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 connect_fts_in_prio(struct mlx5_core_dev *dev,
 514                                struct fs_prio *prio,
 515                                struct mlx5_flow_table *ft)
 516 {
 517         struct mlx5_flow_table *iter;
 518         int i = 0;
 519         int err;
 520
 521         fs_for_each_ft(iter, prio) {
 522                 i++;
 523                 err = mlx5_cmd_modify_flow_table(dev,
 524                                                  iter,
 525                                                  ft);
 526                 if (err) {
 527                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 528                                        iter->id);
 529                         /* The driver is out of sync with the FW */
 530                         if (i > 1)
 531                                 WARN_ON(true);
 532                         return err;
 533                 }
 534         }
 535         return 0;
 536 }
 537
 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)
 542 {
 543         struct mlx5_flow_table *prev_ft;
 544
 545         prev_ft = find_prev_chained_ft(prio);
 546         if (prev_ft) {
 547                 struct fs_prio *prev_prio;
 548
 549                 fs_get_obj(prev_prio, prev_ft->node.parent);
 550                 return connect_fts_in_prio(dev, prev_prio, ft);
 551         }
 552         return 0;
 553 }
 554
 555 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 556                                  *prio)
 557 {
 558         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 559         int min_level = INT_MAX;
 560         int err;
 561
 562         if (root->root_ft)
 563                 min_level = root->root_ft->level;
 564
 565         if (ft->level >= min_level)
 566                 return 0;
 567
 568         err = mlx5_cmd_update_root_ft(root->dev, ft);
 569         if (err)
 570                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 571                                ft->id);
 572         else
 573                 root->root_ft = ft;
 574
 575         return err;
 576 }
 577
 578 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 579                               struct fs_prio *prio)
 580 {
 581         int err = 0;
 582
 583         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 584
 585         if (list_empty(&prio->node.children)) {
 586                 err = connect_prev_fts(dev, ft, prio);
 587                 if (err)
 588                         return err;
 589         }
 590
 591         if (MLX5_CAP_FLOWTABLE(dev,
 592                                flow_table_properties_nic_receive.modify_root))
 593                 err = update_root_ft_create(ft, prio);
 594         return err;
 595 }
 596
 597 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 598                                                int prio,
 599                                                int max_fte)
 600 {
 601         struct mlx5_flow_table *next_ft = NULL;
 602         struct mlx5_flow_table *ft;
 603         int err;
 604         int log_table_sz;
 605         struct mlx5_flow_root_namespace *root =
 606                 find_root(&ns->node);
 607         struct fs_prio *fs_prio = NULL;
 608
 609         if (!root) {
 610                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 611                 return ERR_PTR(-ENODEV);
 612         }
 613
 614         mutex_lock(&root->chain_lock);
 615         fs_prio = find_prio(ns, prio);
 616         if (!fs_prio) {
 617                 err = -EINVAL;
 618                 goto unlock_root;
 619         }
 620         if (fs_prio->num_ft == fs_prio->max_ft) {
 621                 err = -ENOSPC;
 622                 goto unlock_root;
 623         }
 624
 625         ft = alloc_flow_table(find_next_free_level(fs_prio),
 626                               roundup_pow_of_two(max_fte),
 627                               root->table_type);
 628         if (!ft) {
 629                 err = -ENOMEM;
 630                 goto unlock_root;
 631         }
 632
 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);
 638         if (err)
 639                 goto free_ft;
 640
 641         err = connect_flow_table(root->dev, ft, fs_prio);
 642         if (err)
 643                 goto destroy_ft;
 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);
 647         fs_prio->num_ft++;
 648         unlock_ref_node(&fs_prio->node);
 649         mutex_unlock(&root->chain_lock);
 650         return ft;
 651 destroy_ft:
 652         mlx5_cmd_destroy_flow_table(root->dev, ft);
 653 free_ft:
 654         kfree(ft);
 655 unlock_root:
 656         mutex_unlock(&root->chain_lock);
 657         return ERR_PTR(err);
 658 }
 659
 660 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 661                                                             int prio,
 662                                                             int num_flow_table_entries,
 663                                                             int max_num_groups)
 664 {
 665         struct mlx5_flow_table *ft;
 666
 667         if (max_num_groups > num_flow_table_entries)
 668                 return ERR_PTR(-EINVAL);
 669
 670         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries);
 671         if (IS_ERR(ft))
 672                 return ft;
 673
 674         ft->autogroup.active = true;
 675         ft->autogroup.required_groups = max_num_groups;
 676
 677         return ft;
 678 }
 679
 680 /* Flow table should be locked */
 681 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 682                                                         u32 *fg_in,
 683                                                         struct list_head
 684                                                         *prev_fg,
 685                                                         bool is_auto_fg)
 686 {
 687         struct mlx5_flow_group *fg;
 688         struct mlx5_core_dev *dev = get_dev(&ft->node);
 689         int err;
 690
 691         if (!dev)
 692                 return ERR_PTR(-ENODEV);
 693
 694         fg = alloc_flow_group(fg_in);
 695         if (IS_ERR(fg))
 696                 return fg;
 697
 698         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 699         if (err) {
 700                 kfree(fg);
 701                 return ERR_PTR(err);
 702         }
 703
 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);
 711
 712         return fg;
 713 }
 714
 715 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 716                                                u32 *fg_in)
 717 {
 718         struct mlx5_flow_group *fg;
 719
 720         if (ft->autogroup.active)
 721                 return ERR_PTR(-EPERM);
 722
 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);
 726
 727         return fg;
 728 }
 729
 730 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 731 {
 732         struct mlx5_flow_rule *rule;
 733
 734         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 735         if (!rule)
 736                 return NULL;
 737
 738         rule->node.type = FS_TYPE_FLOW_DEST;
 739         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 740
 741         return rule;
 742 }
 743
 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)
 748 {
 749         struct mlx5_flow_table *ft;
 750         struct mlx5_flow_rule *rule;
 751         int err;
 752
 753         rule = alloc_rule(dest);
 754         if (!rule)
 755                 return ERR_PTR(-ENOMEM);
 756
 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);
 761         fte->dests_size++;
 762         if (fte->dests_size == 1)
 763                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 764                                           ft, fg->id, fte);
 765         else
 766                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 767                                           ft, fg->id, fte);
 768         if (err)
 769                 goto free_rule;
 770
 771         fte->status |= FS_FTE_STATUS_EXISTING;
 772
 773         return rule;
 774
 775 free_rule:
 776         list_del(&rule->node.list);
 777         kfree(rule);
 778         fte->dests_size--;
 779         return ERR_PTR(err);
 780 }
 781
 782 /* Assumed fg is locked */
 783 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 784                                        struct list_head **prev)
 785 {
 786         struct fs_fte *fte;
 787         unsigned int start = fg->start_index;
 788
 789         if (prev)
 790                 *prev = &fg->node.children;
 791
 792         /* assumed list is sorted by index */
 793         fs_for_each_fte(fte, fg) {
 794                 if (fte->index != start)
 795                         return start;
 796                 start++;
 797                 if (prev)
 798                         *prev = &fte->node.list;
 799         }
 800
 801         return start;
 802 }
 803
 804 /* prev is output, prev->next = new_fte */
 805 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
 806                                  u32 *match_value,
 807                                  u8 action,
 808                                  u32 flow_tag,
 809                                  struct list_head **prev)
 810 {
 811         struct fs_fte *fte;
 812         int index;
 813
 814         index = get_free_fte_index(fg, prev);
 815         fte = alloc_fte(action, flow_tag, match_value, index);
 816         if (IS_ERR(fte))
 817                 return fte;
 818
 819         return fte;
 820 }
 821
 822 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
 823                                                 u8 match_criteria_enable,
 824                                                 u32 *match_criteria)
 825 {
 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;
 832         u32 *in;
 833
 834         if (!ft->autogroup.active)
 835                 return ERR_PTR(-ENOENT);
 836
 837         in = mlx5_vzalloc(inlen);
 838         if (!in)
 839                 return ERR_PTR(-ENOMEM);
 840
 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);
 844
 845         /*  ft->max_fte == ft->autogroup.max_types */
 846         if (group_size == 0)
 847                 group_size = 1;
 848
 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;
 853                 else
 854                         break;
 855                 prev = &fg->node.list;
 856         }
 857
 858         if (candidate_index + group_size > ft->max_fte) {
 859                 fg = ERR_PTR(-ENOSPC);
 860                 goto out;
 861         }
 862
 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 +
 867                  group_size - 1);
 868         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
 869                                            in, match_criteria);
 870         memcpy(match_criteria_addr, match_criteria,
 871                MLX5_ST_SZ_BYTES(fte_match_param));
 872
 873         fg = create_flow_group_common(ft, in, prev, true);
 874 out:
 875         kvfree(in);
 876         return fg;
 877 }
 878
 879 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
 880                                           u32 *match_value,
 881                                           u8 action,
 882                                           u32 flow_tag,
 883                                           struct mlx5_flow_destination *dest)
 884 {
 885         struct fs_fte *fte;
 886         struct mlx5_flow_rule *rule;
 887         struct mlx5_flow_table *ft;
 888         struct list_head *prev;
 889
 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);
 897                         if (IS_ERR(rule))
 898                                 goto unlock_fg;
 899                         else
 900                                 goto add_rule;
 901                 }
 902                 unlock_ref_node(&fte->node);
 903         }
 904         fs_get_obj(ft, fg->node.parent);
 905         if (fg->num_ftes >= fg->max_ftes) {
 906                 rule = ERR_PTR(-ENOSPC);
 907                 goto unlock_fg;
 908         }
 909
 910         fte = create_fte(fg, match_value, action, flow_tag, &prev);
 911         if (IS_ERR(fte)) {
 912                 rule = (void *)fte;
 913                 goto unlock_fg;
 914         }
 915         tree_init_node(&fte->node, 0, del_fte);
 916         rule = add_rule_fte(fte, fg, dest);
 917         if (IS_ERR(rule)) {
 918                 kfree(fte);
 919                 goto unlock_fg;
 920         }
 921
 922         fg->num_ftes++;
 923
 924         tree_add_node(&fte->node, &fg->node);
 925         list_add(&fte->node.list, prev);
 926 add_rule:
 927         tree_add_node(&rule->node, &fte->node);
 928 unlock_fg:
 929         unlock_ref_node(&fg->node);
 930         return rule;
 931 }
 932
 933 static struct mlx5_flow_rule *add_rule_to_auto_fg(struct mlx5_flow_table *ft,
 934                                                   u8 match_criteria_enable,
 935                                                   u32 *match_criteria,
 936                                                   u32 *match_value,
 937                                                   u8 action,
 938                                                   u32 flow_tag,
 939                                                   struct mlx5_flow_destination *dest)
 940 {
 941         struct mlx5_flow_rule *rule;
 942         struct mlx5_flow_group *g;
 943
 944         g = create_autogroup(ft, match_criteria_enable, match_criteria);
 945         if (IS_ERR(g))
 946                 return (void *)g;
 947
 948         rule = add_rule_fg(g, match_value,
 949                            action, flow_tag, dest);
 950         if (IS_ERR(rule)) {
 951                 /* Remove assumes refcount > 0 and autogroup creates a group
 952                  * with a refcount = 0.
 953                  */
 954                 tree_get_node(&g->node);
 955                 tree_remove_node(&g->node);
 956         }
 957         return rule;
 958 }
 959
 960 struct mlx5_flow_rule *
 961 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
 962                    u8 match_criteria_enable,
 963                    u32 *match_criteria,
 964                    u32 *match_value,
 965                    u32 action,
 966                    u32 flow_tag,
 967                    struct mlx5_flow_destination *dest)
 968 {
 969         struct mlx5_flow_group *g;
 970         struct mlx5_flow_rule *rule;
 971
 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,
 977                                            match_criteria)) {
 978                         rule = add_rule_fg(g, match_value,
 979                                            action, flow_tag, dest);
 980                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
 981                                 goto unlock;
 982                 }
 983
 984         rule = add_rule_to_auto_fg(ft, match_criteria_enable, match_criteria,
 985                                    match_value, action, flow_tag, dest);
 986 unlock:
 987         unlock_ref_node(&ft->node);
 988         return rule;
 989 }
 990
 991 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
 992 {
 993         tree_remove_node(&rule->node);
 994 }
 995
 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)
 998 {
 999         struct fs_prio *prio;
1000
1001         fs_get_obj(prio, ft->node.parent);
1002
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);
1006 }
1007
1008 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1009 {
1010         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1011         struct mlx5_flow_table *new_root_ft = NULL;
1012
1013         if (root->root_ft != ft)
1014                 return 0;
1015
1016         new_root_ft = find_next_ft(ft);
1017         if (new_root_ft) {
1018                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1019
1020                 if (err) {
1021                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1022                                        ft->id);
1023                         return err;
1024                 }
1025                 root->root_ft = new_root_ft;
1026         }
1027         return 0;
1028 }
1029
1030 /* Connect flow table from previous priority to
1031  * the next flow table.
1032  */
1033 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1034 {
1035         struct mlx5_core_dev *dev = get_dev(&ft->node);
1036         struct mlx5_flow_table *next_ft;
1037         struct fs_prio *prio;
1038         int err = 0;
1039
1040         err = update_root_ft_destroy(ft);
1041         if (err)
1042                 return err;
1043
1044         fs_get_obj(prio, ft->node.parent);
1045         if  (!(list_first_entry(&prio->node.children,
1046                                 struct mlx5_flow_table,
1047                                 node.list) == ft))
1048                 return 0;
1049
1050         next_ft = find_next_chained_ft(prio);
1051         err = connect_prev_fts(dev, next_ft, prio);
1052         if (err)
1053                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1054                                ft->id);
1055         return err;
1056 }
1057
1058 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1059 {
1060         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1061         int err = 0;
1062
1063         mutex_lock(&root->chain_lock);
1064         err = disconnect_flow_table(ft);
1065         if (err) {
1066                 mutex_unlock(&root->chain_lock);
1067                 return err;
1068         }
1069         if (tree_remove_node(&ft->node))
1070                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1071                                ft->id);
1072         mutex_unlock(&root->chain_lock);
1073
1074         return err;
1075 }
1076
1077 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1078 {
1079         if (tree_remove_node(&fg->node))
1080                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1081                                fg->id);
1082 }
1083
1084 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1085                                                     enum mlx5_flow_namespace_type type)
1086 {
1087         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1088         int prio;
1089         static struct fs_prio *fs_prio;
1090         struct mlx5_flow_namespace *ns;
1091
1092         if (!root_ns)
1093                 return NULL;
1094
1095         switch (type) {
1096         case MLX5_FLOW_NAMESPACE_KERNEL:
1097                 prio = 0;
1098                 break;
1099         case MLX5_FLOW_NAMESPACE_FDB:
1100                 if (dev->priv.fdb_root_ns)
1101                         return &dev->priv.fdb_root_ns->ns;
1102                 else
1103                         return NULL;
1104         default:
1105                 return NULL;
1106         }
1107
1108         fs_prio = find_prio(&root_ns->ns, prio);
1109         if (!fs_prio)
1110                 return NULL;
1111
1112         ns = list_first_entry(&fs_prio->node.children,
1113                               typeof(*ns),
1114                               node.list);
1115
1116         return ns;
1117 }
1118
1119 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1120                                       unsigned prio, int max_ft,
1121                                       int start_level)
1122 {
1123         struct fs_prio *fs_prio;
1124
1125         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1126         if (!fs_prio)
1127                 return ERR_PTR(-ENOMEM);
1128
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);
1136
1137         return fs_prio;
1138 }
1139
1140 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1141                                                      *ns)
1142 {
1143         ns->node.type = FS_TYPE_NAMESPACE;
1144
1145         return ns;
1146 }
1147
1148 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1149 {
1150         struct mlx5_flow_namespace      *ns;
1151
1152         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1153         if (!ns)
1154                 return ERR_PTR(-ENOMEM);
1155
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);
1160
1161         return ns;
1162 }
1163
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,
1167                                     int index)
1168 {
1169         struct mlx5_flow_namespace *fs_ns;
1170         struct fs_prio *fs_prio;
1171         struct fs_node *base;
1172         int i;
1173         int err;
1174
1175         if (init_node->type == FS_TYPE_PRIO) {
1176                 if (init_node->min_ft_level > max_ft_level)
1177                         return -ENOTSUPP;
1178
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);
1188                 if (IS_ERR(fs_ns))
1189                         return PTR_ERR(fs_ns);
1190                 base = &fs_ns->node;
1191         } else {
1192                 return -EINVAL;
1193         }
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,
1197                                                init_node, i);
1198                 if (err)
1199                         return err;
1200         }
1201
1202         return 0;
1203 }
1204
1205 static int init_root_tree(int max_ft_level, struct init_tree_node *init_node,
1206                           struct fs_node *fs_parent_node)
1207 {
1208         int i;
1209         struct mlx5_flow_namespace *fs_ns;
1210         int err;
1211
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],
1216                                                &fs_ns->node,
1217                                                init_node, i);
1218                 if (err)
1219                         return err;
1220         }
1221         return 0;
1222 }
1223
1224 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_core_dev *dev,
1225                                                        enum fs_flow_table_type
1226                                                        table_type)
1227 {
1228         struct mlx5_flow_root_namespace *root_ns;
1229         struct mlx5_flow_namespace *ns;
1230
1231         /* Create the root namespace */
1232         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1233         if (!root_ns)
1234                 return NULL;
1235
1236         root_ns->dev = dev;
1237         root_ns->table_type = table_type;
1238
1239         ns = &root_ns->ns;
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);
1244
1245         return root_ns;
1246 }
1247
1248 static int init_root_ns(struct mlx5_core_dev *dev)
1249 {
1250         int max_ft_level = MLX5_CAP_FLOWTABLE(dev,
1251                                               flow_table_properties_nic_receive.
1252                                               max_ft_level);
1253
1254         dev->priv.root_ns = create_root_ns(dev, FS_FT_NIC_RX);
1255         if (IS_ERR_OR_NULL(dev->priv.root_ns))
1256                 goto cleanup;
1257
1258         if (init_root_tree(max_ft_level, &root_fs, &dev->priv.root_ns->ns.node))
1259                 goto cleanup;
1260
1261         return 0;
1262
1263 cleanup:
1264         mlx5_cleanup_fs(dev);
1265         return -ENOMEM;
1266 }
1267
1268 static void cleanup_single_prio_root_ns(struct mlx5_core_dev *dev,
1269                                         struct mlx5_flow_root_namespace *root_ns)
1270 {
1271         struct fs_node *prio;
1272
1273         if (!root_ns)
1274                 return;
1275
1276         if (!list_empty(&root_ns->ns.node.children)) {
1277                 prio = list_first_entry(&root_ns->ns.node.children,
1278                                         struct fs_node,
1279                                  list);
1280                 if (tree_remove_node(prio))
1281                         mlx5_core_warn(dev,
1282                                        "Flow steering priority wasn't destroyed, refcount > 1\n");
1283         }
1284         if (tree_remove_node(&root_ns->ns.node))
1285                 mlx5_core_warn(dev,
1286                                "Flow steering namespace wasn't destroyed, refcount > 1\n");
1287         root_ns = NULL;
1288 }
1289
1290 static void cleanup_root_ns(struct mlx5_core_dev *dev)
1291 {
1292         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1293         struct fs_prio *iter_prio;
1294
1295         if (!MLX5_CAP_GEN(dev, nic_flow_table))
1296                 return;
1297
1298         if (!root_ns)
1299                 return;
1300
1301         /* stage 1 */
1302         fs_for_each_prio(iter_prio, &root_ns->ns) {
1303                 struct fs_node *node;
1304                 struct mlx5_flow_namespace *iter_ns;
1305
1306                 fs_for_each_ns_or_ft(node, iter_prio) {
1307                         if (node->type == FS_TYPE_FLOW_TABLE)
1308                                 continue;
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,
1314                                                          struct fs_node,
1315                                                          list);
1316
1317                                 fs_get_obj(obj_iter_prio2, iter_prio2);
1318                                 if (tree_remove_node(iter_prio2)) {
1319                                         mlx5_core_warn(dev,
1320                                                        "Priority %d wasn't destroyed, refcount > 1\n",
1321                                                        obj_iter_prio2->prio);
1322                                         return;
1323                                 }
1324                         }
1325                 }
1326         }
1327
1328         /* stage 2 */
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,
1333                                                  struct fs_node,
1334                                                  list);
1335                         if (tree_remove_node(iter_ns)) {
1336                                 mlx5_core_warn(dev,
1337                                                "Namespace wasn't destroyed, refcount > 1\n");
1338                                 return;
1339                         }
1340                 }
1341         }
1342
1343         /* stage 3 */
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,
1348                                          struct fs_node,
1349                                          list);
1350
1351                 fs_get_obj(obj_prio_node, prio_node);
1352                 if (tree_remove_node(prio_node)) {
1353                         mlx5_core_warn(dev,
1354                                        "Priority %d wasn't destroyed, refcount > 1\n",
1355                                        obj_prio_node->prio);
1356                         return;
1357                 }
1358         }
1359
1360         if (tree_remove_node(&root_ns->ns.node)) {
1361                 mlx5_core_warn(dev,
1362                                "root namespace wasn't destroyed, refcount > 1\n");
1363                 return;
1364         }
1365
1366         dev->priv.root_ns = NULL;
1367 }
1368
1369 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1370 {
1371         cleanup_root_ns(dev);
1372         cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1373 }
1374
1375 static int init_fdb_root_ns(struct mlx5_core_dev *dev)
1376 {
1377         struct fs_prio *prio;
1378
1379         dev->priv.fdb_root_ns = create_root_ns(dev, FS_FT_FDB);
1380         if (!dev->priv.fdb_root_ns)
1381                 return -ENOMEM;
1382
1383         /* Create single prio */
1384         prio = fs_create_prio(&dev->priv.fdb_root_ns->ns, 0, 1, 0);
1385         if (IS_ERR(prio)) {
1386                 cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1387                 return PTR_ERR(prio);
1388         } else {
1389                 return 0;
1390         }
1391 }
1392
1393 int mlx5_init_fs(struct mlx5_core_dev *dev)
1394 {
1395         int err = 0;
1396
1397         if (MLX5_CAP_GEN(dev, nic_flow_table)) {
1398                 err = init_root_ns(dev);
1399                 if (err)
1400                         return err;
1401         }
1402         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1403                 err = init_fdb_root_ns(dev);
1404                 if (err)
1405                         cleanup_root_ns(dev);
1406         }
1407
1408         return err;
1409 }