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 LEFTOVERS_NUM_LEVELS 1
  71 #define LEFTOVERS_NUM_PRIOS 1
  72
  73 #define BY_PASS_PRIO_NUM_LEVELS 1
  74 #define BY_PASS_MIN_LEVEL (KERNEL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  75                            LEFTOVERS_NUM_PRIOS)
  76
  77 /* Vlan, mac, ttc, aRFS */
  78 #define KERNEL_NIC_PRIO_NUM_LEVELS 4
  79 #define KERNEL_NIC_NUM_PRIOS 1
  80 /* One more level for tc */
  81 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
  82
  83 #define ANCHOR_NUM_LEVELS 1
  84 #define ANCHOR_NUM_PRIOS 1
  85 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
  86
  87 #define OFFLOADS_MAX_FT 1
  88 #define OFFLOADS_NUM_PRIOS 1
  89 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
  90
  91 struct node_caps {
  92         size_t  arr_sz;
  93         long    *caps;
  94 };
  95 static struct init_tree_node {
  96         enum fs_node_type       type;
  97         struct init_tree_node *children;
  98         int ar_size;
  99         struct node_caps caps;
 100         int min_ft_level;
 101         int num_leaf_prios;
 102         int prio;
 103         int num_levels;
 104 } root_fs = {
 105         .type = FS_TYPE_NAMESPACE,
 106         .ar_size = 5,
 107         .children = (struct init_tree_node[]) {
 108                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 109                          FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en),
 110                                           FS_CAP(flow_table_properties_nic_receive.modify_root),
 111                                           FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode),
 112                                           FS_CAP(flow_table_properties_nic_receive.flow_table_modify)),
 113                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 114                                                   BY_PASS_PRIO_NUM_LEVELS))),
 115                 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
 116                          ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
 117
 118                 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 119                          ADD_NS(ADD_MULTIPLE_PRIO(1, 1),
 120                                 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 121                                                   KERNEL_NIC_PRIO_NUM_LEVELS))),
 122                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 123                          FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en),
 124                                           FS_CAP(flow_table_properties_nic_receive.modify_root),
 125                                           FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode),
 126                                           FS_CAP(flow_table_properties_nic_receive.flow_table_modify)),
 127                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
 128                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 129                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
 130         }
 131 };
 132
 133 enum fs_i_mutex_lock_class {
 134         FS_MUTEX_GRANDPARENT,
 135         FS_MUTEX_PARENT,
 136         FS_MUTEX_CHILD
 137 };
 138
 139 static void del_rule(struct fs_node *node);
 140 static void del_flow_table(struct fs_node *node);
 141 static void del_flow_group(struct fs_node *node);
 142 static void del_fte(struct fs_node *node);
 143
 144 static void tree_init_node(struct fs_node *node,
 145                            unsigned int refcount,
 146                            void (*remove_func)(struct fs_node *))
 147 {
 148         atomic_set(&node->refcount, refcount);
 149         INIT_LIST_HEAD(&node->list);
 150         INIT_LIST_HEAD(&node->children);
 151         mutex_init(&node->lock);
 152         node->remove_func = remove_func;
 153 }
 154
 155 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 156 {
 157         if (parent)
 158                 atomic_inc(&parent->refcount);
 159         node->parent = parent;
 160
 161         /* Parent is the root */
 162         if (!parent)
 163                 node->root = node;
 164         else
 165                 node->root = parent->root;
 166 }
 167
 168 static void tree_get_node(struct fs_node *node)
 169 {
 170         atomic_inc(&node->refcount);
 171 }
 172
 173 static void nested_lock_ref_node(struct fs_node *node,
 174                                  enum fs_i_mutex_lock_class class)
 175 {
 176         if (node) {
 177                 mutex_lock_nested(&node->lock, class);
 178                 atomic_inc(&node->refcount);
 179         }
 180 }
 181
 182 static void lock_ref_node(struct fs_node *node)
 183 {
 184         if (node) {
 185                 mutex_lock(&node->lock);
 186                 atomic_inc(&node->refcount);
 187         }
 188 }
 189
 190 static void unlock_ref_node(struct fs_node *node)
 191 {
 192         if (node) {
 193                 atomic_dec(&node->refcount);
 194                 mutex_unlock(&node->lock);
 195         }
 196 }
 197
 198 static void tree_put_node(struct fs_node *node)
 199 {
 200         struct fs_node *parent_node = node->parent;
 201
 202         lock_ref_node(parent_node);
 203         if (atomic_dec_and_test(&node->refcount)) {
 204                 if (parent_node)
 205                         list_del_init(&node->list);
 206                 if (node->remove_func)
 207                         node->remove_func(node);
 208                 kfree(node);
 209                 node = NULL;
 210         }
 211         unlock_ref_node(parent_node);
 212         if (!node && parent_node)
 213                 tree_put_node(parent_node);
 214 }
 215
 216 static int tree_remove_node(struct fs_node *node)
 217 {
 218         if (atomic_read(&node->refcount) > 1) {
 219                 atomic_dec(&node->refcount);
 220                 return -EEXIST;
 221         }
 222         tree_put_node(node);
 223         return 0;
 224 }
 225
 226 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 227                                  unsigned int prio)
 228 {
 229         struct fs_prio *iter_prio;
 230
 231         fs_for_each_prio(iter_prio, ns) {
 232                 if (iter_prio->prio == prio)
 233                         return iter_prio;
 234         }
 235
 236         return NULL;
 237 }
 238
 239 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 240 {
 241         unsigned int i;
 242
 243         for (i = 0; i < size; i++, mask++, val1++, val2++)
 244                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 245                     ((*(u8 *)val2) & (*(u8 *)mask)))
 246                         return false;
 247
 248         return true;
 249 }
 250
 251 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 252                                 void *fte_param1, void *fte_param2)
 253 {
 254         if (mask->match_criteria_enable &
 255             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 256                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 257                                                 fte_param1, outer_headers);
 258                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 259                                                 fte_param2, outer_headers);
 260                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 261                                               mask->match_criteria, outer_headers);
 262
 263                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 264                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 265                         return false;
 266         }
 267
 268         if (mask->match_criteria_enable &
 269             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 270                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 271                                                 fte_param1, misc_parameters);
 272                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 273                                                 fte_param2, misc_parameters);
 274                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 275                                           mask->match_criteria, misc_parameters);
 276
 277                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 278                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 279                         return false;
 280         }
 281
 282         if (mask->match_criteria_enable &
 283             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 284                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 285                                                 fte_param1, inner_headers);
 286                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 287                                                 fte_param2, inner_headers);
 288                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 289                                           mask->match_criteria, inner_headers);
 290
 291                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 292                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 293                         return false;
 294         }
 295         return true;
 296 }
 297
 298 static bool compare_match_criteria(u8 match_criteria_enable1,
 299                                    u8 match_criteria_enable2,
 300                                    void *mask1, void *mask2)
 301 {
 302         return match_criteria_enable1 == match_criteria_enable2 &&
 303                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 304 }
 305
 306 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 307 {
 308         struct fs_node *root;
 309         struct mlx5_flow_namespace *ns;
 310
 311         root = node->root;
 312
 313         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 314                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 315                 return NULL;
 316         }
 317
 318         ns = container_of(root, struct mlx5_flow_namespace, node);
 319         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 320 }
 321
 322 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 323 {
 324         struct mlx5_flow_root_namespace *root = find_root(node);
 325
 326         if (root)
 327                 return root->dev;
 328         return NULL;
 329 }
 330
 331 static void del_flow_table(struct fs_node *node)
 332 {
 333         struct mlx5_flow_table *ft;
 334         struct mlx5_core_dev *dev;
 335         struct fs_prio *prio;
 336         int err;
 337
 338         fs_get_obj(ft, node);
 339         dev = get_dev(&ft->node);
 340
 341         err = mlx5_cmd_destroy_flow_table(dev, ft);
 342         if (err)
 343                 pr_warn("flow steering can't destroy ft\n");
 344         fs_get_obj(prio, ft->node.parent);
 345         prio->num_ft--;
 346 }
 347
 348 static void del_rule(struct fs_node *node)
 349 {
 350         struct mlx5_flow_rule *rule;
 351         struct mlx5_flow_table *ft;
 352         struct mlx5_flow_group *fg;
 353         struct fs_fte *fte;
 354         u32     *match_value;
 355         int modify_mask;
 356         struct mlx5_core_dev *dev = get_dev(node);
 357         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 358         int err;
 359
 360         match_value = mlx5_vzalloc(match_len);
 361         if (!match_value) {
 362                 pr_warn("failed to allocate inbox\n");
 363                 return;
 364         }
 365
 366         fs_get_obj(rule, node);
 367         fs_get_obj(fte, rule->node.parent);
 368         fs_get_obj(fg, fte->node.parent);
 369         memcpy(match_value, fte->val, sizeof(fte->val));
 370         fs_get_obj(ft, fg->node.parent);
 371         list_del(&rule->node.list);
 372         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 373                 mutex_lock(&rule->dest_attr.ft->lock);
 374                 list_del(&rule->next_ft);
 375                 mutex_unlock(&rule->dest_attr.ft->lock);
 376         }
 377         if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 378             --fte->dests_size) {
 379                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST),
 380                 err = mlx5_cmd_update_fte(dev, ft,
 381                                           fg->id,
 382                                           modify_mask,
 383                                           fte);
 384                 if (err)
 385                         pr_warn("%s can't del rule fg id=%d fte_index=%d\n",
 386                                 __func__, fg->id, fte->index);
 387         }
 388         kvfree(match_value);
 389 }
 390
 391 static void del_fte(struct fs_node *node)
 392 {
 393         struct mlx5_flow_table *ft;
 394         struct mlx5_flow_group *fg;
 395         struct mlx5_core_dev *dev;
 396         struct fs_fte *fte;
 397         int err;
 398
 399         fs_get_obj(fte, node);
 400         fs_get_obj(fg, fte->node.parent);
 401         fs_get_obj(ft, fg->node.parent);
 402
 403         dev = get_dev(&ft->node);
 404         err = mlx5_cmd_delete_fte(dev, ft,
 405                                   fte->index);
 406         if (err)
 407                 pr_warn("flow steering can't delete fte in index %d of flow group id %d\n",
 408                         fte->index, fg->id);
 409
 410         fte->status = 0;
 411         fg->num_ftes--;
 412 }
 413
 414 static void del_flow_group(struct fs_node *node)
 415 {
 416         struct mlx5_flow_group *fg;
 417         struct mlx5_flow_table *ft;
 418         struct mlx5_core_dev *dev;
 419
 420         fs_get_obj(fg, node);
 421         fs_get_obj(ft, fg->node.parent);
 422         dev = get_dev(&ft->node);
 423
 424         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 425                 pr_warn("flow steering can't destroy fg %d of ft %d\n",
 426                         fg->id, ft->id);
 427 }
 428
 429 static struct fs_fte *alloc_fte(u8 action,
 430                                 u32 flow_tag,
 431                                 u32 *match_value,
 432                                 unsigned int index)
 433 {
 434         struct fs_fte *fte;
 435
 436         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 437         if (!fte)
 438                 return ERR_PTR(-ENOMEM);
 439
 440         memcpy(fte->val, match_value, sizeof(fte->val));
 441         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 442         fte->flow_tag = flow_tag;
 443         fte->index = index;
 444         fte->action = action;
 445
 446         return fte;
 447 }
 448
 449 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 450 {
 451         struct mlx5_flow_group *fg;
 452         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 453                                             create_fg_in, match_criteria);
 454         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 455                                             create_fg_in,
 456                                             match_criteria_enable);
 457         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 458         if (!fg)
 459                 return ERR_PTR(-ENOMEM);
 460
 461         fg->mask.match_criteria_enable = match_criteria_enable;
 462         memcpy(&fg->mask.match_criteria, match_criteria,
 463                sizeof(fg->mask.match_criteria));
 464         fg->node.type =  FS_TYPE_FLOW_GROUP;
 465         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 466                                    start_flow_index);
 467         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 468                                 end_flow_index) - fg->start_index + 1;
 469         return fg;
 470 }
 471
 472 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 473                                                 enum fs_flow_table_type table_type)
 474 {
 475         struct mlx5_flow_table *ft;
 476
 477         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 478         if (!ft)
 479                 return NULL;
 480
 481         ft->level = level;
 482         ft->node.type = FS_TYPE_FLOW_TABLE;
 483         ft->type = table_type;
 484         ft->vport = vport;
 485         ft->max_fte = max_fte;
 486         INIT_LIST_HEAD(&ft->fwd_rules);
 487         mutex_init(&ft->lock);
 488
 489         return ft;
 490 }
 491
 492 /* If reverse is false, then we search for the first flow table in the
 493  * root sub-tree from start(closest from right), else we search for the
 494  * last flow table in the root sub-tree till start(closest from left).
 495  */
 496 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 497                                                          struct list_head *start,
 498                                                          bool reverse)
 499 {
 500 #define list_advance_entry(pos, reverse)                \
 501         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 502
 503 #define list_for_each_advance_continue(pos, head, reverse)      \
 504         for (pos = list_advance_entry(pos, reverse);            \
 505              &pos->list != (head);                              \
 506              pos = list_advance_entry(pos, reverse))
 507
 508         struct fs_node *iter = list_entry(start, struct fs_node, list);
 509         struct mlx5_flow_table *ft = NULL;
 510
 511         if (!root)
 512                 return NULL;
 513
 514         list_for_each_advance_continue(iter, &root->children, reverse) {
 515                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 516                         fs_get_obj(ft, iter);
 517                         return ft;
 518                 }
 519                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 520                 if (ft)
 521                         return ft;
 522         }
 523
 524         return ft;
 525 }
 526
 527 /* If reverse if false then return the first flow table in next priority of
 528  * prio in the tree, else return the last flow table in the previous priority
 529  * of prio in the tree.
 530  */
 531 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 532 {
 533         struct mlx5_flow_table *ft = NULL;
 534         struct fs_node *curr_node;
 535         struct fs_node *parent;
 536
 537         parent = prio->node.parent;
 538         curr_node = &prio->node;
 539         while (!ft && parent) {
 540                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 541                 curr_node = parent;
 542                 parent = curr_node->parent;
 543         }
 544         return ft;
 545 }
 546
 547 /* Assuming all the tree is locked by mutex chain lock */
 548 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 549 {
 550         return find_closest_ft(prio, false);
 551 }
 552
 553 /* Assuming all the tree is locked by mutex chain lock */
 554 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 555 {
 556         return find_closest_ft(prio, true);
 557 }
 558
 559 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 560                                struct fs_prio *prio,
 561                                struct mlx5_flow_table *ft)
 562 {
 563         struct mlx5_flow_table *iter;
 564         int i = 0;
 565         int err;
 566
 567         fs_for_each_ft(iter, prio) {
 568                 i++;
 569                 err = mlx5_cmd_modify_flow_table(dev,
 570                                                  iter,
 571                                                  ft);
 572                 if (err) {
 573                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 574                                        iter->id);
 575                         /* The driver is out of sync with the FW */
 576                         if (i > 1)
 577                                 WARN_ON(true);
 578                         return err;
 579                 }
 580         }
 581         return 0;
 582 }
 583
 584 /* Connect flow tables from previous priority of prio to ft */
 585 static int connect_prev_fts(struct mlx5_core_dev *dev,
 586                             struct mlx5_flow_table *ft,
 587                             struct fs_prio *prio)
 588 {
 589         struct mlx5_flow_table *prev_ft;
 590
 591         prev_ft = find_prev_chained_ft(prio);
 592         if (prev_ft) {
 593                 struct fs_prio *prev_prio;
 594
 595                 fs_get_obj(prev_prio, prev_ft->node.parent);
 596                 return connect_fts_in_prio(dev, prev_prio, ft);
 597         }
 598         return 0;
 599 }
 600
 601 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 602                                  *prio)
 603 {
 604         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 605         int min_level = INT_MAX;
 606         int err;
 607
 608         if (root->root_ft)
 609                 min_level = root->root_ft->level;
 610
 611         if (ft->level >= min_level)
 612                 return 0;
 613
 614         err = mlx5_cmd_update_root_ft(root->dev, ft);
 615         if (err)
 616                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 617                                ft->id);
 618         else
 619                 root->root_ft = ft;
 620
 621         return err;
 622 }
 623
 624 int mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 625                                  struct mlx5_flow_destination *dest)
 626 {
 627         struct mlx5_flow_table *ft;
 628         struct mlx5_flow_group *fg;
 629         struct fs_fte *fte;
 630         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 631         int err = 0;
 632
 633         fs_get_obj(fte, rule->node.parent);
 634         if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 635                 return -EINVAL;
 636         lock_ref_node(&fte->node);
 637         fs_get_obj(fg, fte->node.parent);
 638         fs_get_obj(ft, fg->node.parent);
 639
 640         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 641         err = mlx5_cmd_update_fte(get_dev(&ft->node),
 642                                   ft, fg->id,
 643                                   modify_mask,
 644                                   fte);
 645         unlock_ref_node(&fte->node);
 646
 647         return err;
 648 }
 649
 650 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 651 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 652                              struct mlx5_flow_table *new_next_ft,
 653                              struct mlx5_flow_table *old_next_ft)
 654 {
 655         struct mlx5_flow_destination dest;
 656         struct mlx5_flow_rule *iter;
 657         int err = 0;
 658
 659         /* new_next_ft and old_next_ft could be NULL only
 660          * when we create/destroy the anchor flow table.
 661          */
 662         if (!new_next_ft || !old_next_ft)
 663                 return 0;
 664
 665         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 666         dest.ft = new_next_ft;
 667
 668         mutex_lock(&old_next_ft->lock);
 669         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 670         mutex_unlock(&old_next_ft->lock);
 671         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 672                 err = mlx5_modify_rule_destination(iter, &dest);
 673                 if (err)
 674                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 675                                new_next_ft->id);
 676         }
 677         return 0;
 678 }
 679
 680 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 681                               struct fs_prio *prio)
 682 {
 683         struct mlx5_flow_table *next_ft;
 684         int err = 0;
 685
 686         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 687
 688         if (list_empty(&prio->node.children)) {
 689                 err = connect_prev_fts(dev, ft, prio);
 690                 if (err)
 691                         return err;
 692
 693                 next_ft = find_next_chained_ft(prio);
 694                 err = connect_fwd_rules(dev, ft, next_ft);
 695                 if (err)
 696                         return err;
 697         }
 698
 699         if (MLX5_CAP_FLOWTABLE(dev,
 700                                flow_table_properties_nic_receive.modify_root))
 701                 err = update_root_ft_create(ft, prio);
 702         return err;
 703 }
 704
 705 static void list_add_flow_table(struct mlx5_flow_table *ft,
 706                                 struct fs_prio *prio)
 707 {
 708         struct list_head *prev = &prio->node.children;
 709         struct mlx5_flow_table *iter;
 710
 711         fs_for_each_ft(iter, prio) {
 712                 if (iter->level > ft->level)
 713                         break;
 714                 prev = &iter->node.list;
 715         }
 716         list_add(&ft->node.list, prev);
 717 }
 718
 719 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 720                                                         u16 vport, int prio,
 721                                                         int max_fte, u32 level)
 722 {
 723         struct mlx5_flow_table *next_ft = NULL;
 724         struct mlx5_flow_table *ft;
 725         int err;
 726         int log_table_sz;
 727         struct mlx5_flow_root_namespace *root =
 728                 find_root(&ns->node);
 729         struct fs_prio *fs_prio = NULL;
 730
 731         if (!root) {
 732                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 733                 return ERR_PTR(-ENODEV);
 734         }
 735
 736         mutex_lock(&root->chain_lock);
 737         fs_prio = find_prio(ns, prio);
 738         if (!fs_prio) {
 739                 err = -EINVAL;
 740                 goto unlock_root;
 741         }
 742         if (level >= fs_prio->num_levels) {
 743                 err = -ENOSPC;
 744                 goto unlock_root;
 745         }
 746         /* The level is related to the
 747          * priority level range.
 748          */
 749         level += fs_prio->start_level;
 750         ft = alloc_flow_table(level,
 751                               vport,
 752                               roundup_pow_of_two(max_fte),
 753                               root->table_type);
 754         if (!ft) {
 755                 err = -ENOMEM;
 756                 goto unlock_root;
 757         }
 758
 759         tree_init_node(&ft->node, 1, del_flow_table);
 760         log_table_sz = ilog2(ft->max_fte);
 761         next_ft = find_next_chained_ft(fs_prio);
 762         err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->type, ft->level,
 763                                          log_table_sz, next_ft, &ft->id);
 764         if (err)
 765                 goto free_ft;
 766
 767         err = connect_flow_table(root->dev, ft, fs_prio);
 768         if (err)
 769                 goto destroy_ft;
 770         lock_ref_node(&fs_prio->node);
 771         tree_add_node(&ft->node, &fs_prio->node);
 772         list_add_flow_table(ft, fs_prio);
 773         fs_prio->num_ft++;
 774         unlock_ref_node(&fs_prio->node);
 775         mutex_unlock(&root->chain_lock);
 776         return ft;
 777 destroy_ft:
 778         mlx5_cmd_destroy_flow_table(root->dev, ft);
 779 free_ft:
 780         kfree(ft);
 781 unlock_root:
 782         mutex_unlock(&root->chain_lock);
 783         return ERR_PTR(err);
 784 }
 785
 786 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 787                                                int prio, int max_fte,
 788                                                u32 level)
 789 {
 790         return __mlx5_create_flow_table(ns, 0, prio, max_fte, level);
 791 }
 792
 793 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
 794                                                      int prio, int max_fte,
 795                                                      u32 level, u16 vport)
 796 {
 797         return __mlx5_create_flow_table(ns, vport, prio, max_fte, level);
 798 }
 799
 800 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 801                                                             int prio,
 802                                                             int num_flow_table_entries,
 803                                                             int max_num_groups,
 804                                                             u32 level)
 805 {
 806         struct mlx5_flow_table *ft;
 807
 808         if (max_num_groups > num_flow_table_entries)
 809                 return ERR_PTR(-EINVAL);
 810
 811         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries, level);
 812         if (IS_ERR(ft))
 813                 return ft;
 814
 815         ft->autogroup.active = true;
 816         ft->autogroup.required_groups = max_num_groups;
 817
 818         return ft;
 819 }
 820 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 821
 822 /* Flow table should be locked */
 823 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 824                                                         u32 *fg_in,
 825                                                         struct list_head
 826                                                         *prev_fg,
 827                                                         bool is_auto_fg)
 828 {
 829         struct mlx5_flow_group *fg;
 830         struct mlx5_core_dev *dev = get_dev(&ft->node);
 831         int err;
 832
 833         if (!dev)
 834                 return ERR_PTR(-ENODEV);
 835
 836         fg = alloc_flow_group(fg_in);
 837         if (IS_ERR(fg))
 838                 return fg;
 839
 840         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 841         if (err) {
 842                 kfree(fg);
 843                 return ERR_PTR(err);
 844         }
 845
 846         if (ft->autogroup.active)
 847                 ft->autogroup.num_groups++;
 848         /* Add node to tree */
 849         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 850         tree_add_node(&fg->node, &ft->node);
 851         /* Add node to group list */
 852         list_add(&fg->node.list, ft->node.children.prev);
 853
 854         return fg;
 855 }
 856
 857 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 858                                                u32 *fg_in)
 859 {
 860         struct mlx5_flow_group *fg;
 861
 862         if (ft->autogroup.active)
 863                 return ERR_PTR(-EPERM);
 864
 865         lock_ref_node(&ft->node);
 866         fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
 867         unlock_ref_node(&ft->node);
 868
 869         return fg;
 870 }
 871
 872 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 873 {
 874         struct mlx5_flow_rule *rule;
 875
 876         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 877         if (!rule)
 878                 return NULL;
 879
 880         INIT_LIST_HEAD(&rule->next_ft);
 881         rule->node.type = FS_TYPE_FLOW_DEST;
 882         if (dest)
 883                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 884
 885         return rule;
 886 }
 887
 888 /* fte should not be deleted while calling this function */
 889 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
 890                                            struct mlx5_flow_group *fg,
 891                                            struct mlx5_flow_destination *dest)
 892 {
 893         struct mlx5_flow_table *ft;
 894         struct mlx5_flow_rule *rule;
 895         int modify_mask = 0;
 896         int err;
 897
 898         rule = alloc_rule(dest);
 899         if (!rule)
 900                 return ERR_PTR(-ENOMEM);
 901
 902         fs_get_obj(ft, fg->node.parent);
 903         /* Add dest to dests list- we need flow tables to be in the
 904          * end of the list for forward to next prio rules.
 905          */
 906         tree_init_node(&rule->node, 1, del_rule);
 907         if (dest && dest->type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
 908                 list_add(&rule->node.list, &fte->node.children);
 909         else
 910                 list_add_tail(&rule->node.list, &fte->node.children);
 911         if (dest) {
 912                 fte->dests_size++;
 913
 914                 modify_mask |= dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER ?
 915                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS) :
 916                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 917         }
 918
 919         if (fte->dests_size == 1 || !dest)
 920                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 921                                           ft, fg->id, fte);
 922         else
 923                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 924                                           ft, fg->id, modify_mask, fte);
 925         if (err)
 926                 goto free_rule;
 927
 928         fte->status |= FS_FTE_STATUS_EXISTING;
 929
 930         return rule;
 931
 932 free_rule:
 933         list_del(&rule->node.list);
 934         kfree(rule);
 935         if (dest)
 936                 fte->dests_size--;
 937         return ERR_PTR(err);
 938 }
 939
 940 /* Assumed fg is locked */
 941 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 942                                        struct list_head **prev)
 943 {
 944         struct fs_fte *fte;
 945         unsigned int start = fg->start_index;
 946
 947         if (prev)
 948                 *prev = &fg->node.children;
 949
 950         /* assumed list is sorted by index */
 951         fs_for_each_fte(fte, fg) {
 952                 if (fte->index != start)
 953                         return start;
 954                 start++;
 955                 if (prev)
 956                         *prev = &fte->node.list;
 957         }
 958
 959         return start;
 960 }
 961
 962 /* prev is output, prev->next = new_fte */
 963 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
 964                                  u32 *match_value,
 965                                  u8 action,
 966                                  u32 flow_tag,
 967                                  struct list_head **prev)
 968 {
 969         struct fs_fte *fte;
 970         int index;
 971
 972         index = get_free_fte_index(fg, prev);
 973         fte = alloc_fte(action, flow_tag, match_value, index);
 974         if (IS_ERR(fte))
 975                 return fte;
 976
 977         return fte;
 978 }
 979
 980 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
 981                                                 u8 match_criteria_enable,
 982                                                 u32 *match_criteria)
 983 {
 984         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 985         struct list_head *prev = &ft->node.children;
 986         unsigned int candidate_index = 0;
 987         struct mlx5_flow_group *fg;
 988         void *match_criteria_addr;
 989         unsigned int group_size = 0;
 990         u32 *in;
 991
 992         if (!ft->autogroup.active)
 993                 return ERR_PTR(-ENOENT);
 994
 995         in = mlx5_vzalloc(inlen);
 996         if (!in)
 997                 return ERR_PTR(-ENOMEM);
 998
 999         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1000                 /* We save place for flow groups in addition to max types */
1001                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1002
1003         /*  ft->max_fte == ft->autogroup.max_types */
1004         if (group_size == 0)
1005                 group_size = 1;
1006
1007         /* sorted by start_index */
1008         fs_for_each_fg(fg, ft) {
1009                 if (candidate_index + group_size > fg->start_index)
1010                         candidate_index = fg->start_index + fg->max_ftes;
1011                 else
1012                         break;
1013                 prev = &fg->node.list;
1014         }
1015
1016         if (candidate_index + group_size > ft->max_fte) {
1017                 fg = ERR_PTR(-ENOSPC);
1018                 goto out;
1019         }
1020
1021         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1022                  match_criteria_enable);
1023         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1024         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
1025                  group_size - 1);
1026         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1027                                            in, match_criteria);
1028         memcpy(match_criteria_addr, match_criteria,
1029                MLX5_ST_SZ_BYTES(fte_match_param));
1030
1031         fg = create_flow_group_common(ft, in, prev, true);
1032 out:
1033         kvfree(in);
1034         return fg;
1035 }
1036
1037 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1038                                              struct mlx5_flow_destination *dest)
1039 {
1040         struct mlx5_flow_rule *rule;
1041
1042         list_for_each_entry(rule, &fte->node.children, node.list) {
1043                 if (rule->dest_attr.type == dest->type) {
1044                         if ((dest->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1045                              dest->vport_num == rule->dest_attr.vport_num) ||
1046                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1047                              dest->ft == rule->dest_attr.ft) ||
1048                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1049                              dest->tir_num == rule->dest_attr.tir_num))
1050                                 return rule;
1051                 }
1052         }
1053         return NULL;
1054 }
1055
1056 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
1057                                           u32 *match_value,
1058                                           u8 action,
1059                                           u32 flow_tag,
1060                                           struct mlx5_flow_destination *dest)
1061 {
1062         struct fs_fte *fte;
1063         struct mlx5_flow_rule *rule;
1064         struct mlx5_flow_table *ft;
1065         struct list_head *prev;
1066
1067         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1068         fs_for_each_fte(fte, fg) {
1069                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1070                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1071                     action == fte->action && flow_tag == fte->flow_tag) {
1072                         rule = find_flow_rule(fte, dest);
1073                         if (rule) {
1074                                 atomic_inc(&rule->node.refcount);
1075                                 unlock_ref_node(&fte->node);
1076                                 unlock_ref_node(&fg->node);
1077                                 return rule;
1078                         }
1079                         rule = add_rule_fte(fte, fg, dest);
1080                         unlock_ref_node(&fte->node);
1081                         if (IS_ERR(rule))
1082                                 goto unlock_fg;
1083                         else
1084                                 goto add_rule;
1085                 }
1086                 unlock_ref_node(&fte->node);
1087         }
1088         fs_get_obj(ft, fg->node.parent);
1089         if (fg->num_ftes >= fg->max_ftes) {
1090                 rule = ERR_PTR(-ENOSPC);
1091                 goto unlock_fg;
1092         }
1093
1094         fte = create_fte(fg, match_value, action, flow_tag, &prev);
1095         if (IS_ERR(fte)) {
1096                 rule = (void *)fte;
1097                 goto unlock_fg;
1098         }
1099         tree_init_node(&fte->node, 0, del_fte);
1100         rule = add_rule_fte(fte, fg, dest);
1101         if (IS_ERR(rule)) {
1102                 kfree(fte);
1103                 goto unlock_fg;
1104         }
1105
1106         fg->num_ftes++;
1107
1108         tree_add_node(&fte->node, &fg->node);
1109         list_add(&fte->node.list, prev);
1110 add_rule:
1111         tree_add_node(&rule->node, &fte->node);
1112 unlock_fg:
1113         unlock_ref_node(&fg->node);
1114         return rule;
1115 }
1116
1117 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_rule *rule)
1118 {
1119         struct mlx5_flow_rule *dst;
1120         struct fs_fte *fte;
1121
1122         fs_get_obj(fte, rule->node.parent);
1123
1124         fs_for_each_dst(dst, fte) {
1125                 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1126                         return dst->dest_attr.counter;
1127         }
1128
1129         return NULL;
1130 }
1131
1132 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1133 {
1134         if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1135                 return !counter;
1136
1137         if (!counter)
1138                 return false;
1139
1140         /* Hardware support counter for a drop action only */
1141         return action == (MLX5_FLOW_CONTEXT_ACTION_DROP | MLX5_FLOW_CONTEXT_ACTION_COUNT);
1142 }
1143
1144 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1145                           u32 action,
1146                           struct mlx5_flow_table *ft)
1147 {
1148         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1149                 return counter_is_valid(dest->counter, action);
1150
1151         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1152                 return true;
1153
1154         if (!dest || ((dest->type ==
1155             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1156             (dest->ft->level <= ft->level)))
1157                 return false;
1158         return true;
1159 }
1160
1161 static struct mlx5_flow_rule *
1162 _mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1163                    struct mlx5_flow_spec *spec,
1164                     u32 action,
1165                     u32 flow_tag,
1166                     struct mlx5_flow_destination *dest)
1167 {
1168         struct mlx5_flow_group *g;
1169         struct mlx5_flow_rule *rule;
1170
1171         if (!dest_is_valid(dest, action, ft))
1172                 return ERR_PTR(-EINVAL);
1173
1174         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1175         fs_for_each_fg(g, ft)
1176                 if (compare_match_criteria(g->mask.match_criteria_enable,
1177                                            spec->match_criteria_enable,
1178                                            g->mask.match_criteria,
1179                                            spec->match_criteria)) {
1180                         rule = add_rule_fg(g, spec->match_value,
1181                                            action, flow_tag, dest);
1182                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1183                                 goto unlock;
1184                 }
1185
1186         g = create_autogroup(ft, spec->match_criteria_enable,
1187                              spec->match_criteria);
1188         if (IS_ERR(g)) {
1189                 rule = (void *)g;
1190                 goto unlock;
1191         }
1192
1193         rule = add_rule_fg(g, spec->match_value,
1194                            action, flow_tag, dest);
1195         if (IS_ERR(rule)) {
1196                 /* Remove assumes refcount > 0 and autogroup creates a group
1197                  * with a refcount = 0.
1198                  */
1199                 unlock_ref_node(&ft->node);
1200                 tree_get_node(&g->node);
1201                 tree_remove_node(&g->node);
1202                 return rule;
1203         }
1204 unlock:
1205         unlock_ref_node(&ft->node);
1206         return rule;
1207 }
1208
1209 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1210 {
1211         return ((ft->type == FS_FT_NIC_RX) &&
1212                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1213 }
1214
1215 struct mlx5_flow_rule *
1216 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1217                    struct mlx5_flow_spec *spec,
1218                    u32 action,
1219                    u32 flow_tag,
1220                    struct mlx5_flow_destination *dest)
1221 {
1222         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1223         struct mlx5_flow_destination gen_dest;
1224         struct mlx5_flow_table *next_ft = NULL;
1225         struct mlx5_flow_rule *rule = NULL;
1226         u32 sw_action = action;
1227         struct fs_prio *prio;
1228
1229         fs_get_obj(prio, ft->node.parent);
1230         if (action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1231                 if (!fwd_next_prio_supported(ft))
1232                         return ERR_PTR(-EOPNOTSUPP);
1233                 if (dest)
1234                         return ERR_PTR(-EINVAL);
1235                 mutex_lock(&root->chain_lock);
1236                 next_ft = find_next_chained_ft(prio);
1237                 if (next_ft) {
1238                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1239                         gen_dest.ft = next_ft;
1240                         dest = &gen_dest;
1241                         action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1242                 } else {
1243                         mutex_unlock(&root->chain_lock);
1244                         return ERR_PTR(-EOPNOTSUPP);
1245                 }
1246         }
1247
1248         rule = _mlx5_add_flow_rule(ft, spec, action, flow_tag, dest);
1249
1250         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1251                 if (!IS_ERR_OR_NULL(rule) &&
1252                     (list_empty(&rule->next_ft))) {
1253                         mutex_lock(&next_ft->lock);
1254                         list_add(&rule->next_ft, &next_ft->fwd_rules);
1255                         mutex_unlock(&next_ft->lock);
1256                         rule->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1257                 }
1258                 mutex_unlock(&root->chain_lock);
1259         }
1260         return rule;
1261 }
1262 EXPORT_SYMBOL(mlx5_add_flow_rule);
1263
1264 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
1265 {
1266         tree_remove_node(&rule->node);
1267 }
1268 EXPORT_SYMBOL(mlx5_del_flow_rule);
1269
1270 /* Assuming prio->node.children(flow tables) is sorted by level */
1271 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1272 {
1273         struct fs_prio *prio;
1274
1275         fs_get_obj(prio, ft->node.parent);
1276
1277         if (!list_is_last(&ft->node.list, &prio->node.children))
1278                 return list_next_entry(ft, node.list);
1279         return find_next_chained_ft(prio);
1280 }
1281
1282 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1283 {
1284         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1285         struct mlx5_flow_table *new_root_ft = NULL;
1286
1287         if (root->root_ft != ft)
1288                 return 0;
1289
1290         new_root_ft = find_next_ft(ft);
1291         if (new_root_ft) {
1292                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1293
1294                 if (err) {
1295                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1296                                        ft->id);
1297                         return err;
1298                 }
1299         }
1300         root->root_ft = new_root_ft;
1301         return 0;
1302 }
1303
1304 /* Connect flow table from previous priority to
1305  * the next flow table.
1306  */
1307 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1308 {
1309         struct mlx5_core_dev *dev = get_dev(&ft->node);
1310         struct mlx5_flow_table *next_ft;
1311         struct fs_prio *prio;
1312         int err = 0;
1313
1314         err = update_root_ft_destroy(ft);
1315         if (err)
1316                 return err;
1317
1318         fs_get_obj(prio, ft->node.parent);
1319         if  (!(list_first_entry(&prio->node.children,
1320                                 struct mlx5_flow_table,
1321                                 node.list) == ft))
1322                 return 0;
1323
1324         next_ft = find_next_chained_ft(prio);
1325         err = connect_fwd_rules(dev, next_ft, ft);
1326         if (err)
1327                 return err;
1328
1329         err = connect_prev_fts(dev, next_ft, prio);
1330         if (err)
1331                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1332                                ft->id);
1333         return err;
1334 }
1335
1336 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1337 {
1338         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1339         int err = 0;
1340
1341         mutex_lock(&root->chain_lock);
1342         err = disconnect_flow_table(ft);
1343         if (err) {
1344                 mutex_unlock(&root->chain_lock);
1345                 return err;
1346         }
1347         if (tree_remove_node(&ft->node))
1348                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1349                                ft->id);
1350         mutex_unlock(&root->chain_lock);
1351
1352         return err;
1353 }
1354 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1355
1356 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1357 {
1358         if (tree_remove_node(&fg->node))
1359                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1360                                fg->id);
1361 }
1362
1363 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1364                                                     enum mlx5_flow_namespace_type type)
1365 {
1366         struct mlx5_flow_steering *steering = dev->priv.steering;
1367         struct mlx5_flow_root_namespace *root_ns;
1368         int prio;
1369         struct fs_prio *fs_prio;
1370         struct mlx5_flow_namespace *ns;
1371
1372         if (!steering)
1373                 return NULL;
1374
1375         switch (type) {
1376         case MLX5_FLOW_NAMESPACE_BYPASS:
1377         case MLX5_FLOW_NAMESPACE_OFFLOADS:
1378         case MLX5_FLOW_NAMESPACE_KERNEL:
1379         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1380         case MLX5_FLOW_NAMESPACE_ANCHOR:
1381                 prio = type;
1382                 break;
1383         case MLX5_FLOW_NAMESPACE_FDB:
1384                 if (steering->fdb_root_ns)
1385                         return &steering->fdb_root_ns->ns;
1386                 else
1387                         return NULL;
1388         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1389                 if (steering->esw_egress_root_ns)
1390                         return &steering->esw_egress_root_ns->ns;
1391                 else
1392                         return NULL;
1393         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1394                 if (steering->esw_ingress_root_ns)
1395                         return &steering->esw_ingress_root_ns->ns;
1396                 else
1397                         return NULL;
1398         default:
1399                 return NULL;
1400         }
1401
1402         root_ns = steering->root_ns;
1403         if (!root_ns)
1404                 return NULL;
1405
1406         fs_prio = find_prio(&root_ns->ns, prio);
1407         if (!fs_prio)
1408                 return NULL;
1409
1410         ns = list_first_entry(&fs_prio->node.children,
1411                               typeof(*ns),
1412                               node.list);
1413
1414         return ns;
1415 }
1416 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1417
1418 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1419                                       unsigned int prio, int num_levels)
1420 {
1421         struct fs_prio *fs_prio;
1422
1423         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1424         if (!fs_prio)
1425                 return ERR_PTR(-ENOMEM);
1426
1427         fs_prio->node.type = FS_TYPE_PRIO;
1428         tree_init_node(&fs_prio->node, 1, NULL);
1429         tree_add_node(&fs_prio->node, &ns->node);
1430         fs_prio->num_levels = num_levels;
1431         fs_prio->prio = prio;
1432         list_add_tail(&fs_prio->node.list, &ns->node.children);
1433
1434         return fs_prio;
1435 }
1436
1437 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1438                                                      *ns)
1439 {
1440         ns->node.type = FS_TYPE_NAMESPACE;
1441
1442         return ns;
1443 }
1444
1445 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1446 {
1447         struct mlx5_flow_namespace      *ns;
1448
1449         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1450         if (!ns)
1451                 return ERR_PTR(-ENOMEM);
1452
1453         fs_init_namespace(ns);
1454         tree_init_node(&ns->node, 1, NULL);
1455         tree_add_node(&ns->node, &prio->node);
1456         list_add_tail(&ns->node.list, &prio->node.children);
1457
1458         return ns;
1459 }
1460
1461 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1462                              struct init_tree_node *prio_metadata)
1463 {
1464         struct fs_prio *fs_prio;
1465         int i;
1466
1467         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1468                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1469                 if (IS_ERR(fs_prio))
1470                         return PTR_ERR(fs_prio);
1471         }
1472         return 0;
1473 }
1474
1475 #define FLOW_TABLE_BIT_SZ 1
1476 #define GET_FLOW_TABLE_CAP(dev, offset) \
1477         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1478                         offset / 32)) >>                                        \
1479           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1480 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1481 {
1482         int i;
1483
1484         for (i = 0; i < caps->arr_sz; i++) {
1485                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1486                         return false;
1487         }
1488         return true;
1489 }
1490
1491 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
1492                                     struct init_tree_node *init_node,
1493                                     struct fs_node *fs_parent_node,
1494                                     struct init_tree_node *init_parent_node,
1495                                     int prio)
1496 {
1497         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
1498                                               flow_table_properties_nic_receive.
1499                                               max_ft_level);
1500         struct mlx5_flow_namespace *fs_ns;
1501         struct fs_prio *fs_prio;
1502         struct fs_node *base;
1503         int i;
1504         int err;
1505
1506         if (init_node->type == FS_TYPE_PRIO) {
1507                 if ((init_node->min_ft_level > max_ft_level) ||
1508                     !has_required_caps(steering->dev, &init_node->caps))
1509                         return 0;
1510
1511                 fs_get_obj(fs_ns, fs_parent_node);
1512                 if (init_node->num_leaf_prios)
1513                         return create_leaf_prios(fs_ns, prio, init_node);
1514                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1515                 if (IS_ERR(fs_prio))
1516                         return PTR_ERR(fs_prio);
1517                 base = &fs_prio->node;
1518         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1519                 fs_get_obj(fs_prio, fs_parent_node);
1520                 fs_ns = fs_create_namespace(fs_prio);
1521                 if (IS_ERR(fs_ns))
1522                         return PTR_ERR(fs_ns);
1523                 base = &fs_ns->node;
1524         } else {
1525                 return -EINVAL;
1526         }
1527         prio = 0;
1528         for (i = 0; i < init_node->ar_size; i++) {
1529                 err = init_root_tree_recursive(steering, &init_node->children[i],
1530                                                base, init_node, prio);
1531                 if (err)
1532                         return err;
1533                 if (init_node->children[i].type == FS_TYPE_PRIO &&
1534                     init_node->children[i].num_leaf_prios) {
1535                         prio += init_node->children[i].num_leaf_prios;
1536                 }
1537         }
1538
1539         return 0;
1540 }
1541
1542 static int init_root_tree(struct mlx5_flow_steering *steering,
1543                           struct init_tree_node *init_node,
1544                           struct fs_node *fs_parent_node)
1545 {
1546         int i;
1547         struct mlx5_flow_namespace *fs_ns;
1548         int err;
1549
1550         fs_get_obj(fs_ns, fs_parent_node);
1551         for (i = 0; i < init_node->ar_size; i++) {
1552                 err = init_root_tree_recursive(steering, &init_node->children[i],
1553                                                &fs_ns->node,
1554                                                init_node, i);
1555                 if (err)
1556                         return err;
1557         }
1558         return 0;
1559 }
1560
1561 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
1562                                                        enum fs_flow_table_type
1563                                                        table_type)
1564 {
1565         struct mlx5_flow_root_namespace *root_ns;
1566         struct mlx5_flow_namespace *ns;
1567
1568         /* Create the root namespace */
1569         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1570         if (!root_ns)
1571                 return NULL;
1572
1573         root_ns->dev = steering->dev;
1574         root_ns->table_type = table_type;
1575
1576         ns = &root_ns->ns;
1577         fs_init_namespace(ns);
1578         mutex_init(&root_ns->chain_lock);
1579         tree_init_node(&ns->node, 1, NULL);
1580         tree_add_node(&ns->node, NULL);
1581
1582         return root_ns;
1583 }
1584
1585 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1586
1587 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1588 {
1589         struct fs_prio *prio;
1590
1591         fs_for_each_prio(prio, ns) {
1592                  /* This updates prio start_level and num_levels */
1593                 set_prio_attrs_in_prio(prio, acc_level);
1594                 acc_level += prio->num_levels;
1595         }
1596         return acc_level;
1597 }
1598
1599 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1600 {
1601         struct mlx5_flow_namespace *ns;
1602         int acc_level_ns = acc_level;
1603
1604         prio->start_level = acc_level;
1605         fs_for_each_ns(ns, prio)
1606                 /* This updates start_level and num_levels of ns's priority descendants */
1607                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1608         if (!prio->num_levels)
1609                 prio->num_levels = acc_level_ns - prio->start_level;
1610         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1611 }
1612
1613 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1614 {
1615         struct mlx5_flow_namespace *ns = &root_ns->ns;
1616         struct fs_prio *prio;
1617         int start_level = 0;
1618
1619         fs_for_each_prio(prio, ns) {
1620                 set_prio_attrs_in_prio(prio, start_level);
1621                 start_level += prio->num_levels;
1622         }
1623 }
1624
1625 #define ANCHOR_PRIO 0
1626 #define ANCHOR_SIZE 1
1627 #define ANCHOR_LEVEL 0
1628 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
1629 {
1630         struct mlx5_flow_namespace *ns = NULL;
1631         struct mlx5_flow_table *ft;
1632
1633         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1634         if (!ns)
1635                 return -EINVAL;
1636         ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL);
1637         if (IS_ERR(ft)) {
1638                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
1639                 return PTR_ERR(ft);
1640         }
1641         return 0;
1642 }
1643
1644 static int init_root_ns(struct mlx5_flow_steering *steering)
1645 {
1646
1647         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
1648         if (IS_ERR_OR_NULL(steering->root_ns))
1649                 goto cleanup;
1650
1651         if (init_root_tree(steering, &root_fs, &steering->root_ns->ns.node))
1652                 goto cleanup;
1653
1654         set_prio_attrs(steering->root_ns);
1655
1656         if (create_anchor_flow_table(steering))
1657                 goto cleanup;
1658
1659         return 0;
1660
1661 cleanup:
1662         mlx5_cleanup_fs(steering->dev);
1663         return -ENOMEM;
1664 }
1665
1666 static void cleanup_single_prio_root_ns(struct mlx5_flow_steering *steering,
1667                                         struct mlx5_flow_root_namespace *root_ns)
1668 {
1669         struct fs_node *prio;
1670
1671         if (!root_ns)
1672                 return;
1673
1674         if (!list_empty(&root_ns->ns.node.children)) {
1675                 prio = list_first_entry(&root_ns->ns.node.children,
1676                                         struct fs_node,
1677                                  list);
1678                 if (tree_remove_node(prio))
1679                         mlx5_core_warn(steering->dev,
1680                                        "Flow steering priority wasn't destroyed, refcount > 1\n");
1681         }
1682         if (tree_remove_node(&root_ns->ns.node))
1683                 mlx5_core_warn(steering->dev,
1684                                "Flow steering namespace wasn't destroyed, refcount > 1\n");
1685         root_ns = NULL;
1686 }
1687
1688 static void destroy_flow_tables(struct fs_prio *prio)
1689 {
1690         struct mlx5_flow_table *iter;
1691         struct mlx5_flow_table *tmp;
1692
1693         fs_for_each_ft_safe(iter, tmp, prio)
1694                 mlx5_destroy_flow_table(iter);
1695 }
1696
1697 static void cleanup_root_ns(struct mlx5_flow_steering *steering)
1698 {
1699         struct mlx5_flow_root_namespace *root_ns = steering->root_ns;
1700         struct fs_prio *iter_prio;
1701
1702         if (!MLX5_CAP_GEN(steering->dev, nic_flow_table))
1703                 return;
1704
1705         if (!root_ns)
1706                 return;
1707
1708         /* stage 1 */
1709         fs_for_each_prio(iter_prio, &root_ns->ns) {
1710                 struct fs_node *node;
1711                 struct mlx5_flow_namespace *iter_ns;
1712
1713                 fs_for_each_ns_or_ft(node, iter_prio) {
1714                         if (node->type == FS_TYPE_FLOW_TABLE)
1715                                 continue;
1716                         fs_get_obj(iter_ns, node);
1717                         while (!list_empty(&iter_ns->node.children)) {
1718                                 struct fs_prio *obj_iter_prio2;
1719                                 struct fs_node *iter_prio2 =
1720                                         list_first_entry(&iter_ns->node.children,
1721                                                          struct fs_node,
1722                                                          list);
1723
1724                                 fs_get_obj(obj_iter_prio2, iter_prio2);
1725                                 destroy_flow_tables(obj_iter_prio2);
1726                                 if (tree_remove_node(iter_prio2)) {
1727                                         mlx5_core_warn(steering->dev,
1728                                                        "Priority %d wasn't destroyed, refcount > 1\n",
1729                                                        obj_iter_prio2->prio);
1730                                         return;
1731                                 }
1732                         }
1733                 }
1734         }
1735
1736         /* stage 2 */
1737         fs_for_each_prio(iter_prio, &root_ns->ns) {
1738                 while (!list_empty(&iter_prio->node.children)) {
1739                         struct fs_node *iter_ns =
1740                                 list_first_entry(&iter_prio->node.children,
1741                                                  struct fs_node,
1742                                                  list);
1743                         if (tree_remove_node(iter_ns)) {
1744                                 mlx5_core_warn(steering->dev,
1745                                                "Namespace wasn't destroyed, refcount > 1\n");
1746                                 return;
1747                         }
1748                 }
1749         }
1750
1751         /* stage 3 */
1752         while (!list_empty(&root_ns->ns.node.children)) {
1753                 struct fs_prio *obj_prio_node;
1754                 struct fs_node *prio_node =
1755                         list_first_entry(&root_ns->ns.node.children,
1756                                          struct fs_node,
1757                                          list);
1758
1759                 fs_get_obj(obj_prio_node, prio_node);
1760                 if (tree_remove_node(prio_node)) {
1761                         mlx5_core_warn(steering->dev,
1762                                        "Priority %d wasn't destroyed, refcount > 1\n",
1763                                        obj_prio_node->prio);
1764                         return;
1765                 }
1766         }
1767
1768         if (tree_remove_node(&root_ns->ns.node)) {
1769                 mlx5_core_warn(steering->dev,
1770                                "root namespace wasn't destroyed, refcount > 1\n");
1771                 return;
1772         }
1773
1774         steering->root_ns = NULL;
1775 }
1776
1777 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1778 {
1779         struct mlx5_flow_steering *steering = dev->priv.steering;
1780
1781         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1782                 return;
1783
1784         cleanup_root_ns(steering);
1785         cleanup_single_prio_root_ns(steering, steering->esw_egress_root_ns);
1786         cleanup_single_prio_root_ns(steering, steering->esw_ingress_root_ns);
1787         cleanup_single_prio_root_ns(steering, steering->fdb_root_ns);
1788         mlx5_cleanup_fc_stats(dev);
1789         kfree(steering);
1790 }
1791
1792 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
1793 {
1794         struct fs_prio *prio;
1795
1796         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
1797         if (!steering->fdb_root_ns)
1798                 return -ENOMEM;
1799
1800         /* Create single prio */
1801         prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
1802         if (IS_ERR(prio)) {
1803                 cleanup_single_prio_root_ns(steering, steering->fdb_root_ns);
1804                 return PTR_ERR(prio);
1805         } else {
1806                 return 0;
1807         }
1808 }
1809
1810 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
1811 {
1812         struct fs_prio *prio;
1813
1814         steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
1815         if (!steering->esw_egress_root_ns)
1816                 return -ENOMEM;
1817
1818         /* create 1 prio*/
1819         prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
1820                               MLX5_TOTAL_VPORTS(steering->dev));
1821         if (IS_ERR(prio))
1822                 return PTR_ERR(prio);
1823         else
1824                 return 0;
1825 }
1826
1827 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
1828 {
1829         struct fs_prio *prio;
1830
1831         steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
1832         if (!steering->esw_ingress_root_ns)
1833                 return -ENOMEM;
1834
1835         /* create 1 prio*/
1836         prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
1837                               MLX5_TOTAL_VPORTS(steering->dev));
1838         if (IS_ERR(prio))
1839                 return PTR_ERR(prio);
1840         else
1841                 return 0;
1842 }
1843
1844 int mlx5_init_fs(struct mlx5_core_dev *dev)
1845 {
1846         struct mlx5_flow_steering *steering;
1847         int err = 0;
1848
1849         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1850                 return 0;
1851
1852         err = mlx5_init_fc_stats(dev);
1853         if (err)
1854                 return err;
1855
1856         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
1857         if (!steering)
1858                 return -ENOMEM;
1859         steering->dev = dev;
1860         dev->priv.steering = steering;
1861
1862         if (MLX5_CAP_GEN(dev, nic_flow_table) &&
1863             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
1864                 err = init_root_ns(steering);
1865                 if (err)
1866                         goto err;
1867         }
1868
1869         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1870                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
1871                         err = init_fdb_root_ns(steering);
1872                         if (err)
1873                                 goto err;
1874                 }
1875                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
1876                         err = init_egress_acl_root_ns(steering);
1877                         if (err)
1878                                 goto err;
1879                 }
1880                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
1881                         err = init_ingress_acl_root_ns(steering);
1882                         if (err)
1883                                 goto err;
1884                 }
1885         }
1886
1887         return 0;
1888 err:
1889         mlx5_cleanup_fs(dev);
1890         return err;
1891 }