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, num_levels_val, caps_val,\
  44                  ...) {.type = FS_TYPE_PRIO,\
  45         .min_ft_level = min_level_val,\
  46         .num_levels = num_levels_val,\
  47         .num_leaf_prios = num_prios_val,\
  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, num_levels_val, ...)\
  54         ADD_PRIO(num_prios_val, 0, num_levels_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 FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  71                                            FS_CAP(flow_table_properties_nic_receive.modify_root), \
  72                                            FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  73                                            FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  74
  75 #define LEFTOVERS_NUM_LEVELS 1
  76 #define LEFTOVERS_NUM_PRIOS 1
  77
  78 #define BY_PASS_PRIO_NUM_LEVELS 1
  79 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  80                            LEFTOVERS_NUM_PRIOS)
  81
  82 #define ETHTOOL_PRIO_NUM_LEVELS 1
  83 #define ETHTOOL_NUM_PRIOS 11
  84 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
  85 /* Vlan, mac, ttc, aRFS */
  86 #define KERNEL_NIC_PRIO_NUM_LEVELS 4
  87 #define KERNEL_NIC_NUM_PRIOS 1
  88 /* One more level for tc */
  89 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
  90
  91 #define ANCHOR_NUM_LEVELS 1
  92 #define ANCHOR_NUM_PRIOS 1
  93 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
  94
  95 #define OFFLOADS_MAX_FT 1
  96 #define OFFLOADS_NUM_PRIOS 1
  97 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
  98
  99 #define LAG_PRIO_NUM_LEVELS 1
 100 #define LAG_NUM_PRIOS 1
 101 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 102
 103 struct node_caps {
 104         size_t  arr_sz;
 105         long    *caps;
 106 };
 107 static struct init_tree_node {
 108         enum fs_node_type       type;
 109         struct init_tree_node *children;
 110         int ar_size;
 111         struct node_caps caps;
 112         int min_ft_level;
 113         int num_leaf_prios;
 114         int prio;
 115         int num_levels;
 116 } root_fs = {
 117         .type = FS_TYPE_NAMESPACE,
 118         .ar_size = 7,
 119         .children = (struct init_tree_node[]) {
 120                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 121                          FS_CHAINING_CAPS,
 122                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 123                                                   BY_PASS_PRIO_NUM_LEVELS))),
 124                 ADD_PRIO(0, LAG_MIN_LEVEL, 0,
 125                          FS_CHAINING_CAPS,
 126                          ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 127                                                   LAG_PRIO_NUM_LEVELS))),
 128                 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
 129                          ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
 130                 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
 131                          FS_CHAINING_CAPS,
 132                          ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 133                                                   ETHTOOL_PRIO_NUM_LEVELS))),
 134                 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 135                          ADD_NS(ADD_MULTIPLE_PRIO(1, 1),
 136                                 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 137                                                   KERNEL_NIC_PRIO_NUM_LEVELS))),
 138                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 139                          FS_CHAINING_CAPS,
 140                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
 141                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 142                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
 143         }
 144 };
 145
 146 enum fs_i_mutex_lock_class {
 147         FS_MUTEX_GRANDPARENT,
 148         FS_MUTEX_PARENT,
 149         FS_MUTEX_CHILD
 150 };
 151
 152 static void del_rule(struct fs_node *node);
 153 static void del_flow_table(struct fs_node *node);
 154 static void del_flow_group(struct fs_node *node);
 155 static void del_fte(struct fs_node *node);
 156
 157 static void tree_init_node(struct fs_node *node,
 158                            unsigned int refcount,
 159                            void (*remove_func)(struct fs_node *))
 160 {
 161         atomic_set(&node->refcount, refcount);
 162         INIT_LIST_HEAD(&node->list);
 163         INIT_LIST_HEAD(&node->children);
 164         mutex_init(&node->lock);
 165         node->remove_func = remove_func;
 166 }
 167
 168 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 169 {
 170         if (parent)
 171                 atomic_inc(&parent->refcount);
 172         node->parent = parent;
 173
 174         /* Parent is the root */
 175         if (!parent)
 176                 node->root = node;
 177         else
 178                 node->root = parent->root;
 179 }
 180
 181 static void tree_get_node(struct fs_node *node)
 182 {
 183         atomic_inc(&node->refcount);
 184 }
 185
 186 static void nested_lock_ref_node(struct fs_node *node,
 187                                  enum fs_i_mutex_lock_class class)
 188 {
 189         if (node) {
 190                 mutex_lock_nested(&node->lock, class);
 191                 atomic_inc(&node->refcount);
 192         }
 193 }
 194
 195 static void lock_ref_node(struct fs_node *node)
 196 {
 197         if (node) {
 198                 mutex_lock(&node->lock);
 199                 atomic_inc(&node->refcount);
 200         }
 201 }
 202
 203 static void unlock_ref_node(struct fs_node *node)
 204 {
 205         if (node) {
 206                 atomic_dec(&node->refcount);
 207                 mutex_unlock(&node->lock);
 208         }
 209 }
 210
 211 static void tree_put_node(struct fs_node *node)
 212 {
 213         struct fs_node *parent_node = node->parent;
 214
 215         lock_ref_node(parent_node);
 216         if (atomic_dec_and_test(&node->refcount)) {
 217                 if (parent_node)
 218                         list_del_init(&node->list);
 219                 if (node->remove_func)
 220                         node->remove_func(node);
 221                 kfree(node);
 222                 node = NULL;
 223         }
 224         unlock_ref_node(parent_node);
 225         if (!node && parent_node)
 226                 tree_put_node(parent_node);
 227 }
 228
 229 static int tree_remove_node(struct fs_node *node)
 230 {
 231         if (atomic_read(&node->refcount) > 1) {
 232                 atomic_dec(&node->refcount);
 233                 return -EEXIST;
 234         }
 235         tree_put_node(node);
 236         return 0;
 237 }
 238
 239 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 240                                  unsigned int prio)
 241 {
 242         struct fs_prio *iter_prio;
 243
 244         fs_for_each_prio(iter_prio, ns) {
 245                 if (iter_prio->prio == prio)
 246                         return iter_prio;
 247         }
 248
 249         return NULL;
 250 }
 251
 252 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 253 {
 254         unsigned int i;
 255
 256         for (i = 0; i < size; i++, mask++, val1++, val2++)
 257                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 258                     ((*(u8 *)val2) & (*(u8 *)mask)))
 259                         return false;
 260
 261         return true;
 262 }
 263
 264 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 265                                 void *fte_param1, void *fte_param2)
 266 {
 267         if (mask->match_criteria_enable &
 268             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 269                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 270                                                 fte_param1, outer_headers);
 271                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 272                                                 fte_param2, outer_headers);
 273                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 274                                               mask->match_criteria, outer_headers);
 275
 276                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 277                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 278                         return false;
 279         }
 280
 281         if (mask->match_criteria_enable &
 282             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 283                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 284                                                 fte_param1, misc_parameters);
 285                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 286                                                 fte_param2, misc_parameters);
 287                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 288                                           mask->match_criteria, misc_parameters);
 289
 290                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 291                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 292                         return false;
 293         }
 294
 295         if (mask->match_criteria_enable &
 296             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 297                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 298                                                 fte_param1, inner_headers);
 299                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 300                                                 fte_param2, inner_headers);
 301                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 302                                           mask->match_criteria, inner_headers);
 303
 304                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 305                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 306                         return false;
 307         }
 308         return true;
 309 }
 310
 311 static bool compare_match_criteria(u8 match_criteria_enable1,
 312                                    u8 match_criteria_enable2,
 313                                    void *mask1, void *mask2)
 314 {
 315         return match_criteria_enable1 == match_criteria_enable2 &&
 316                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 317 }
 318
 319 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 320 {
 321         struct fs_node *root;
 322         struct mlx5_flow_namespace *ns;
 323
 324         root = node->root;
 325
 326         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 327                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 328                 return NULL;
 329         }
 330
 331         ns = container_of(root, struct mlx5_flow_namespace, node);
 332         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 333 }
 334
 335 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 336 {
 337         struct mlx5_flow_root_namespace *root = find_root(node);
 338
 339         if (root)
 340                 return root->dev;
 341         return NULL;
 342 }
 343
 344 static void del_flow_table(struct fs_node *node)
 345 {
 346         struct mlx5_flow_table *ft;
 347         struct mlx5_core_dev *dev;
 348         struct fs_prio *prio;
 349         int err;
 350
 351         fs_get_obj(ft, node);
 352         dev = get_dev(&ft->node);
 353
 354         err = mlx5_cmd_destroy_flow_table(dev, ft);
 355         if (err)
 356                 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 357         fs_get_obj(prio, ft->node.parent);
 358         prio->num_ft--;
 359 }
 360
 361 static void del_rule(struct fs_node *node)
 362 {
 363         struct mlx5_flow_rule *rule;
 364         struct mlx5_flow_table *ft;
 365         struct mlx5_flow_group *fg;
 366         struct fs_fte *fte;
 367         u32     *match_value;
 368         int modify_mask;
 369         struct mlx5_core_dev *dev = get_dev(node);
 370         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 371         int err;
 372
 373         match_value = mlx5_vzalloc(match_len);
 374         if (!match_value) {
 375                 mlx5_core_warn(dev, "failed to allocate inbox\n");
 376                 return;
 377         }
 378
 379         fs_get_obj(rule, node);
 380         fs_get_obj(fte, rule->node.parent);
 381         fs_get_obj(fg, fte->node.parent);
 382         memcpy(match_value, fte->val, sizeof(fte->val));
 383         fs_get_obj(ft, fg->node.parent);
 384         list_del(&rule->node.list);
 385         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 386                 mutex_lock(&rule->dest_attr.ft->lock);
 387                 list_del(&rule->next_ft);
 388                 mutex_unlock(&rule->dest_attr.ft->lock);
 389         }
 390         if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 391             --fte->dests_size) {
 392                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST),
 393                 err = mlx5_cmd_update_fte(dev, ft,
 394                                           fg->id,
 395                                           modify_mask,
 396                                           fte);
 397                 if (err)
 398                         mlx5_core_warn(dev,
 399                                        "%s can't del rule fg id=%d fte_index=%d\n",
 400                                        __func__, fg->id, fte->index);
 401         }
 402         kvfree(match_value);
 403 }
 404
 405 static void del_fte(struct fs_node *node)
 406 {
 407         struct mlx5_flow_table *ft;
 408         struct mlx5_flow_group *fg;
 409         struct mlx5_core_dev *dev;
 410         struct fs_fte *fte;
 411         int err;
 412
 413         fs_get_obj(fte, node);
 414         fs_get_obj(fg, fte->node.parent);
 415         fs_get_obj(ft, fg->node.parent);
 416
 417         dev = get_dev(&ft->node);
 418         err = mlx5_cmd_delete_fte(dev, ft,
 419                                   fte->index);
 420         if (err)
 421                 mlx5_core_warn(dev,
 422                                "flow steering can't delete fte in index %d of flow group id %d\n",
 423                                fte->index, fg->id);
 424
 425         fte->status = 0;
 426         fg->num_ftes--;
 427 }
 428
 429 static void del_flow_group(struct fs_node *node)
 430 {
 431         struct mlx5_flow_group *fg;
 432         struct mlx5_flow_table *ft;
 433         struct mlx5_core_dev *dev;
 434
 435         fs_get_obj(fg, node);
 436         fs_get_obj(ft, fg->node.parent);
 437         dev = get_dev(&ft->node);
 438
 439         if (ft->autogroup.active)
 440                 ft->autogroup.num_groups--;
 441
 442         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 443                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 444                                fg->id, ft->id);
 445 }
 446
 447 static struct fs_fte *alloc_fte(u8 action,
 448                                 u32 flow_tag,
 449                                 u32 *match_value,
 450                                 unsigned int index)
 451 {
 452         struct fs_fte *fte;
 453
 454         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 455         if (!fte)
 456                 return ERR_PTR(-ENOMEM);
 457
 458         memcpy(fte->val, match_value, sizeof(fte->val));
 459         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 460         fte->flow_tag = flow_tag;
 461         fte->index = index;
 462         fte->action = action;
 463
 464         return fte;
 465 }
 466
 467 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 468 {
 469         struct mlx5_flow_group *fg;
 470         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 471                                             create_fg_in, match_criteria);
 472         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 473                                             create_fg_in,
 474                                             match_criteria_enable);
 475         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 476         if (!fg)
 477                 return ERR_PTR(-ENOMEM);
 478
 479         fg->mask.match_criteria_enable = match_criteria_enable;
 480         memcpy(&fg->mask.match_criteria, match_criteria,
 481                sizeof(fg->mask.match_criteria));
 482         fg->node.type =  FS_TYPE_FLOW_GROUP;
 483         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 484                                    start_flow_index);
 485         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 486                                 end_flow_index) - fg->start_index + 1;
 487         return fg;
 488 }
 489
 490 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 491                                                 enum fs_flow_table_type table_type,
 492                                                 enum fs_flow_table_op_mod op_mod)
 493 {
 494         struct mlx5_flow_table *ft;
 495
 496         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 497         if (!ft)
 498                 return NULL;
 499
 500         ft->level = level;
 501         ft->node.type = FS_TYPE_FLOW_TABLE;
 502         ft->op_mod = op_mod;
 503         ft->type = table_type;
 504         ft->vport = vport;
 505         ft->max_fte = max_fte;
 506         INIT_LIST_HEAD(&ft->fwd_rules);
 507         mutex_init(&ft->lock);
 508
 509         return ft;
 510 }
 511
 512 /* If reverse is false, then we search for the first flow table in the
 513  * root sub-tree from start(closest from right), else we search for the
 514  * last flow table in the root sub-tree till start(closest from left).
 515  */
 516 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 517                                                          struct list_head *start,
 518                                                          bool reverse)
 519 {
 520 #define list_advance_entry(pos, reverse)                \
 521         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 522
 523 #define list_for_each_advance_continue(pos, head, reverse)      \
 524         for (pos = list_advance_entry(pos, reverse);            \
 525              &pos->list != (head);                              \
 526              pos = list_advance_entry(pos, reverse))
 527
 528         struct fs_node *iter = list_entry(start, struct fs_node, list);
 529         struct mlx5_flow_table *ft = NULL;
 530
 531         if (!root)
 532                 return NULL;
 533
 534         list_for_each_advance_continue(iter, &root->children, reverse) {
 535                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 536                         fs_get_obj(ft, iter);
 537                         return ft;
 538                 }
 539                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 540                 if (ft)
 541                         return ft;
 542         }
 543
 544         return ft;
 545 }
 546
 547 /* If reverse if false then return the first flow table in next priority of
 548  * prio in the tree, else return the last flow table in the previous priority
 549  * of prio in the tree.
 550  */
 551 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 552 {
 553         struct mlx5_flow_table *ft = NULL;
 554         struct fs_node *curr_node;
 555         struct fs_node *parent;
 556
 557         parent = prio->node.parent;
 558         curr_node = &prio->node;
 559         while (!ft && parent) {
 560                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 561                 curr_node = parent;
 562                 parent = curr_node->parent;
 563         }
 564         return ft;
 565 }
 566
 567 /* Assuming all the tree is locked by mutex chain lock */
 568 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 569 {
 570         return find_closest_ft(prio, false);
 571 }
 572
 573 /* Assuming all the tree is locked by mutex chain lock */
 574 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 575 {
 576         return find_closest_ft(prio, true);
 577 }
 578
 579 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 580                                struct fs_prio *prio,
 581                                struct mlx5_flow_table *ft)
 582 {
 583         struct mlx5_flow_table *iter;
 584         int i = 0;
 585         int err;
 586
 587         fs_for_each_ft(iter, prio) {
 588                 i++;
 589                 err = mlx5_cmd_modify_flow_table(dev,
 590                                                  iter,
 591                                                  ft);
 592                 if (err) {
 593                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 594                                        iter->id);
 595                         /* The driver is out of sync with the FW */
 596                         if (i > 1)
 597                                 WARN_ON(true);
 598                         return err;
 599                 }
 600         }
 601         return 0;
 602 }
 603
 604 /* Connect flow tables from previous priority of prio to ft */
 605 static int connect_prev_fts(struct mlx5_core_dev *dev,
 606                             struct mlx5_flow_table *ft,
 607                             struct fs_prio *prio)
 608 {
 609         struct mlx5_flow_table *prev_ft;
 610
 611         prev_ft = find_prev_chained_ft(prio);
 612         if (prev_ft) {
 613                 struct fs_prio *prev_prio;
 614
 615                 fs_get_obj(prev_prio, prev_ft->node.parent);
 616                 return connect_fts_in_prio(dev, prev_prio, ft);
 617         }
 618         return 0;
 619 }
 620
 621 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 622                                  *prio)
 623 {
 624         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 625         int min_level = INT_MAX;
 626         int err;
 627
 628         if (root->root_ft)
 629                 min_level = root->root_ft->level;
 630
 631         if (ft->level >= min_level)
 632                 return 0;
 633
 634         err = mlx5_cmd_update_root_ft(root->dev, ft);
 635         if (err)
 636                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 637                                ft->id);
 638         else
 639                 root->root_ft = ft;
 640
 641         return err;
 642 }
 643
 644 int mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 645                                  struct mlx5_flow_destination *dest)
 646 {
 647         struct mlx5_flow_table *ft;
 648         struct mlx5_flow_group *fg;
 649         struct fs_fte *fte;
 650         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 651         int err = 0;
 652
 653         fs_get_obj(fte, rule->node.parent);
 654         if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 655                 return -EINVAL;
 656         lock_ref_node(&fte->node);
 657         fs_get_obj(fg, fte->node.parent);
 658         fs_get_obj(ft, fg->node.parent);
 659
 660         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 661         err = mlx5_cmd_update_fte(get_dev(&ft->node),
 662                                   ft, fg->id,
 663                                   modify_mask,
 664                                   fte);
 665         unlock_ref_node(&fte->node);
 666
 667         return err;
 668 }
 669
 670 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 671 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 672                              struct mlx5_flow_table *new_next_ft,
 673                              struct mlx5_flow_table *old_next_ft)
 674 {
 675         struct mlx5_flow_destination dest;
 676         struct mlx5_flow_rule *iter;
 677         int err = 0;
 678
 679         /* new_next_ft and old_next_ft could be NULL only
 680          * when we create/destroy the anchor flow table.
 681          */
 682         if (!new_next_ft || !old_next_ft)
 683                 return 0;
 684
 685         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 686         dest.ft = new_next_ft;
 687
 688         mutex_lock(&old_next_ft->lock);
 689         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 690         mutex_unlock(&old_next_ft->lock);
 691         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 692                 err = mlx5_modify_rule_destination(iter, &dest);
 693                 if (err)
 694                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 695                                new_next_ft->id);
 696         }
 697         return 0;
 698 }
 699
 700 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 701                               struct fs_prio *prio)
 702 {
 703         struct mlx5_flow_table *next_ft;
 704         int err = 0;
 705
 706         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 707
 708         if (list_empty(&prio->node.children)) {
 709                 err = connect_prev_fts(dev, ft, prio);
 710                 if (err)
 711                         return err;
 712
 713                 next_ft = find_next_chained_ft(prio);
 714                 err = connect_fwd_rules(dev, ft, next_ft);
 715                 if (err)
 716                         return err;
 717         }
 718
 719         if (MLX5_CAP_FLOWTABLE(dev,
 720                                flow_table_properties_nic_receive.modify_root))
 721                 err = update_root_ft_create(ft, prio);
 722         return err;
 723 }
 724
 725 static void list_add_flow_table(struct mlx5_flow_table *ft,
 726                                 struct fs_prio *prio)
 727 {
 728         struct list_head *prev = &prio->node.children;
 729         struct mlx5_flow_table *iter;
 730
 731         fs_for_each_ft(iter, prio) {
 732                 if (iter->level > ft->level)
 733                         break;
 734                 prev = &iter->node.list;
 735         }
 736         list_add(&ft->node.list, prev);
 737 }
 738
 739 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 740                                                         enum fs_flow_table_op_mod op_mod,
 741                                                         u16 vport, int prio,
 742                                                         int max_fte, u32 level)
 743 {
 744         struct mlx5_flow_table *next_ft = NULL;
 745         struct mlx5_flow_table *ft;
 746         int err;
 747         int log_table_sz;
 748         struct mlx5_flow_root_namespace *root =
 749                 find_root(&ns->node);
 750         struct fs_prio *fs_prio = NULL;
 751
 752         if (!root) {
 753                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 754                 return ERR_PTR(-ENODEV);
 755         }
 756
 757         mutex_lock(&root->chain_lock);
 758         fs_prio = find_prio(ns, prio);
 759         if (!fs_prio) {
 760                 err = -EINVAL;
 761                 goto unlock_root;
 762         }
 763         if (level >= fs_prio->num_levels) {
 764                 err = -ENOSPC;
 765                 goto unlock_root;
 766         }
 767         /* The level is related to the
 768          * priority level range.
 769          */
 770         level += fs_prio->start_level;
 771         ft = alloc_flow_table(level,
 772                               vport,
 773                               max_fte ? roundup_pow_of_two(max_fte) : 0,
 774                               root->table_type,
 775                               op_mod);
 776         if (!ft) {
 777                 err = -ENOMEM;
 778                 goto unlock_root;
 779         }
 780
 781         tree_init_node(&ft->node, 1, del_flow_table);
 782         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
 783         next_ft = find_next_chained_ft(fs_prio);
 784         err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->op_mod, ft->type,
 785                                          ft->level, log_table_sz, next_ft, &ft->id);
 786         if (err)
 787                 goto free_ft;
 788
 789         err = connect_flow_table(root->dev, ft, fs_prio);
 790         if (err)
 791                 goto destroy_ft;
 792         lock_ref_node(&fs_prio->node);
 793         tree_add_node(&ft->node, &fs_prio->node);
 794         list_add_flow_table(ft, fs_prio);
 795         fs_prio->num_ft++;
 796         unlock_ref_node(&fs_prio->node);
 797         mutex_unlock(&root->chain_lock);
 798         return ft;
 799 destroy_ft:
 800         mlx5_cmd_destroy_flow_table(root->dev, ft);
 801 free_ft:
 802         kfree(ft);
 803 unlock_root:
 804         mutex_unlock(&root->chain_lock);
 805         return ERR_PTR(err);
 806 }
 807
 808 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 809                                                int prio, int max_fte,
 810                                                u32 level)
 811 {
 812         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, 0, prio,
 813                                         max_fte, level);
 814 }
 815
 816 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
 817                                                      int prio, int max_fte,
 818                                                      u32 level, u16 vport)
 819 {
 820         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, vport, prio,
 821                                         max_fte, level);
 822 }
 823
 824 struct mlx5_flow_table *mlx5_create_lag_demux_flow_table(
 825                                                struct mlx5_flow_namespace *ns,
 826                                                int prio, u32 level)
 827 {
 828         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_LAG_DEMUX, 0, prio, 0,
 829                                         level);
 830 }
 831 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
 832
 833 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 834                                                             int prio,
 835                                                             int num_flow_table_entries,
 836                                                             int max_num_groups,
 837                                                             u32 level)
 838 {
 839         struct mlx5_flow_table *ft;
 840
 841         if (max_num_groups > num_flow_table_entries)
 842                 return ERR_PTR(-EINVAL);
 843
 844         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries, level);
 845         if (IS_ERR(ft))
 846                 return ft;
 847
 848         ft->autogroup.active = true;
 849         ft->autogroup.required_groups = max_num_groups;
 850
 851         return ft;
 852 }
 853 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 854
 855 /* Flow table should be locked */
 856 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 857                                                         u32 *fg_in,
 858                                                         struct list_head
 859                                                         *prev_fg,
 860                                                         bool is_auto_fg)
 861 {
 862         struct mlx5_flow_group *fg;
 863         struct mlx5_core_dev *dev = get_dev(&ft->node);
 864         int err;
 865
 866         if (!dev)
 867                 return ERR_PTR(-ENODEV);
 868
 869         fg = alloc_flow_group(fg_in);
 870         if (IS_ERR(fg))
 871                 return fg;
 872
 873         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 874         if (err) {
 875                 kfree(fg);
 876                 return ERR_PTR(err);
 877         }
 878
 879         if (ft->autogroup.active)
 880                 ft->autogroup.num_groups++;
 881         /* Add node to tree */
 882         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 883         tree_add_node(&fg->node, &ft->node);
 884         /* Add node to group list */
 885         list_add(&fg->node.list, prev_fg);
 886
 887         return fg;
 888 }
 889
 890 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 891                                                u32 *fg_in)
 892 {
 893         struct mlx5_flow_group *fg;
 894
 895         if (ft->autogroup.active)
 896                 return ERR_PTR(-EPERM);
 897
 898         lock_ref_node(&ft->node);
 899         fg = create_flow_group_common(ft, fg_in, ft->node.children.prev, false);
 900         unlock_ref_node(&ft->node);
 901
 902         return fg;
 903 }
 904
 905 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 906 {
 907         struct mlx5_flow_rule *rule;
 908
 909         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 910         if (!rule)
 911                 return NULL;
 912
 913         INIT_LIST_HEAD(&rule->next_ft);
 914         rule->node.type = FS_TYPE_FLOW_DEST;
 915         if (dest)
 916                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 917
 918         return rule;
 919 }
 920
 921 /* fte should not be deleted while calling this function */
 922 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
 923                                            struct mlx5_flow_group *fg,
 924                                            struct mlx5_flow_destination *dest)
 925 {
 926         struct mlx5_flow_table *ft;
 927         struct mlx5_flow_rule *rule;
 928         int modify_mask = 0;
 929         int err;
 930
 931         rule = alloc_rule(dest);
 932         if (!rule)
 933                 return ERR_PTR(-ENOMEM);
 934
 935         fs_get_obj(ft, fg->node.parent);
 936         /* Add dest to dests list- we need flow tables to be in the
 937          * end of the list for forward to next prio rules.
 938          */
 939         tree_init_node(&rule->node, 1, del_rule);
 940         if (dest && dest->type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
 941                 list_add(&rule->node.list, &fte->node.children);
 942         else
 943                 list_add_tail(&rule->node.list, &fte->node.children);
 944         if (dest) {
 945                 fte->dests_size++;
 946
 947                 modify_mask |= dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER ?
 948                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS) :
 949                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 950         }
 951
 952         if (fte->dests_size == 1 || !dest)
 953                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 954                                           ft, fg->id, fte);
 955         else
 956                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 957                                           ft, fg->id, modify_mask, fte);
 958         if (err)
 959                 goto free_rule;
 960
 961         fte->status |= FS_FTE_STATUS_EXISTING;
 962
 963         return rule;
 964
 965 free_rule:
 966         list_del(&rule->node.list);
 967         kfree(rule);
 968         if (dest)
 969                 fte->dests_size--;
 970         return ERR_PTR(err);
 971 }
 972
 973 /* Assumed fg is locked */
 974 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 975                                        struct list_head **prev)
 976 {
 977         struct fs_fte *fte;
 978         unsigned int start = fg->start_index;
 979
 980         if (prev)
 981                 *prev = &fg->node.children;
 982
 983         /* assumed list is sorted by index */
 984         fs_for_each_fte(fte, fg) {
 985                 if (fte->index != start)
 986                         return start;
 987                 start++;
 988                 if (prev)
 989                         *prev = &fte->node.list;
 990         }
 991
 992         return start;
 993 }
 994
 995 /* prev is output, prev->next = new_fte */
 996 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
 997                                  u32 *match_value,
 998                                  u8 action,
 999                                  u32 flow_tag,
1000                                  struct list_head **prev)
1001 {
1002         struct fs_fte *fte;
1003         int index;
1004
1005         index = get_free_fte_index(fg, prev);
1006         fte = alloc_fte(action, flow_tag, match_value, index);
1007         if (IS_ERR(fte))
1008                 return fte;
1009
1010         return fte;
1011 }
1012
1013 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
1014                                                 u8 match_criteria_enable,
1015                                                 u32 *match_criteria)
1016 {
1017         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1018         struct list_head *prev = ft->node.children.prev;
1019         unsigned int candidate_index = 0;
1020         struct mlx5_flow_group *fg;
1021         void *match_criteria_addr;
1022         unsigned int group_size = 0;
1023         u32 *in;
1024
1025         if (!ft->autogroup.active)
1026                 return ERR_PTR(-ENOENT);
1027
1028         in = mlx5_vzalloc(inlen);
1029         if (!in)
1030                 return ERR_PTR(-ENOMEM);
1031
1032         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1033                 /* We save place for flow groups in addition to max types */
1034                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1035
1036         /*  ft->max_fte == ft->autogroup.max_types */
1037         if (group_size == 0)
1038                 group_size = 1;
1039
1040         /* sorted by start_index */
1041         fs_for_each_fg(fg, ft) {
1042                 if (candidate_index + group_size > fg->start_index)
1043                         candidate_index = fg->start_index + fg->max_ftes;
1044                 else
1045                         break;
1046                 prev = &fg->node.list;
1047         }
1048
1049         if (candidate_index + group_size > ft->max_fte) {
1050                 fg = ERR_PTR(-ENOSPC);
1051                 goto out;
1052         }
1053
1054         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1055                  match_criteria_enable);
1056         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1057         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
1058                  group_size - 1);
1059         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1060                                            in, match_criteria);
1061         memcpy(match_criteria_addr, match_criteria,
1062                MLX5_ST_SZ_BYTES(fte_match_param));
1063
1064         fg = create_flow_group_common(ft, in, prev, true);
1065 out:
1066         kvfree(in);
1067         return fg;
1068 }
1069
1070 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1071                                              struct mlx5_flow_destination *dest)
1072 {
1073         struct mlx5_flow_rule *rule;
1074
1075         list_for_each_entry(rule, &fte->node.children, node.list) {
1076                 if (rule->dest_attr.type == dest->type) {
1077                         if ((dest->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1078                              dest->vport_num == rule->dest_attr.vport_num) ||
1079                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1080                              dest->ft == rule->dest_attr.ft) ||
1081                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1082                              dest->tir_num == rule->dest_attr.tir_num))
1083                                 return rule;
1084                 }
1085         }
1086         return NULL;
1087 }
1088
1089 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
1090                                           u32 *match_value,
1091                                           u8 action,
1092                                           u32 flow_tag,
1093                                           struct mlx5_flow_destination *dest)
1094 {
1095         struct fs_fte *fte;
1096         struct mlx5_flow_rule *rule;
1097         struct mlx5_flow_table *ft;
1098         struct list_head *prev;
1099
1100         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1101         fs_for_each_fte(fte, fg) {
1102                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1103                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1104                     action == fte->action && flow_tag == fte->flow_tag) {
1105                         rule = find_flow_rule(fte, dest);
1106                         if (rule) {
1107                                 atomic_inc(&rule->node.refcount);
1108                                 unlock_ref_node(&fte->node);
1109                                 unlock_ref_node(&fg->node);
1110                                 return rule;
1111                         }
1112                         rule = add_rule_fte(fte, fg, dest);
1113                         unlock_ref_node(&fte->node);
1114                         if (IS_ERR(rule))
1115                                 goto unlock_fg;
1116                         else
1117                                 goto add_rule;
1118                 }
1119                 unlock_ref_node(&fte->node);
1120         }
1121         fs_get_obj(ft, fg->node.parent);
1122         if (fg->num_ftes >= fg->max_ftes) {
1123                 rule = ERR_PTR(-ENOSPC);
1124                 goto unlock_fg;
1125         }
1126
1127         fte = create_fte(fg, match_value, action, flow_tag, &prev);
1128         if (IS_ERR(fte)) {
1129                 rule = (void *)fte;
1130                 goto unlock_fg;
1131         }
1132         tree_init_node(&fte->node, 0, del_fte);
1133         rule = add_rule_fte(fte, fg, dest);
1134         if (IS_ERR(rule)) {
1135                 kfree(fte);
1136                 goto unlock_fg;
1137         }
1138
1139         fg->num_ftes++;
1140
1141         tree_add_node(&fte->node, &fg->node);
1142         list_add(&fte->node.list, prev);
1143 add_rule:
1144         tree_add_node(&rule->node, &fte->node);
1145 unlock_fg:
1146         unlock_ref_node(&fg->node);
1147         return rule;
1148 }
1149
1150 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_rule *rule)
1151 {
1152         struct mlx5_flow_rule *dst;
1153         struct fs_fte *fte;
1154
1155         fs_get_obj(fte, rule->node.parent);
1156
1157         fs_for_each_dst(dst, fte) {
1158                 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1159                         return dst->dest_attr.counter;
1160         }
1161
1162         return NULL;
1163 }
1164
1165 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1166 {
1167         if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1168                 return !counter;
1169
1170         if (!counter)
1171                 return false;
1172
1173         /* Hardware support counter for a drop action only */
1174         return action == (MLX5_FLOW_CONTEXT_ACTION_DROP | MLX5_FLOW_CONTEXT_ACTION_COUNT);
1175 }
1176
1177 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1178                           u32 action,
1179                           struct mlx5_flow_table *ft)
1180 {
1181         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1182                 return counter_is_valid(dest->counter, action);
1183
1184         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1185                 return true;
1186
1187         if (!dest || ((dest->type ==
1188             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1189             (dest->ft->level <= ft->level)))
1190                 return false;
1191         return true;
1192 }
1193
1194 static struct mlx5_flow_rule *
1195 _mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1196                    struct mlx5_flow_spec *spec,
1197                     u32 action,
1198                     u32 flow_tag,
1199                     struct mlx5_flow_destination *dest)
1200 {
1201         struct mlx5_flow_group *g;
1202         struct mlx5_flow_rule *rule;
1203
1204         if (!dest_is_valid(dest, action, ft))
1205                 return ERR_PTR(-EINVAL);
1206
1207         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1208         fs_for_each_fg(g, ft)
1209                 if (compare_match_criteria(g->mask.match_criteria_enable,
1210                                            spec->match_criteria_enable,
1211                                            g->mask.match_criteria,
1212                                            spec->match_criteria)) {
1213                         rule = add_rule_fg(g, spec->match_value,
1214                                            action, flow_tag, dest);
1215                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1216                                 goto unlock;
1217                 }
1218
1219         g = create_autogroup(ft, spec->match_criteria_enable,
1220                              spec->match_criteria);
1221         if (IS_ERR(g)) {
1222                 rule = (void *)g;
1223                 goto unlock;
1224         }
1225
1226         rule = add_rule_fg(g, spec->match_value,
1227                            action, flow_tag, dest);
1228         if (IS_ERR(rule)) {
1229                 /* Remove assumes refcount > 0 and autogroup creates a group
1230                  * with a refcount = 0.
1231                  */
1232                 unlock_ref_node(&ft->node);
1233                 tree_get_node(&g->node);
1234                 tree_remove_node(&g->node);
1235                 return rule;
1236         }
1237 unlock:
1238         unlock_ref_node(&ft->node);
1239         return rule;
1240 }
1241
1242 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1243 {
1244         return ((ft->type == FS_FT_NIC_RX) &&
1245                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1246 }
1247
1248 struct mlx5_flow_rule *
1249 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1250                    struct mlx5_flow_spec *spec,
1251                    u32 action,
1252                    u32 flow_tag,
1253                    struct mlx5_flow_destination *dest)
1254 {
1255         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1256         struct mlx5_flow_destination gen_dest;
1257         struct mlx5_flow_table *next_ft = NULL;
1258         struct mlx5_flow_rule *rule = NULL;
1259         u32 sw_action = action;
1260         struct fs_prio *prio;
1261
1262         fs_get_obj(prio, ft->node.parent);
1263         if (action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1264                 if (!fwd_next_prio_supported(ft))
1265                         return ERR_PTR(-EOPNOTSUPP);
1266                 if (dest)
1267                         return ERR_PTR(-EINVAL);
1268                 mutex_lock(&root->chain_lock);
1269                 next_ft = find_next_chained_ft(prio);
1270                 if (next_ft) {
1271                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1272                         gen_dest.ft = next_ft;
1273                         dest = &gen_dest;
1274                         action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1275                 } else {
1276                         mutex_unlock(&root->chain_lock);
1277                         return ERR_PTR(-EOPNOTSUPP);
1278                 }
1279         }
1280
1281         rule = _mlx5_add_flow_rule(ft, spec, action, flow_tag, dest);
1282
1283         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1284                 if (!IS_ERR_OR_NULL(rule) &&
1285                     (list_empty(&rule->next_ft))) {
1286                         mutex_lock(&next_ft->lock);
1287                         list_add(&rule->next_ft, &next_ft->fwd_rules);
1288                         mutex_unlock(&next_ft->lock);
1289                         rule->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1290                 }
1291                 mutex_unlock(&root->chain_lock);
1292         }
1293         return rule;
1294 }
1295 EXPORT_SYMBOL(mlx5_add_flow_rule);
1296
1297 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
1298 {
1299         tree_remove_node(&rule->node);
1300 }
1301 EXPORT_SYMBOL(mlx5_del_flow_rule);
1302
1303 /* Assuming prio->node.children(flow tables) is sorted by level */
1304 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1305 {
1306         struct fs_prio *prio;
1307
1308         fs_get_obj(prio, ft->node.parent);
1309
1310         if (!list_is_last(&ft->node.list, &prio->node.children))
1311                 return list_next_entry(ft, node.list);
1312         return find_next_chained_ft(prio);
1313 }
1314
1315 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1316 {
1317         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1318         struct mlx5_flow_table *new_root_ft = NULL;
1319
1320         if (root->root_ft != ft)
1321                 return 0;
1322
1323         new_root_ft = find_next_ft(ft);
1324         if (new_root_ft) {
1325                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1326
1327                 if (err) {
1328                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1329                                        ft->id);
1330                         return err;
1331                 }
1332         }
1333         root->root_ft = new_root_ft;
1334         return 0;
1335 }
1336
1337 /* Connect flow table from previous priority to
1338  * the next flow table.
1339  */
1340 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1341 {
1342         struct mlx5_core_dev *dev = get_dev(&ft->node);
1343         struct mlx5_flow_table *next_ft;
1344         struct fs_prio *prio;
1345         int err = 0;
1346
1347         err = update_root_ft_destroy(ft);
1348         if (err)
1349                 return err;
1350
1351         fs_get_obj(prio, ft->node.parent);
1352         if  (!(list_first_entry(&prio->node.children,
1353                                 struct mlx5_flow_table,
1354                                 node.list) == ft))
1355                 return 0;
1356
1357         next_ft = find_next_chained_ft(prio);
1358         err = connect_fwd_rules(dev, next_ft, ft);
1359         if (err)
1360                 return err;
1361
1362         err = connect_prev_fts(dev, next_ft, prio);
1363         if (err)
1364                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1365                                ft->id);
1366         return err;
1367 }
1368
1369 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1370 {
1371         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1372         int err = 0;
1373
1374         mutex_lock(&root->chain_lock);
1375         err = disconnect_flow_table(ft);
1376         if (err) {
1377                 mutex_unlock(&root->chain_lock);
1378                 return err;
1379         }
1380         if (tree_remove_node(&ft->node))
1381                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1382                                ft->id);
1383         mutex_unlock(&root->chain_lock);
1384
1385         return err;
1386 }
1387 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1388
1389 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1390 {
1391         if (tree_remove_node(&fg->node))
1392                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1393                                fg->id);
1394 }
1395
1396 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1397                                                     enum mlx5_flow_namespace_type type)
1398 {
1399         struct mlx5_flow_steering *steering = dev->priv.steering;
1400         struct mlx5_flow_root_namespace *root_ns;
1401         int prio;
1402         struct fs_prio *fs_prio;
1403         struct mlx5_flow_namespace *ns;
1404
1405         if (!steering)
1406                 return NULL;
1407
1408         switch (type) {
1409         case MLX5_FLOW_NAMESPACE_BYPASS:
1410         case MLX5_FLOW_NAMESPACE_LAG:
1411         case MLX5_FLOW_NAMESPACE_OFFLOADS:
1412         case MLX5_FLOW_NAMESPACE_ETHTOOL:
1413         case MLX5_FLOW_NAMESPACE_KERNEL:
1414         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1415         case MLX5_FLOW_NAMESPACE_ANCHOR:
1416                 prio = type;
1417                 break;
1418         case MLX5_FLOW_NAMESPACE_FDB:
1419                 if (steering->fdb_root_ns)
1420                         return &steering->fdb_root_ns->ns;
1421                 else
1422                         return NULL;
1423         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1424                 if (steering->esw_egress_root_ns)
1425                         return &steering->esw_egress_root_ns->ns;
1426                 else
1427                         return NULL;
1428         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1429                 if (steering->esw_ingress_root_ns)
1430                         return &steering->esw_ingress_root_ns->ns;
1431                 else
1432                         return NULL;
1433         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
1434                 if (steering->sniffer_rx_root_ns)
1435                         return &steering->sniffer_rx_root_ns->ns;
1436                 else
1437                         return NULL;
1438         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
1439                 if (steering->sniffer_tx_root_ns)
1440                         return &steering->sniffer_tx_root_ns->ns;
1441                 else
1442                         return NULL;
1443         default:
1444                 return NULL;
1445         }
1446
1447         root_ns = steering->root_ns;
1448         if (!root_ns)
1449                 return NULL;
1450
1451         fs_prio = find_prio(&root_ns->ns, prio);
1452         if (!fs_prio)
1453                 return NULL;
1454
1455         ns = list_first_entry(&fs_prio->node.children,
1456                               typeof(*ns),
1457                               node.list);
1458
1459         return ns;
1460 }
1461 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1462
1463 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1464                                       unsigned int prio, int num_levels)
1465 {
1466         struct fs_prio *fs_prio;
1467
1468         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1469         if (!fs_prio)
1470                 return ERR_PTR(-ENOMEM);
1471
1472         fs_prio->node.type = FS_TYPE_PRIO;
1473         tree_init_node(&fs_prio->node, 1, NULL);
1474         tree_add_node(&fs_prio->node, &ns->node);
1475         fs_prio->num_levels = num_levels;
1476         fs_prio->prio = prio;
1477         list_add_tail(&fs_prio->node.list, &ns->node.children);
1478
1479         return fs_prio;
1480 }
1481
1482 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1483                                                      *ns)
1484 {
1485         ns->node.type = FS_TYPE_NAMESPACE;
1486
1487         return ns;
1488 }
1489
1490 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1491 {
1492         struct mlx5_flow_namespace      *ns;
1493
1494         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1495         if (!ns)
1496                 return ERR_PTR(-ENOMEM);
1497
1498         fs_init_namespace(ns);
1499         tree_init_node(&ns->node, 1, NULL);
1500         tree_add_node(&ns->node, &prio->node);
1501         list_add_tail(&ns->node.list, &prio->node.children);
1502
1503         return ns;
1504 }
1505
1506 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1507                              struct init_tree_node *prio_metadata)
1508 {
1509         struct fs_prio *fs_prio;
1510         int i;
1511
1512         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1513                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1514                 if (IS_ERR(fs_prio))
1515                         return PTR_ERR(fs_prio);
1516         }
1517         return 0;
1518 }
1519
1520 #define FLOW_TABLE_BIT_SZ 1
1521 #define GET_FLOW_TABLE_CAP(dev, offset) \
1522         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1523                         offset / 32)) >>                                        \
1524           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1525 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1526 {
1527         int i;
1528
1529         for (i = 0; i < caps->arr_sz; i++) {
1530                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1531                         return false;
1532         }
1533         return true;
1534 }
1535
1536 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
1537                                     struct init_tree_node *init_node,
1538                                     struct fs_node *fs_parent_node,
1539                                     struct init_tree_node *init_parent_node,
1540                                     int prio)
1541 {
1542         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
1543                                               flow_table_properties_nic_receive.
1544                                               max_ft_level);
1545         struct mlx5_flow_namespace *fs_ns;
1546         struct fs_prio *fs_prio;
1547         struct fs_node *base;
1548         int i;
1549         int err;
1550
1551         if (init_node->type == FS_TYPE_PRIO) {
1552                 if ((init_node->min_ft_level > max_ft_level) ||
1553                     !has_required_caps(steering->dev, &init_node->caps))
1554                         return 0;
1555
1556                 fs_get_obj(fs_ns, fs_parent_node);
1557                 if (init_node->num_leaf_prios)
1558                         return create_leaf_prios(fs_ns, prio, init_node);
1559                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1560                 if (IS_ERR(fs_prio))
1561                         return PTR_ERR(fs_prio);
1562                 base = &fs_prio->node;
1563         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1564                 fs_get_obj(fs_prio, fs_parent_node);
1565                 fs_ns = fs_create_namespace(fs_prio);
1566                 if (IS_ERR(fs_ns))
1567                         return PTR_ERR(fs_ns);
1568                 base = &fs_ns->node;
1569         } else {
1570                 return -EINVAL;
1571         }
1572         prio = 0;
1573         for (i = 0; i < init_node->ar_size; i++) {
1574                 err = init_root_tree_recursive(steering, &init_node->children[i],
1575                                                base, init_node, prio);
1576                 if (err)
1577                         return err;
1578                 if (init_node->children[i].type == FS_TYPE_PRIO &&
1579                     init_node->children[i].num_leaf_prios) {
1580                         prio += init_node->children[i].num_leaf_prios;
1581                 }
1582         }
1583
1584         return 0;
1585 }
1586
1587 static int init_root_tree(struct mlx5_flow_steering *steering,
1588                           struct init_tree_node *init_node,
1589                           struct fs_node *fs_parent_node)
1590 {
1591         int i;
1592         struct mlx5_flow_namespace *fs_ns;
1593         int err;
1594
1595         fs_get_obj(fs_ns, fs_parent_node);
1596         for (i = 0; i < init_node->ar_size; i++) {
1597                 err = init_root_tree_recursive(steering, &init_node->children[i],
1598                                                &fs_ns->node,
1599                                                init_node, i);
1600                 if (err)
1601                         return err;
1602         }
1603         return 0;
1604 }
1605
1606 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
1607                                                        enum fs_flow_table_type
1608                                                        table_type)
1609 {
1610         struct mlx5_flow_root_namespace *root_ns;
1611         struct mlx5_flow_namespace *ns;
1612
1613         /* Create the root namespace */
1614         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1615         if (!root_ns)
1616                 return NULL;
1617
1618         root_ns->dev = steering->dev;
1619         root_ns->table_type = table_type;
1620
1621         ns = &root_ns->ns;
1622         fs_init_namespace(ns);
1623         mutex_init(&root_ns->chain_lock);
1624         tree_init_node(&ns->node, 1, NULL);
1625         tree_add_node(&ns->node, NULL);
1626
1627         return root_ns;
1628 }
1629
1630 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1631
1632 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1633 {
1634         struct fs_prio *prio;
1635
1636         fs_for_each_prio(prio, ns) {
1637                  /* This updates prio start_level and num_levels */
1638                 set_prio_attrs_in_prio(prio, acc_level);
1639                 acc_level += prio->num_levels;
1640         }
1641         return acc_level;
1642 }
1643
1644 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1645 {
1646         struct mlx5_flow_namespace *ns;
1647         int acc_level_ns = acc_level;
1648
1649         prio->start_level = acc_level;
1650         fs_for_each_ns(ns, prio)
1651                 /* This updates start_level and num_levels of ns's priority descendants */
1652                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1653         if (!prio->num_levels)
1654                 prio->num_levels = acc_level_ns - prio->start_level;
1655         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1656 }
1657
1658 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1659 {
1660         struct mlx5_flow_namespace *ns = &root_ns->ns;
1661         struct fs_prio *prio;
1662         int start_level = 0;
1663
1664         fs_for_each_prio(prio, ns) {
1665                 set_prio_attrs_in_prio(prio, start_level);
1666                 start_level += prio->num_levels;
1667         }
1668 }
1669
1670 #define ANCHOR_PRIO 0
1671 #define ANCHOR_SIZE 1
1672 #define ANCHOR_LEVEL 0
1673 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
1674 {
1675         struct mlx5_flow_namespace *ns = NULL;
1676         struct mlx5_flow_table *ft;
1677
1678         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1679         if (!ns)
1680                 return -EINVAL;
1681         ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL);
1682         if (IS_ERR(ft)) {
1683                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
1684                 return PTR_ERR(ft);
1685         }
1686         return 0;
1687 }
1688
1689 static int init_root_ns(struct mlx5_flow_steering *steering)
1690 {
1691
1692         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
1693         if (IS_ERR_OR_NULL(steering->root_ns))
1694                 goto cleanup;
1695
1696         if (init_root_tree(steering, &root_fs, &steering->root_ns->ns.node))
1697                 goto cleanup;
1698
1699         set_prio_attrs(steering->root_ns);
1700
1701         if (create_anchor_flow_table(steering))
1702                 goto cleanup;
1703
1704         return 0;
1705
1706 cleanup:
1707         mlx5_cleanup_fs(steering->dev);
1708         return -ENOMEM;
1709 }
1710
1711 static void clean_tree(struct fs_node *node)
1712 {
1713         if (node) {
1714                 struct fs_node *iter;
1715                 struct fs_node *temp;
1716
1717                 list_for_each_entry_safe(iter, temp, &node->children, list)
1718                         clean_tree(iter);
1719                 tree_remove_node(node);
1720         }
1721 }
1722
1723 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
1724 {
1725         if (!root_ns)
1726                 return;
1727
1728         clean_tree(&root_ns->ns.node);
1729 }
1730
1731 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1732 {
1733         struct mlx5_flow_steering *steering = dev->priv.steering;
1734
1735         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1736                 return;
1737
1738         cleanup_root_ns(steering->root_ns);
1739         cleanup_root_ns(steering->esw_egress_root_ns);
1740         cleanup_root_ns(steering->esw_ingress_root_ns);
1741         cleanup_root_ns(steering->fdb_root_ns);
1742         cleanup_root_ns(steering->sniffer_rx_root_ns);
1743         cleanup_root_ns(steering->sniffer_tx_root_ns);
1744         mlx5_cleanup_fc_stats(dev);
1745         kfree(steering);
1746 }
1747
1748 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
1749 {
1750         struct fs_prio *prio;
1751
1752         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
1753         if (!steering->sniffer_tx_root_ns)
1754                 return -ENOMEM;
1755
1756         /* Create single prio */
1757         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
1758         if (IS_ERR(prio)) {
1759                 cleanup_root_ns(steering->sniffer_tx_root_ns);
1760                 return PTR_ERR(prio);
1761         }
1762         return 0;
1763 }
1764
1765 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
1766 {
1767         struct fs_prio *prio;
1768
1769         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
1770         if (!steering->sniffer_rx_root_ns)
1771                 return -ENOMEM;
1772
1773         /* Create single prio */
1774         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
1775         if (IS_ERR(prio)) {
1776                 cleanup_root_ns(steering->sniffer_rx_root_ns);
1777                 return PTR_ERR(prio);
1778         }
1779         return 0;
1780 }
1781
1782 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
1783 {
1784         struct fs_prio *prio;
1785
1786         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
1787         if (!steering->fdb_root_ns)
1788                 return -ENOMEM;
1789
1790         prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
1791         if (IS_ERR(prio))
1792                 goto out_err;
1793
1794         prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
1795         if (IS_ERR(prio))
1796                 goto out_err;
1797
1798         set_prio_attrs(steering->fdb_root_ns);
1799         return 0;
1800
1801 out_err:
1802         cleanup_root_ns(steering->fdb_root_ns);
1803         steering->fdb_root_ns = NULL;
1804         return PTR_ERR(prio);
1805 }
1806
1807 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
1808 {
1809         struct fs_prio *prio;
1810
1811         steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
1812         if (!steering->esw_egress_root_ns)
1813                 return -ENOMEM;
1814
1815         /* create 1 prio*/
1816         prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
1817                               MLX5_TOTAL_VPORTS(steering->dev));
1818         return PTR_ERR_OR_ZERO(prio);
1819 }
1820
1821 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
1822 {
1823         struct fs_prio *prio;
1824
1825         steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
1826         if (!steering->esw_ingress_root_ns)
1827                 return -ENOMEM;
1828
1829         /* create 1 prio*/
1830         prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
1831                               MLX5_TOTAL_VPORTS(steering->dev));
1832         return PTR_ERR_OR_ZERO(prio);
1833 }
1834
1835 int mlx5_init_fs(struct mlx5_core_dev *dev)
1836 {
1837         struct mlx5_flow_steering *steering;
1838         int err = 0;
1839
1840         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1841                 return 0;
1842
1843         err = mlx5_init_fc_stats(dev);
1844         if (err)
1845                 return err;
1846
1847         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
1848         if (!steering)
1849                 return -ENOMEM;
1850         steering->dev = dev;
1851         dev->priv.steering = steering;
1852
1853         if (MLX5_CAP_GEN(dev, nic_flow_table) &&
1854             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
1855                 err = init_root_ns(steering);
1856                 if (err)
1857                         goto err;
1858         }
1859
1860         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1861                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
1862                         err = init_fdb_root_ns(steering);
1863                         if (err)
1864                                 goto err;
1865                 }
1866                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
1867                         err = init_egress_acl_root_ns(steering);
1868                         if (err)
1869                                 goto err;
1870                 }
1871                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
1872                         err = init_ingress_acl_root_ns(steering);
1873                         if (err)
1874                                 goto err;
1875                 }
1876         }
1877
1878         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
1879                 err = init_sniffer_rx_root_ns(steering);
1880                 if (err)
1881                         goto err;
1882         }
1883
1884         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
1885                 err = init_sniffer_tx_root_ns(steering);
1886                 if (err)
1887                         goto err;
1888         }
1889
1890         return 0;
1891 err:
1892         mlx5_cleanup_fs(dev);
1893         return err;
1894 }