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 bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 157                                 struct mlx5_flow_destination *d2);
 158 static struct mlx5_flow_rule *
 159 find_flow_rule(struct fs_fte *fte,
 160                struct mlx5_flow_destination *dest);
 161
 162 static void tree_init_node(struct fs_node *node,
 163                            unsigned int refcount,
 164                            void (*remove_func)(struct fs_node *))
 165 {
 166         atomic_set(&node->refcount, refcount);
 167         INIT_LIST_HEAD(&node->list);
 168         INIT_LIST_HEAD(&node->children);
 169         mutex_init(&node->lock);
 170         node->remove_func = remove_func;
 171 }
 172
 173 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 174 {
 175         if (parent)
 176                 atomic_inc(&parent->refcount);
 177         node->parent = parent;
 178
 179         /* Parent is the root */
 180         if (!parent)
 181                 node->root = node;
 182         else
 183                 node->root = parent->root;
 184 }
 185
 186 static void tree_get_node(struct fs_node *node)
 187 {
 188         atomic_inc(&node->refcount);
 189 }
 190
 191 static void nested_lock_ref_node(struct fs_node *node,
 192                                  enum fs_i_mutex_lock_class class)
 193 {
 194         if (node) {
 195                 mutex_lock_nested(&node->lock, class);
 196                 atomic_inc(&node->refcount);
 197         }
 198 }
 199
 200 static void lock_ref_node(struct fs_node *node)
 201 {
 202         if (node) {
 203                 mutex_lock(&node->lock);
 204                 atomic_inc(&node->refcount);
 205         }
 206 }
 207
 208 static void unlock_ref_node(struct fs_node *node)
 209 {
 210         if (node) {
 211                 atomic_dec(&node->refcount);
 212                 mutex_unlock(&node->lock);
 213         }
 214 }
 215
 216 static void tree_put_node(struct fs_node *node)
 217 {
 218         struct fs_node *parent_node = node->parent;
 219
 220         lock_ref_node(parent_node);
 221         if (atomic_dec_and_test(&node->refcount)) {
 222                 if (parent_node)
 223                         list_del_init(&node->list);
 224                 if (node->remove_func)
 225                         node->remove_func(node);
 226                 kfree(node);
 227                 node = NULL;
 228         }
 229         unlock_ref_node(parent_node);
 230         if (!node && parent_node)
 231                 tree_put_node(parent_node);
 232 }
 233
 234 static int tree_remove_node(struct fs_node *node)
 235 {
 236         if (atomic_read(&node->refcount) > 1) {
 237                 atomic_dec(&node->refcount);
 238                 return -EEXIST;
 239         }
 240         tree_put_node(node);
 241         return 0;
 242 }
 243
 244 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 245                                  unsigned int prio)
 246 {
 247         struct fs_prio *iter_prio;
 248
 249         fs_for_each_prio(iter_prio, ns) {
 250                 if (iter_prio->prio == prio)
 251                         return iter_prio;
 252         }
 253
 254         return NULL;
 255 }
 256
 257 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 258 {
 259         unsigned int i;
 260
 261         for (i = 0; i < size; i++, mask++, val1++, val2++)
 262                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 263                     ((*(u8 *)val2) & (*(u8 *)mask)))
 264                         return false;
 265
 266         return true;
 267 }
 268
 269 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 270                                 void *fte_param1, void *fte_param2)
 271 {
 272         if (mask->match_criteria_enable &
 273             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 274                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 275                                                 fte_param1, outer_headers);
 276                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 277                                                 fte_param2, outer_headers);
 278                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 279                                               mask->match_criteria, outer_headers);
 280
 281                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 282                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 283                         return false;
 284         }
 285
 286         if (mask->match_criteria_enable &
 287             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 288                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 289                                                 fte_param1, misc_parameters);
 290                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 291                                                 fte_param2, misc_parameters);
 292                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 293                                           mask->match_criteria, misc_parameters);
 294
 295                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 296                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 297                         return false;
 298         }
 299
 300         if (mask->match_criteria_enable &
 301             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 302                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 303                                                 fte_param1, inner_headers);
 304                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 305                                                 fte_param2, inner_headers);
 306                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 307                                           mask->match_criteria, inner_headers);
 308
 309                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 310                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 311                         return false;
 312         }
 313         return true;
 314 }
 315
 316 static bool compare_match_criteria(u8 match_criteria_enable1,
 317                                    u8 match_criteria_enable2,
 318                                    void *mask1, void *mask2)
 319 {
 320         return match_criteria_enable1 == match_criteria_enable2 &&
 321                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 322 }
 323
 324 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 325 {
 326         struct fs_node *root;
 327         struct mlx5_flow_namespace *ns;
 328
 329         root = node->root;
 330
 331         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 332                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 333                 return NULL;
 334         }
 335
 336         ns = container_of(root, struct mlx5_flow_namespace, node);
 337         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 338 }
 339
 340 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 341 {
 342         struct mlx5_flow_root_namespace *root = find_root(node);
 343
 344         if (root)
 345                 return root->dev;
 346         return NULL;
 347 }
 348
 349 static void del_flow_table(struct fs_node *node)
 350 {
 351         struct mlx5_flow_table *ft;
 352         struct mlx5_core_dev *dev;
 353         struct fs_prio *prio;
 354         int err;
 355
 356         fs_get_obj(ft, node);
 357         dev = get_dev(&ft->node);
 358
 359         err = mlx5_cmd_destroy_flow_table(dev, ft);
 360         if (err)
 361                 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 362         fs_get_obj(prio, ft->node.parent);
 363         prio->num_ft--;
 364 }
 365
 366 static void del_rule(struct fs_node *node)
 367 {
 368         struct mlx5_flow_rule *rule;
 369         struct mlx5_flow_table *ft;
 370         struct mlx5_flow_group *fg;
 371         struct fs_fte *fte;
 372         u32     *match_value;
 373         int modify_mask;
 374         struct mlx5_core_dev *dev = get_dev(node);
 375         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 376         int err;
 377         bool update_fte = false;
 378
 379         match_value = mlx5_vzalloc(match_len);
 380         if (!match_value) {
 381                 mlx5_core_warn(dev, "failed to allocate inbox\n");
 382                 return;
 383         }
 384
 385         fs_get_obj(rule, node);
 386         fs_get_obj(fte, rule->node.parent);
 387         fs_get_obj(fg, fte->node.parent);
 388         memcpy(match_value, fte->val, sizeof(fte->val));
 389         fs_get_obj(ft, fg->node.parent);
 390         list_del(&rule->node.list);
 391         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 392                 mutex_lock(&rule->dest_attr.ft->lock);
 393                 list_del(&rule->next_ft);
 394                 mutex_unlock(&rule->dest_attr.ft->lock);
 395         }
 396
 397         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 398             --fte->dests_size) {
 399                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
 400                 fte->action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 401                 update_fte = true;
 402                 goto out;
 403         }
 404
 405         if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 406             --fte->dests_size) {
 407                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST),
 408                 update_fte = true;
 409         }
 410 out:
 411         if (update_fte && fte->dests_size) {
 412                 err = mlx5_cmd_update_fte(dev, ft, fg->id, modify_mask, fte);
 413                 if (err)
 414                         mlx5_core_warn(dev,
 415                                        "%s can't del rule fg id=%d fte_index=%d\n",
 416                                        __func__, fg->id, fte->index);
 417         }
 418         kvfree(match_value);
 419 }
 420
 421 static void del_fte(struct fs_node *node)
 422 {
 423         struct mlx5_flow_table *ft;
 424         struct mlx5_flow_group *fg;
 425         struct mlx5_core_dev *dev;
 426         struct fs_fte *fte;
 427         int err;
 428
 429         fs_get_obj(fte, node);
 430         fs_get_obj(fg, fte->node.parent);
 431         fs_get_obj(ft, fg->node.parent);
 432
 433         dev = get_dev(&ft->node);
 434         err = mlx5_cmd_delete_fte(dev, ft,
 435                                   fte->index);
 436         if (err)
 437                 mlx5_core_warn(dev,
 438                                "flow steering can't delete fte in index %d of flow group id %d\n",
 439                                fte->index, fg->id);
 440
 441         fte->status = 0;
 442         fg->num_ftes--;
 443 }
 444
 445 static void del_flow_group(struct fs_node *node)
 446 {
 447         struct mlx5_flow_group *fg;
 448         struct mlx5_flow_table *ft;
 449         struct mlx5_core_dev *dev;
 450
 451         fs_get_obj(fg, node);
 452         fs_get_obj(ft, fg->node.parent);
 453         dev = get_dev(&ft->node);
 454
 455         if (ft->autogroup.active)
 456                 ft->autogroup.num_groups--;
 457
 458         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 459                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 460                                fg->id, ft->id);
 461 }
 462
 463 static struct fs_fte *alloc_fte(struct mlx5_flow_act *flow_act,
 464                                 u32 *match_value,
 465                                 unsigned int index)
 466 {
 467         struct fs_fte *fte;
 468
 469         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 470         if (!fte)
 471                 return ERR_PTR(-ENOMEM);
 472
 473         memcpy(fte->val, match_value, sizeof(fte->val));
 474         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 475         fte->flow_tag = flow_act->flow_tag;
 476         fte->index = index;
 477         fte->action = flow_act->action;
 478         fte->encap_id = flow_act->encap_id;
 479
 480         return fte;
 481 }
 482
 483 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 484 {
 485         struct mlx5_flow_group *fg;
 486         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 487                                             create_fg_in, match_criteria);
 488         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 489                                             create_fg_in,
 490                                             match_criteria_enable);
 491         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 492         if (!fg)
 493                 return ERR_PTR(-ENOMEM);
 494
 495         fg->mask.match_criteria_enable = match_criteria_enable;
 496         memcpy(&fg->mask.match_criteria, match_criteria,
 497                sizeof(fg->mask.match_criteria));
 498         fg->node.type =  FS_TYPE_FLOW_GROUP;
 499         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 500                                    start_flow_index);
 501         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 502                                 end_flow_index) - fg->start_index + 1;
 503         return fg;
 504 }
 505
 506 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 507                                                 enum fs_flow_table_type table_type,
 508                                                 enum fs_flow_table_op_mod op_mod,
 509                                                 u32 flags)
 510 {
 511         struct mlx5_flow_table *ft;
 512
 513         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 514         if (!ft)
 515                 return NULL;
 516
 517         ft->level = level;
 518         ft->node.type = FS_TYPE_FLOW_TABLE;
 519         ft->op_mod = op_mod;
 520         ft->type = table_type;
 521         ft->vport = vport;
 522         ft->max_fte = max_fte;
 523         ft->flags = flags;
 524         INIT_LIST_HEAD(&ft->fwd_rules);
 525         mutex_init(&ft->lock);
 526
 527         return ft;
 528 }
 529
 530 /* If reverse is false, then we search for the first flow table in the
 531  * root sub-tree from start(closest from right), else we search for the
 532  * last flow table in the root sub-tree till start(closest from left).
 533  */
 534 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 535                                                          struct list_head *start,
 536                                                          bool reverse)
 537 {
 538 #define list_advance_entry(pos, reverse)                \
 539         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 540
 541 #define list_for_each_advance_continue(pos, head, reverse)      \
 542         for (pos = list_advance_entry(pos, reverse);            \
 543              &pos->list != (head);                              \
 544              pos = list_advance_entry(pos, reverse))
 545
 546         struct fs_node *iter = list_entry(start, struct fs_node, list);
 547         struct mlx5_flow_table *ft = NULL;
 548
 549         if (!root)
 550                 return NULL;
 551
 552         list_for_each_advance_continue(iter, &root->children, reverse) {
 553                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 554                         fs_get_obj(ft, iter);
 555                         return ft;
 556                 }
 557                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 558                 if (ft)
 559                         return ft;
 560         }
 561
 562         return ft;
 563 }
 564
 565 /* If reverse if false then return the first flow table in next priority of
 566  * prio in the tree, else return the last flow table in the previous priority
 567  * of prio in the tree.
 568  */
 569 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 570 {
 571         struct mlx5_flow_table *ft = NULL;
 572         struct fs_node *curr_node;
 573         struct fs_node *parent;
 574
 575         parent = prio->node.parent;
 576         curr_node = &prio->node;
 577         while (!ft && parent) {
 578                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 579                 curr_node = parent;
 580                 parent = curr_node->parent;
 581         }
 582         return ft;
 583 }
 584
 585 /* Assuming all the tree is locked by mutex chain lock */
 586 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 587 {
 588         return find_closest_ft(prio, false);
 589 }
 590
 591 /* Assuming all the tree is locked by mutex chain lock */
 592 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 593 {
 594         return find_closest_ft(prio, true);
 595 }
 596
 597 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 598                                struct fs_prio *prio,
 599                                struct mlx5_flow_table *ft)
 600 {
 601         struct mlx5_flow_table *iter;
 602         int i = 0;
 603         int err;
 604
 605         fs_for_each_ft(iter, prio) {
 606                 i++;
 607                 err = mlx5_cmd_modify_flow_table(dev,
 608                                                  iter,
 609                                                  ft);
 610                 if (err) {
 611                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 612                                        iter->id);
 613                         /* The driver is out of sync with the FW */
 614                         if (i > 1)
 615                                 WARN_ON(true);
 616                         return err;
 617                 }
 618         }
 619         return 0;
 620 }
 621
 622 /* Connect flow tables from previous priority of prio to ft */
 623 static int connect_prev_fts(struct mlx5_core_dev *dev,
 624                             struct mlx5_flow_table *ft,
 625                             struct fs_prio *prio)
 626 {
 627         struct mlx5_flow_table *prev_ft;
 628
 629         prev_ft = find_prev_chained_ft(prio);
 630         if (prev_ft) {
 631                 struct fs_prio *prev_prio;
 632
 633                 fs_get_obj(prev_prio, prev_ft->node.parent);
 634                 return connect_fts_in_prio(dev, prev_prio, ft);
 635         }
 636         return 0;
 637 }
 638
 639 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 640                                  *prio)
 641 {
 642         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 643         int min_level = INT_MAX;
 644         int err;
 645
 646         if (root->root_ft)
 647                 min_level = root->root_ft->level;
 648
 649         if (ft->level >= min_level)
 650                 return 0;
 651
 652         err = mlx5_cmd_update_root_ft(root->dev, ft);
 653         if (err)
 654                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 655                                ft->id);
 656         else
 657                 root->root_ft = ft;
 658
 659         return err;
 660 }
 661
 662 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 663                                          struct mlx5_flow_destination *dest)
 664 {
 665         struct mlx5_flow_table *ft;
 666         struct mlx5_flow_group *fg;
 667         struct fs_fte *fte;
 668         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 669         int err = 0;
 670
 671         fs_get_obj(fte, rule->node.parent);
 672         if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 673                 return -EINVAL;
 674         lock_ref_node(&fte->node);
 675         fs_get_obj(fg, fte->node.parent);
 676         fs_get_obj(ft, fg->node.parent);
 677
 678         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 679         err = mlx5_cmd_update_fte(get_dev(&ft->node),
 680                                   ft, fg->id,
 681                                   modify_mask,
 682                                   fte);
 683         unlock_ref_node(&fte->node);
 684
 685         return err;
 686 }
 687
 688 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
 689                                  struct mlx5_flow_destination *new_dest,
 690                                  struct mlx5_flow_destination *old_dest)
 691 {
 692         int i;
 693
 694         if (!old_dest) {
 695                 if (handle->num_rules != 1)
 696                         return -EINVAL;
 697                 return _mlx5_modify_rule_destination(handle->rule[0],
 698                                                      new_dest);
 699         }
 700
 701         for (i = 0; i < handle->num_rules; i++) {
 702                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
 703                         return _mlx5_modify_rule_destination(handle->rule[i],
 704                                                              new_dest);
 705         }
 706
 707         return -EINVAL;
 708 }
 709
 710 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 711 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 712                              struct mlx5_flow_table *new_next_ft,
 713                              struct mlx5_flow_table *old_next_ft)
 714 {
 715         struct mlx5_flow_destination dest;
 716         struct mlx5_flow_rule *iter;
 717         int err = 0;
 718
 719         /* new_next_ft and old_next_ft could be NULL only
 720          * when we create/destroy the anchor flow table.
 721          */
 722         if (!new_next_ft || !old_next_ft)
 723                 return 0;
 724
 725         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 726         dest.ft = new_next_ft;
 727
 728         mutex_lock(&old_next_ft->lock);
 729         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 730         mutex_unlock(&old_next_ft->lock);
 731         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 732                 err = _mlx5_modify_rule_destination(iter, &dest);
 733                 if (err)
 734                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 735                                new_next_ft->id);
 736         }
 737         return 0;
 738 }
 739
 740 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 741                               struct fs_prio *prio)
 742 {
 743         struct mlx5_flow_table *next_ft;
 744         int err = 0;
 745
 746         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 747
 748         if (list_empty(&prio->node.children)) {
 749                 err = connect_prev_fts(dev, ft, prio);
 750                 if (err)
 751                         return err;
 752
 753                 next_ft = find_next_chained_ft(prio);
 754                 err = connect_fwd_rules(dev, ft, next_ft);
 755                 if (err)
 756                         return err;
 757         }
 758
 759         if (MLX5_CAP_FLOWTABLE(dev,
 760                                flow_table_properties_nic_receive.modify_root))
 761                 err = update_root_ft_create(ft, prio);
 762         return err;
 763 }
 764
 765 static void list_add_flow_table(struct mlx5_flow_table *ft,
 766                                 struct fs_prio *prio)
 767 {
 768         struct list_head *prev = &prio->node.children;
 769         struct mlx5_flow_table *iter;
 770
 771         fs_for_each_ft(iter, prio) {
 772                 if (iter->level > ft->level)
 773                         break;
 774                 prev = &iter->node.list;
 775         }
 776         list_add(&ft->node.list, prev);
 777 }
 778
 779 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 780                                                         enum fs_flow_table_op_mod op_mod,
 781                                                         u16 vport, int prio,
 782                                                         int max_fte, u32 level,
 783                                                         u32 flags)
 784 {
 785         struct mlx5_flow_table *next_ft = NULL;
 786         struct mlx5_flow_table *ft;
 787         int err;
 788         int log_table_sz;
 789         struct mlx5_flow_root_namespace *root =
 790                 find_root(&ns->node);
 791         struct fs_prio *fs_prio = NULL;
 792
 793         if (!root) {
 794                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 795                 return ERR_PTR(-ENODEV);
 796         }
 797
 798         mutex_lock(&root->chain_lock);
 799         fs_prio = find_prio(ns, prio);
 800         if (!fs_prio) {
 801                 err = -EINVAL;
 802                 goto unlock_root;
 803         }
 804         if (level >= fs_prio->num_levels) {
 805                 err = -ENOSPC;
 806                 goto unlock_root;
 807         }
 808         /* The level is related to the
 809          * priority level range.
 810          */
 811         level += fs_prio->start_level;
 812         ft = alloc_flow_table(level,
 813                               vport,
 814                               max_fte ? roundup_pow_of_two(max_fte) : 0,
 815                               root->table_type,
 816                               op_mod, flags);
 817         if (!ft) {
 818                 err = -ENOMEM;
 819                 goto unlock_root;
 820         }
 821
 822         tree_init_node(&ft->node, 1, del_flow_table);
 823         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
 824         next_ft = find_next_chained_ft(fs_prio);
 825         err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->op_mod, ft->type,
 826                                          ft->level, log_table_sz, next_ft, &ft->id,
 827                                          ft->flags);
 828         if (err)
 829                 goto free_ft;
 830
 831         err = connect_flow_table(root->dev, ft, fs_prio);
 832         if (err)
 833                 goto destroy_ft;
 834         lock_ref_node(&fs_prio->node);
 835         tree_add_node(&ft->node, &fs_prio->node);
 836         list_add_flow_table(ft, fs_prio);
 837         fs_prio->num_ft++;
 838         unlock_ref_node(&fs_prio->node);
 839         mutex_unlock(&root->chain_lock);
 840         return ft;
 841 destroy_ft:
 842         mlx5_cmd_destroy_flow_table(root->dev, ft);
 843 free_ft:
 844         kfree(ft);
 845 unlock_root:
 846         mutex_unlock(&root->chain_lock);
 847         return ERR_PTR(err);
 848 }
 849
 850 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 851                                                int prio, int max_fte,
 852                                                u32 level,
 853                                                u32 flags)
 854 {
 855         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, 0, prio,
 856                                         max_fte, level, flags);
 857 }
 858
 859 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
 860                                                      int prio, int max_fte,
 861                                                      u32 level, u16 vport)
 862 {
 863         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, vport, prio,
 864                                         max_fte, level, 0);
 865 }
 866
 867 struct mlx5_flow_table *mlx5_create_lag_demux_flow_table(
 868                                                struct mlx5_flow_namespace *ns,
 869                                                int prio, u32 level)
 870 {
 871         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_LAG_DEMUX, 0, prio, 0,
 872                                         level, 0);
 873 }
 874 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
 875
 876 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 877                                                             int prio,
 878                                                             int num_flow_table_entries,
 879                                                             int max_num_groups,
 880                                                             u32 level,
 881                                                             u32 flags)
 882 {
 883         struct mlx5_flow_table *ft;
 884
 885         if (max_num_groups > num_flow_table_entries)
 886                 return ERR_PTR(-EINVAL);
 887
 888         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries, level, flags);
 889         if (IS_ERR(ft))
 890                 return ft;
 891
 892         ft->autogroup.active = true;
 893         ft->autogroup.required_groups = max_num_groups;
 894
 895         return ft;
 896 }
 897 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 898
 899 /* Flow table should be locked */
 900 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 901                                                         u32 *fg_in,
 902                                                         struct list_head
 903                                                         *prev_fg,
 904                                                         bool is_auto_fg)
 905 {
 906         struct mlx5_flow_group *fg;
 907         struct mlx5_core_dev *dev = get_dev(&ft->node);
 908         int err;
 909
 910         if (!dev)
 911                 return ERR_PTR(-ENODEV);
 912
 913         fg = alloc_flow_group(fg_in);
 914         if (IS_ERR(fg))
 915                 return fg;
 916
 917         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 918         if (err) {
 919                 kfree(fg);
 920                 return ERR_PTR(err);
 921         }
 922
 923         if (ft->autogroup.active)
 924                 ft->autogroup.num_groups++;
 925         /* Add node to tree */
 926         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 927         tree_add_node(&fg->node, &ft->node);
 928         /* Add node to group list */
 929         list_add(&fg->node.list, prev_fg);
 930
 931         return fg;
 932 }
 933
 934 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 935                                                u32 *fg_in)
 936 {
 937         struct mlx5_flow_group *fg;
 938
 939         if (ft->autogroup.active)
 940                 return ERR_PTR(-EPERM);
 941
 942         lock_ref_node(&ft->node);
 943         fg = create_flow_group_common(ft, fg_in, ft->node.children.prev, false);
 944         unlock_ref_node(&ft->node);
 945
 946         return fg;
 947 }
 948
 949 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 950 {
 951         struct mlx5_flow_rule *rule;
 952
 953         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 954         if (!rule)
 955                 return NULL;
 956
 957         INIT_LIST_HEAD(&rule->next_ft);
 958         rule->node.type = FS_TYPE_FLOW_DEST;
 959         if (dest)
 960                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 961
 962         return rule;
 963 }
 964
 965 static struct mlx5_flow_handle *alloc_handle(int num_rules)
 966 {
 967         struct mlx5_flow_handle *handle;
 968
 969         handle = kzalloc(sizeof(*handle) + sizeof(handle->rule[0]) *
 970                           num_rules, GFP_KERNEL);
 971         if (!handle)
 972                 return NULL;
 973
 974         handle->num_rules = num_rules;
 975
 976         return handle;
 977 }
 978
 979 static void destroy_flow_handle(struct fs_fte *fte,
 980                                 struct mlx5_flow_handle *handle,
 981                                 struct mlx5_flow_destination *dest,
 982                                 int i)
 983 {
 984         for (; --i >= 0;) {
 985                 if (atomic_dec_and_test(&handle->rule[i]->node.refcount)) {
 986                         fte->dests_size--;
 987                         list_del(&handle->rule[i]->node.list);
 988                         kfree(handle->rule[i]);
 989                 }
 990         }
 991         kfree(handle);
 992 }
 993
 994 static struct mlx5_flow_handle *
 995 create_flow_handle(struct fs_fte *fte,
 996                    struct mlx5_flow_destination *dest,
 997                    int dest_num,
 998                    int *modify_mask,
 999                    bool *new_rule)
1000 {
1001         struct mlx5_flow_handle *handle;
1002         struct mlx5_flow_rule *rule = NULL;
1003         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1004         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1005         int type;
1006         int i = 0;
1007
1008         handle = alloc_handle((dest_num) ? dest_num : 1);
1009         if (!handle)
1010                 return ERR_PTR(-ENOMEM);
1011
1012         do {
1013                 if (dest) {
1014                         rule = find_flow_rule(fte, dest + i);
1015                         if (rule) {
1016                                 atomic_inc(&rule->node.refcount);
1017                                 goto rule_found;
1018                         }
1019                 }
1020
1021                 *new_rule = true;
1022                 rule = alloc_rule(dest + i);
1023                 if (!rule)
1024                         goto free_rules;
1025
1026                 /* Add dest to dests list- we need flow tables to be in the
1027                  * end of the list for forward to next prio rules.
1028                  */
1029                 tree_init_node(&rule->node, 1, del_rule);
1030                 if (dest &&
1031                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1032                         list_add(&rule->node.list, &fte->node.children);
1033                 else
1034                         list_add_tail(&rule->node.list, &fte->node.children);
1035                 if (dest) {
1036                         fte->dests_size++;
1037
1038                         type = dest[i].type ==
1039                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1040                         *modify_mask |= type ? count : dst;
1041                 }
1042 rule_found:
1043                 handle->rule[i] = rule;
1044         } while (++i < dest_num);
1045
1046         return handle;
1047
1048 free_rules:
1049         destroy_flow_handle(fte, handle, dest, i);
1050         return ERR_PTR(-ENOMEM);
1051 }
1052
1053 /* fte should not be deleted while calling this function */
1054 static struct mlx5_flow_handle *
1055 add_rule_fte(struct fs_fte *fte,
1056              struct mlx5_flow_group *fg,
1057              struct mlx5_flow_destination *dest,
1058              int dest_num,
1059              bool update_action)
1060 {
1061         struct mlx5_flow_handle *handle;
1062         struct mlx5_flow_table *ft;
1063         int modify_mask = 0;
1064         int err;
1065         bool new_rule = false;
1066
1067         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1068                                     &new_rule);
1069         if (IS_ERR(handle) || !new_rule)
1070                 goto out;
1071
1072         if (update_action)
1073                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1074
1075         fs_get_obj(ft, fg->node.parent);
1076         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1077                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
1078                                           ft, fg->id, fte);
1079         else
1080                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
1081                                           ft, fg->id, modify_mask, fte);
1082         if (err)
1083                 goto free_handle;
1084
1085         fte->status |= FS_FTE_STATUS_EXISTING;
1086
1087 out:
1088         return handle;
1089
1090 free_handle:
1091         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1092         return ERR_PTR(err);
1093 }
1094
1095 /* Assumed fg is locked */
1096 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
1097                                        struct list_head **prev)
1098 {
1099         struct fs_fte *fte;
1100         unsigned int start = fg->start_index;
1101
1102         if (prev)
1103                 *prev = &fg->node.children;
1104
1105         /* assumed list is sorted by index */
1106         fs_for_each_fte(fte, fg) {
1107                 if (fte->index != start)
1108                         return start;
1109                 start++;
1110                 if (prev)
1111                         *prev = &fte->node.list;
1112         }
1113
1114         return start;
1115 }
1116
1117 /* prev is output, prev->next = new_fte */
1118 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
1119                                  u32 *match_value,
1120                                  struct mlx5_flow_act *flow_act,
1121                                  struct list_head **prev)
1122 {
1123         struct fs_fte *fte;
1124         int index;
1125
1126         index = get_free_fte_index(fg, prev);
1127         fte = alloc_fte(flow_act, match_value, index);
1128         if (IS_ERR(fte))
1129                 return fte;
1130
1131         return fte;
1132 }
1133
1134 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
1135                                                 u8 match_criteria_enable,
1136                                                 u32 *match_criteria)
1137 {
1138         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1139         struct list_head *prev = ft->node.children.prev;
1140         unsigned int candidate_index = 0;
1141         struct mlx5_flow_group *fg;
1142         void *match_criteria_addr;
1143         unsigned int group_size = 0;
1144         u32 *in;
1145
1146         if (!ft->autogroup.active)
1147                 return ERR_PTR(-ENOENT);
1148
1149         in = mlx5_vzalloc(inlen);
1150         if (!in)
1151                 return ERR_PTR(-ENOMEM);
1152
1153         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1154                 /* We save place for flow groups in addition to max types */
1155                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1156
1157         /*  ft->max_fte == ft->autogroup.max_types */
1158         if (group_size == 0)
1159                 group_size = 1;
1160
1161         /* sorted by start_index */
1162         fs_for_each_fg(fg, ft) {
1163                 if (candidate_index + group_size > fg->start_index)
1164                         candidate_index = fg->start_index + fg->max_ftes;
1165                 else
1166                         break;
1167                 prev = &fg->node.list;
1168         }
1169
1170         if (candidate_index + group_size > ft->max_fte) {
1171                 fg = ERR_PTR(-ENOSPC);
1172                 goto out;
1173         }
1174
1175         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1176                  match_criteria_enable);
1177         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1178         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
1179                  group_size - 1);
1180         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1181                                            in, match_criteria);
1182         memcpy(match_criteria_addr, match_criteria,
1183                MLX5_ST_SZ_BYTES(fte_match_param));
1184
1185         fg = create_flow_group_common(ft, in, prev, true);
1186 out:
1187         kvfree(in);
1188         return fg;
1189 }
1190
1191 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1192                                 struct mlx5_flow_destination *d2)
1193 {
1194         if (d1->type == d2->type) {
1195                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1196                      d1->vport_num == d2->vport_num) ||
1197                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1198                      d1->ft == d2->ft) ||
1199                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1200                      d1->tir_num == d2->tir_num))
1201                         return true;
1202         }
1203
1204         return false;
1205 }
1206
1207 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1208                                              struct mlx5_flow_destination *dest)
1209 {
1210         struct mlx5_flow_rule *rule;
1211
1212         list_for_each_entry(rule, &fte->node.children, node.list) {
1213                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1214                         return rule;
1215         }
1216         return NULL;
1217 }
1218
1219 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1220                                             u32 *match_value,
1221                                             struct mlx5_flow_act *flow_act,
1222                                             struct mlx5_flow_destination *dest,
1223                                             int dest_num)
1224 {
1225         struct mlx5_flow_handle *handle;
1226         struct mlx5_flow_table *ft;
1227         struct list_head *prev;
1228         struct fs_fte *fte;
1229         int i;
1230
1231         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1232         fs_for_each_fte(fte, fg) {
1233                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1234                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1235                     (flow_act->action & fte->action) &&
1236                     flow_act->flow_tag == fte->flow_tag) {
1237                         int old_action = fte->action;
1238
1239                         fte->action |= flow_act->action;
1240                         handle = add_rule_fte(fte, fg, dest, dest_num,
1241                                               old_action != flow_act->action);
1242                         if (IS_ERR(handle)) {
1243                                 fte->action = old_action;
1244                                 goto unlock_fte;
1245                         } else {
1246                                 goto add_rules;
1247                         }
1248                 }
1249                 unlock_ref_node(&fte->node);
1250         }
1251         fs_get_obj(ft, fg->node.parent);
1252         if (fg->num_ftes >= fg->max_ftes) {
1253                 handle = ERR_PTR(-ENOSPC);
1254                 goto unlock_fg;
1255         }
1256
1257         fte = create_fte(fg, match_value, flow_act, &prev);
1258         if (IS_ERR(fte)) {
1259                 handle = (void *)fte;
1260                 goto unlock_fg;
1261         }
1262         tree_init_node(&fte->node, 0, del_fte);
1263         nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1264         handle = add_rule_fte(fte, fg, dest, dest_num, false);
1265         if (IS_ERR(handle)) {
1266                 unlock_ref_node(&fte->node);
1267                 kfree(fte);
1268                 goto unlock_fg;
1269         }
1270
1271         fg->num_ftes++;
1272
1273         tree_add_node(&fte->node, &fg->node);
1274         list_add(&fte->node.list, prev);
1275 add_rules:
1276         for (i = 0; i < handle->num_rules; i++) {
1277                 if (atomic_read(&handle->rule[i]->node.refcount) == 1)
1278                         tree_add_node(&handle->rule[i]->node, &fte->node);
1279         }
1280 unlock_fte:
1281         unlock_ref_node(&fte->node);
1282 unlock_fg:
1283         unlock_ref_node(&fg->node);
1284         return handle;
1285 }
1286
1287 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_handle *handle)
1288 {
1289         struct mlx5_flow_rule *dst;
1290         struct fs_fte *fte;
1291
1292         fs_get_obj(fte, handle->rule[0]->node.parent);
1293
1294         fs_for_each_dst(dst, fte) {
1295                 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1296                         return dst->dest_attr.counter;
1297         }
1298
1299         return NULL;
1300 }
1301
1302 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1303 {
1304         if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1305                 return !counter;
1306
1307         if (!counter)
1308                 return false;
1309
1310         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1311                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1312 }
1313
1314 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1315                           u32 action,
1316                           struct mlx5_flow_table *ft)
1317 {
1318         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1319                 return counter_is_valid(dest->counter, action);
1320
1321         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1322                 return true;
1323
1324         if (!dest || ((dest->type ==
1325             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1326             (dest->ft->level <= ft->level)))
1327                 return false;
1328         return true;
1329 }
1330
1331 static struct mlx5_flow_handle *
1332 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1333                      struct mlx5_flow_spec *spec,
1334                      struct mlx5_flow_act *flow_act,
1335                      struct mlx5_flow_destination *dest,
1336                      int dest_num)
1337
1338 {
1339         struct mlx5_flow_group *g;
1340         struct mlx5_flow_handle *rule;
1341         int i;
1342
1343         for (i = 0; i < dest_num; i++) {
1344                 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1345                         return ERR_PTR(-EINVAL);
1346         }
1347
1348         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1349         fs_for_each_fg(g, ft)
1350                 if (compare_match_criteria(g->mask.match_criteria_enable,
1351                                            spec->match_criteria_enable,
1352                                            g->mask.match_criteria,
1353                                            spec->match_criteria)) {
1354                         rule = add_rule_fg(g, spec->match_value,
1355                                            flow_act, dest, dest_num);
1356                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1357                                 goto unlock;
1358                 }
1359
1360         g = create_autogroup(ft, spec->match_criteria_enable,
1361                              spec->match_criteria);
1362         if (IS_ERR(g)) {
1363                 rule = (void *)g;
1364                 goto unlock;
1365         }
1366
1367         rule = add_rule_fg(g, spec->match_value, flow_act, dest, dest_num);
1368         if (IS_ERR(rule)) {
1369                 /* Remove assumes refcount > 0 and autogroup creates a group
1370                  * with a refcount = 0.
1371                  */
1372                 unlock_ref_node(&ft->node);
1373                 tree_get_node(&g->node);
1374                 tree_remove_node(&g->node);
1375                 return rule;
1376         }
1377 unlock:
1378         unlock_ref_node(&ft->node);
1379         return rule;
1380 }
1381
1382 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1383 {
1384         return ((ft->type == FS_FT_NIC_RX) &&
1385                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1386 }
1387
1388 struct mlx5_flow_handle *
1389 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1390                     struct mlx5_flow_spec *spec,
1391                     struct mlx5_flow_act *flow_act,
1392                     struct mlx5_flow_destination *dest,
1393                     int dest_num)
1394 {
1395         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1396         struct mlx5_flow_destination gen_dest;
1397         struct mlx5_flow_table *next_ft = NULL;
1398         struct mlx5_flow_handle *handle = NULL;
1399         u32 sw_action = flow_act->action;
1400         struct fs_prio *prio;
1401
1402         fs_get_obj(prio, ft->node.parent);
1403         if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1404                 if (!fwd_next_prio_supported(ft))
1405                         return ERR_PTR(-EOPNOTSUPP);
1406                 if (dest)
1407                         return ERR_PTR(-EINVAL);
1408                 mutex_lock(&root->chain_lock);
1409                 next_ft = find_next_chained_ft(prio);
1410                 if (next_ft) {
1411                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1412                         gen_dest.ft = next_ft;
1413                         dest = &gen_dest;
1414                         dest_num = 1;
1415                         flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1416                 } else {
1417                         mutex_unlock(&root->chain_lock);
1418                         return ERR_PTR(-EOPNOTSUPP);
1419                 }
1420         }
1421
1422         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, dest_num);
1423
1424         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1425                 if (!IS_ERR_OR_NULL(handle) &&
1426                     (list_empty(&handle->rule[0]->next_ft))) {
1427                         mutex_lock(&next_ft->lock);
1428                         list_add(&handle->rule[0]->next_ft,
1429                                  &next_ft->fwd_rules);
1430                         mutex_unlock(&next_ft->lock);
1431                         handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1432                 }
1433                 mutex_unlock(&root->chain_lock);
1434         }
1435         return handle;
1436 }
1437 EXPORT_SYMBOL(mlx5_add_flow_rules);
1438
1439 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1440 {
1441         int i;
1442
1443         for (i = handle->num_rules - 1; i >= 0; i--)
1444                 tree_remove_node(&handle->rule[i]->node);
1445         kfree(handle);
1446 }
1447 EXPORT_SYMBOL(mlx5_del_flow_rules);
1448
1449 /* Assuming prio->node.children(flow tables) is sorted by level */
1450 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1451 {
1452         struct fs_prio *prio;
1453
1454         fs_get_obj(prio, ft->node.parent);
1455
1456         if (!list_is_last(&ft->node.list, &prio->node.children))
1457                 return list_next_entry(ft, node.list);
1458         return find_next_chained_ft(prio);
1459 }
1460
1461 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1462 {
1463         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1464         struct mlx5_flow_table *new_root_ft = NULL;
1465
1466         if (root->root_ft != ft)
1467                 return 0;
1468
1469         new_root_ft = find_next_ft(ft);
1470         if (new_root_ft) {
1471                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1472
1473                 if (err) {
1474                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1475                                        ft->id);
1476                         return err;
1477                 }
1478         }
1479         root->root_ft = new_root_ft;
1480         return 0;
1481 }
1482
1483 /* Connect flow table from previous priority to
1484  * the next flow table.
1485  */
1486 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1487 {
1488         struct mlx5_core_dev *dev = get_dev(&ft->node);
1489         struct mlx5_flow_table *next_ft;
1490         struct fs_prio *prio;
1491         int err = 0;
1492
1493         err = update_root_ft_destroy(ft);
1494         if (err)
1495                 return err;
1496
1497         fs_get_obj(prio, ft->node.parent);
1498         if  (!(list_first_entry(&prio->node.children,
1499                                 struct mlx5_flow_table,
1500                                 node.list) == ft))
1501                 return 0;
1502
1503         next_ft = find_next_chained_ft(prio);
1504         err = connect_fwd_rules(dev, next_ft, ft);
1505         if (err)
1506                 return err;
1507
1508         err = connect_prev_fts(dev, next_ft, prio);
1509         if (err)
1510                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1511                                ft->id);
1512         return err;
1513 }
1514
1515 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1516 {
1517         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1518         int err = 0;
1519
1520         mutex_lock(&root->chain_lock);
1521         err = disconnect_flow_table(ft);
1522         if (err) {
1523                 mutex_unlock(&root->chain_lock);
1524                 return err;
1525         }
1526         if (tree_remove_node(&ft->node))
1527                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1528                                ft->id);
1529         mutex_unlock(&root->chain_lock);
1530
1531         return err;
1532 }
1533 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1534
1535 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1536 {
1537         if (tree_remove_node(&fg->node))
1538                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1539                                fg->id);
1540 }
1541
1542 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1543                                                     enum mlx5_flow_namespace_type type)
1544 {
1545         struct mlx5_flow_steering *steering = dev->priv.steering;
1546         struct mlx5_flow_root_namespace *root_ns;
1547         int prio;
1548         struct fs_prio *fs_prio;
1549         struct mlx5_flow_namespace *ns;
1550
1551         if (!steering)
1552                 return NULL;
1553
1554         switch (type) {
1555         case MLX5_FLOW_NAMESPACE_BYPASS:
1556         case MLX5_FLOW_NAMESPACE_LAG:
1557         case MLX5_FLOW_NAMESPACE_OFFLOADS:
1558         case MLX5_FLOW_NAMESPACE_ETHTOOL:
1559         case MLX5_FLOW_NAMESPACE_KERNEL:
1560         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1561         case MLX5_FLOW_NAMESPACE_ANCHOR:
1562                 prio = type;
1563                 break;
1564         case MLX5_FLOW_NAMESPACE_FDB:
1565                 if (steering->fdb_root_ns)
1566                         return &steering->fdb_root_ns->ns;
1567                 else
1568                         return NULL;
1569         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1570                 if (steering->esw_egress_root_ns)
1571                         return &steering->esw_egress_root_ns->ns;
1572                 else
1573                         return NULL;
1574         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1575                 if (steering->esw_ingress_root_ns)
1576                         return &steering->esw_ingress_root_ns->ns;
1577                 else
1578                         return NULL;
1579         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
1580                 if (steering->sniffer_rx_root_ns)
1581                         return &steering->sniffer_rx_root_ns->ns;
1582                 else
1583                         return NULL;
1584         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
1585                 if (steering->sniffer_tx_root_ns)
1586                         return &steering->sniffer_tx_root_ns->ns;
1587                 else
1588                         return NULL;
1589         default:
1590                 return NULL;
1591         }
1592
1593         root_ns = steering->root_ns;
1594         if (!root_ns)
1595                 return NULL;
1596
1597         fs_prio = find_prio(&root_ns->ns, prio);
1598         if (!fs_prio)
1599                 return NULL;
1600
1601         ns = list_first_entry(&fs_prio->node.children,
1602                               typeof(*ns),
1603                               node.list);
1604
1605         return ns;
1606 }
1607 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1608
1609 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1610                                       unsigned int prio, int num_levels)
1611 {
1612         struct fs_prio *fs_prio;
1613
1614         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1615         if (!fs_prio)
1616                 return ERR_PTR(-ENOMEM);
1617
1618         fs_prio->node.type = FS_TYPE_PRIO;
1619         tree_init_node(&fs_prio->node, 1, NULL);
1620         tree_add_node(&fs_prio->node, &ns->node);
1621         fs_prio->num_levels = num_levels;
1622         fs_prio->prio = prio;
1623         list_add_tail(&fs_prio->node.list, &ns->node.children);
1624
1625         return fs_prio;
1626 }
1627
1628 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1629                                                      *ns)
1630 {
1631         ns->node.type = FS_TYPE_NAMESPACE;
1632
1633         return ns;
1634 }
1635
1636 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1637 {
1638         struct mlx5_flow_namespace      *ns;
1639
1640         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1641         if (!ns)
1642                 return ERR_PTR(-ENOMEM);
1643
1644         fs_init_namespace(ns);
1645         tree_init_node(&ns->node, 1, NULL);
1646         tree_add_node(&ns->node, &prio->node);
1647         list_add_tail(&ns->node.list, &prio->node.children);
1648
1649         return ns;
1650 }
1651
1652 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1653                              struct init_tree_node *prio_metadata)
1654 {
1655         struct fs_prio *fs_prio;
1656         int i;
1657
1658         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1659                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1660                 if (IS_ERR(fs_prio))
1661                         return PTR_ERR(fs_prio);
1662         }
1663         return 0;
1664 }
1665
1666 #define FLOW_TABLE_BIT_SZ 1
1667 #define GET_FLOW_TABLE_CAP(dev, offset) \
1668         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1669                         offset / 32)) >>                                        \
1670           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1671 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1672 {
1673         int i;
1674
1675         for (i = 0; i < caps->arr_sz; i++) {
1676                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1677                         return false;
1678         }
1679         return true;
1680 }
1681
1682 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
1683                                     struct init_tree_node *init_node,
1684                                     struct fs_node *fs_parent_node,
1685                                     struct init_tree_node *init_parent_node,
1686                                     int prio)
1687 {
1688         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
1689                                               flow_table_properties_nic_receive.
1690                                               max_ft_level);
1691         struct mlx5_flow_namespace *fs_ns;
1692         struct fs_prio *fs_prio;
1693         struct fs_node *base;
1694         int i;
1695         int err;
1696
1697         if (init_node->type == FS_TYPE_PRIO) {
1698                 if ((init_node->min_ft_level > max_ft_level) ||
1699                     !has_required_caps(steering->dev, &init_node->caps))
1700                         return 0;
1701
1702                 fs_get_obj(fs_ns, fs_parent_node);
1703                 if (init_node->num_leaf_prios)
1704                         return create_leaf_prios(fs_ns, prio, init_node);
1705                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1706                 if (IS_ERR(fs_prio))
1707                         return PTR_ERR(fs_prio);
1708                 base = &fs_prio->node;
1709         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1710                 fs_get_obj(fs_prio, fs_parent_node);
1711                 fs_ns = fs_create_namespace(fs_prio);
1712                 if (IS_ERR(fs_ns))
1713                         return PTR_ERR(fs_ns);
1714                 base = &fs_ns->node;
1715         } else {
1716                 return -EINVAL;
1717         }
1718         prio = 0;
1719         for (i = 0; i < init_node->ar_size; i++) {
1720                 err = init_root_tree_recursive(steering, &init_node->children[i],
1721                                                base, init_node, prio);
1722                 if (err)
1723                         return err;
1724                 if (init_node->children[i].type == FS_TYPE_PRIO &&
1725                     init_node->children[i].num_leaf_prios) {
1726                         prio += init_node->children[i].num_leaf_prios;
1727                 }
1728         }
1729
1730         return 0;
1731 }
1732
1733 static int init_root_tree(struct mlx5_flow_steering *steering,
1734                           struct init_tree_node *init_node,
1735                           struct fs_node *fs_parent_node)
1736 {
1737         int i;
1738         struct mlx5_flow_namespace *fs_ns;
1739         int err;
1740
1741         fs_get_obj(fs_ns, fs_parent_node);
1742         for (i = 0; i < init_node->ar_size; i++) {
1743                 err = init_root_tree_recursive(steering, &init_node->children[i],
1744                                                &fs_ns->node,
1745                                                init_node, i);
1746                 if (err)
1747                         return err;
1748         }
1749         return 0;
1750 }
1751
1752 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
1753                                                        enum fs_flow_table_type
1754                                                        table_type)
1755 {
1756         struct mlx5_flow_root_namespace *root_ns;
1757         struct mlx5_flow_namespace *ns;
1758
1759         /* Create the root namespace */
1760         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1761         if (!root_ns)
1762                 return NULL;
1763
1764         root_ns->dev = steering->dev;
1765         root_ns->table_type = table_type;
1766
1767         ns = &root_ns->ns;
1768         fs_init_namespace(ns);
1769         mutex_init(&root_ns->chain_lock);
1770         tree_init_node(&ns->node, 1, NULL);
1771         tree_add_node(&ns->node, NULL);
1772
1773         return root_ns;
1774 }
1775
1776 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1777
1778 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1779 {
1780         struct fs_prio *prio;
1781
1782         fs_for_each_prio(prio, ns) {
1783                  /* This updates prio start_level and num_levels */
1784                 set_prio_attrs_in_prio(prio, acc_level);
1785                 acc_level += prio->num_levels;
1786         }
1787         return acc_level;
1788 }
1789
1790 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1791 {
1792         struct mlx5_flow_namespace *ns;
1793         int acc_level_ns = acc_level;
1794
1795         prio->start_level = acc_level;
1796         fs_for_each_ns(ns, prio)
1797                 /* This updates start_level and num_levels of ns's priority descendants */
1798                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1799         if (!prio->num_levels)
1800                 prio->num_levels = acc_level_ns - prio->start_level;
1801         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1802 }
1803
1804 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1805 {
1806         struct mlx5_flow_namespace *ns = &root_ns->ns;
1807         struct fs_prio *prio;
1808         int start_level = 0;
1809
1810         fs_for_each_prio(prio, ns) {
1811                 set_prio_attrs_in_prio(prio, start_level);
1812                 start_level += prio->num_levels;
1813         }
1814 }
1815
1816 #define ANCHOR_PRIO 0
1817 #define ANCHOR_SIZE 1
1818 #define ANCHOR_LEVEL 0
1819 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
1820 {
1821         struct mlx5_flow_namespace *ns = NULL;
1822         struct mlx5_flow_table *ft;
1823
1824         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1825         if (!ns)
1826                 return -EINVAL;
1827         ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL, 0);
1828         if (IS_ERR(ft)) {
1829                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
1830                 return PTR_ERR(ft);
1831         }
1832         return 0;
1833 }
1834
1835 static int init_root_ns(struct mlx5_flow_steering *steering)
1836 {
1837
1838         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
1839         if (!steering->root_ns)
1840                 goto cleanup;
1841
1842         if (init_root_tree(steering, &root_fs, &steering->root_ns->ns.node))
1843                 goto cleanup;
1844
1845         set_prio_attrs(steering->root_ns);
1846
1847         if (create_anchor_flow_table(steering))
1848                 goto cleanup;
1849
1850         return 0;
1851
1852 cleanup:
1853         mlx5_cleanup_fs(steering->dev);
1854         return -ENOMEM;
1855 }
1856
1857 static void clean_tree(struct fs_node *node)
1858 {
1859         if (node) {
1860                 struct fs_node *iter;
1861                 struct fs_node *temp;
1862
1863                 list_for_each_entry_safe(iter, temp, &node->children, list)
1864                         clean_tree(iter);
1865                 tree_remove_node(node);
1866         }
1867 }
1868
1869 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
1870 {
1871         if (!root_ns)
1872                 return;
1873
1874         clean_tree(&root_ns->ns.node);
1875 }
1876
1877 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1878 {
1879         struct mlx5_flow_steering *steering = dev->priv.steering;
1880
1881         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1882                 return;
1883
1884         cleanup_root_ns(steering->root_ns);
1885         cleanup_root_ns(steering->esw_egress_root_ns);
1886         cleanup_root_ns(steering->esw_ingress_root_ns);
1887         cleanup_root_ns(steering->fdb_root_ns);
1888         cleanup_root_ns(steering->sniffer_rx_root_ns);
1889         cleanup_root_ns(steering->sniffer_tx_root_ns);
1890         mlx5_cleanup_fc_stats(dev);
1891         kfree(steering);
1892 }
1893
1894 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
1895 {
1896         struct fs_prio *prio;
1897
1898         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
1899         if (!steering->sniffer_tx_root_ns)
1900                 return -ENOMEM;
1901
1902         /* Create single prio */
1903         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
1904         if (IS_ERR(prio)) {
1905                 cleanup_root_ns(steering->sniffer_tx_root_ns);
1906                 return PTR_ERR(prio);
1907         }
1908         return 0;
1909 }
1910
1911 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
1912 {
1913         struct fs_prio *prio;
1914
1915         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
1916         if (!steering->sniffer_rx_root_ns)
1917                 return -ENOMEM;
1918
1919         /* Create single prio */
1920         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
1921         if (IS_ERR(prio)) {
1922                 cleanup_root_ns(steering->sniffer_rx_root_ns);
1923                 return PTR_ERR(prio);
1924         }
1925         return 0;
1926 }
1927
1928 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
1929 {
1930         struct fs_prio *prio;
1931
1932         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
1933         if (!steering->fdb_root_ns)
1934                 return -ENOMEM;
1935
1936         prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
1937         if (IS_ERR(prio))
1938                 goto out_err;
1939
1940         prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
1941         if (IS_ERR(prio))
1942                 goto out_err;
1943
1944         set_prio_attrs(steering->fdb_root_ns);
1945         return 0;
1946
1947 out_err:
1948         cleanup_root_ns(steering->fdb_root_ns);
1949         steering->fdb_root_ns = NULL;
1950         return PTR_ERR(prio);
1951 }
1952
1953 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
1954 {
1955         struct fs_prio *prio;
1956
1957         steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
1958         if (!steering->esw_egress_root_ns)
1959                 return -ENOMEM;
1960
1961         /* create 1 prio*/
1962         prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
1963                               MLX5_TOTAL_VPORTS(steering->dev));
1964         return PTR_ERR_OR_ZERO(prio);
1965 }
1966
1967 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
1968 {
1969         struct fs_prio *prio;
1970
1971         steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
1972         if (!steering->esw_ingress_root_ns)
1973                 return -ENOMEM;
1974
1975         /* create 1 prio*/
1976         prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
1977                               MLX5_TOTAL_VPORTS(steering->dev));
1978         return PTR_ERR_OR_ZERO(prio);
1979 }
1980
1981 int mlx5_init_fs(struct mlx5_core_dev *dev)
1982 {
1983         struct mlx5_flow_steering *steering;
1984         int err = 0;
1985
1986         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1987                 return 0;
1988
1989         err = mlx5_init_fc_stats(dev);
1990         if (err)
1991                 return err;
1992
1993         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
1994         if (!steering)
1995                 return -ENOMEM;
1996         steering->dev = dev;
1997         dev->priv.steering = steering;
1998
1999         if (MLX5_CAP_GEN(dev, nic_flow_table) &&
2000             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2001                 err = init_root_ns(steering);
2002                 if (err)
2003                         goto err;
2004         }
2005
2006         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
2007                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2008                         err = init_fdb_root_ns(steering);
2009                         if (err)
2010                                 goto err;
2011                 }
2012                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2013                         err = init_egress_acl_root_ns(steering);
2014                         if (err)
2015                                 goto err;
2016                 }
2017                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2018                         err = init_ingress_acl_root_ns(steering);
2019                         if (err)
2020                                 goto err;
2021                 }
2022         }
2023
2024         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2025                 err = init_sniffer_rx_root_ns(steering);
2026                 if (err)
2027                         goto err;
2028         }
2029
2030         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2031                 err = init_sniffer_tx_root_ns(steering);
2032                 if (err)
2033                         goto err;
2034         }
2035
2036         return 0;
2037 err:
2038         mlx5_cleanup_fs(dev);
2039         return err;
2040 }