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 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 157
 158 static void tree_init_node(struct fs_node *node,
 159                            unsigned int refcount,
 160                            void (*remove_func)(struct fs_node *))
 161 {
 162         atomic_set(&node->refcount, refcount);
 163         INIT_LIST_HEAD(&node->list);
 164         INIT_LIST_HEAD(&node->children);
 165         mutex_init(&node->lock);
 166         node->remove_func = remove_func;
 167 }
 168
 169 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 170 {
 171         if (parent)
 172                 atomic_inc(&parent->refcount);
 173         node->parent = parent;
 174
 175         /* Parent is the root */
 176         if (!parent)
 177                 node->root = node;
 178         else
 179                 node->root = parent->root;
 180 }
 181
 182 static void tree_get_node(struct fs_node *node)
 183 {
 184         atomic_inc(&node->refcount);
 185 }
 186
 187 static void nested_lock_ref_node(struct fs_node *node,
 188                                  enum fs_i_mutex_lock_class class)
 189 {
 190         if (node) {
 191                 mutex_lock_nested(&node->lock, class);
 192                 atomic_inc(&node->refcount);
 193         }
 194 }
 195
 196 static void lock_ref_node(struct fs_node *node)
 197 {
 198         if (node) {
 199                 mutex_lock(&node->lock);
 200                 atomic_inc(&node->refcount);
 201         }
 202 }
 203
 204 static void unlock_ref_node(struct fs_node *node)
 205 {
 206         if (node) {
 207                 atomic_dec(&node->refcount);
 208                 mutex_unlock(&node->lock);
 209         }
 210 }
 211
 212 static void tree_put_node(struct fs_node *node)
 213 {
 214         struct fs_node *parent_node = node->parent;
 215
 216         lock_ref_node(parent_node);
 217         if (atomic_dec_and_test(&node->refcount)) {
 218                 if (parent_node)
 219                         list_del_init(&node->list);
 220                 if (node->remove_func)
 221                         node->remove_func(node);
 222                 kfree(node);
 223                 node = NULL;
 224         }
 225         unlock_ref_node(parent_node);
 226         if (!node && parent_node)
 227                 tree_put_node(parent_node);
 228 }
 229
 230 static int tree_remove_node(struct fs_node *node)
 231 {
 232         if (atomic_read(&node->refcount) > 1) {
 233                 atomic_dec(&node->refcount);
 234                 return -EEXIST;
 235         }
 236         tree_put_node(node);
 237         return 0;
 238 }
 239
 240 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 241                                  unsigned int prio)
 242 {
 243         struct fs_prio *iter_prio;
 244
 245         fs_for_each_prio(iter_prio, ns) {
 246                 if (iter_prio->prio == prio)
 247                         return iter_prio;
 248         }
 249
 250         return NULL;
 251 }
 252
 253 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 254 {
 255         unsigned int i;
 256
 257         for (i = 0; i < size; i++, mask++, val1++, val2++)
 258                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 259                     ((*(u8 *)val2) & (*(u8 *)mask)))
 260                         return false;
 261
 262         return true;
 263 }
 264
 265 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 266                                 void *fte_param1, void *fte_param2)
 267 {
 268         if (mask->match_criteria_enable &
 269             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 270                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 271                                                 fte_param1, outer_headers);
 272                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 273                                                 fte_param2, outer_headers);
 274                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 275                                               mask->match_criteria, outer_headers);
 276
 277                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 278                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 279                         return false;
 280         }
 281
 282         if (mask->match_criteria_enable &
 283             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 284                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 285                                                 fte_param1, misc_parameters);
 286                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 287                                                 fte_param2, misc_parameters);
 288                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 289                                           mask->match_criteria, misc_parameters);
 290
 291                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 292                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 293                         return false;
 294         }
 295
 296         if (mask->match_criteria_enable &
 297             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 298                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 299                                                 fte_param1, inner_headers);
 300                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 301                                                 fte_param2, inner_headers);
 302                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 303                                           mask->match_criteria, inner_headers);
 304
 305                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 306                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 307                         return false;
 308         }
 309         return true;
 310 }
 311
 312 static bool compare_match_criteria(u8 match_criteria_enable1,
 313                                    u8 match_criteria_enable2,
 314                                    void *mask1, void *mask2)
 315 {
 316         return match_criteria_enable1 == match_criteria_enable2 &&
 317                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 318 }
 319
 320 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 321 {
 322         struct fs_node *root;
 323         struct mlx5_flow_namespace *ns;
 324
 325         root = node->root;
 326
 327         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 328                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 329                 return NULL;
 330         }
 331
 332         ns = container_of(root, struct mlx5_flow_namespace, node);
 333         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 334 }
 335
 336 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 337 {
 338         struct mlx5_flow_root_namespace *root = find_root(node);
 339
 340         if (root)
 341                 return root->dev;
 342         return NULL;
 343 }
 344
 345 static void del_flow_table(struct fs_node *node)
 346 {
 347         struct mlx5_flow_table *ft;
 348         struct mlx5_core_dev *dev;
 349         struct fs_prio *prio;
 350         int err;
 351
 352         fs_get_obj(ft, node);
 353         dev = get_dev(&ft->node);
 354
 355         err = mlx5_cmd_destroy_flow_table(dev, ft);
 356         if (err)
 357                 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 358         fs_get_obj(prio, ft->node.parent);
 359         prio->num_ft--;
 360 }
 361
 362 static void del_rule(struct fs_node *node)
 363 {
 364         struct mlx5_flow_rule *rule;
 365         struct mlx5_flow_table *ft;
 366         struct mlx5_flow_group *fg;
 367         struct fs_fte *fte;
 368         u32     *match_value;
 369         int modify_mask;
 370         struct mlx5_core_dev *dev = get_dev(node);
 371         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 372         int err;
 373
 374         match_value = mlx5_vzalloc(match_len);
 375         if (!match_value) {
 376                 mlx5_core_warn(dev, "failed to allocate inbox\n");
 377                 return;
 378         }
 379
 380         fs_get_obj(rule, node);
 381         fs_get_obj(fte, rule->node.parent);
 382         fs_get_obj(fg, fte->node.parent);
 383         memcpy(match_value, fte->val, sizeof(fte->val));
 384         fs_get_obj(ft, fg->node.parent);
 385         list_del(&rule->node.list);
 386         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 387                 mutex_lock(&rule->dest_attr.ft->lock);
 388                 list_del(&rule->next_ft);
 389                 mutex_unlock(&rule->dest_attr.ft->lock);
 390         }
 391         if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 392             --fte->dests_size) {
 393                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 394                 err = mlx5_cmd_update_fte(dev, ft,
 395                                           fg->id,
 396                                           modify_mask,
 397                                           fte);
 398                 if (err)
 399                         mlx5_core_warn(dev,
 400                                        "%s can't del rule fg id=%d fte_index=%d\n",
 401                                        __func__, fg->id, fte->index);
 402         }
 403         kvfree(match_value);
 404 }
 405
 406 static void del_fte(struct fs_node *node)
 407 {
 408         struct mlx5_flow_table *ft;
 409         struct mlx5_flow_group *fg;
 410         struct mlx5_core_dev *dev;
 411         struct fs_fte *fte;
 412         int err;
 413
 414         fs_get_obj(fte, node);
 415         fs_get_obj(fg, fte->node.parent);
 416         fs_get_obj(ft, fg->node.parent);
 417
 418         dev = get_dev(&ft->node);
 419         err = mlx5_cmd_delete_fte(dev, ft,
 420                                   fte->index);
 421         if (err)
 422                 mlx5_core_warn(dev,
 423                                "flow steering can't delete fte in index %d of flow group id %d\n",
 424                                fte->index, fg->id);
 425
 426         fte->status = 0;
 427         fg->num_ftes--;
 428 }
 429
 430 static void del_flow_group(struct fs_node *node)
 431 {
 432         struct mlx5_flow_group *fg;
 433         struct mlx5_flow_table *ft;
 434         struct mlx5_core_dev *dev;
 435
 436         fs_get_obj(fg, node);
 437         fs_get_obj(ft, fg->node.parent);
 438         dev = get_dev(&ft->node);
 439
 440         if (ft->autogroup.active)
 441                 ft->autogroup.num_groups--;
 442
 443         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 444                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 445                                fg->id, ft->id);
 446 }
 447
 448 static struct fs_fte *alloc_fte(u8 action,
 449                                 u32 flow_tag,
 450                                 u32 *match_value,
 451                                 unsigned int index)
 452 {
 453         struct fs_fte *fte;
 454
 455         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 456         if (!fte)
 457                 return ERR_PTR(-ENOMEM);
 458
 459         memcpy(fte->val, match_value, sizeof(fte->val));
 460         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 461         fte->flow_tag = flow_tag;
 462         fte->index = index;
 463         fte->action = action;
 464
 465         return fte;
 466 }
 467
 468 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 469 {
 470         struct mlx5_flow_group *fg;
 471         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 472                                             create_fg_in, match_criteria);
 473         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 474                                             create_fg_in,
 475                                             match_criteria_enable);
 476         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 477         if (!fg)
 478                 return ERR_PTR(-ENOMEM);
 479
 480         fg->mask.match_criteria_enable = match_criteria_enable;
 481         memcpy(&fg->mask.match_criteria, match_criteria,
 482                sizeof(fg->mask.match_criteria));
 483         fg->node.type =  FS_TYPE_FLOW_GROUP;
 484         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 485                                    start_flow_index);
 486         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 487                                 end_flow_index) - fg->start_index + 1;
 488         return fg;
 489 }
 490
 491 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 492                                                 enum fs_flow_table_type table_type,
 493                                                 enum fs_flow_table_op_mod op_mod)
 494 {
 495         struct mlx5_flow_table *ft;
 496
 497         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 498         if (!ft)
 499                 return NULL;
 500
 501         ft->level = level;
 502         ft->node.type = FS_TYPE_FLOW_TABLE;
 503         ft->op_mod = op_mod;
 504         ft->type = table_type;
 505         ft->vport = vport;
 506         ft->max_fte = max_fte;
 507         INIT_LIST_HEAD(&ft->fwd_rules);
 508         mutex_init(&ft->lock);
 509
 510         return ft;
 511 }
 512
 513 /* If reverse is false, then we search for the first flow table in the
 514  * root sub-tree from start(closest from right), else we search for the
 515  * last flow table in the root sub-tree till start(closest from left).
 516  */
 517 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 518                                                          struct list_head *start,
 519                                                          bool reverse)
 520 {
 521 #define list_advance_entry(pos, reverse)                \
 522         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 523
 524 #define list_for_each_advance_continue(pos, head, reverse)      \
 525         for (pos = list_advance_entry(pos, reverse);            \
 526              &pos->list != (head);                              \
 527              pos = list_advance_entry(pos, reverse))
 528
 529         struct fs_node *iter = list_entry(start, struct fs_node, list);
 530         struct mlx5_flow_table *ft = NULL;
 531
 532         if (!root)
 533                 return NULL;
 534
 535         list_for_each_advance_continue(iter, &root->children, reverse) {
 536                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 537                         fs_get_obj(ft, iter);
 538                         return ft;
 539                 }
 540                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 541                 if (ft)
 542                         return ft;
 543         }
 544
 545         return ft;
 546 }
 547
 548 /* If reverse if false then return the first flow table in next priority of
 549  * prio in the tree, else return the last flow table in the previous priority
 550  * of prio in the tree.
 551  */
 552 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 553 {
 554         struct mlx5_flow_table *ft = NULL;
 555         struct fs_node *curr_node;
 556         struct fs_node *parent;
 557
 558         parent = prio->node.parent;
 559         curr_node = &prio->node;
 560         while (!ft && parent) {
 561                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 562                 curr_node = parent;
 563                 parent = curr_node->parent;
 564         }
 565         return ft;
 566 }
 567
 568 /* Assuming all the tree is locked by mutex chain lock */
 569 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 570 {
 571         return find_closest_ft(prio, false);
 572 }
 573
 574 /* Assuming all the tree is locked by mutex chain lock */
 575 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 576 {
 577         return find_closest_ft(prio, true);
 578 }
 579
 580 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 581                                struct fs_prio *prio,
 582                                struct mlx5_flow_table *ft)
 583 {
 584         struct mlx5_flow_table *iter;
 585         int i = 0;
 586         int err;
 587
 588         fs_for_each_ft(iter, prio) {
 589                 i++;
 590                 err = mlx5_cmd_modify_flow_table(dev,
 591                                                  iter,
 592                                                  ft);
 593                 if (err) {
 594                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 595                                        iter->id);
 596                         /* The driver is out of sync with the FW */
 597                         if (i > 1)
 598                                 WARN_ON(true);
 599                         return err;
 600                 }
 601         }
 602         return 0;
 603 }
 604
 605 /* Connect flow tables from previous priority of prio to ft */
 606 static int connect_prev_fts(struct mlx5_core_dev *dev,
 607                             struct mlx5_flow_table *ft,
 608                             struct fs_prio *prio)
 609 {
 610         struct mlx5_flow_table *prev_ft;
 611
 612         prev_ft = find_prev_chained_ft(prio);
 613         if (prev_ft) {
 614                 struct fs_prio *prev_prio;
 615
 616                 fs_get_obj(prev_prio, prev_ft->node.parent);
 617                 return connect_fts_in_prio(dev, prev_prio, ft);
 618         }
 619         return 0;
 620 }
 621
 622 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 623                                  *prio)
 624 {
 625         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 626         int min_level = INT_MAX;
 627         int err;
 628
 629         if (root->root_ft)
 630                 min_level = root->root_ft->level;
 631
 632         if (ft->level >= min_level)
 633                 return 0;
 634
 635         err = mlx5_cmd_update_root_ft(root->dev, ft);
 636         if (err)
 637                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 638                                ft->id);
 639         else
 640                 root->root_ft = ft;
 641
 642         return err;
 643 }
 644
 645 int mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 646                                  struct mlx5_flow_destination *dest)
 647 {
 648         struct mlx5_flow_table *ft;
 649         struct mlx5_flow_group *fg;
 650         struct fs_fte *fte;
 651         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 652         int err = 0;
 653
 654         fs_get_obj(fte, rule->node.parent);
 655         if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 656                 return -EINVAL;
 657         lock_ref_node(&fte->node);
 658         fs_get_obj(fg, fte->node.parent);
 659         fs_get_obj(ft, fg->node.parent);
 660
 661         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 662         err = mlx5_cmd_update_fte(get_dev(&ft->node),
 663                                   ft, fg->id,
 664                                   modify_mask,
 665                                   fte);
 666         unlock_ref_node(&fte->node);
 667
 668         return err;
 669 }
 670
 671 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 672 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 673                              struct mlx5_flow_table *new_next_ft,
 674                              struct mlx5_flow_table *old_next_ft)
 675 {
 676         struct mlx5_flow_destination dest;
 677         struct mlx5_flow_rule *iter;
 678         int err = 0;
 679
 680         /* new_next_ft and old_next_ft could be NULL only
 681          * when we create/destroy the anchor flow table.
 682          */
 683         if (!new_next_ft || !old_next_ft)
 684                 return 0;
 685
 686         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 687         dest.ft = new_next_ft;
 688
 689         mutex_lock(&old_next_ft->lock);
 690         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 691         mutex_unlock(&old_next_ft->lock);
 692         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 693                 err = mlx5_modify_rule_destination(iter, &dest);
 694                 if (err)
 695                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 696                                new_next_ft->id);
 697         }
 698         return 0;
 699 }
 700
 701 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 702                               struct fs_prio *prio)
 703 {
 704         struct mlx5_flow_table *next_ft, *first_ft;
 705         int err = 0;
 706
 707         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 708
 709         first_ft = list_first_entry_or_null(&prio->node.children,
 710                                             struct mlx5_flow_table, node.list);
 711         if (!first_ft || first_ft->level > ft->level) {
 712                 err = connect_prev_fts(dev, ft, prio);
 713                 if (err)
 714                         return err;
 715
 716                 next_ft = first_ft ? first_ft : find_next_chained_ft(prio);
 717                 err = connect_fwd_rules(dev, ft, next_ft);
 718                 if (err)
 719                         return err;
 720         }
 721
 722         if (MLX5_CAP_FLOWTABLE(dev,
 723                                flow_table_properties_nic_receive.modify_root))
 724                 err = update_root_ft_create(ft, prio);
 725         return err;
 726 }
 727
 728 static void list_add_flow_table(struct mlx5_flow_table *ft,
 729                                 struct fs_prio *prio)
 730 {
 731         struct list_head *prev = &prio->node.children;
 732         struct mlx5_flow_table *iter;
 733
 734         fs_for_each_ft(iter, prio) {
 735                 if (iter->level > ft->level)
 736                         break;
 737                 prev = &iter->node.list;
 738         }
 739         list_add(&ft->node.list, prev);
 740 }
 741
 742 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 743                                                         enum fs_flow_table_op_mod op_mod,
 744                                                         u16 vport, int prio,
 745                                                         int max_fte, u32 level)
 746 {
 747         struct mlx5_flow_table *next_ft = NULL;
 748         struct mlx5_flow_table *ft;
 749         int err;
 750         int log_table_sz;
 751         struct mlx5_flow_root_namespace *root =
 752                 find_root(&ns->node);
 753         struct fs_prio *fs_prio = NULL;
 754
 755         if (!root) {
 756                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 757                 return ERR_PTR(-ENODEV);
 758         }
 759
 760         mutex_lock(&root->chain_lock);
 761         fs_prio = find_prio(ns, prio);
 762         if (!fs_prio) {
 763                 err = -EINVAL;
 764                 goto unlock_root;
 765         }
 766         if (level >= fs_prio->num_levels) {
 767                 err = -ENOSPC;
 768                 goto unlock_root;
 769         }
 770         /* The level is related to the
 771          * priority level range.
 772          */
 773         level += fs_prio->start_level;
 774         ft = alloc_flow_table(level,
 775                               vport,
 776                               max_fte ? roundup_pow_of_two(max_fte) : 0,
 777                               root->table_type,
 778                               op_mod);
 779         if (!ft) {
 780                 err = -ENOMEM;
 781                 goto unlock_root;
 782         }
 783
 784         tree_init_node(&ft->node, 1, del_flow_table);
 785         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
 786         next_ft = find_next_chained_ft(fs_prio);
 787         err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->op_mod, ft->type,
 788                                          ft->level, log_table_sz, next_ft, &ft->id);
 789         if (err)
 790                 goto free_ft;
 791
 792         err = connect_flow_table(root->dev, ft, fs_prio);
 793         if (err)
 794                 goto destroy_ft;
 795         lock_ref_node(&fs_prio->node);
 796         tree_add_node(&ft->node, &fs_prio->node);
 797         list_add_flow_table(ft, fs_prio);
 798         fs_prio->num_ft++;
 799         unlock_ref_node(&fs_prio->node);
 800         mutex_unlock(&root->chain_lock);
 801         return ft;
 802 destroy_ft:
 803         mlx5_cmd_destroy_flow_table(root->dev, ft);
 804 free_ft:
 805         kfree(ft);
 806 unlock_root:
 807         mutex_unlock(&root->chain_lock);
 808         return ERR_PTR(err);
 809 }
 810
 811 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 812                                                int prio, int max_fte,
 813                                                u32 level)
 814 {
 815         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, 0, prio,
 816                                         max_fte, level);
 817 }
 818
 819 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
 820                                                      int prio, int max_fte,
 821                                                      u32 level, u16 vport)
 822 {
 823         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, vport, prio,
 824                                         max_fte, level);
 825 }
 826
 827 struct mlx5_flow_table *mlx5_create_lag_demux_flow_table(
 828                                                struct mlx5_flow_namespace *ns,
 829                                                int prio, u32 level)
 830 {
 831         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_LAG_DEMUX, 0, prio, 0,
 832                                         level);
 833 }
 834 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
 835
 836 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 837                                                             int prio,
 838                                                             int num_flow_table_entries,
 839                                                             int max_num_groups,
 840                                                             u32 level)
 841 {
 842         struct mlx5_flow_table *ft;
 843
 844         if (max_num_groups > num_flow_table_entries)
 845                 return ERR_PTR(-EINVAL);
 846
 847         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries, level);
 848         if (IS_ERR(ft))
 849                 return ft;
 850
 851         ft->autogroup.active = true;
 852         ft->autogroup.required_groups = max_num_groups;
 853
 854         return ft;
 855 }
 856 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 857
 858 /* Flow table should be locked */
 859 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 860                                                         u32 *fg_in,
 861                                                         struct list_head
 862                                                         *prev_fg,
 863                                                         bool is_auto_fg)
 864 {
 865         struct mlx5_flow_group *fg;
 866         struct mlx5_core_dev *dev = get_dev(&ft->node);
 867         int err;
 868
 869         if (!dev)
 870                 return ERR_PTR(-ENODEV);
 871
 872         fg = alloc_flow_group(fg_in);
 873         if (IS_ERR(fg))
 874                 return fg;
 875
 876         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 877         if (err) {
 878                 kfree(fg);
 879                 return ERR_PTR(err);
 880         }
 881
 882         if (ft->autogroup.active)
 883                 ft->autogroup.num_groups++;
 884         /* Add node to tree */
 885         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 886         tree_add_node(&fg->node, &ft->node);
 887         /* Add node to group list */
 888         list_add(&fg->node.list, prev_fg);
 889
 890         return fg;
 891 }
 892
 893 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 894                                                u32 *fg_in)
 895 {
 896         struct mlx5_flow_group *fg;
 897
 898         if (ft->autogroup.active)
 899                 return ERR_PTR(-EPERM);
 900
 901         lock_ref_node(&ft->node);
 902         fg = create_flow_group_common(ft, fg_in, ft->node.children.prev, false);
 903         unlock_ref_node(&ft->node);
 904
 905         return fg;
 906 }
 907
 908 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 909 {
 910         struct mlx5_flow_rule *rule;
 911
 912         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 913         if (!rule)
 914                 return NULL;
 915
 916         INIT_LIST_HEAD(&rule->next_ft);
 917         rule->node.type = FS_TYPE_FLOW_DEST;
 918         if (dest)
 919                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 920
 921         return rule;
 922 }
 923
 924 /* fte should not be deleted while calling this function */
 925 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
 926                                            struct mlx5_flow_group *fg,
 927                                            struct mlx5_flow_destination *dest)
 928 {
 929         struct mlx5_flow_table *ft;
 930         struct mlx5_flow_rule *rule;
 931         int modify_mask = 0;
 932         int err;
 933
 934         rule = alloc_rule(dest);
 935         if (!rule)
 936                 return ERR_PTR(-ENOMEM);
 937
 938         fs_get_obj(ft, fg->node.parent);
 939         /* Add dest to dests list- we need flow tables to be in the
 940          * end of the list for forward to next prio rules.
 941          */
 942         tree_init_node(&rule->node, 1, del_rule);
 943         if (dest && dest->type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
 944                 list_add(&rule->node.list, &fte->node.children);
 945         else
 946                 list_add_tail(&rule->node.list, &fte->node.children);
 947         if (dest) {
 948                 fte->dests_size++;
 949
 950                 modify_mask |= dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER ?
 951                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS) :
 952                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 953         }
 954
 955         if (fte->dests_size == 1 || !dest)
 956                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 957                                           ft, fg->id, fte);
 958         else
 959                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 960                                           ft, fg->id, modify_mask, fte);
 961         if (err)
 962                 goto free_rule;
 963
 964         fte->status |= FS_FTE_STATUS_EXISTING;
 965
 966         return rule;
 967
 968 free_rule:
 969         list_del(&rule->node.list);
 970         kfree(rule);
 971         if (dest)
 972                 fte->dests_size--;
 973         return ERR_PTR(err);
 974 }
 975
 976 /* Assumed fg is locked */
 977 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 978                                        struct list_head **prev)
 979 {
 980         struct fs_fte *fte;
 981         unsigned int start = fg->start_index;
 982
 983         if (prev)
 984                 *prev = &fg->node.children;
 985
 986         /* assumed list is sorted by index */
 987         fs_for_each_fte(fte, fg) {
 988                 if (fte->index != start)
 989                         return start;
 990                 start++;
 991                 if (prev)
 992                         *prev = &fte->node.list;
 993         }
 994
 995         return start;
 996 }
 997
 998 /* prev is output, prev->next = new_fte */
 999 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
1000                                  u32 *match_value,
1001                                  u8 action,
1002                                  u32 flow_tag,
1003                                  struct list_head **prev)
1004 {
1005         struct fs_fte *fte;
1006         int index;
1007
1008         index = get_free_fte_index(fg, prev);
1009         fte = alloc_fte(action, flow_tag, match_value, index);
1010         if (IS_ERR(fte))
1011                 return fte;
1012
1013         return fte;
1014 }
1015
1016 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
1017                                                 u8 match_criteria_enable,
1018                                                 u32 *match_criteria)
1019 {
1020         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1021         struct list_head *prev = &ft->node.children;
1022         unsigned int candidate_index = 0;
1023         struct mlx5_flow_group *fg;
1024         void *match_criteria_addr;
1025         unsigned int group_size = 0;
1026         u32 *in;
1027
1028         if (!ft->autogroup.active)
1029                 return ERR_PTR(-ENOENT);
1030
1031         in = mlx5_vzalloc(inlen);
1032         if (!in)
1033                 return ERR_PTR(-ENOMEM);
1034
1035         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1036                 /* We save place for flow groups in addition to max types */
1037                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1038
1039         /*  ft->max_fte == ft->autogroup.max_types */
1040         if (group_size == 0)
1041                 group_size = 1;
1042
1043         /* sorted by start_index */
1044         fs_for_each_fg(fg, ft) {
1045                 if (candidate_index + group_size > fg->start_index)
1046                         candidate_index = fg->start_index + fg->max_ftes;
1047                 else
1048                         break;
1049                 prev = &fg->node.list;
1050         }
1051
1052         if (candidate_index + group_size > ft->max_fte) {
1053                 fg = ERR_PTR(-ENOSPC);
1054                 goto out;
1055         }
1056
1057         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1058                  match_criteria_enable);
1059         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1060         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
1061                  group_size - 1);
1062         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1063                                            in, match_criteria);
1064         memcpy(match_criteria_addr, match_criteria,
1065                MLX5_ST_SZ_BYTES(fte_match_param));
1066
1067         fg = create_flow_group_common(ft, in, prev, true);
1068 out:
1069         kvfree(in);
1070         return fg;
1071 }
1072
1073 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1074                                              struct mlx5_flow_destination *dest)
1075 {
1076         struct mlx5_flow_rule *rule;
1077
1078         list_for_each_entry(rule, &fte->node.children, node.list) {
1079                 if (rule->dest_attr.type == dest->type) {
1080                         if ((dest->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1081                              dest->vport_num == rule->dest_attr.vport_num) ||
1082                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1083                              dest->ft == rule->dest_attr.ft) ||
1084                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1085                              dest->tir_num == rule->dest_attr.tir_num))
1086                                 return rule;
1087                 }
1088         }
1089         return NULL;
1090 }
1091
1092 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
1093                                           u32 *match_value,
1094                                           u8 action,
1095                                           u32 flow_tag,
1096                                           struct mlx5_flow_destination *dest)
1097 {
1098         struct fs_fte *fte;
1099         struct mlx5_flow_rule *rule;
1100         struct mlx5_flow_table *ft;
1101         struct list_head *prev;
1102
1103         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1104         fs_for_each_fte(fte, fg) {
1105                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1106                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1107                     action == fte->action && flow_tag == fte->flow_tag) {
1108                         rule = find_flow_rule(fte, dest);
1109                         if (rule) {
1110                                 atomic_inc(&rule->node.refcount);
1111                                 unlock_ref_node(&fte->node);
1112                                 unlock_ref_node(&fg->node);
1113                                 return rule;
1114                         }
1115                         rule = add_rule_fte(fte, fg, dest);
1116                         if (IS_ERR(rule))
1117                                 goto unlock_fte;
1118                         else
1119                                 goto add_rule;
1120                 }
1121                 unlock_ref_node(&fte->node);
1122         }
1123         fs_get_obj(ft, fg->node.parent);
1124         if (fg->num_ftes >= fg->max_ftes) {
1125                 rule = ERR_PTR(-ENOSPC);
1126                 goto unlock_fg;
1127         }
1128
1129         fte = create_fte(fg, match_value, action, flow_tag, &prev);
1130         if (IS_ERR(fte)) {
1131                 rule = (void *)fte;
1132                 goto unlock_fg;
1133         }
1134         tree_init_node(&fte->node, 0, del_fte);
1135         nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1136         rule = add_rule_fte(fte, fg, dest);
1137         if (IS_ERR(rule)) {
1138                 kfree(fte);
1139                 goto unlock_fg;
1140         }
1141
1142         fg->num_ftes++;
1143
1144         tree_add_node(&fte->node, &fg->node);
1145         list_add(&fte->node.list, prev);
1146 add_rule:
1147         tree_add_node(&rule->node, &fte->node);
1148 unlock_fte:
1149         unlock_ref_node(&fte->node);
1150 unlock_fg:
1151         unlock_ref_node(&fg->node);
1152         return rule;
1153 }
1154
1155 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_rule *rule)
1156 {
1157         struct mlx5_flow_rule *dst;
1158         struct fs_fte *fte;
1159
1160         fs_get_obj(fte, rule->node.parent);
1161
1162         fs_for_each_dst(dst, fte) {
1163                 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1164                         return dst->dest_attr.counter;
1165         }
1166
1167         return NULL;
1168 }
1169
1170 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1171 {
1172         if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1173                 return !counter;
1174
1175         if (!counter)
1176                 return false;
1177
1178         /* Hardware support counter for a drop action only */
1179         return action == (MLX5_FLOW_CONTEXT_ACTION_DROP | MLX5_FLOW_CONTEXT_ACTION_COUNT);
1180 }
1181
1182 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1183                           u32 action,
1184                           struct mlx5_flow_table *ft)
1185 {
1186         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1187                 return counter_is_valid(dest->counter, action);
1188
1189         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1190                 return true;
1191
1192         if (!dest || ((dest->type ==
1193             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1194             (dest->ft->level <= ft->level)))
1195                 return false;
1196         return true;
1197 }
1198
1199 static struct mlx5_flow_rule *
1200 _mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1201                    struct mlx5_flow_spec *spec,
1202                     u32 action,
1203                     u32 flow_tag,
1204                     struct mlx5_flow_destination *dest)
1205 {
1206         struct mlx5_flow_group *g;
1207         struct mlx5_flow_rule *rule;
1208
1209         if (!dest_is_valid(dest, action, ft))
1210                 return ERR_PTR(-EINVAL);
1211
1212         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1213         fs_for_each_fg(g, ft)
1214                 if (compare_match_criteria(g->mask.match_criteria_enable,
1215                                            spec->match_criteria_enable,
1216                                            g->mask.match_criteria,
1217                                            spec->match_criteria)) {
1218                         rule = add_rule_fg(g, spec->match_value,
1219                                            action, flow_tag, dest);
1220                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1221                                 goto unlock;
1222                 }
1223
1224         g = create_autogroup(ft, spec->match_criteria_enable,
1225                              spec->match_criteria);
1226         if (IS_ERR(g)) {
1227                 rule = (void *)g;
1228                 goto unlock;
1229         }
1230
1231         rule = add_rule_fg(g, spec->match_value,
1232                            action, flow_tag, dest);
1233         if (IS_ERR(rule)) {
1234                 /* Remove assumes refcount > 0 and autogroup creates a group
1235                  * with a refcount = 0.
1236                  */
1237                 unlock_ref_node(&ft->node);
1238                 tree_get_node(&g->node);
1239                 tree_remove_node(&g->node);
1240                 return rule;
1241         }
1242 unlock:
1243         unlock_ref_node(&ft->node);
1244         return rule;
1245 }
1246
1247 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1248 {
1249         return ((ft->type == FS_FT_NIC_RX) &&
1250                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1251 }
1252
1253 struct mlx5_flow_rule *
1254 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1255                    struct mlx5_flow_spec *spec,
1256                    u32 action,
1257                    u32 flow_tag,
1258                    struct mlx5_flow_destination *dest)
1259 {
1260         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1261         struct mlx5_flow_destination gen_dest;
1262         struct mlx5_flow_table *next_ft = NULL;
1263         struct mlx5_flow_rule *rule = NULL;
1264         u32 sw_action = action;
1265         struct fs_prio *prio;
1266
1267         fs_get_obj(prio, ft->node.parent);
1268         if (action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1269                 if (!fwd_next_prio_supported(ft))
1270                         return ERR_PTR(-EOPNOTSUPP);
1271                 if (dest)
1272                         return ERR_PTR(-EINVAL);
1273                 mutex_lock(&root->chain_lock);
1274                 next_ft = find_next_chained_ft(prio);
1275                 if (next_ft) {
1276                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1277                         gen_dest.ft = next_ft;
1278                         dest = &gen_dest;
1279                         action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1280                 } else {
1281                         mutex_unlock(&root->chain_lock);
1282                         return ERR_PTR(-EOPNOTSUPP);
1283                 }
1284         }
1285
1286         rule = _mlx5_add_flow_rule(ft, spec, action, flow_tag, dest);
1287
1288         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1289                 if (!IS_ERR_OR_NULL(rule) &&
1290                     (list_empty(&rule->next_ft))) {
1291                         mutex_lock(&next_ft->lock);
1292                         list_add(&rule->next_ft, &next_ft->fwd_rules);
1293                         mutex_unlock(&next_ft->lock);
1294                         rule->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1295                 }
1296                 mutex_unlock(&root->chain_lock);
1297         }
1298         return rule;
1299 }
1300 EXPORT_SYMBOL(mlx5_add_flow_rule);
1301
1302 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
1303 {
1304         tree_remove_node(&rule->node);
1305 }
1306 EXPORT_SYMBOL(mlx5_del_flow_rule);
1307
1308 /* Assuming prio->node.children(flow tables) is sorted by level */
1309 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1310 {
1311         struct fs_prio *prio;
1312
1313         fs_get_obj(prio, ft->node.parent);
1314
1315         if (!list_is_last(&ft->node.list, &prio->node.children))
1316                 return list_next_entry(ft, node.list);
1317         return find_next_chained_ft(prio);
1318 }
1319
1320 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1321 {
1322         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1323         struct mlx5_flow_table *new_root_ft = NULL;
1324
1325         if (root->root_ft != ft)
1326                 return 0;
1327
1328         new_root_ft = find_next_ft(ft);
1329         if (new_root_ft) {
1330                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1331
1332                 if (err) {
1333                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1334                                        ft->id);
1335                         return err;
1336                 }
1337         }
1338         root->root_ft = new_root_ft;
1339         return 0;
1340 }
1341
1342 /* Connect flow table from previous priority to
1343  * the next flow table.
1344  */
1345 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1346 {
1347         struct mlx5_core_dev *dev = get_dev(&ft->node);
1348         struct mlx5_flow_table *next_ft;
1349         struct fs_prio *prio;
1350         int err = 0;
1351
1352         err = update_root_ft_destroy(ft);
1353         if (err)
1354                 return err;
1355
1356         fs_get_obj(prio, ft->node.parent);
1357         if  (!(list_first_entry(&prio->node.children,
1358                                 struct mlx5_flow_table,
1359                                 node.list) == ft))
1360                 return 0;
1361
1362         next_ft = find_next_ft(ft);
1363         err = connect_fwd_rules(dev, next_ft, ft);
1364         if (err)
1365                 return err;
1366
1367         err = connect_prev_fts(dev, next_ft, prio);
1368         if (err)
1369                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1370                                ft->id);
1371         return err;
1372 }
1373
1374 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1375 {
1376         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1377         int err = 0;
1378
1379         mutex_lock(&root->chain_lock);
1380         err = disconnect_flow_table(ft);
1381         if (err) {
1382                 mutex_unlock(&root->chain_lock);
1383                 return err;
1384         }
1385         if (tree_remove_node(&ft->node))
1386                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1387                                ft->id);
1388         mutex_unlock(&root->chain_lock);
1389
1390         return err;
1391 }
1392 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1393
1394 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1395 {
1396         if (tree_remove_node(&fg->node))
1397                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1398                                fg->id);
1399 }
1400
1401 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1402                                                     enum mlx5_flow_namespace_type type)
1403 {
1404         struct mlx5_flow_steering *steering = dev->priv.steering;
1405         struct mlx5_flow_root_namespace *root_ns;
1406         int prio;
1407         struct fs_prio *fs_prio;
1408         struct mlx5_flow_namespace *ns;
1409
1410         if (!steering)
1411                 return NULL;
1412
1413         switch (type) {
1414         case MLX5_FLOW_NAMESPACE_BYPASS:
1415         case MLX5_FLOW_NAMESPACE_LAG:
1416         case MLX5_FLOW_NAMESPACE_OFFLOADS:
1417         case MLX5_FLOW_NAMESPACE_ETHTOOL:
1418         case MLX5_FLOW_NAMESPACE_KERNEL:
1419         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1420         case MLX5_FLOW_NAMESPACE_ANCHOR:
1421                 prio = type;
1422                 break;
1423         case MLX5_FLOW_NAMESPACE_FDB:
1424                 if (steering->fdb_root_ns)
1425                         return &steering->fdb_root_ns->ns;
1426                 else
1427                         return NULL;
1428         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1429                 if (steering->esw_egress_root_ns)
1430                         return &steering->esw_egress_root_ns->ns;
1431                 else
1432                         return NULL;
1433         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1434                 if (steering->esw_ingress_root_ns)
1435                         return &steering->esw_ingress_root_ns->ns;
1436                 else
1437                         return NULL;
1438         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
1439                 if (steering->sniffer_rx_root_ns)
1440                         return &steering->sniffer_rx_root_ns->ns;
1441                 else
1442                         return NULL;
1443         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
1444                 if (steering->sniffer_tx_root_ns)
1445                         return &steering->sniffer_tx_root_ns->ns;
1446                 else
1447                         return NULL;
1448         default:
1449                 return NULL;
1450         }
1451
1452         root_ns = steering->root_ns;
1453         if (!root_ns)
1454                 return NULL;
1455
1456         fs_prio = find_prio(&root_ns->ns, prio);
1457         if (!fs_prio)
1458                 return NULL;
1459
1460         ns = list_first_entry(&fs_prio->node.children,
1461                               typeof(*ns),
1462                               node.list);
1463
1464         return ns;
1465 }
1466 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1467
1468 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1469                                       unsigned int prio, int num_levels)
1470 {
1471         struct fs_prio *fs_prio;
1472
1473         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1474         if (!fs_prio)
1475                 return ERR_PTR(-ENOMEM);
1476
1477         fs_prio->node.type = FS_TYPE_PRIO;
1478         tree_init_node(&fs_prio->node, 1, NULL);
1479         tree_add_node(&fs_prio->node, &ns->node);
1480         fs_prio->num_levels = num_levels;
1481         fs_prio->prio = prio;
1482         list_add_tail(&fs_prio->node.list, &ns->node.children);
1483
1484         return fs_prio;
1485 }
1486
1487 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1488                                                      *ns)
1489 {
1490         ns->node.type = FS_TYPE_NAMESPACE;
1491
1492         return ns;
1493 }
1494
1495 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1496 {
1497         struct mlx5_flow_namespace      *ns;
1498
1499         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1500         if (!ns)
1501                 return ERR_PTR(-ENOMEM);
1502
1503         fs_init_namespace(ns);
1504         tree_init_node(&ns->node, 1, NULL);
1505         tree_add_node(&ns->node, &prio->node);
1506         list_add_tail(&ns->node.list, &prio->node.children);
1507
1508         return ns;
1509 }
1510
1511 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1512                              struct init_tree_node *prio_metadata)
1513 {
1514         struct fs_prio *fs_prio;
1515         int i;
1516
1517         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1518                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1519                 if (IS_ERR(fs_prio))
1520                         return PTR_ERR(fs_prio);
1521         }
1522         return 0;
1523 }
1524
1525 #define FLOW_TABLE_BIT_SZ 1
1526 #define GET_FLOW_TABLE_CAP(dev, offset) \
1527         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1528                         offset / 32)) >>                                        \
1529           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1530 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1531 {
1532         int i;
1533
1534         for (i = 0; i < caps->arr_sz; i++) {
1535                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1536                         return false;
1537         }
1538         return true;
1539 }
1540
1541 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
1542                                     struct init_tree_node *init_node,
1543                                     struct fs_node *fs_parent_node,
1544                                     struct init_tree_node *init_parent_node,
1545                                     int prio)
1546 {
1547         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
1548                                               flow_table_properties_nic_receive.
1549                                               max_ft_level);
1550         struct mlx5_flow_namespace *fs_ns;
1551         struct fs_prio *fs_prio;
1552         struct fs_node *base;
1553         int i;
1554         int err;
1555
1556         if (init_node->type == FS_TYPE_PRIO) {
1557                 if ((init_node->min_ft_level > max_ft_level) ||
1558                     !has_required_caps(steering->dev, &init_node->caps))
1559                         return 0;
1560
1561                 fs_get_obj(fs_ns, fs_parent_node);
1562                 if (init_node->num_leaf_prios)
1563                         return create_leaf_prios(fs_ns, prio, init_node);
1564                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1565                 if (IS_ERR(fs_prio))
1566                         return PTR_ERR(fs_prio);
1567                 base = &fs_prio->node;
1568         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1569                 fs_get_obj(fs_prio, fs_parent_node);
1570                 fs_ns = fs_create_namespace(fs_prio);
1571                 if (IS_ERR(fs_ns))
1572                         return PTR_ERR(fs_ns);
1573                 base = &fs_ns->node;
1574         } else {
1575                 return -EINVAL;
1576         }
1577         prio = 0;
1578         for (i = 0; i < init_node->ar_size; i++) {
1579                 err = init_root_tree_recursive(steering, &init_node->children[i],
1580                                                base, init_node, prio);
1581                 if (err)
1582                         return err;
1583                 if (init_node->children[i].type == FS_TYPE_PRIO &&
1584                     init_node->children[i].num_leaf_prios) {
1585                         prio += init_node->children[i].num_leaf_prios;
1586                 }
1587         }
1588
1589         return 0;
1590 }
1591
1592 static int init_root_tree(struct mlx5_flow_steering *steering,
1593                           struct init_tree_node *init_node,
1594                           struct fs_node *fs_parent_node)
1595 {
1596         int i;
1597         struct mlx5_flow_namespace *fs_ns;
1598         int err;
1599
1600         fs_get_obj(fs_ns, fs_parent_node);
1601         for (i = 0; i < init_node->ar_size; i++) {
1602                 err = init_root_tree_recursive(steering, &init_node->children[i],
1603                                                &fs_ns->node,
1604                                                init_node, i);
1605                 if (err)
1606                         return err;
1607         }
1608         return 0;
1609 }
1610
1611 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
1612                                                        enum fs_flow_table_type
1613                                                        table_type)
1614 {
1615         struct mlx5_flow_root_namespace *root_ns;
1616         struct mlx5_flow_namespace *ns;
1617
1618         /* Create the root namespace */
1619         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1620         if (!root_ns)
1621                 return NULL;
1622
1623         root_ns->dev = steering->dev;
1624         root_ns->table_type = table_type;
1625
1626         ns = &root_ns->ns;
1627         fs_init_namespace(ns);
1628         mutex_init(&root_ns->chain_lock);
1629         tree_init_node(&ns->node, 1, NULL);
1630         tree_add_node(&ns->node, NULL);
1631
1632         return root_ns;
1633 }
1634
1635 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1636
1637 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1638 {
1639         struct fs_prio *prio;
1640
1641         fs_for_each_prio(prio, ns) {
1642                  /* This updates prio start_level and num_levels */
1643                 set_prio_attrs_in_prio(prio, acc_level);
1644                 acc_level += prio->num_levels;
1645         }
1646         return acc_level;
1647 }
1648
1649 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1650 {
1651         struct mlx5_flow_namespace *ns;
1652         int acc_level_ns = acc_level;
1653
1654         prio->start_level = acc_level;
1655         fs_for_each_ns(ns, prio)
1656                 /* This updates start_level and num_levels of ns's priority descendants */
1657                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1658         if (!prio->num_levels)
1659                 prio->num_levels = acc_level_ns - prio->start_level;
1660         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1661 }
1662
1663 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1664 {
1665         struct mlx5_flow_namespace *ns = &root_ns->ns;
1666         struct fs_prio *prio;
1667         int start_level = 0;
1668
1669         fs_for_each_prio(prio, ns) {
1670                 set_prio_attrs_in_prio(prio, start_level);
1671                 start_level += prio->num_levels;
1672         }
1673 }
1674
1675 #define ANCHOR_PRIO 0
1676 #define ANCHOR_SIZE 1
1677 #define ANCHOR_LEVEL 0
1678 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
1679 {
1680         struct mlx5_flow_namespace *ns = NULL;
1681         struct mlx5_flow_table *ft;
1682
1683         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1684         if (WARN_ON(!ns))
1685                 return -EINVAL;
1686         ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL);
1687         if (IS_ERR(ft)) {
1688                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
1689                 return PTR_ERR(ft);
1690         }
1691         return 0;
1692 }
1693
1694 static int init_root_ns(struct mlx5_flow_steering *steering)
1695 {
1696         int err;
1697
1698         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
1699         if (!steering->root_ns)
1700                 return -ENOMEM;
1701
1702         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
1703         if (err)
1704                 goto out_err;
1705
1706         set_prio_attrs(steering->root_ns);
1707
1708         err = create_anchor_flow_table(steering);
1709         if (err)
1710                 goto out_err;
1711
1712         return 0;
1713
1714 out_err:
1715         cleanup_root_ns(steering->root_ns);
1716         steering->root_ns = NULL;
1717         return err;
1718 }
1719
1720 static void clean_tree(struct fs_node *node)
1721 {
1722         if (node) {
1723                 struct fs_node *iter;
1724                 struct fs_node *temp;
1725
1726                 list_for_each_entry_safe(iter, temp, &node->children, list)
1727                         clean_tree(iter);
1728                 tree_remove_node(node);
1729         }
1730 }
1731
1732 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
1733 {
1734         if (!root_ns)
1735                 return;
1736
1737         clean_tree(&root_ns->ns.node);
1738 }
1739
1740 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1741 {
1742         struct mlx5_flow_steering *steering = dev->priv.steering;
1743
1744         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1745                 return;
1746
1747         cleanup_root_ns(steering->root_ns);
1748         cleanup_root_ns(steering->esw_egress_root_ns);
1749         cleanup_root_ns(steering->esw_ingress_root_ns);
1750         cleanup_root_ns(steering->fdb_root_ns);
1751         cleanup_root_ns(steering->sniffer_rx_root_ns);
1752         cleanup_root_ns(steering->sniffer_tx_root_ns);
1753         mlx5_cleanup_fc_stats(dev);
1754         kfree(steering);
1755 }
1756
1757 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
1758 {
1759         struct fs_prio *prio;
1760
1761         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
1762         if (!steering->sniffer_tx_root_ns)
1763                 return -ENOMEM;
1764
1765         /* Create single prio */
1766         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
1767         if (IS_ERR(prio)) {
1768                 cleanup_root_ns(steering->sniffer_tx_root_ns);
1769                 return PTR_ERR(prio);
1770         }
1771         return 0;
1772 }
1773
1774 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
1775 {
1776         struct fs_prio *prio;
1777
1778         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
1779         if (!steering->sniffer_rx_root_ns)
1780                 return -ENOMEM;
1781
1782         /* Create single prio */
1783         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
1784         if (IS_ERR(prio)) {
1785                 cleanup_root_ns(steering->sniffer_rx_root_ns);
1786                 return PTR_ERR(prio);
1787         }
1788         return 0;
1789 }
1790
1791 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
1792 {
1793         struct fs_prio *prio;
1794
1795         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
1796         if (!steering->fdb_root_ns)
1797                 return -ENOMEM;
1798
1799         prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
1800         if (IS_ERR(prio))
1801                 goto out_err;
1802
1803         prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
1804         if (IS_ERR(prio))
1805                 goto out_err;
1806
1807         set_prio_attrs(steering->fdb_root_ns);
1808         return 0;
1809
1810 out_err:
1811         cleanup_root_ns(steering->fdb_root_ns);
1812         steering->fdb_root_ns = NULL;
1813         return PTR_ERR(prio);
1814 }
1815
1816 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
1817 {
1818         struct fs_prio *prio;
1819
1820         steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
1821         if (!steering->esw_egress_root_ns)
1822                 return -ENOMEM;
1823
1824         /* create 1 prio*/
1825         prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
1826                               MLX5_TOTAL_VPORTS(steering->dev));
1827         return PTR_ERR_OR_ZERO(prio);
1828 }
1829
1830 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
1831 {
1832         struct fs_prio *prio;
1833
1834         steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
1835         if (!steering->esw_ingress_root_ns)
1836                 return -ENOMEM;
1837
1838         /* create 1 prio*/
1839         prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
1840                               MLX5_TOTAL_VPORTS(steering->dev));
1841         return PTR_ERR_OR_ZERO(prio);
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         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
1888                 err = init_sniffer_rx_root_ns(steering);
1889                 if (err)
1890                         goto err;
1891         }
1892
1893         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
1894                 err = init_sniffer_tx_root_ns(steering);
1895                 if (err)
1896                         goto err;
1897         }
1898
1899         return 0;
1900 err:
1901         mlx5_cleanup_fs(dev);
1902         return err;
1903 }