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