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 ADD_PRIO(num_prios_val, min_level_val, max_ft_val, caps_val,\
  44                  ...) {.type = FS_TYPE_PRIO,\
  45         .min_ft_level = min_level_val,\
  46         .max_ft = max_ft_val,\
  47         .num_leaf_prios = num_prios_val,\
  48         .caps = caps_val,\
  49         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  50         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  51 }
  52
  53 #define ADD_MULTIPLE_PRIO(num_prios_val, max_ft_val, ...)\
  54         ADD_PRIO(num_prios_val, 0, max_ft_val, {},\
  55                  __VA_ARGS__)\
  56
  57 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
  58         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  59         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  60 }
  61
  62 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  63                                    sizeof(long))
  64
  65 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  66
  67 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  68                                .caps = (long[]) {__VA_ARGS__} }
  69
  70 #define LEFTOVERS_MAX_FT 1
  71 #define LEFTOVERS_NUM_PRIOS 1
  72 #define BY_PASS_PRIO_MAX_FT 1
  73 #define BY_PASS_MIN_LEVEL (KENREL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  74                            LEFTOVERS_MAX_FT)
  75
  76 #define KERNEL_MAX_FT 2
  77 #define KERNEL_NUM_PRIOS 1
  78 #define KENREL_MIN_LEVEL 2
  79
  80 struct node_caps {
  81         size_t  arr_sz;
  82         long    *caps;
  83 };
  84 static struct init_tree_node {
  85         enum fs_node_type       type;
  86         struct init_tree_node *children;
  87         int ar_size;
  88         struct node_caps caps;
  89         int min_ft_level;
  90         int num_leaf_prios;
  91         int prio;
  92         int max_ft;
  93 } root_fs = {
  94         .type = FS_TYPE_NAMESPACE,
  95         .ar_size = 3,
  96         .children = (struct init_tree_node[]) {
  97                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
  98                          FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en),
  99                                           FS_CAP(flow_table_properties_nic_receive.modify_root),
 100                                           FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode),
 101                                           FS_CAP(flow_table_properties_nic_receive.flow_table_modify)),
 102                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS, BY_PASS_PRIO_MAX_FT))),
 103                 ADD_PRIO(0, KENREL_MIN_LEVEL, 0, {},
 104                          ADD_NS(ADD_MULTIPLE_PRIO(KERNEL_NUM_PRIOS, KERNEL_MAX_FT))),
 105                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 106                          FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en),
 107                                           FS_CAP(flow_table_properties_nic_receive.modify_root),
 108                                           FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode),
 109                                           FS_CAP(flow_table_properties_nic_receive.flow_table_modify)),
 110                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_MAX_FT))),
 111         }
 112 };
 113
 114 enum fs_i_mutex_lock_class {
 115         FS_MUTEX_GRANDPARENT,
 116         FS_MUTEX_PARENT,
 117         FS_MUTEX_CHILD
 118 };
 119
 120 static void del_rule(struct fs_node *node);
 121 static void del_flow_table(struct fs_node *node);
 122 static void del_flow_group(struct fs_node *node);
 123 static void del_fte(struct fs_node *node);
 124
 125 static void tree_init_node(struct fs_node *node,
 126                            unsigned int refcount,
 127                            void (*remove_func)(struct fs_node *))
 128 {
 129         atomic_set(&node->refcount, refcount);
 130         INIT_LIST_HEAD(&node->list);
 131         INIT_LIST_HEAD(&node->children);
 132         mutex_init(&node->lock);
 133         node->remove_func = remove_func;
 134 }
 135
 136 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 137 {
 138         if (parent)
 139                 atomic_inc(&parent->refcount);
 140         node->parent = parent;
 141
 142         /* Parent is the root */
 143         if (!parent)
 144                 node->root = node;
 145         else
 146                 node->root = parent->root;
 147 }
 148
 149 static void tree_get_node(struct fs_node *node)
 150 {
 151         atomic_inc(&node->refcount);
 152 }
 153
 154 static void nested_lock_ref_node(struct fs_node *node,
 155                                  enum fs_i_mutex_lock_class class)
 156 {
 157         if (node) {
 158                 mutex_lock_nested(&node->lock, class);
 159                 atomic_inc(&node->refcount);
 160         }
 161 }
 162
 163 static void lock_ref_node(struct fs_node *node)
 164 {
 165         if (node) {
 166                 mutex_lock(&node->lock);
 167                 atomic_inc(&node->refcount);
 168         }
 169 }
 170
 171 static void unlock_ref_node(struct fs_node *node)
 172 {
 173         if (node) {
 174                 atomic_dec(&node->refcount);
 175                 mutex_unlock(&node->lock);
 176         }
 177 }
 178
 179 static void tree_put_node(struct fs_node *node)
 180 {
 181         struct fs_node *parent_node = node->parent;
 182
 183         lock_ref_node(parent_node);
 184         if (atomic_dec_and_test(&node->refcount)) {
 185                 if (parent_node)
 186                         list_del_init(&node->list);
 187                 if (node->remove_func)
 188                         node->remove_func(node);
 189                 kfree(node);
 190                 node = NULL;
 191         }
 192         unlock_ref_node(parent_node);
 193         if (!node && parent_node)
 194                 tree_put_node(parent_node);
 195 }
 196
 197 static int tree_remove_node(struct fs_node *node)
 198 {
 199         if (atomic_read(&node->refcount) > 1)
 200                 return -EPERM;
 201         tree_put_node(node);
 202         return 0;
 203 }
 204
 205 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 206                                  unsigned int prio)
 207 {
 208         struct fs_prio *iter_prio;
 209
 210         fs_for_each_prio(iter_prio, ns) {
 211                 if (iter_prio->prio == prio)
 212                         return iter_prio;
 213         }
 214
 215         return NULL;
 216 }
 217
 218 static unsigned int find_next_free_level(struct fs_prio *prio)
 219 {
 220         if (!list_empty(&prio->node.children)) {
 221                 struct mlx5_flow_table *ft;
 222
 223                 ft = list_last_entry(&prio->node.children,
 224                                      struct mlx5_flow_table,
 225                                      node.list);
 226                 return ft->level + 1;
 227         }
 228         return prio->start_level;
 229 }
 230
 231 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 232 {
 233         unsigned int i;
 234
 235         for (i = 0; i < size; i++, mask++, val1++, val2++)
 236                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 237                     ((*(u8 *)val2) & (*(u8 *)mask)))
 238                         return false;
 239
 240         return true;
 241 }
 242
 243 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 244                                 void *fte_param1, void *fte_param2)
 245 {
 246         if (mask->match_criteria_enable &
 247             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 248                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 249                                                 fte_param1, outer_headers);
 250                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 251                                                 fte_param2, outer_headers);
 252                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 253                                               mask->match_criteria, outer_headers);
 254
 255                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 256                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 257                         return false;
 258         }
 259
 260         if (mask->match_criteria_enable &
 261             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 262                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 263                                                 fte_param1, misc_parameters);
 264                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 265                                                 fte_param2, misc_parameters);
 266                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 267                                           mask->match_criteria, misc_parameters);
 268
 269                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 270                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 271                         return false;
 272         }
 273
 274         if (mask->match_criteria_enable &
 275             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 276                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 277                                                 fte_param1, inner_headers);
 278                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 279                                                 fte_param2, inner_headers);
 280                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 281                                           mask->match_criteria, inner_headers);
 282
 283                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 284                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 285                         return false;
 286         }
 287         return true;
 288 }
 289
 290 static bool compare_match_criteria(u8 match_criteria_enable1,
 291                                    u8 match_criteria_enable2,
 292                                    void *mask1, void *mask2)
 293 {
 294         return match_criteria_enable1 == match_criteria_enable2 &&
 295                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 296 }
 297
 298 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 299 {
 300         struct fs_node *root;
 301         struct mlx5_flow_namespace *ns;
 302
 303         root = node->root;
 304
 305         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 306                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 307                 return NULL;
 308         }
 309
 310         ns = container_of(root, struct mlx5_flow_namespace, node);
 311         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 312 }
 313
 314 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 315 {
 316         struct mlx5_flow_root_namespace *root = find_root(node);
 317
 318         if (root)
 319                 return root->dev;
 320         return NULL;
 321 }
 322
 323 static void del_flow_table(struct fs_node *node)
 324 {
 325         struct mlx5_flow_table *ft;
 326         struct mlx5_core_dev *dev;
 327         struct fs_prio *prio;
 328         int err;
 329
 330         fs_get_obj(ft, node);
 331         dev = get_dev(&ft->node);
 332
 333         err = mlx5_cmd_destroy_flow_table(dev, ft);
 334         if (err)
 335                 pr_warn("flow steering can't destroy ft\n");
 336         fs_get_obj(prio, ft->node.parent);
 337         prio->num_ft--;
 338 }
 339
 340 static void del_rule(struct fs_node *node)
 341 {
 342         struct mlx5_flow_rule *rule;
 343         struct mlx5_flow_table *ft;
 344         struct mlx5_flow_group *fg;
 345         struct fs_fte *fte;
 346         u32     *match_value;
 347         struct mlx5_core_dev *dev = get_dev(node);
 348         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 349         int err;
 350
 351         match_value = mlx5_vzalloc(match_len);
 352         if (!match_value) {
 353                 pr_warn("failed to allocate inbox\n");
 354                 return;
 355         }
 356
 357         fs_get_obj(rule, node);
 358         fs_get_obj(fte, rule->node.parent);
 359         fs_get_obj(fg, fte->node.parent);
 360         memcpy(match_value, fte->val, sizeof(fte->val));
 361         fs_get_obj(ft, fg->node.parent);
 362         list_del(&rule->node.list);
 363         fte->dests_size--;
 364         if (fte->dests_size) {
 365                 err = mlx5_cmd_update_fte(dev, ft,
 366                                           fg->id, fte);
 367                 if (err)
 368                         pr_warn("%s can't del rule fg id=%d fte_index=%d\n",
 369                                 __func__, fg->id, fte->index);
 370         }
 371         kvfree(match_value);
 372 }
 373
 374 static void del_fte(struct fs_node *node)
 375 {
 376         struct mlx5_flow_table *ft;
 377         struct mlx5_flow_group *fg;
 378         struct mlx5_core_dev *dev;
 379         struct fs_fte *fte;
 380         int err;
 381
 382         fs_get_obj(fte, node);
 383         fs_get_obj(fg, fte->node.parent);
 384         fs_get_obj(ft, fg->node.parent);
 385
 386         dev = get_dev(&ft->node);
 387         err = mlx5_cmd_delete_fte(dev, ft,
 388                                   fte->index);
 389         if (err)
 390                 pr_warn("flow steering can't delete fte in index %d of flow group id %d\n",
 391                         fte->index, fg->id);
 392
 393         fte->status = 0;
 394         fg->num_ftes--;
 395 }
 396
 397 static void del_flow_group(struct fs_node *node)
 398 {
 399         struct mlx5_flow_group *fg;
 400         struct mlx5_flow_table *ft;
 401         struct mlx5_core_dev *dev;
 402
 403         fs_get_obj(fg, node);
 404         fs_get_obj(ft, fg->node.parent);
 405         dev = get_dev(&ft->node);
 406
 407         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 408                 pr_warn("flow steering can't destroy fg %d of ft %d\n",
 409                         fg->id, ft->id);
 410 }
 411
 412 static struct fs_fte *alloc_fte(u8 action,
 413                                 u32 flow_tag,
 414                                 u32 *match_value,
 415                                 unsigned int index)
 416 {
 417         struct fs_fte *fte;
 418
 419         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 420         if (!fte)
 421                 return ERR_PTR(-ENOMEM);
 422
 423         memcpy(fte->val, match_value, sizeof(fte->val));
 424         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 425         fte->flow_tag = flow_tag;
 426         fte->index = index;
 427         fte->action = action;
 428
 429         return fte;
 430 }
 431
 432 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 433 {
 434         struct mlx5_flow_group *fg;
 435         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 436                                             create_fg_in, match_criteria);
 437         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 438                                             create_fg_in,
 439                                             match_criteria_enable);
 440         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 441         if (!fg)
 442                 return ERR_PTR(-ENOMEM);
 443
 444         fg->mask.match_criteria_enable = match_criteria_enable;
 445         memcpy(&fg->mask.match_criteria, match_criteria,
 446                sizeof(fg->mask.match_criteria));
 447         fg->node.type =  FS_TYPE_FLOW_GROUP;
 448         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 449                                    start_flow_index);
 450         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 451                                 end_flow_index) - fg->start_index + 1;
 452         return fg;
 453 }
 454
 455 static struct mlx5_flow_table *alloc_flow_table(int level, int max_fte,
 456                                                 enum fs_flow_table_type table_type)
 457 {
 458         struct mlx5_flow_table *ft;
 459
 460         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 461         if (!ft)
 462                 return NULL;
 463
 464         ft->level = level;
 465         ft->node.type = FS_TYPE_FLOW_TABLE;
 466         ft->type = table_type;
 467         ft->max_fte = max_fte;
 468
 469         return ft;
 470 }
 471
 472 /* If reverse is false, then we search for the first flow table in the
 473  * root sub-tree from start(closest from right), else we search for the
 474  * last flow table in the root sub-tree till start(closest from left).
 475  */
 476 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 477                                                          struct list_head *start,
 478                                                          bool reverse)
 479 {
 480 #define list_advance_entry(pos, reverse)                \
 481         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 482
 483 #define list_for_each_advance_continue(pos, head, reverse)      \
 484         for (pos = list_advance_entry(pos, reverse);            \
 485              &pos->list != (head);                              \
 486              pos = list_advance_entry(pos, reverse))
 487
 488         struct fs_node *iter = list_entry(start, struct fs_node, list);
 489         struct mlx5_flow_table *ft = NULL;
 490
 491         if (!root)
 492                 return NULL;
 493
 494         list_for_each_advance_continue(iter, &root->children, reverse) {
 495                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 496                         fs_get_obj(ft, iter);
 497                         return ft;
 498                 }
 499                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 500                 if (ft)
 501                         return ft;
 502         }
 503
 504         return ft;
 505 }
 506
 507 /* If reverse if false then return the first flow table in next priority of
 508  * prio in the tree, else return the last flow table in the previous priority
 509  * of prio in the tree.
 510  */
 511 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 512 {
 513         struct mlx5_flow_table *ft = NULL;
 514         struct fs_node *curr_node;
 515         struct fs_node *parent;
 516
 517         parent = prio->node.parent;
 518         curr_node = &prio->node;
 519         while (!ft && parent) {
 520                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 521                 curr_node = parent;
 522                 parent = curr_node->parent;
 523         }
 524         return ft;
 525 }
 526
 527 /* Assuming all the tree is locked by mutex chain lock */
 528 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 529 {
 530         return find_closest_ft(prio, false);
 531 }
 532
 533 /* Assuming all the tree is locked by mutex chain lock */
 534 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 535 {
 536         return find_closest_ft(prio, true);
 537 }
 538
 539 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 540                                struct fs_prio *prio,
 541                                struct mlx5_flow_table *ft)
 542 {
 543         struct mlx5_flow_table *iter;
 544         int i = 0;
 545         int err;
 546
 547         fs_for_each_ft(iter, prio) {
 548                 i++;
 549                 err = mlx5_cmd_modify_flow_table(dev,
 550                                                  iter,
 551                                                  ft);
 552                 if (err) {
 553                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 554                                        iter->id);
 555                         /* The driver is out of sync with the FW */
 556                         if (i > 1)
 557                                 WARN_ON(true);
 558                         return err;
 559                 }
 560         }
 561         return 0;
 562 }
 563
 564 /* Connect flow tables from previous priority of prio to ft */
 565 static int connect_prev_fts(struct mlx5_core_dev *dev,
 566                             struct mlx5_flow_table *ft,
 567                             struct fs_prio *prio)
 568 {
 569         struct mlx5_flow_table *prev_ft;
 570
 571         prev_ft = find_prev_chained_ft(prio);
 572         if (prev_ft) {
 573                 struct fs_prio *prev_prio;
 574
 575                 fs_get_obj(prev_prio, prev_ft->node.parent);
 576                 return connect_fts_in_prio(dev, prev_prio, ft);
 577         }
 578         return 0;
 579 }
 580
 581 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 582                                  *prio)
 583 {
 584         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 585         int min_level = INT_MAX;
 586         int err;
 587
 588         if (root->root_ft)
 589                 min_level = root->root_ft->level;
 590
 591         if (ft->level >= min_level)
 592                 return 0;
 593
 594         err = mlx5_cmd_update_root_ft(root->dev, ft);
 595         if (err)
 596                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 597                                ft->id);
 598         else
 599                 root->root_ft = ft;
 600
 601         return err;
 602 }
 603
 604 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 605                               struct fs_prio *prio)
 606 {
 607         int err = 0;
 608
 609         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 610
 611         if (list_empty(&prio->node.children)) {
 612                 err = connect_prev_fts(dev, ft, prio);
 613                 if (err)
 614                         return err;
 615         }
 616
 617         if (MLX5_CAP_FLOWTABLE(dev,
 618                                flow_table_properties_nic_receive.modify_root))
 619                 err = update_root_ft_create(ft, prio);
 620         return err;
 621 }
 622
 623 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 624                                                int prio,
 625                                                int max_fte)
 626 {
 627         struct mlx5_flow_table *next_ft = NULL;
 628         struct mlx5_flow_table *ft;
 629         int err;
 630         int log_table_sz;
 631         struct mlx5_flow_root_namespace *root =
 632                 find_root(&ns->node);
 633         struct fs_prio *fs_prio = NULL;
 634
 635         if (!root) {
 636                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 637                 return ERR_PTR(-ENODEV);
 638         }
 639
 640         mutex_lock(&root->chain_lock);
 641         fs_prio = find_prio(ns, prio);
 642         if (!fs_prio) {
 643                 err = -EINVAL;
 644                 goto unlock_root;
 645         }
 646         if (fs_prio->num_ft == fs_prio->max_ft) {
 647                 err = -ENOSPC;
 648                 goto unlock_root;
 649         }
 650
 651         ft = alloc_flow_table(find_next_free_level(fs_prio),
 652                               roundup_pow_of_two(max_fte),
 653                               root->table_type);
 654         if (!ft) {
 655                 err = -ENOMEM;
 656                 goto unlock_root;
 657         }
 658
 659         tree_init_node(&ft->node, 1, del_flow_table);
 660         log_table_sz = ilog2(ft->max_fte);
 661         next_ft = find_next_chained_ft(fs_prio);
 662         err = mlx5_cmd_create_flow_table(root->dev, ft->type, ft->level,
 663                                          log_table_sz, next_ft, &ft->id);
 664         if (err)
 665                 goto free_ft;
 666
 667         err = connect_flow_table(root->dev, ft, fs_prio);
 668         if (err)
 669                 goto destroy_ft;
 670         lock_ref_node(&fs_prio->node);
 671         tree_add_node(&ft->node, &fs_prio->node);
 672         list_add_tail(&ft->node.list, &fs_prio->node.children);
 673         fs_prio->num_ft++;
 674         unlock_ref_node(&fs_prio->node);
 675         mutex_unlock(&root->chain_lock);
 676         return ft;
 677 destroy_ft:
 678         mlx5_cmd_destroy_flow_table(root->dev, ft);
 679 free_ft:
 680         kfree(ft);
 681 unlock_root:
 682         mutex_unlock(&root->chain_lock);
 683         return ERR_PTR(err);
 684 }
 685
 686 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 687                                                             int prio,
 688                                                             int num_flow_table_entries,
 689                                                             int max_num_groups)
 690 {
 691         struct mlx5_flow_table *ft;
 692
 693         if (max_num_groups > num_flow_table_entries)
 694                 return ERR_PTR(-EINVAL);
 695
 696         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries);
 697         if (IS_ERR(ft))
 698                 return ft;
 699
 700         ft->autogroup.active = true;
 701         ft->autogroup.required_groups = max_num_groups;
 702
 703         return ft;
 704 }
 705 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 706
 707 /* Flow table should be locked */
 708 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 709                                                         u32 *fg_in,
 710                                                         struct list_head
 711                                                         *prev_fg,
 712                                                         bool is_auto_fg)
 713 {
 714         struct mlx5_flow_group *fg;
 715         struct mlx5_core_dev *dev = get_dev(&ft->node);
 716         int err;
 717
 718         if (!dev)
 719                 return ERR_PTR(-ENODEV);
 720
 721         fg = alloc_flow_group(fg_in);
 722         if (IS_ERR(fg))
 723                 return fg;
 724
 725         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 726         if (err) {
 727                 kfree(fg);
 728                 return ERR_PTR(err);
 729         }
 730
 731         if (ft->autogroup.active)
 732                 ft->autogroup.num_groups++;
 733         /* Add node to tree */
 734         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 735         tree_add_node(&fg->node, &ft->node);
 736         /* Add node to group list */
 737         list_add(&fg->node.list, ft->node.children.prev);
 738
 739         return fg;
 740 }
 741
 742 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 743                                                u32 *fg_in)
 744 {
 745         struct mlx5_flow_group *fg;
 746
 747         if (ft->autogroup.active)
 748                 return ERR_PTR(-EPERM);
 749
 750         lock_ref_node(&ft->node);
 751         fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
 752         unlock_ref_node(&ft->node);
 753
 754         return fg;
 755 }
 756
 757 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 758 {
 759         struct mlx5_flow_rule *rule;
 760
 761         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 762         if (!rule)
 763                 return NULL;
 764
 765         rule->node.type = FS_TYPE_FLOW_DEST;
 766         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 767
 768         return rule;
 769 }
 770
 771 /* fte should not be deleted while calling this function */
 772 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
 773                                            struct mlx5_flow_group *fg,
 774                                            struct mlx5_flow_destination *dest)
 775 {
 776         struct mlx5_flow_table *ft;
 777         struct mlx5_flow_rule *rule;
 778         int err;
 779
 780         rule = alloc_rule(dest);
 781         if (!rule)
 782                 return ERR_PTR(-ENOMEM);
 783
 784         fs_get_obj(ft, fg->node.parent);
 785         /* Add dest to dests list- added as first element after the head */
 786         tree_init_node(&rule->node, 1, del_rule);
 787         list_add_tail(&rule->node.list, &fte->node.children);
 788         fte->dests_size++;
 789         if (fte->dests_size == 1)
 790                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 791                                           ft, fg->id, fte);
 792         else
 793                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 794                                           ft, fg->id, fte);
 795         if (err)
 796                 goto free_rule;
 797
 798         fte->status |= FS_FTE_STATUS_EXISTING;
 799
 800         return rule;
 801
 802 free_rule:
 803         list_del(&rule->node.list);
 804         kfree(rule);
 805         fte->dests_size--;
 806         return ERR_PTR(err);
 807 }
 808
 809 /* Assumed fg is locked */
 810 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 811                                        struct list_head **prev)
 812 {
 813         struct fs_fte *fte;
 814         unsigned int start = fg->start_index;
 815
 816         if (prev)
 817                 *prev = &fg->node.children;
 818
 819         /* assumed list is sorted by index */
 820         fs_for_each_fte(fte, fg) {
 821                 if (fte->index != start)
 822                         return start;
 823                 start++;
 824                 if (prev)
 825                         *prev = &fte->node.list;
 826         }
 827
 828         return start;
 829 }
 830
 831 /* prev is output, prev->next = new_fte */
 832 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
 833                                  u32 *match_value,
 834                                  u8 action,
 835                                  u32 flow_tag,
 836                                  struct list_head **prev)
 837 {
 838         struct fs_fte *fte;
 839         int index;
 840
 841         index = get_free_fte_index(fg, prev);
 842         fte = alloc_fte(action, flow_tag, match_value, index);
 843         if (IS_ERR(fte))
 844                 return fte;
 845
 846         return fte;
 847 }
 848
 849 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
 850                                                 u8 match_criteria_enable,
 851                                                 u32 *match_criteria)
 852 {
 853         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 854         struct list_head *prev = &ft->node.children;
 855         unsigned int candidate_index = 0;
 856         struct mlx5_flow_group *fg;
 857         void *match_criteria_addr;
 858         unsigned int group_size = 0;
 859         u32 *in;
 860
 861         if (!ft->autogroup.active)
 862                 return ERR_PTR(-ENOENT);
 863
 864         in = mlx5_vzalloc(inlen);
 865         if (!in)
 866                 return ERR_PTR(-ENOMEM);
 867
 868         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
 869                 /* We save place for flow groups in addition to max types */
 870                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
 871
 872         /*  ft->max_fte == ft->autogroup.max_types */
 873         if (group_size == 0)
 874                 group_size = 1;
 875
 876         /* sorted by start_index */
 877         fs_for_each_fg(fg, ft) {
 878                 if (candidate_index + group_size > fg->start_index)
 879                         candidate_index = fg->start_index + fg->max_ftes;
 880                 else
 881                         break;
 882                 prev = &fg->node.list;
 883         }
 884
 885         if (candidate_index + group_size > ft->max_fte) {
 886                 fg = ERR_PTR(-ENOSPC);
 887                 goto out;
 888         }
 889
 890         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
 891                  match_criteria_enable);
 892         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
 893         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
 894                  group_size - 1);
 895         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
 896                                            in, match_criteria);
 897         memcpy(match_criteria_addr, match_criteria,
 898                MLX5_ST_SZ_BYTES(fte_match_param));
 899
 900         fg = create_flow_group_common(ft, in, prev, true);
 901 out:
 902         kvfree(in);
 903         return fg;
 904 }
 905
 906 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
 907                                           u32 *match_value,
 908                                           u8 action,
 909                                           u32 flow_tag,
 910                                           struct mlx5_flow_destination *dest)
 911 {
 912         struct fs_fte *fte;
 913         struct mlx5_flow_rule *rule;
 914         struct mlx5_flow_table *ft;
 915         struct list_head *prev;
 916
 917         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
 918         fs_for_each_fte(fte, fg) {
 919                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
 920                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
 921                     action == fte->action && flow_tag == fte->flow_tag) {
 922                         rule = add_rule_fte(fte, fg, dest);
 923                         unlock_ref_node(&fte->node);
 924                         if (IS_ERR(rule))
 925                                 goto unlock_fg;
 926                         else
 927                                 goto add_rule;
 928                 }
 929                 unlock_ref_node(&fte->node);
 930         }
 931         fs_get_obj(ft, fg->node.parent);
 932         if (fg->num_ftes >= fg->max_ftes) {
 933                 rule = ERR_PTR(-ENOSPC);
 934                 goto unlock_fg;
 935         }
 936
 937         fte = create_fte(fg, match_value, action, flow_tag, &prev);
 938         if (IS_ERR(fte)) {
 939                 rule = (void *)fte;
 940                 goto unlock_fg;
 941         }
 942         tree_init_node(&fte->node, 0, del_fte);
 943         rule = add_rule_fte(fte, fg, dest);
 944         if (IS_ERR(rule)) {
 945                 kfree(fte);
 946                 goto unlock_fg;
 947         }
 948
 949         fg->num_ftes++;
 950
 951         tree_add_node(&fte->node, &fg->node);
 952         list_add(&fte->node.list, prev);
 953 add_rule:
 954         tree_add_node(&rule->node, &fte->node);
 955 unlock_fg:
 956         unlock_ref_node(&fg->node);
 957         return rule;
 958 }
 959
 960 static struct mlx5_flow_rule *add_rule_to_auto_fg(struct mlx5_flow_table *ft,
 961                                                   u8 match_criteria_enable,
 962                                                   u32 *match_criteria,
 963                                                   u32 *match_value,
 964                                                   u8 action,
 965                                                   u32 flow_tag,
 966                                                   struct mlx5_flow_destination *dest)
 967 {
 968         struct mlx5_flow_rule *rule;
 969         struct mlx5_flow_group *g;
 970
 971         g = create_autogroup(ft, match_criteria_enable, match_criteria);
 972         if (IS_ERR(g))
 973                 return (void *)g;
 974
 975         rule = add_rule_fg(g, match_value,
 976                            action, flow_tag, dest);
 977         if (IS_ERR(rule)) {
 978                 /* Remove assumes refcount > 0 and autogroup creates a group
 979                  * with a refcount = 0.
 980                  */
 981                 tree_get_node(&g->node);
 982                 tree_remove_node(&g->node);
 983         }
 984         return rule;
 985 }
 986
 987 struct mlx5_flow_rule *
 988 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
 989                    u8 match_criteria_enable,
 990                    u32 *match_criteria,
 991                    u32 *match_value,
 992                    u32 action,
 993                    u32 flow_tag,
 994                    struct mlx5_flow_destination *dest)
 995 {
 996         struct mlx5_flow_group *g;
 997         struct mlx5_flow_rule *rule;
 998
 999         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1000         fs_for_each_fg(g, ft)
1001                 if (compare_match_criteria(g->mask.match_criteria_enable,
1002                                            match_criteria_enable,
1003                                            g->mask.match_criteria,
1004                                            match_criteria)) {
1005                         rule = add_rule_fg(g, match_value,
1006                                            action, flow_tag, dest);
1007                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1008                                 goto unlock;
1009                 }
1010
1011         rule = add_rule_to_auto_fg(ft, match_criteria_enable, match_criteria,
1012                                    match_value, action, flow_tag, dest);
1013 unlock:
1014         unlock_ref_node(&ft->node);
1015         return rule;
1016 }
1017 EXPORT_SYMBOL(mlx5_add_flow_rule);
1018
1019 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
1020 {
1021         tree_remove_node(&rule->node);
1022 }
1023 EXPORT_SYMBOL(mlx5_del_flow_rule);
1024
1025 /* Assuming prio->node.children(flow tables) is sorted by level */
1026 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1027 {
1028         struct fs_prio *prio;
1029
1030         fs_get_obj(prio, ft->node.parent);
1031
1032         if (!list_is_last(&ft->node.list, &prio->node.children))
1033                 return list_next_entry(ft, node.list);
1034         return find_next_chained_ft(prio);
1035 }
1036
1037 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1038 {
1039         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1040         struct mlx5_flow_table *new_root_ft = NULL;
1041
1042         if (root->root_ft != ft)
1043                 return 0;
1044
1045         new_root_ft = find_next_ft(ft);
1046         if (new_root_ft) {
1047                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1048
1049                 if (err) {
1050                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1051                                        ft->id);
1052                         return err;
1053                 }
1054                 root->root_ft = new_root_ft;
1055         }
1056         return 0;
1057 }
1058
1059 /* Connect flow table from previous priority to
1060  * the next flow table.
1061  */
1062 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1063 {
1064         struct mlx5_core_dev *dev = get_dev(&ft->node);
1065         struct mlx5_flow_table *next_ft;
1066         struct fs_prio *prio;
1067         int err = 0;
1068
1069         err = update_root_ft_destroy(ft);
1070         if (err)
1071                 return err;
1072
1073         fs_get_obj(prio, ft->node.parent);
1074         if  (!(list_first_entry(&prio->node.children,
1075                                 struct mlx5_flow_table,
1076                                 node.list) == ft))
1077                 return 0;
1078
1079         next_ft = find_next_chained_ft(prio);
1080         err = connect_prev_fts(dev, next_ft, prio);
1081         if (err)
1082                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1083                                ft->id);
1084         return err;
1085 }
1086
1087 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1088 {
1089         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1090         int err = 0;
1091
1092         mutex_lock(&root->chain_lock);
1093         err = disconnect_flow_table(ft);
1094         if (err) {
1095                 mutex_unlock(&root->chain_lock);
1096                 return err;
1097         }
1098         if (tree_remove_node(&ft->node))
1099                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1100                                ft->id);
1101         mutex_unlock(&root->chain_lock);
1102
1103         return err;
1104 }
1105 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1106
1107 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1108 {
1109         if (tree_remove_node(&fg->node))
1110                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1111                                fg->id);
1112 }
1113
1114 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1115                                                     enum mlx5_flow_namespace_type type)
1116 {
1117         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1118         int prio;
1119         static struct fs_prio *fs_prio;
1120         struct mlx5_flow_namespace *ns;
1121
1122         if (!root_ns)
1123                 return NULL;
1124
1125         switch (type) {
1126         case MLX5_FLOW_NAMESPACE_BYPASS:
1127         case MLX5_FLOW_NAMESPACE_KERNEL:
1128         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1129                 prio = type;
1130                 break;
1131         case MLX5_FLOW_NAMESPACE_FDB:
1132                 if (dev->priv.fdb_root_ns)
1133                         return &dev->priv.fdb_root_ns->ns;
1134                 else
1135                         return NULL;
1136         default:
1137                 return NULL;
1138         }
1139
1140         fs_prio = find_prio(&root_ns->ns, prio);
1141         if (!fs_prio)
1142                 return NULL;
1143
1144         ns = list_first_entry(&fs_prio->node.children,
1145                               typeof(*ns),
1146                               node.list);
1147
1148         return ns;
1149 }
1150 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1151
1152 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1153                                       unsigned prio, int max_ft)
1154 {
1155         struct fs_prio *fs_prio;
1156
1157         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1158         if (!fs_prio)
1159                 return ERR_PTR(-ENOMEM);
1160
1161         fs_prio->node.type = FS_TYPE_PRIO;
1162         tree_init_node(&fs_prio->node, 1, NULL);
1163         tree_add_node(&fs_prio->node, &ns->node);
1164         fs_prio->max_ft = max_ft;
1165         fs_prio->prio = prio;
1166         list_add_tail(&fs_prio->node.list, &ns->node.children);
1167
1168         return fs_prio;
1169 }
1170
1171 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1172                                                      *ns)
1173 {
1174         ns->node.type = FS_TYPE_NAMESPACE;
1175
1176         return ns;
1177 }
1178
1179 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1180 {
1181         struct mlx5_flow_namespace      *ns;
1182
1183         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1184         if (!ns)
1185                 return ERR_PTR(-ENOMEM);
1186
1187         fs_init_namespace(ns);
1188         tree_init_node(&ns->node, 1, NULL);
1189         tree_add_node(&ns->node, &prio->node);
1190         list_add_tail(&ns->node.list, &prio->node.children);
1191
1192         return ns;
1193 }
1194
1195 static int create_leaf_prios(struct mlx5_flow_namespace *ns, struct init_tree_node
1196                              *prio_metadata)
1197 {
1198         struct fs_prio *fs_prio;
1199         int i;
1200
1201         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1202                 fs_prio = fs_create_prio(ns, i, prio_metadata->max_ft);
1203                 if (IS_ERR(fs_prio))
1204                         return PTR_ERR(fs_prio);
1205         }
1206         return 0;
1207 }
1208
1209 #define FLOW_TABLE_BIT_SZ 1
1210 #define GET_FLOW_TABLE_CAP(dev, offset) \
1211         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1212                         offset / 32)) >>                                        \
1213           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1214 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1215 {
1216         int i;
1217
1218         for (i = 0; i < caps->arr_sz; i++) {
1219                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1220                         return false;
1221         }
1222         return true;
1223 }
1224
1225 static int init_root_tree_recursive(struct mlx5_core_dev *dev,
1226                                     struct init_tree_node *init_node,
1227                                     struct fs_node *fs_parent_node,
1228                                     struct init_tree_node *init_parent_node,
1229                                     int index)
1230 {
1231         int max_ft_level = MLX5_CAP_FLOWTABLE(dev,
1232                                               flow_table_properties_nic_receive.
1233                                               max_ft_level);
1234         struct mlx5_flow_namespace *fs_ns;
1235         struct fs_prio *fs_prio;
1236         struct fs_node *base;
1237         int i;
1238         int err;
1239
1240         if (init_node->type == FS_TYPE_PRIO) {
1241                 if ((init_node->min_ft_level > max_ft_level) ||
1242                     !has_required_caps(dev, &init_node->caps))
1243                         return 0;
1244
1245                 fs_get_obj(fs_ns, fs_parent_node);
1246                 if (init_node->num_leaf_prios)
1247                         return create_leaf_prios(fs_ns, init_node);
1248                 fs_prio = fs_create_prio(fs_ns, index, init_node->max_ft);
1249                 if (IS_ERR(fs_prio))
1250                         return PTR_ERR(fs_prio);
1251                 base = &fs_prio->node;
1252         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1253                 fs_get_obj(fs_prio, fs_parent_node);
1254                 fs_ns = fs_create_namespace(fs_prio);
1255                 if (IS_ERR(fs_ns))
1256                         return PTR_ERR(fs_ns);
1257                 base = &fs_ns->node;
1258         } else {
1259                 return -EINVAL;
1260         }
1261         for (i = 0; i < init_node->ar_size; i++) {
1262                 err = init_root_tree_recursive(dev, &init_node->children[i],
1263                                                base, init_node, i);
1264                 if (err)
1265                         return err;
1266         }
1267
1268         return 0;
1269 }
1270
1271 static int init_root_tree(struct mlx5_core_dev *dev,
1272                           struct init_tree_node *init_node,
1273                           struct fs_node *fs_parent_node)
1274 {
1275         int i;
1276         struct mlx5_flow_namespace *fs_ns;
1277         int err;
1278
1279         fs_get_obj(fs_ns, fs_parent_node);
1280         for (i = 0; i < init_node->ar_size; i++) {
1281                 err = init_root_tree_recursive(dev, &init_node->children[i],
1282                                                &fs_ns->node,
1283                                                init_node, i);
1284                 if (err)
1285                         return err;
1286         }
1287         return 0;
1288 }
1289
1290 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_core_dev *dev,
1291                                                        enum fs_flow_table_type
1292                                                        table_type)
1293 {
1294         struct mlx5_flow_root_namespace *root_ns;
1295         struct mlx5_flow_namespace *ns;
1296
1297         /* Create the root namespace */
1298         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1299         if (!root_ns)
1300                 return NULL;
1301
1302         root_ns->dev = dev;
1303         root_ns->table_type = table_type;
1304
1305         ns = &root_ns->ns;
1306         fs_init_namespace(ns);
1307         mutex_init(&root_ns->chain_lock);
1308         tree_init_node(&ns->node, 1, NULL);
1309         tree_add_node(&ns->node, NULL);
1310
1311         return root_ns;
1312 }
1313
1314 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1315
1316 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1317 {
1318         struct fs_prio *prio;
1319
1320         fs_for_each_prio(prio, ns) {
1321                  /* This updates prio start_level and max_ft */
1322                 set_prio_attrs_in_prio(prio, acc_level);
1323                 acc_level += prio->max_ft;
1324         }
1325         return acc_level;
1326 }
1327
1328 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1329 {
1330         struct mlx5_flow_namespace *ns;
1331         int acc_level_ns = acc_level;
1332
1333         prio->start_level = acc_level;
1334         fs_for_each_ns(ns, prio)
1335                 /* This updates start_level and max_ft of ns's priority descendants */
1336                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1337         if (!prio->max_ft)
1338                 prio->max_ft = acc_level_ns - prio->start_level;
1339         WARN_ON(prio->max_ft < acc_level_ns - prio->start_level);
1340 }
1341
1342 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1343 {
1344         struct mlx5_flow_namespace *ns = &root_ns->ns;
1345         struct fs_prio *prio;
1346         int start_level = 0;
1347
1348         fs_for_each_prio(prio, ns) {
1349                 set_prio_attrs_in_prio(prio, start_level);
1350                 start_level += prio->max_ft;
1351         }
1352 }
1353
1354 static int init_root_ns(struct mlx5_core_dev *dev)
1355 {
1356
1357         dev->priv.root_ns = create_root_ns(dev, FS_FT_NIC_RX);
1358         if (IS_ERR_OR_NULL(dev->priv.root_ns))
1359                 goto cleanup;
1360
1361         if (init_root_tree(dev, &root_fs, &dev->priv.root_ns->ns.node))
1362                 goto cleanup;
1363
1364         set_prio_attrs(dev->priv.root_ns);
1365
1366         return 0;
1367
1368 cleanup:
1369         mlx5_cleanup_fs(dev);
1370         return -ENOMEM;
1371 }
1372
1373 static void cleanup_single_prio_root_ns(struct mlx5_core_dev *dev,
1374                                         struct mlx5_flow_root_namespace *root_ns)
1375 {
1376         struct fs_node *prio;
1377
1378         if (!root_ns)
1379                 return;
1380
1381         if (!list_empty(&root_ns->ns.node.children)) {
1382                 prio = list_first_entry(&root_ns->ns.node.children,
1383                                         struct fs_node,
1384                                  list);
1385                 if (tree_remove_node(prio))
1386                         mlx5_core_warn(dev,
1387                                        "Flow steering priority wasn't destroyed, refcount > 1\n");
1388         }
1389         if (tree_remove_node(&root_ns->ns.node))
1390                 mlx5_core_warn(dev,
1391                                "Flow steering namespace wasn't destroyed, refcount > 1\n");
1392         root_ns = NULL;
1393 }
1394
1395 static void cleanup_root_ns(struct mlx5_core_dev *dev)
1396 {
1397         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1398         struct fs_prio *iter_prio;
1399
1400         if (!MLX5_CAP_GEN(dev, nic_flow_table))
1401                 return;
1402
1403         if (!root_ns)
1404                 return;
1405
1406         /* stage 1 */
1407         fs_for_each_prio(iter_prio, &root_ns->ns) {
1408                 struct fs_node *node;
1409                 struct mlx5_flow_namespace *iter_ns;
1410
1411                 fs_for_each_ns_or_ft(node, iter_prio) {
1412                         if (node->type == FS_TYPE_FLOW_TABLE)
1413                                 continue;
1414                         fs_get_obj(iter_ns, node);
1415                         while (!list_empty(&iter_ns->node.children)) {
1416                                 struct fs_prio *obj_iter_prio2;
1417                                 struct fs_node *iter_prio2 =
1418                                         list_first_entry(&iter_ns->node.children,
1419                                                          struct fs_node,
1420                                                          list);
1421
1422                                 fs_get_obj(obj_iter_prio2, iter_prio2);
1423                                 if (tree_remove_node(iter_prio2)) {
1424                                         mlx5_core_warn(dev,
1425                                                        "Priority %d wasn't destroyed, refcount > 1\n",
1426                                                        obj_iter_prio2->prio);
1427                                         return;
1428                                 }
1429                         }
1430                 }
1431         }
1432
1433         /* stage 2 */
1434         fs_for_each_prio(iter_prio, &root_ns->ns) {
1435                 while (!list_empty(&iter_prio->node.children)) {
1436                         struct fs_node *iter_ns =
1437                                 list_first_entry(&iter_prio->node.children,
1438                                                  struct fs_node,
1439                                                  list);
1440                         if (tree_remove_node(iter_ns)) {
1441                                 mlx5_core_warn(dev,
1442                                                "Namespace wasn't destroyed, refcount > 1\n");
1443                                 return;
1444                         }
1445                 }
1446         }
1447
1448         /* stage 3 */
1449         while (!list_empty(&root_ns->ns.node.children)) {
1450                 struct fs_prio *obj_prio_node;
1451                 struct fs_node *prio_node =
1452                         list_first_entry(&root_ns->ns.node.children,
1453                                          struct fs_node,
1454                                          list);
1455
1456                 fs_get_obj(obj_prio_node, prio_node);
1457                 if (tree_remove_node(prio_node)) {
1458                         mlx5_core_warn(dev,
1459                                        "Priority %d wasn't destroyed, refcount > 1\n",
1460                                        obj_prio_node->prio);
1461                         return;
1462                 }
1463         }
1464
1465         if (tree_remove_node(&root_ns->ns.node)) {
1466                 mlx5_core_warn(dev,
1467                                "root namespace wasn't destroyed, refcount > 1\n");
1468                 return;
1469         }
1470
1471         dev->priv.root_ns = NULL;
1472 }
1473
1474 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1475 {
1476         cleanup_root_ns(dev);
1477         cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1478 }
1479
1480 static int init_fdb_root_ns(struct mlx5_core_dev *dev)
1481 {
1482         struct fs_prio *prio;
1483
1484         dev->priv.fdb_root_ns = create_root_ns(dev, FS_FT_FDB);
1485         if (!dev->priv.fdb_root_ns)
1486                 return -ENOMEM;
1487
1488         /* Create single prio */
1489         prio = fs_create_prio(&dev->priv.fdb_root_ns->ns, 0, 1);
1490         if (IS_ERR(prio)) {
1491                 cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1492                 return PTR_ERR(prio);
1493         } else {
1494                 return 0;
1495         }
1496 }
1497
1498 int mlx5_init_fs(struct mlx5_core_dev *dev)
1499 {
1500         int err = 0;
1501
1502         if (MLX5_CAP_GEN(dev, nic_flow_table)) {
1503                 err = init_root_ns(dev);
1504                 if (err)
1505                         return err;
1506         }
1507         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1508                 err = init_fdb_root_ns(dev);
1509                 if (err)
1510                         cleanup_root_ns(dev);
1511         }
1512
1513         return err;
1514 }