net/mlx5_core: Managing root flow table
[cascardo/linux.git] / drivers / net / ethernet / mellanox / mlx5 / core / fs_core.c
1 /*
2  * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
3  *
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:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
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.
22  *
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
30  * SOFTWARE.
31  */
32
33 #include <linux/mutex.h>
34 #include <linux/mlx5/driver.h>
35
36 #include "mlx5_core.h"
37 #include "fs_core.h"
38 #include "fs_cmd.h"
39
40 #define INIT_TREE_NODE_ARRAY_SIZE(...)  (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
41                                          sizeof(struct init_tree_node))
42
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__) \
50 }
51
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,\
54                   __VA_ARGS__)\
55
56 #define ADD_FT_PRIO(max_ft_val, start_level_val, ...)\
57         INIT_PRIO(0, max_ft_val, start_level_val,\
58                   __VA_ARGS__)\
59
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__) \
63 }
64
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;
72         int ar_size;
73         int min_ft_level;
74         int prio;
75         int max_ft;
76         int start_level;
77 } root_fs = {
78         .type = FS_TYPE_NAMESPACE,
79         .ar_size = 1,
80         .children = (struct init_tree_node[]) {
81                 ADD_PRIO(KENREL_MIN_LEVEL, KERNEL_MAX_FT,
82                          KERNEL_START_LEVEL,
83                          ADD_NS(ADD_FT_PRIO(KERNEL_MAX_FT,
84                                             KERNEL_P0_START_LEVEL))),
85         }
86 };
87
88 enum fs_i_mutex_lock_class {
89         FS_MUTEX_GRANDPARENT,
90         FS_MUTEX_PARENT,
91         FS_MUTEX_CHILD
92 };
93
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);
98
99 static void tree_init_node(struct fs_node *node,
100                            unsigned int refcount,
101                            void (*remove_func)(struct fs_node *))
102 {
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;
108 }
109
110 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
111 {
112         if (parent)
113                 atomic_inc(&parent->refcount);
114         node->parent = parent;
115
116         /* Parent is the root */
117         if (!parent)
118                 node->root = node;
119         else
120                 node->root = parent->root;
121 }
122
123 static void tree_get_node(struct fs_node *node)
124 {
125         atomic_inc(&node->refcount);
126 }
127
128 static void nested_lock_ref_node(struct fs_node *node,
129                                  enum fs_i_mutex_lock_class class)
130 {
131         if (node) {
132                 mutex_lock_nested(&node->lock, class);
133                 atomic_inc(&node->refcount);
134         }
135 }
136
137 static void lock_ref_node(struct fs_node *node)
138 {
139         if (node) {
140                 mutex_lock(&node->lock);
141                 atomic_inc(&node->refcount);
142         }
143 }
144
145 static void unlock_ref_node(struct fs_node *node)
146 {
147         if (node) {
148                 atomic_dec(&node->refcount);
149                 mutex_unlock(&node->lock);
150         }
151 }
152
153 static void tree_put_node(struct fs_node *node)
154 {
155         struct fs_node *parent_node = node->parent;
156
157         lock_ref_node(parent_node);
158         if (atomic_dec_and_test(&node->refcount)) {
159                 if (parent_node)
160                         list_del_init(&node->list);
161                 if (node->remove_func)
162                         node->remove_func(node);
163                 kfree(node);
164                 node = NULL;
165         }
166         unlock_ref_node(parent_node);
167         if (!node && parent_node)
168                 tree_put_node(parent_node);
169 }
170
171 static int tree_remove_node(struct fs_node *node)
172 {
173         if (atomic_read(&node->refcount) > 1)
174                 return -EPERM;
175         tree_put_node(node);
176         return 0;
177 }
178
179 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
180                                  unsigned int prio)
181 {
182         struct fs_prio *iter_prio;
183
184         fs_for_each_prio(iter_prio, ns) {
185                 if (iter_prio->prio == prio)
186                         return iter_prio;
187         }
188
189         return NULL;
190 }
191
192 static unsigned int find_next_free_level(struct fs_prio *prio)
193 {
194         if (!list_empty(&prio->node.children)) {
195                 struct mlx5_flow_table *ft;
196
197                 ft = list_last_entry(&prio->node.children,
198                                      struct mlx5_flow_table,
199                                      node.list);
200                 return ft->level + 1;
201         }
202         return prio->start_level;
203 }
204
205 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
206 {
207         unsigned int i;
208
209         for (i = 0; i < size; i++, mask++, val1++, val2++)
210                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
211                     ((*(u8 *)val2) & (*(u8 *)mask)))
212                         return false;
213
214         return true;
215 }
216
217 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
218                                 void *fte_param1, void *fte_param2)
219 {
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);
228
229                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
230                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
231                         return false;
232         }
233
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);
242
243                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
244                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
245                         return false;
246         }
247
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);
256
257                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
258                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
259                         return false;
260         }
261         return true;
262 }
263
264 static bool compare_match_criteria(u8 match_criteria_enable1,
265                                    u8 match_criteria_enable2,
266                                    void *mask1, void *mask2)
267 {
268         return match_criteria_enable1 == match_criteria_enable2 &&
269                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
270 }
271
272 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
273 {
274         struct fs_node *root;
275         struct mlx5_flow_namespace *ns;
276
277         root = node->root;
278
279         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
280                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
281                 return NULL;
282         }
283
284         ns = container_of(root, struct mlx5_flow_namespace, node);
285         return container_of(ns, struct mlx5_flow_root_namespace, ns);
286 }
287
288 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
289 {
290         struct mlx5_flow_root_namespace *root = find_root(node);
291
292         if (root)
293                 return root->dev;
294         return NULL;
295 }
296
297 static void del_flow_table(struct fs_node *node)
298 {
299         struct mlx5_flow_table *ft;
300         struct mlx5_core_dev *dev;
301         struct fs_prio *prio;
302         int err;
303
304         fs_get_obj(ft, node);
305         dev = get_dev(&ft->node);
306
307         err = mlx5_cmd_destroy_flow_table(dev, ft);
308         if (err)
309                 pr_warn("flow steering can't destroy ft\n");
310         fs_get_obj(prio, ft->node.parent);
311         prio->num_ft--;
312 }
313
314 static void del_rule(struct fs_node *node)
315 {
316         struct mlx5_flow_rule *rule;
317         struct mlx5_flow_table *ft;
318         struct mlx5_flow_group *fg;
319         struct fs_fte *fte;
320         u32     *match_value;
321         struct mlx5_core_dev *dev = get_dev(node);
322         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
323         int err;
324
325         match_value = mlx5_vzalloc(match_len);
326         if (!match_value) {
327                 pr_warn("failed to allocate inbox\n");
328                 return;
329         }
330
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);
337         fte->dests_size--;
338         if (fte->dests_size) {
339                 err = mlx5_cmd_update_fte(dev, ft,
340                                           fg->id, fte);
341                 if (err)
342                         pr_warn("%s can't del rule fg id=%d fte_index=%d\n",
343                                 __func__, fg->id, fte->index);
344         }
345         kvfree(match_value);
346 }
347
348 static void del_fte(struct fs_node *node)
349 {
350         struct mlx5_flow_table *ft;
351         struct mlx5_flow_group *fg;
352         struct mlx5_core_dev *dev;
353         struct fs_fte *fte;
354         int err;
355
356         fs_get_obj(fte, node);
357         fs_get_obj(fg, fte->node.parent);
358         fs_get_obj(ft, fg->node.parent);
359
360         dev = get_dev(&ft->node);
361         err = mlx5_cmd_delete_fte(dev, ft,
362                                   fte->index);
363         if (err)
364                 pr_warn("flow steering can't delete fte in index %d of flow group id %d\n",
365                         fte->index, fg->id);
366
367         fte->status = 0;
368         fg->num_ftes--;
369 }
370
371 static void del_flow_group(struct fs_node *node)
372 {
373         struct mlx5_flow_group *fg;
374         struct mlx5_flow_table *ft;
375         struct mlx5_core_dev *dev;
376
377         fs_get_obj(fg, node);
378         fs_get_obj(ft, fg->node.parent);
379         dev = get_dev(&ft->node);
380
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",
383                         fg->id, ft->id);
384 }
385
386 static struct fs_fte *alloc_fte(u8 action,
387                                 u32 flow_tag,
388                                 u32 *match_value,
389                                 unsigned int index)
390 {
391         struct fs_fte *fte;
392
393         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
394         if (!fte)
395                 return ERR_PTR(-ENOMEM);
396
397         memcpy(fte->val, match_value, sizeof(fte->val));
398         fte->node.type =  FS_TYPE_FLOW_ENTRY;
399         fte->flow_tag = flow_tag;
400         fte->index = index;
401         fte->action = action;
402
403         return fte;
404 }
405
406 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
407 {
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,
412                                             create_fg_in,
413                                             match_criteria_enable);
414         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
415         if (!fg)
416                 return ERR_PTR(-ENOMEM);
417
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,
423                                    start_flow_index);
424         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
425                                 end_flow_index) - fg->start_index + 1;
426         return fg;
427 }
428
429 static struct mlx5_flow_table *alloc_flow_table(int level, int max_fte,
430                                                 enum fs_flow_table_type table_type)
431 {
432         struct mlx5_flow_table *ft;
433
434         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
435         if (!ft)
436                 return NULL;
437
438         ft->level = level;
439         ft->node.type = FS_TYPE_FLOW_TABLE;
440         ft->type = table_type;
441         ft->max_fte = max_fte;
442
443         return ft;
444 }
445
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).
449  */
450 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
451                                                          struct list_head *start,
452                                                          bool reverse)
453 {
454 #define list_advance_entry(pos, reverse)                \
455         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
456
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))
461
462         struct fs_node *iter = list_entry(start, struct fs_node, list);
463         struct mlx5_flow_table *ft = NULL;
464
465         if (!root)
466                 return NULL;
467
468         list_for_each_advance_continue(iter, &root->children, reverse) {
469                 if (iter->type == FS_TYPE_FLOW_TABLE) {
470                         fs_get_obj(ft, iter);
471                         return ft;
472                 }
473                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
474                 if (ft)
475                         return ft;
476         }
477
478         return ft;
479 }
480
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.
484  */
485 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
486 {
487         struct mlx5_flow_table *ft = NULL;
488         struct fs_node *curr_node;
489         struct fs_node *parent;
490
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);
495                 curr_node = parent;
496                 parent = curr_node->parent;
497         }
498         return ft;
499 }
500
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)
503 {
504         return find_closest_ft(prio, false);
505 }
506
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)
509 {
510         return find_closest_ft(prio, true);
511 }
512
513 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
514                                  *prio)
515 {
516         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
517         int min_level = INT_MAX;
518         int err;
519
520         if (root->root_ft)
521                 min_level = root->root_ft->level;
522
523         if (ft->level >= min_level)
524                 return 0;
525
526         err = mlx5_cmd_update_root_ft(root->dev, ft);
527         if (err)
528                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
529                                ft->id);
530         else
531                 root->root_ft = ft;
532
533         return err;
534 }
535
536 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
537                                                int prio,
538                                                int max_fte)
539 {
540         struct mlx5_flow_table *ft;
541         int err;
542         int log_table_sz;
543         struct mlx5_flow_root_namespace *root =
544                 find_root(&ns->node);
545         struct fs_prio *fs_prio = NULL;
546
547         if (!root) {
548                 pr_err("mlx5: flow steering failed to find root of namespace\n");
549                 return ERR_PTR(-ENODEV);
550         }
551
552         mutex_lock(&root->chain_lock);
553         fs_prio = find_prio(ns, prio);
554         if (!fs_prio) {
555                 err = -EINVAL;
556                 goto unlock_root;
557         }
558         if (fs_prio->num_ft == fs_prio->max_ft) {
559                 err = -ENOSPC;
560                 goto unlock_root;
561         }
562
563         ft = alloc_flow_table(find_next_free_level(fs_prio),
564                               roundup_pow_of_two(max_fte),
565                               root->table_type);
566         if (!ft) {
567                 err = -ENOMEM;
568                 goto unlock_root;
569         }
570
571         tree_init_node(&ft->node, 1, del_flow_table);
572         log_table_sz = ilog2(ft->max_fte);
573         err = mlx5_cmd_create_flow_table(root->dev, ft->type, ft->level,
574                                          log_table_sz, &ft->id);
575         if (err)
576                 goto free_ft;
577
578         if (MLX5_CAP_FLOWTABLE(root->dev,
579                                flow_table_properties_nic_receive.modify_root)) {
580                 err = update_root_ft_create(ft, fs_prio);
581                 if (err)
582                         goto destroy_ft;
583         }
584         lock_ref_node(&fs_prio->node);
585         tree_add_node(&ft->node, &fs_prio->node);
586         list_add_tail(&ft->node.list, &fs_prio->node.children);
587         fs_prio->num_ft++;
588         unlock_ref_node(&fs_prio->node);
589         mutex_unlock(&root->chain_lock);
590         return ft;
591 destroy_ft:
592         mlx5_cmd_destroy_flow_table(root->dev, ft);
593 free_ft:
594         kfree(ft);
595 unlock_root:
596         mutex_unlock(&root->chain_lock);
597         return ERR_PTR(err);
598 }
599
600 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
601                                                             int prio,
602                                                             int num_flow_table_entries,
603                                                             int max_num_groups)
604 {
605         struct mlx5_flow_table *ft;
606
607         if (max_num_groups > num_flow_table_entries)
608                 return ERR_PTR(-EINVAL);
609
610         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries);
611         if (IS_ERR(ft))
612                 return ft;
613
614         ft->autogroup.active = true;
615         ft->autogroup.required_groups = max_num_groups;
616
617         return ft;
618 }
619
620 /* Flow table should be locked */
621 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
622                                                         u32 *fg_in,
623                                                         struct list_head
624                                                         *prev_fg,
625                                                         bool is_auto_fg)
626 {
627         struct mlx5_flow_group *fg;
628         struct mlx5_core_dev *dev = get_dev(&ft->node);
629         int err;
630
631         if (!dev)
632                 return ERR_PTR(-ENODEV);
633
634         fg = alloc_flow_group(fg_in);
635         if (IS_ERR(fg))
636                 return fg;
637
638         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
639         if (err) {
640                 kfree(fg);
641                 return ERR_PTR(err);
642         }
643
644         if (ft->autogroup.active)
645                 ft->autogroup.num_groups++;
646         /* Add node to tree */
647         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
648         tree_add_node(&fg->node, &ft->node);
649         /* Add node to group list */
650         list_add(&fg->node.list, ft->node.children.prev);
651
652         return fg;
653 }
654
655 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
656                                                u32 *fg_in)
657 {
658         struct mlx5_flow_group *fg;
659
660         if (ft->autogroup.active)
661                 return ERR_PTR(-EPERM);
662
663         lock_ref_node(&ft->node);
664         fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
665         unlock_ref_node(&ft->node);
666
667         return fg;
668 }
669
670 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
671 {
672         struct mlx5_flow_rule *rule;
673
674         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
675         if (!rule)
676                 return NULL;
677
678         rule->node.type = FS_TYPE_FLOW_DEST;
679         memcpy(&rule->dest_attr, dest, sizeof(*dest));
680
681         return rule;
682 }
683
684 /* fte should not be deleted while calling this function */
685 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
686                                            struct mlx5_flow_group *fg,
687                                            struct mlx5_flow_destination *dest)
688 {
689         struct mlx5_flow_table *ft;
690         struct mlx5_flow_rule *rule;
691         int err;
692
693         rule = alloc_rule(dest);
694         if (!rule)
695                 return ERR_PTR(-ENOMEM);
696
697         fs_get_obj(ft, fg->node.parent);
698         /* Add dest to dests list- added as first element after the head */
699         tree_init_node(&rule->node, 1, del_rule);
700         list_add_tail(&rule->node.list, &fte->node.children);
701         fte->dests_size++;
702         if (fte->dests_size == 1)
703                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
704                                           ft, fg->id, fte);
705         else
706                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
707                                           ft, fg->id, fte);
708         if (err)
709                 goto free_rule;
710
711         fte->status |= FS_FTE_STATUS_EXISTING;
712
713         return rule;
714
715 free_rule:
716         list_del(&rule->node.list);
717         kfree(rule);
718         fte->dests_size--;
719         return ERR_PTR(err);
720 }
721
722 /* Assumed fg is locked */
723 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
724                                        struct list_head **prev)
725 {
726         struct fs_fte *fte;
727         unsigned int start = fg->start_index;
728
729         if (prev)
730                 *prev = &fg->node.children;
731
732         /* assumed list is sorted by index */
733         fs_for_each_fte(fte, fg) {
734                 if (fte->index != start)
735                         return start;
736                 start++;
737                 if (prev)
738                         *prev = &fte->node.list;
739         }
740
741         return start;
742 }
743
744 /* prev is output, prev->next = new_fte */
745 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
746                                  u32 *match_value,
747                                  u8 action,
748                                  u32 flow_tag,
749                                  struct list_head **prev)
750 {
751         struct fs_fte *fte;
752         int index;
753
754         index = get_free_fte_index(fg, prev);
755         fte = alloc_fte(action, flow_tag, match_value, index);
756         if (IS_ERR(fte))
757                 return fte;
758
759         return fte;
760 }
761
762 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
763                                                 u8 match_criteria_enable,
764                                                 u32 *match_criteria)
765 {
766         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
767         struct list_head *prev = &ft->node.children;
768         unsigned int candidate_index = 0;
769         struct mlx5_flow_group *fg;
770         void *match_criteria_addr;
771         unsigned int group_size = 0;
772         u32 *in;
773
774         if (!ft->autogroup.active)
775                 return ERR_PTR(-ENOENT);
776
777         in = mlx5_vzalloc(inlen);
778         if (!in)
779                 return ERR_PTR(-ENOMEM);
780
781         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
782                 /* We save place for flow groups in addition to max types */
783                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
784
785         /*  ft->max_fte == ft->autogroup.max_types */
786         if (group_size == 0)
787                 group_size = 1;
788
789         /* sorted by start_index */
790         fs_for_each_fg(fg, ft) {
791                 if (candidate_index + group_size > fg->start_index)
792                         candidate_index = fg->start_index + fg->max_ftes;
793                 else
794                         break;
795                 prev = &fg->node.list;
796         }
797
798         if (candidate_index + group_size > ft->max_fte) {
799                 fg = ERR_PTR(-ENOSPC);
800                 goto out;
801         }
802
803         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
804                  match_criteria_enable);
805         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
806         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
807                  group_size - 1);
808         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
809                                            in, match_criteria);
810         memcpy(match_criteria_addr, match_criteria,
811                MLX5_ST_SZ_BYTES(fte_match_param));
812
813         fg = create_flow_group_common(ft, in, prev, true);
814 out:
815         kvfree(in);
816         return fg;
817 }
818
819 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
820                                           u32 *match_value,
821                                           u8 action,
822                                           u32 flow_tag,
823                                           struct mlx5_flow_destination *dest)
824 {
825         struct fs_fte *fte;
826         struct mlx5_flow_rule *rule;
827         struct mlx5_flow_table *ft;
828         struct list_head *prev;
829
830         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
831         fs_for_each_fte(fte, fg) {
832                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
833                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
834                     action == fte->action && flow_tag == fte->flow_tag) {
835                         rule = add_rule_fte(fte, fg, dest);
836                         unlock_ref_node(&fte->node);
837                         if (IS_ERR(rule))
838                                 goto unlock_fg;
839                         else
840                                 goto add_rule;
841                 }
842                 unlock_ref_node(&fte->node);
843         }
844         fs_get_obj(ft, fg->node.parent);
845         if (fg->num_ftes >= fg->max_ftes) {
846                 rule = ERR_PTR(-ENOSPC);
847                 goto unlock_fg;
848         }
849
850         fte = create_fte(fg, match_value, action, flow_tag, &prev);
851         if (IS_ERR(fte)) {
852                 rule = (void *)fte;
853                 goto unlock_fg;
854         }
855         tree_init_node(&fte->node, 0, del_fte);
856         rule = add_rule_fte(fte, fg, dest);
857         if (IS_ERR(rule)) {
858                 kfree(fte);
859                 goto unlock_fg;
860         }
861
862         fg->num_ftes++;
863
864         tree_add_node(&fte->node, &fg->node);
865         list_add(&fte->node.list, prev);
866 add_rule:
867         tree_add_node(&rule->node, &fte->node);
868 unlock_fg:
869         unlock_ref_node(&fg->node);
870         return rule;
871 }
872
873 static struct mlx5_flow_rule *add_rule_to_auto_fg(struct mlx5_flow_table *ft,
874                                                   u8 match_criteria_enable,
875                                                   u32 *match_criteria,
876                                                   u32 *match_value,
877                                                   u8 action,
878                                                   u32 flow_tag,
879                                                   struct mlx5_flow_destination *dest)
880 {
881         struct mlx5_flow_rule *rule;
882         struct mlx5_flow_group *g;
883
884         g = create_autogroup(ft, match_criteria_enable, match_criteria);
885         if (IS_ERR(g))
886                 return (void *)g;
887
888         rule = add_rule_fg(g, match_value,
889                            action, flow_tag, dest);
890         if (IS_ERR(rule)) {
891                 /* Remove assumes refcount > 0 and autogroup creates a group
892                  * with a refcount = 0.
893                  */
894                 tree_get_node(&g->node);
895                 tree_remove_node(&g->node);
896         }
897         return rule;
898 }
899
900 struct mlx5_flow_rule *
901 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
902                    u8 match_criteria_enable,
903                    u32 *match_criteria,
904                    u32 *match_value,
905                    u32 action,
906                    u32 flow_tag,
907                    struct mlx5_flow_destination *dest)
908 {
909         struct mlx5_flow_group *g;
910         struct mlx5_flow_rule *rule;
911
912         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
913         fs_for_each_fg(g, ft)
914                 if (compare_match_criteria(g->mask.match_criteria_enable,
915                                            match_criteria_enable,
916                                            g->mask.match_criteria,
917                                            match_criteria)) {
918                         rule = add_rule_fg(g, match_value,
919                                            action, flow_tag, dest);
920                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
921                                 goto unlock;
922                 }
923
924         rule = add_rule_to_auto_fg(ft, match_criteria_enable, match_criteria,
925                                    match_value, action, flow_tag, dest);
926 unlock:
927         unlock_ref_node(&ft->node);
928         return rule;
929 }
930
931 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
932 {
933         tree_remove_node(&rule->node);
934 }
935
936 /* Assuming prio->node.children(flow tables) is sorted by level */
937 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
938 {
939         struct fs_prio *prio;
940
941         fs_get_obj(prio, ft->node.parent);
942
943         if (!list_is_last(&ft->node.list, &prio->node.children))
944                 return list_next_entry(ft, node.list);
945         return find_next_chained_ft(prio);
946 }
947
948 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
949 {
950         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
951         struct mlx5_flow_table *new_root_ft = NULL;
952
953         if (root->root_ft != ft)
954                 return 0;
955
956         new_root_ft = find_next_ft(ft);
957         if (new_root_ft) {
958                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
959
960                 if (err) {
961                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
962                                        ft->id);
963                         return err;
964                 }
965                 root->root_ft = new_root_ft;
966         }
967         return 0;
968 }
969
970 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
971 {
972         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
973         int err = 0;
974
975         mutex_lock(&root->chain_lock);
976         err = update_root_ft_destroy(ft);
977         if (err) {
978                 mutex_unlock(&root->chain_lock);
979                 return err;
980         }
981         if (tree_remove_node(&ft->node))
982                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
983                                ft->id);
984         mutex_unlock(&root->chain_lock);
985
986         return err;
987 }
988
989 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
990 {
991         if (tree_remove_node(&fg->node))
992                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
993                                fg->id);
994 }
995
996 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
997                                                     enum mlx5_flow_namespace_type type)
998 {
999         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1000         int prio;
1001         static struct fs_prio *fs_prio;
1002         struct mlx5_flow_namespace *ns;
1003
1004         if (!root_ns)
1005                 return NULL;
1006
1007         switch (type) {
1008         case MLX5_FLOW_NAMESPACE_KERNEL:
1009                 prio = 0;
1010                 break;
1011         case MLX5_FLOW_NAMESPACE_FDB:
1012                 if (dev->priv.fdb_root_ns)
1013                         return &dev->priv.fdb_root_ns->ns;
1014                 else
1015                         return NULL;
1016         default:
1017                 return NULL;
1018         }
1019
1020         fs_prio = find_prio(&root_ns->ns, prio);
1021         if (!fs_prio)
1022                 return NULL;
1023
1024         ns = list_first_entry(&fs_prio->node.children,
1025                               typeof(*ns),
1026                               node.list);
1027
1028         return ns;
1029 }
1030
1031 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1032                                       unsigned prio, int max_ft,
1033                                       int start_level)
1034 {
1035         struct fs_prio *fs_prio;
1036
1037         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1038         if (!fs_prio)
1039                 return ERR_PTR(-ENOMEM);
1040
1041         fs_prio->node.type = FS_TYPE_PRIO;
1042         tree_init_node(&fs_prio->node, 1, NULL);
1043         tree_add_node(&fs_prio->node, &ns->node);
1044         fs_prio->max_ft = max_ft;
1045         fs_prio->prio = prio;
1046         fs_prio->start_level = start_level;
1047         list_add_tail(&fs_prio->node.list, &ns->node.children);
1048
1049         return fs_prio;
1050 }
1051
1052 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1053                                                      *ns)
1054 {
1055         ns->node.type = FS_TYPE_NAMESPACE;
1056
1057         return ns;
1058 }
1059
1060 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1061 {
1062         struct mlx5_flow_namespace      *ns;
1063
1064         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1065         if (!ns)
1066                 return ERR_PTR(-ENOMEM);
1067
1068         fs_init_namespace(ns);
1069         tree_init_node(&ns->node, 1, NULL);
1070         tree_add_node(&ns->node, &prio->node);
1071         list_add_tail(&ns->node.list, &prio->node.children);
1072
1073         return ns;
1074 }
1075
1076 static int init_root_tree_recursive(int max_ft_level, struct init_tree_node *init_node,
1077                                     struct fs_node *fs_parent_node,
1078                                     struct init_tree_node *init_parent_node,
1079                                     int index)
1080 {
1081         struct mlx5_flow_namespace *fs_ns;
1082         struct fs_prio *fs_prio;
1083         struct fs_node *base;
1084         int i;
1085         int err;
1086
1087         if (init_node->type == FS_TYPE_PRIO) {
1088                 if (init_node->min_ft_level > max_ft_level)
1089                         return -ENOTSUPP;
1090
1091                 fs_get_obj(fs_ns, fs_parent_node);
1092                 fs_prio = fs_create_prio(fs_ns, index, init_node->max_ft,
1093                                          init_node->start_level);
1094                 if (IS_ERR(fs_prio))
1095                         return PTR_ERR(fs_prio);
1096                 base = &fs_prio->node;
1097         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1098                 fs_get_obj(fs_prio, fs_parent_node);
1099                 fs_ns = fs_create_namespace(fs_prio);
1100                 if (IS_ERR(fs_ns))
1101                         return PTR_ERR(fs_ns);
1102                 base = &fs_ns->node;
1103         } else {
1104                 return -EINVAL;
1105         }
1106         for (i = 0; i < init_node->ar_size; i++) {
1107                 err = init_root_tree_recursive(max_ft_level,
1108                                                &init_node->children[i], base,
1109                                                init_node, i);
1110                 if (err)
1111                         return err;
1112         }
1113
1114         return 0;
1115 }
1116
1117 static int init_root_tree(int max_ft_level, struct init_tree_node *init_node,
1118                           struct fs_node *fs_parent_node)
1119 {
1120         int i;
1121         struct mlx5_flow_namespace *fs_ns;
1122         int err;
1123
1124         fs_get_obj(fs_ns, fs_parent_node);
1125         for (i = 0; i < init_node->ar_size; i++) {
1126                 err = init_root_tree_recursive(max_ft_level,
1127                                                &init_node->children[i],
1128                                                &fs_ns->node,
1129                                                init_node, i);
1130                 if (err)
1131                         return err;
1132         }
1133         return 0;
1134 }
1135
1136 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_core_dev *dev,
1137                                                        enum fs_flow_table_type
1138                                                        table_type)
1139 {
1140         struct mlx5_flow_root_namespace *root_ns;
1141         struct mlx5_flow_namespace *ns;
1142
1143         /* Create the root namespace */
1144         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1145         if (!root_ns)
1146                 return NULL;
1147
1148         root_ns->dev = dev;
1149         root_ns->table_type = table_type;
1150
1151         ns = &root_ns->ns;
1152         fs_init_namespace(ns);
1153         mutex_init(&root_ns->chain_lock);
1154         tree_init_node(&ns->node, 1, NULL);
1155         tree_add_node(&ns->node, NULL);
1156
1157         return root_ns;
1158 }
1159
1160 static int init_root_ns(struct mlx5_core_dev *dev)
1161 {
1162         int max_ft_level = MLX5_CAP_FLOWTABLE(dev,
1163                                               flow_table_properties_nic_receive.
1164                                               max_ft_level);
1165
1166         dev->priv.root_ns = create_root_ns(dev, FS_FT_NIC_RX);
1167         if (IS_ERR_OR_NULL(dev->priv.root_ns))
1168                 goto cleanup;
1169
1170         if (init_root_tree(max_ft_level, &root_fs, &dev->priv.root_ns->ns.node))
1171                 goto cleanup;
1172
1173         return 0;
1174
1175 cleanup:
1176         mlx5_cleanup_fs(dev);
1177         return -ENOMEM;
1178 }
1179
1180 static void cleanup_single_prio_root_ns(struct mlx5_core_dev *dev,
1181                                         struct mlx5_flow_root_namespace *root_ns)
1182 {
1183         struct fs_node *prio;
1184
1185         if (!root_ns)
1186                 return;
1187
1188         if (!list_empty(&root_ns->ns.node.children)) {
1189                 prio = list_first_entry(&root_ns->ns.node.children,
1190                                         struct fs_node,
1191                                  list);
1192                 if (tree_remove_node(prio))
1193                         mlx5_core_warn(dev,
1194                                        "Flow steering priority wasn't destroyed, refcount > 1\n");
1195         }
1196         if (tree_remove_node(&root_ns->ns.node))
1197                 mlx5_core_warn(dev,
1198                                "Flow steering namespace wasn't destroyed, refcount > 1\n");
1199         root_ns = NULL;
1200 }
1201
1202 static void cleanup_root_ns(struct mlx5_core_dev *dev)
1203 {
1204         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1205         struct fs_prio *iter_prio;
1206
1207         if (!MLX5_CAP_GEN(dev, nic_flow_table))
1208                 return;
1209
1210         if (!root_ns)
1211                 return;
1212
1213         /* stage 1 */
1214         fs_for_each_prio(iter_prio, &root_ns->ns) {
1215                 struct fs_node *node;
1216                 struct mlx5_flow_namespace *iter_ns;
1217
1218                 fs_for_each_ns_or_ft(node, iter_prio) {
1219                         if (node->type == FS_TYPE_FLOW_TABLE)
1220                                 continue;
1221                         fs_get_obj(iter_ns, node);
1222                         while (!list_empty(&iter_ns->node.children)) {
1223                                 struct fs_prio *obj_iter_prio2;
1224                                 struct fs_node *iter_prio2 =
1225                                         list_first_entry(&iter_ns->node.children,
1226                                                          struct fs_node,
1227                                                          list);
1228
1229                                 fs_get_obj(obj_iter_prio2, iter_prio2);
1230                                 if (tree_remove_node(iter_prio2)) {
1231                                         mlx5_core_warn(dev,
1232                                                        "Priority %d wasn't destroyed, refcount > 1\n",
1233                                                        obj_iter_prio2->prio);
1234                                         return;
1235                                 }
1236                         }
1237                 }
1238         }
1239
1240         /* stage 2 */
1241         fs_for_each_prio(iter_prio, &root_ns->ns) {
1242                 while (!list_empty(&iter_prio->node.children)) {
1243                         struct fs_node *iter_ns =
1244                                 list_first_entry(&iter_prio->node.children,
1245                                                  struct fs_node,
1246                                                  list);
1247                         if (tree_remove_node(iter_ns)) {
1248                                 mlx5_core_warn(dev,
1249                                                "Namespace wasn't destroyed, refcount > 1\n");
1250                                 return;
1251                         }
1252                 }
1253         }
1254
1255         /* stage 3 */
1256         while (!list_empty(&root_ns->ns.node.children)) {
1257                 struct fs_prio *obj_prio_node;
1258                 struct fs_node *prio_node =
1259                         list_first_entry(&root_ns->ns.node.children,
1260                                          struct fs_node,
1261                                          list);
1262
1263                 fs_get_obj(obj_prio_node, prio_node);
1264                 if (tree_remove_node(prio_node)) {
1265                         mlx5_core_warn(dev,
1266                                        "Priority %d wasn't destroyed, refcount > 1\n",
1267                                        obj_prio_node->prio);
1268                         return;
1269                 }
1270         }
1271
1272         if (tree_remove_node(&root_ns->ns.node)) {
1273                 mlx5_core_warn(dev,
1274                                "root namespace wasn't destroyed, refcount > 1\n");
1275                 return;
1276         }
1277
1278         dev->priv.root_ns = NULL;
1279 }
1280
1281 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1282 {
1283         cleanup_root_ns(dev);
1284         cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1285 }
1286
1287 static int init_fdb_root_ns(struct mlx5_core_dev *dev)
1288 {
1289         struct fs_prio *prio;
1290
1291         dev->priv.fdb_root_ns = create_root_ns(dev, FS_FT_FDB);
1292         if (!dev->priv.fdb_root_ns)
1293                 return -ENOMEM;
1294
1295         /* Create single prio */
1296         prio = fs_create_prio(&dev->priv.fdb_root_ns->ns, 0, 1, 0);
1297         if (IS_ERR(prio)) {
1298                 cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1299                 return PTR_ERR(prio);
1300         } else {
1301                 return 0;
1302         }
1303 }
1304
1305 int mlx5_init_fs(struct mlx5_core_dev *dev)
1306 {
1307         int err = 0;
1308
1309         if (MLX5_CAP_GEN(dev, nic_flow_table)) {
1310                 err = init_root_ns(dev);
1311                 if (err)
1312                         return err;
1313         }
1314         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1315                 err = init_fdb_root_ns(dev);
1316                 if (err)
1317                         cleanup_root_ns(dev);
1318         }
1319
1320         return err;
1321 }