fg
,
883 struct mlx5_flow_destination
*dest
)
886 struct mlx5_flow_rule
*rule
;
887 struct mlx5_flow_table
*ft
;
888 struct list_head
*prev
;
890 nested_lock_ref_node(&fg
->node
, FS_MUTEX_PARENT
);
891 fs_for_each_fte(fte
, fg
) {
892 nested_lock_ref_node(&fte
->node
, FS_MUTEX_CHILD
);
893 if (compare_match_value(&fg
->mask
, match_value
, &fte
->val
) &&
894 action
== fte
->action
&& flow_tag
== fte
->flow_tag
) {
895 rule
= add_rule_fte(fte
, fg
, dest
);
896 unlock_ref_node(&fte
->node
);
902 unlock_ref_node(&fte
->node
);
904 fs_get_obj(ft
, fg
->node
.parent
);
905 if (fg
->num_ftes
>= fg
->max_ftes
) {
906 rule
= ERR_PTR(-ENOSPC
);
910 fte
= create_fte(fg
, match_value
, action
, flow_tag
, &prev
);
915 tree_init_node(&fte
->node
, 0, del_fte
);
916 rule
= add_rule_fte(fte
, fg
, dest
);
924 tree_add_node(&fte
->node
, &fg
->node
);
925 list_add(&fte
->node
.list
, prev
);
927 tree_add_node(&rule
->node
, &fte
->node
);
929 unlock_ref_node(&fg
->node
);
933 static struct mlx5_flow_rule
*add_rule_to_auto_fg(struct mlx5_flow_table
*ft
,
934 u8 match_criteria_enable
,
939 struct mlx5_flow_destination
*dest
)
941 struct mlx5_flow_rule
*rule
;
942 struct mlx5_flow_group
*g
;
944 g
= create_autogroup(ft
, match_criteria_enable
, match_criteria
);
948 rule
= add_rule_fg(g
, match_value
,
949 action
, flow_tag
, dest
);
951 /* Remove assumes refcount > 0 and autogroup creates a group
952 * with a refcount = 0.
954 tree_get_node(&g
->node
);
955 tree_remove_node(&g
->node
);
960 struct mlx5_flow_rule
*
961 mlx5_add_flow_rule(struct mlx5_flow_table
*ft
,
962 u8 match_criteria_enable
,
967 struct mlx5_flow_destination
*dest
)
969 struct mlx5_flow_group
*g
;
970 struct mlx5_flow_rule
*rule
;
972 nested_lock_ref_node(&ft
->node
, FS_MUTEX_GRANDPARENT
);
973 fs_for_each_fg(g
, ft
)
974 if (compare_match_criteria(g
->mask
.match_criteria_enable
,
975 match_criteria_enable
,
976 g
->mask
.match_criteria
,
978 rule
= add_rule_fg(g
, match_value
,
979 action
, flow_tag
, dest
);
980 if (!IS_ERR(rule
) || PTR_ERR(rule
) != -ENOSPC
)
984 rule
= add_rule_to_auto_fg(ft
, match_criteria_enable
, match_criteria
,
985 match_value
, action
, flow_tag
, dest
);
987 unlock_ref_node(&ft
->node
);
991 void mlx5_del_flow_rule(struct mlx5_flow_rule
*rule
)
993 tree_remove_node(&rule
->node
);
996 /* Assuming prio->node.children(flow tables) is sorted by level */
997 static struct mlx5_flow_table
*find_next_ft(struct mlx5_flow_table
*ft
)
999 struct fs_prio
*prio
;
1001 fs_get_obj(prio
, ft
->node
.parent
);
1003 if (!list_is_last(&ft
->node
.list
, &prio
->node
.children
))
1004 return list_next_entry(ft
, node
.list
);
1005 return find_next_chained_ft(prio
);
1008 static int update_root_ft_destroy(struct mlx5_flow_table
*ft
)
1010 struct mlx5_flow_root_namespace
*root
= find_root(&ft
->node
);
1011 struct mlx5_flow_table
*new_root_ft
= NULL
;
1013 if (root
->root_ft
!= ft
)
1016 new_root_ft
= find_next_ft(ft
);
1018 int err
= mlx5_cmd_update_root_ft(root
->dev
, new_root_ft
);
1021 mlx5_core_warn(root
->dev
, "Update root flow table of id=%u failed\n",
1025 root
->root_ft
= new_root_ft
;
1030 /* Connect flow table from previous priority to
1031 * the next flow table.
1033 static int disconnect_flow_table(struct mlx5_flow_table
*ft
)
1035 struct mlx5_core_dev
*dev
= get_dev(&ft
->node
);
1036 struct mlx5_flow_table
*next_ft
;
1037 struct fs_prio
*prio
;
1040 err
= update_root_ft_destroy(ft
);
1044 fs_get_obj(prio
, ft
->node
.parent
);
1045 if (!(list_first_entry(&prio
->node
.children
,
1046 struct mlx5_flow_table
,
1050 next_ft
= find_next_chained_ft(prio
);
1051 err
= connect_prev_fts(dev
, next_ft
, prio
);
1053 mlx5_core_warn(dev
, "Failed to disconnect flow table %d\n",
1058 int mlx5_destroy_flow_table(struct mlx5_flow_table
*ft
)
1060 struct mlx5_flow_root_namespace
*root
= find_root(&ft
->node
);
1063 mutex_lock(&root
->chain_lock
);
1064 err
= disconnect_flow_table(ft
);
1066 mutex_unlock(&root
->chain_lock
);
1069 if (tree_remove_node(&ft
->node
))
1070 mlx5_core_warn(get_dev(&ft
->node
), "Flow table %d wasn't destroyed, refcount > 1\n",
1072 mutex_unlock(&root
->chain_lock
);
1077 void mlx5_destroy_flow_group(struct mlx5_flow_group
*fg
)
1079 if (tree_remove_node(&fg
->node
))
1080 mlx5_core_warn(get_dev(&fg
->node
), "Flow group %d wasn't destroyed, refcount > 1\n",
1084 struct mlx5_flow_namespace
*mlx5_get_flow_namespace(struct mlx5_core_dev
*dev
,
1085 enum mlx5_flow_namespace_type type
)
1087 struct mlx5_flow_root_namespace
*root_ns
= dev
->priv
.root_ns
;
1089 static struct fs_prio
*fs_prio
;
1090 struct mlx5_flow_namespace
*ns
;
1096 case MLX5_FLOW_NAMESPACE_KERNEL
:
1099 case MLX5_FLOW_NAMESPACE_FDB
:
1100 if (dev
->priv
.fdb_root_ns
)
1101 return &dev
->priv
.fdb_root_ns
->ns
;
1108 fs_prio
= find_prio(&root_ns
->ns
, prio
);
1112 ns
= list_first_entry(&fs_prio
->node
.children
,
1119 static struct fs_prio
*fs_create_prio(struct mlx5_flow_namespace
*ns
,
1120 unsigned prio
, int max_ft
,
1123 struct fs_prio
*fs_prio
;
1125 fs_prio
= kzalloc(sizeof(*fs_prio
), GFP_KERNEL
);
1127 return ERR_PTR(-ENOMEM
);
1129 fs_prio
->node
.type
= FS_TYPE_PRIO
;
1130 tree_init_node(&fs_prio
->node
, 1, NULL
);
1131 tree_add_node(&fs_prio
->node
, &ns
->node
);
1132 fs_prio
->max_ft
= max_ft
;
1133 fs_prio
->prio
= prio
;
1134 fs_prio
->start_level
= start_level
;
1135 list_add_tail(&fs_prio
->node
.list
, &ns
->node
.children
);
1140 static struct mlx5_flow_namespace
*fs_init_namespace(struct mlx5_flow_namespace
1143 ns
->node
.type
= FS_TYPE_NAMESPACE
;
1148 static struct mlx5_flow_namespace
*fs_create_namespace(struct fs_prio
*prio
)
1150 struct mlx5_flow_namespace
*ns
;
1152 ns
= kzalloc(sizeof(*ns
), GFP_KERNEL
);
1154 return ERR_PTR(-ENOMEM
);
1156 fs_init_namespace(ns
);
1157 tree_init_node(&ns
->node
, 1, NULL
);
1158 tree_add_node(&ns
->node
, &prio
->node
);
1159 list_add_tail(&ns
->node
.list
, &prio
->node
.children
);
1164 static int init_root_tree_recursive(int max_ft_level
, struct init_tree_node
*init_node
,
1165 struct fs_node
*fs_parent_node
,
1166 struct init_tree_node
*init_parent_node
,
1169 struct mlx5_flow_namespace
*fs_ns
;
1170 struct fs_prio
*fs_prio
;
1171 struct fs_node
*base
;
1175 if (init_node
->type
== FS_TYPE_PRIO
) {
1176 if (init_node
->min_ft_level
> max_ft_level
)
1179 fs_get_obj(fs_ns
, fs_parent_node
);
1180 fs_prio
= fs_create_prio(fs_ns
, index
, init_node
->max_ft
,
1181 init_node
->start_level
);
1182 if (IS_ERR(fs_prio
))
1183 return PTR_ERR(fs_prio
);
1184 base
= &fs_prio
->node
;
1185 } else if (init_node
->type
== FS_TYPE_NAMESPACE
) {
1186 fs_get_obj(fs_prio
, fs_parent_node
);
1187 fs_ns
= fs_create_namespace(fs_prio
);
1189 return PTR_ERR(fs_ns
);
1190 base
= &fs_ns
->node
;
1194 for (i
= 0; i
< init_node
->ar_size
; i
++) {
1195 err
= init_root_tree_recursive(max_ft_level
,
1196 &init_node
->children
[i
], base
,
1205 static int init_root_tree(int max_ft_level
, struct init_tree_node
*init_node
,
1206 struct fs_node
*fs_parent_node
)
1209 struct mlx5_flow_namespace
*fs_ns
;
1212 fs_get_obj(fs_ns
, fs_parent_node
);
1213 for (i
= 0; i
< init_node
->ar_size
; i
++) {
1214 err
= init_root_tree_recursive(max_ft_level
,
1215 &init_node
->children
[i
],
1224 static struct mlx5_flow_root_namespace
*create_root_ns(struct mlx5_core_dev
*dev
,
1225 enum fs_flow_table_type
1228 struct mlx5_flow_root_namespace
*root_ns
;
1229 struct mlx5_flow_namespace
*ns
;
1231 /* Create the root namespace */
1232 root_ns
= mlx5_vzalloc(sizeof(*root_ns
));
1237 root_ns
->table_type
= table_type
;
1240 fs_init_namespace(ns
);
1241 mutex_init(&root_ns
->chain_lock
);
1242 tree_init_node(&ns
->node
, 1, NULL
);
1243 tree_add_node(&ns
->node
, NULL
);
1248 static int init_root_ns(struct mlx5_core_dev
*dev
)
1250 int max_ft_level
= MLX5_CAP_FLOWTABLE(dev
,
1251 flow_table_properties_nic_receive
.
1254 dev
->priv
.root_ns
= create_root_ns(dev
, FS_FT_NIC_RX
);
1255 if (IS_ERR_OR_NULL(dev
->priv
.root_ns
))
1258 if (init_root_tree(max_ft_level
, &root_fs
, &dev
->priv
.root_ns
->ns
.node
))
1264 mlx5_cleanup_fs(dev
);
1268 static void cleanup_single_prio_root_ns(struct mlx5_core_dev
*dev
,
1269 struct mlx5_flow_root_namespace
*root_ns
)
1271 struct fs_node
*prio
;
1276 if (!list_empty(&root_ns
->ns
.node
.children
)) {
1277 prio
= list_first_entry(&root_ns
->ns
.node
.children
,
1280 if (tree_remove_node(prio
))
1282 "Flow steering priority wasn't destroyed, refcount > 1\n");
1284 if (tree_remove_node(&root_ns
->ns
.node
))
1286 "Flow steering namespace wasn't destroyed, refcount > 1\n");
1290 static void cleanup_root_ns(struct mlx5_core_dev
*dev
)
1292 struct mlx5_flow_root_namespace
*root_ns
= dev
->priv
.root_ns
;
1293 struct fs_prio
*iter_prio
;
1295 if (!MLX5_CAP_GEN(dev
, nic_flow_table
))
1302 fs_for_each_prio(iter_prio
, &root_ns
->ns
) {
1303 struct fs_node
*node
;
1304 struct mlx5_flow_namespace
*iter_ns
;
1306 fs_for_each_ns_or_ft(node
, iter_prio
) {
1307 if (node
->type
== FS_TYPE_FLOW_TABLE
)
1309 fs_get_obj(iter_ns
, node
);
1310 while (!list_empty(&iter_ns
->node
.children
)) {
1311 struct fs_prio
*obj_iter_prio2
;
1312 struct fs_node
*iter_prio2
=
1313 list_first_entry(&iter_ns
->node
.children
,
1317 fs_get_obj(obj_iter_prio2
, iter_prio2
);
1318 if (tree_remove_node(iter_prio2
)) {
1320 "Priority %d wasn't destroyed, refcount > 1\n",
1321 obj_iter_prio2
->prio
);
1329 fs_for_each_prio(iter_prio
, &root_ns
->ns
) {
1330 while (!list_empty(&iter_prio
->node
.children
)) {
1331 struct fs_node
*iter_ns
=
1332 list_first_entry(&iter_prio
->node
.children
,
1335 if (tree_remove_node(iter_ns
)) {
1337 "Namespace wasn't destroyed, refcount > 1\n");
1344 while (!list_empty(&root_ns
->ns
.node
.children
)) {
1345 struct fs_prio
*obj_prio_node
;
1346 struct fs_node
*prio_node
=
1347 list_first_entry(&root_ns
->ns
.node
.children
,
1351 fs_get_obj(obj_prio_node
, prio_node
);
1352 if (tree_remove_node(prio_node
)) {
1354 "Priority %d wasn't destroyed, refcount > 1\n",
1355 obj_prio_node
->prio
);
1360 if (tree_remove_node(&root_ns
->ns
.node
)) {
1362 "root namespace wasn't destroyed, refcount > 1\n");
1366 dev
->priv
.root_ns
= NULL
;
1369 void mlx5_cleanup_fs(struct mlx5_core_dev
*dev
)
1371 cleanup_root_ns(dev
);
1372 cleanup_single_prio_root_ns(dev
, dev
->priv
.fdb_root_ns
);
1375 static int init_fdb_root_ns(struct mlx5_core_dev
*dev
)
1377 struct fs_prio
*prio
;
1379 dev
->priv
.fdb_root_ns
= create_root_ns(dev
, FS_FT_FDB
);
1380 if (!dev
->priv
.fdb_root_ns
)
1383 /* Create single prio */
1384 prio
= fs_create_prio(&dev
->priv
.fdb_root_ns
->ns
, 0, 1, 0);
1386 cleanup_single_prio_root_ns(dev
, dev
->priv
.fdb_root_ns
);
1387 return PTR_ERR(prio
);
1393 int mlx5_init_fs(struct mlx5_core_dev
*dev
)
1397 if (MLX5_CAP_GEN(dev
, nic_flow_table
)) {
1398 err
= init_root_ns(dev
);
1402 if (MLX5_CAP_GEN(dev
, eswitch_flow_table
)) {
1403 err
= init_fdb_root_ns(dev
);
1405 cleanup_root_ns(dev
);