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
 108 static struct init_tree_node {
 109         enum fs_node_type       type;
 110         struct init_tree_node *children;
 111         int ar_size;
 112         struct node_caps caps;
 113         int min_ft_level;
 114         int num_leaf_prios;
 115         int prio;
 116         int num_levels;
 117 } root_fs = {
 118         .type = FS_TYPE_NAMESPACE,
 119         .ar_size = 7,
 120         .children = (struct init_tree_node[]) {
 121                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 122                          FS_CHAINING_CAPS,
 123                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 124                                                   BY_PASS_PRIO_NUM_LEVELS))),
 125                 ADD_PRIO(0, LAG_MIN_LEVEL, 0,
 126                          FS_CHAINING_CAPS,
 127                          ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 128                                                   LAG_PRIO_NUM_LEVELS))),
 129                 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
 130                          ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
 131                 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
 132                          FS_CHAINING_CAPS,
 133                          ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 134                                                   ETHTOOL_PRIO_NUM_LEVELS))),
 135                 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 136                          ADD_NS(ADD_MULTIPLE_PRIO(1, 1),
 137                                 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 138                                                   KERNEL_NIC_PRIO_NUM_LEVELS))),
 139                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 140                          FS_CHAINING_CAPS,
 141                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
 142                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 143                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
 144         }
 145 };
 146
 147 enum fs_i_mutex_lock_class {
 148         FS_MUTEX_GRANDPARENT,
 149         FS_MUTEX_PARENT,
 150         FS_MUTEX_CHILD
 151 };
 152
 153 static void del_rule(struct fs_node *node);
 154 static void del_flow_table(struct fs_node *node);
 155 static void del_flow_group(struct fs_node *node);
 156 static void del_fte(struct fs_node *node);
 157 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 158                                 struct mlx5_flow_destination *d2);
 159 static struct mlx5_flow_rule *
 160 find_flow_rule(struct fs_fte *fte,
 161                struct mlx5_flow_destination *dest);
 162
 163 static void tree_init_node(struct fs_node *node,
 164                            unsigned int refcount,
 165                            void (*remove_func)(struct fs_node *))
 166 {
 167         atomic_set(&node->refcount, refcount);
 168         INIT_LIST_HEAD(&node->list);
 169         INIT_LIST_HEAD(&node->children);
 170         mutex_init(&node->lock);
 171         node->remove_func = remove_func;
 172 }
 173
 174 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 175 {
 176         if (parent)
 177                 atomic_inc(&parent->refcount);
 178         node->parent = parent;
 179
 180         /* Parent is the root */
 181         if (!parent)
 182                 node->root = node;
 183         else
 184                 node->root = parent->root;
 185 }
 186
 187 static void tree_get_node(struct fs_node *node)
 188 {
 189         atomic_inc(&node->refcount);
 190 }
 191
 192 static void nested_lock_ref_node(struct fs_node *node,
 193                                  enum fs_i_mutex_lock_class class)
 194 {
 195         if (node) {
 196                 mutex_lock_nested(&node->lock, class);
 197                 atomic_inc(&node->refcount);
 198         }
 199 }
 200
 201 static void lock_ref_node(struct fs_node *node)
 202 {
 203         if (node) {
 204                 mutex_lock(&node->lock);
 205                 atomic_inc(&node->refcount);
 206         }
 207 }
 208
 209 static void unlock_ref_node(struct fs_node *node)
 210 {
 211         if (node) {
 212                 atomic_dec(&node->refcount);
 213                 mutex_unlock(&node->lock);
 214         }
 215 }
 216
 217 static void tree_put_node(struct fs_node *node)
 218 {
 219         struct fs_node *parent_node = node->parent;
 220
 221         lock_ref_node(parent_node);
 222         if (atomic_dec_and_test(&node->refcount)) {
 223                 if (parent_node)
 224                         list_del_init(&node->list);
 225                 if (node->remove_func)
 226                         node->remove_func(node);
 227                 kfree(node);
 228                 node = NULL;
 229         }
 230         unlock_ref_node(parent_node);
 231         if (!node && parent_node)
 232                 tree_put_node(parent_node);
 233 }
 234
 235 static int tree_remove_node(struct fs_node *node)
 236 {
 237         if (atomic_read(&node->refcount) > 1) {
 238                 atomic_dec(&node->refcount);
 239                 return -EEXIST;
 240         }
 241         tree_put_node(node);
 242         return 0;
 243 }
 244
 245 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 246                                  unsigned int prio)
 247 {
 248         struct fs_prio *iter_prio;
 249
 250         fs_for_each_prio(iter_prio, ns) {
 251                 if (iter_prio->prio == prio)
 252                         return iter_prio;
 253         }
 254
 255         return NULL;
 256 }
 257
 258 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 259 {
 260         unsigned int i;
 261
 262         for (i = 0; i < size; i++, mask++, val1++, val2++)
 263                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 264                     ((*(u8 *)val2) & (*(u8 *)mask)))
 265                         return false;
 266
 267         return true;
 268 }
 269
 270 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 271                                 void *fte_param1, void *fte_param2)
 272 {
 273         if (mask->match_criteria_enable &
 274             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 275                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 276                                                 fte_param1, outer_headers);
 277                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 278                                                 fte_param2, outer_headers);
 279                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 280                                               mask->match_criteria, outer_headers);
 281
 282                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 283                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 284                         return false;
 285         }
 286
 287         if (mask->match_criteria_enable &
 288             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 289                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 290                                                 fte_param1, misc_parameters);
 291                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 292                                                 fte_param2, misc_parameters);
 293                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 294                                           mask->match_criteria, misc_parameters);
 295
 296                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 297                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 298                         return false;
 299         }
 300
 301         if (mask->match_criteria_enable &
 302             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 303                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 304                                                 fte_param1, inner_headers);
 305                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 306                                                 fte_param2, inner_headers);
 307                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 308                                           mask->match_criteria, inner_headers);
 309
 310                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 311                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 312                         return false;
 313         }
 314         return true;
 315 }
 316
 317 static bool compare_match_criteria(u8 match_criteria_enable1,
 318                                    u8 match_criteria_enable2,
 319                                    void *mask1, void *mask2)
 320 {
 321         return match_criteria_enable1 == match_criteria_enable2 &&
 322                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 323 }
 324
 325 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 326 {
 327         struct fs_node *root;
 328         struct mlx5_flow_namespace *ns;
 329
 330         root = node->root;
 331
 332         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 333                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 334                 return NULL;
 335         }
 336
 337         ns = container_of(root, struct mlx5_flow_namespace, node);
 338         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 339 }
 340
 341 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 342 {
 343         struct mlx5_flow_root_namespace *root = find_root(node);
 344
 345         if (root)
 346                 return root->dev;
 347         return NULL;
 348 }
 349
 350 static void del_flow_table(struct fs_node *node)
 351 {
 352         struct mlx5_flow_table *ft;
 353         struct mlx5_core_dev *dev;
 354         struct fs_prio *prio;
 355         int err;
 356
 357         fs_get_obj(ft, node);
 358         dev = get_dev(&ft->node);
 359
 360         err = mlx5_cmd_destroy_flow_table(dev, ft);
 361         if (err)
 362                 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 363         fs_get_obj(prio, ft->node.parent);
 364         prio->num_ft--;
 365 }
 366
 367 static void del_rule(struct fs_node *node)
 368 {
 369         struct mlx5_flow_rule *rule;
 370         struct mlx5_flow_table *ft;
 371         struct mlx5_flow_group *fg;
 372         struct fs_fte *fte;
 373         u32     *match_value;
 374         int modify_mask;
 375         struct mlx5_core_dev *dev = get_dev(node);
 376         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 377         int err;
 378         bool update_fte = false;
 379
 380         match_value = kvzalloc(match_len, GFP_KERNEL);
 381         if (!match_value)
 382                 return;
 383
 384         fs_get_obj(rule, node);
 385         fs_get_obj(fte, rule->node.parent);
 386         fs_get_obj(fg, fte->node.parent);
 387         memcpy(match_value, fte->val, sizeof(fte->val));
 388         fs_get_obj(ft, fg->node.parent);
 389         list_del(&rule->node.list);
 390         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 391                 mutex_lock(&rule->dest_attr.ft->lock);
 392                 list_del(&rule->next_ft);
 393                 mutex_unlock(&rule->dest_attr.ft->lock);
 394         }
 395
 396         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 397             --fte->dests_size) {
 398                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
 399                 fte->action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 400                 update_fte = true;
 401                 goto out;
 402         }
 403
 404         if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 405             --fte->dests_size) {
 406                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST),
 407                 update_fte = true;
 408         }
 409 out:
 410         if (update_fte && fte->dests_size) {
 411                 err = mlx5_cmd_update_fte(dev, ft, fg->id, modify_mask, fte);
 412                 if (err)
 413                         mlx5_core_warn(dev,
 414                                        "%s can't del rule fg id=%d fte_index=%d\n",
 415                                        __func__, fg->id, fte->index);
 416         }
 417         kvfree(match_value);
 418 }
 419
 420 static void del_fte(struct fs_node *node)
 421 {
 422         struct mlx5_flow_table *ft;
 423         struct mlx5_flow_group *fg;
 424         struct mlx5_core_dev *dev;
 425         struct fs_fte *fte;
 426         int err;
 427
 428         fs_get_obj(fte, node);
 429         fs_get_obj(fg, fte->node.parent);
 430         fs_get_obj(ft, fg->node.parent);
 431
 432         dev = get_dev(&ft->node);
 433         err = mlx5_cmd_delete_fte(dev, ft,
 434                                   fte->index);
 435         if (err)
 436                 mlx5_core_warn(dev,
 437                                "flow steering can't delete fte in index %d of flow group id %d\n",
 438                                fte->index, fg->id);
 439
 440         fte->status = 0;
 441         fg->num_ftes--;
 442 }
 443
 444 static void del_flow_group(struct fs_node *node)
 445 {
 446         struct mlx5_flow_group *fg;
 447         struct mlx5_flow_table *ft;
 448         struct mlx5_core_dev *dev;
 449
 450         fs_get_obj(fg, node);
 451         fs_get_obj(ft, fg->node.parent);
 452         dev = get_dev(&ft->node);
 453
 454         if (ft->autogroup.active)
 455                 ft->autogroup.num_groups--;
 456
 457         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 458                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 459                                fg->id, ft->id);
 460 }
 461
 462 static struct fs_fte *alloc_fte(struct mlx5_flow_act *flow_act,
 463                                 u32 *match_value,
 464                                 unsigned int index)
 465 {
 466         struct fs_fte *fte;
 467
 468         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 469         if (!fte)
 470                 return ERR_PTR(-ENOMEM);
 471
 472         memcpy(fte->val, match_value, sizeof(fte->val));
 473         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 474         fte->flow_tag = flow_act->flow_tag;
 475         fte->index = index;
 476         fte->action = flow_act->action;
 477         fte->encap_id = flow_act->encap_id;
 478         fte->modify_id = flow_act->modify_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, root->underlay_qpn);
 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                                                         struct mlx5_flow_table_attr *ft_attr,
 781                                                         enum fs_flow_table_op_mod op_mod,
 782                                                         u16 vport)
 783 {
 784         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
 785         struct mlx5_flow_table *next_ft = NULL;
 786         struct fs_prio *fs_prio = NULL;
 787         struct mlx5_flow_table *ft;
 788         int log_table_sz;
 789         int err;
 790
 791         if (!root) {
 792                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 793                 return ERR_PTR(-ENODEV);
 794         }
 795
 796         mutex_lock(&root->chain_lock);
 797         fs_prio = find_prio(ns, ft_attr->prio);
 798         if (!fs_prio) {
 799                 err = -EINVAL;
 800                 goto unlock_root;
 801         }
 802         if (ft_attr->level >= fs_prio->num_levels) {
 803                 err = -ENOSPC;
 804                 goto unlock_root;
 805         }
 806         /* The level is related to the
 807          * priority level range.
 808          */
 809         ft_attr->level += fs_prio->start_level;
 810         ft = alloc_flow_table(ft_attr->level,
 811                               vport,
 812                               ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
 813                               root->table_type,
 814                               op_mod, ft_attr->flags);
 815         if (!ft) {
 816                 err = -ENOMEM;
 817                 goto unlock_root;
 818         }
 819
 820         tree_init_node(&ft->node, 1, del_flow_table);
 821         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
 822         next_ft = find_next_chained_ft(fs_prio);
 823         err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->op_mod, ft->type,
 824                                          ft->level, log_table_sz, next_ft, &ft->id,
 825                                          ft->flags);
 826         if (err)
 827                 goto free_ft;
 828
 829         err = connect_flow_table(root->dev, ft, fs_prio);
 830         if (err)
 831                 goto destroy_ft;
 832         lock_ref_node(&fs_prio->node);
 833         tree_add_node(&ft->node, &fs_prio->node);
 834         list_add_flow_table(ft, fs_prio);
 835         fs_prio->num_ft++;
 836         unlock_ref_node(&fs_prio->node);
 837         mutex_unlock(&root->chain_lock);
 838         return ft;
 839 destroy_ft:
 840         mlx5_cmd_destroy_flow_table(root->dev, ft);
 841 free_ft:
 842         kfree(ft);
 843 unlock_root:
 844         mutex_unlock(&root->chain_lock);
 845         return ERR_PTR(err);
 846 }
 847
 848 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 849                                                struct mlx5_flow_table_attr *ft_attr)
 850 {
 851         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
 852 }
 853
 854 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
 855                                                      int prio, int max_fte,
 856                                                      u32 level, u16 vport)
 857 {
 858         struct mlx5_flow_table_attr ft_attr = {};
 859
 860         ft_attr.max_fte = max_fte;
 861         ft_attr.level   = level;
 862         ft_attr.prio    = prio;
 863
 864         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, vport);
 865 }
 866
 867 struct mlx5_flow_table*
 868 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
 869                                  int prio, u32 level)
 870 {
 871         struct mlx5_flow_table_attr ft_attr = {};
 872
 873         ft_attr.level = level;
 874         ft_attr.prio  = prio;
 875         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
 876 }
 877 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
 878
 879 struct mlx5_flow_table*
 880 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 881                                     int prio,
 882                                     int num_flow_table_entries,
 883                                     int max_num_groups,
 884                                     u32 level,
 885                                     u32 flags)
 886 {
 887         struct mlx5_flow_table_attr ft_attr = {};
 888         struct mlx5_flow_table *ft;
 889
 890         if (max_num_groups > num_flow_table_entries)
 891                 return ERR_PTR(-EINVAL);
 892
 893         ft_attr.max_fte = num_flow_table_entries;
 894         ft_attr.prio    = prio;
 895         ft_attr.level   = level;
 896         ft_attr.flags   = flags;
 897
 898         ft = mlx5_create_flow_table(ns, &ft_attr);
 899         if (IS_ERR(ft))
 900                 return ft;
 901
 902         ft->autogroup.active = true;
 903         ft->autogroup.required_groups = max_num_groups;
 904
 905         return ft;
 906 }
 907 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 908
 909 /* Flow table should be locked */
 910 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 911                                                         u32 *fg_in,
 912                                                         struct list_head
 913                                                         *prev_fg,
 914                                                         bool is_auto_fg)
 915 {
 916         struct mlx5_flow_group *fg;
 917         struct mlx5_core_dev *dev = get_dev(&ft->node);
 918         int err;
 919
 920         if (!dev)
 921                 return ERR_PTR(-ENODEV);
 922
 923         fg = alloc_flow_group(fg_in);
 924         if (IS_ERR(fg))
 925                 return fg;
 926
 927         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 928         if (err) {
 929                 kfree(fg);
 930                 return ERR_PTR(err);
 931         }
 932
 933         if (ft->autogroup.active)
 934                 ft->autogroup.num_groups++;
 935         /* Add node to tree */
 936         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 937         tree_add_node(&fg->node, &ft->node);
 938         /* Add node to group list */
 939         list_add(&fg->node.list, prev_fg);
 940
 941         return fg;
 942 }
 943
 944 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 945                                                u32 *fg_in)
 946 {
 947         struct mlx5_flow_group *fg;
 948
 949         if (ft->autogroup.active)
 950                 return ERR_PTR(-EPERM);
 951
 952         lock_ref_node(&ft->node);
 953         fg = create_flow_group_common(ft, fg_in, ft->node.children.prev, false);
 954         unlock_ref_node(&ft->node);
 955
 956         return fg;
 957 }
 958
 959 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 960 {
 961         struct mlx5_flow_rule *rule;
 962
 963         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 964         if (!rule)
 965                 return NULL;
 966
 967         INIT_LIST_HEAD(&rule->next_ft);
 968         rule->node.type = FS_TYPE_FLOW_DEST;
 969         if (dest)
 970                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 971
 972         return rule;
 973 }
 974
 975 static struct mlx5_flow_handle *alloc_handle(int num_rules)
 976 {
 977         struct mlx5_flow_handle *handle;
 978
 979         handle = kzalloc(sizeof(*handle) + sizeof(handle->rule[0]) *
 980                           num_rules, GFP_KERNEL);
 981         if (!handle)
 982                 return NULL;
 983
 984         handle->num_rules = num_rules;
 985
 986         return handle;
 987 }
 988
 989 static void destroy_flow_handle(struct fs_fte *fte,
 990                                 struct mlx5_flow_handle *handle,
 991                                 struct mlx5_flow_destination *dest,
 992                                 int i)
 993 {
 994         for (; --i >= 0;) {
 995                 if (atomic_dec_and_test(&handle->rule[i]->node.refcount)) {
 996                         fte->dests_size--;
 997                         list_del(&handle->rule[i]->node.list);
 998                         kfree(handle->rule[i]);
 999                 }
1000         }
1001         kfree(handle);
1002 }
1003
1004 static struct mlx5_flow_handle *
1005 create_flow_handle(struct fs_fte *fte,
1006                    struct mlx5_flow_destination *dest,
1007                    int dest_num,
1008                    int *modify_mask,
1009                    bool *new_rule)
1010 {
1011         struct mlx5_flow_handle *handle;
1012         struct mlx5_flow_rule *rule = NULL;
1013         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1014         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1015         int type;
1016         int i = 0;
1017
1018         handle = alloc_handle((dest_num) ? dest_num : 1);
1019         if (!handle)
1020                 return ERR_PTR(-ENOMEM);
1021
1022         do {
1023                 if (dest) {
1024                         rule = find_flow_rule(fte, dest + i);
1025                         if (rule) {
1026                                 atomic_inc(&rule->node.refcount);
1027                                 goto rule_found;
1028                         }
1029                 }
1030
1031                 *new_rule = true;
1032                 rule = alloc_rule(dest + i);
1033                 if (!rule)
1034                         goto free_rules;
1035
1036                 /* Add dest to dests list- we need flow tables to be in the
1037                  * end of the list for forward to next prio rules.
1038                  */
1039                 tree_init_node(&rule->node, 1, del_rule);
1040                 if (dest &&
1041                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1042                         list_add(&rule->node.list, &fte->node.children);
1043                 else
1044                         list_add_tail(&rule->node.list, &fte->node.children);
1045                 if (dest) {
1046                         fte->dests_size++;
1047
1048                         type = dest[i].type ==
1049                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1050                         *modify_mask |= type ? count : dst;
1051                 }
1052 rule_found:
1053                 handle->rule[i] = rule;
1054         } while (++i < dest_num);
1055
1056         return handle;
1057
1058 free_rules:
1059         destroy_flow_handle(fte, handle, dest, i);
1060         return ERR_PTR(-ENOMEM);
1061 }
1062
1063 /* fte should not be deleted while calling this function */
1064 static struct mlx5_flow_handle *
1065 add_rule_fte(struct fs_fte *fte,
1066              struct mlx5_flow_group *fg,
1067              struct mlx5_flow_destination *dest,
1068              int dest_num,
1069              bool update_action)
1070 {
1071         struct mlx5_flow_handle *handle;
1072         struct mlx5_flow_table *ft;
1073         int modify_mask = 0;
1074         int err;
1075         bool new_rule = false;
1076
1077         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1078                                     &new_rule);
1079         if (IS_ERR(handle) || !new_rule)
1080                 goto out;
1081
1082         if (update_action)
1083                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1084
1085         fs_get_obj(ft, fg->node.parent);
1086         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1087                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
1088                                           ft, fg->id, fte);
1089         else
1090                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
1091                                           ft, fg->id, modify_mask, fte);
1092         if (err)
1093                 goto free_handle;
1094
1095         fte->status |= FS_FTE_STATUS_EXISTING;
1096
1097 out:
1098         return handle;
1099
1100 free_handle:
1101         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1102         return ERR_PTR(err);
1103 }
1104
1105 /* Assumed fg is locked */
1106 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
1107                                        struct list_head **prev)
1108 {
1109         struct fs_fte *fte;
1110         unsigned int start = fg->start_index;
1111
1112         if (prev)
1113                 *prev = &fg->node.children;
1114
1115         /* assumed list is sorted by index */
1116         fs_for_each_fte(fte, fg) {
1117                 if (fte->index != start)
1118                         return start;
1119                 start++;
1120                 if (prev)
1121                         *prev = &fte->node.list;
1122         }
1123
1124         return start;
1125 }
1126
1127 /* prev is output, prev->next = new_fte */
1128 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
1129                                  u32 *match_value,
1130                                  struct mlx5_flow_act *flow_act,
1131                                  struct list_head **prev)
1132 {
1133         struct fs_fte *fte;
1134         int index;
1135
1136         index = get_free_fte_index(fg, prev);
1137         fte = alloc_fte(flow_act, match_value, index);
1138         if (IS_ERR(fte))
1139                 return fte;
1140
1141         return fte;
1142 }
1143
1144 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
1145                                                 u8 match_criteria_enable,
1146                                                 u32 *match_criteria)
1147 {
1148         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1149         struct list_head *prev = &ft->node.children;
1150         unsigned int candidate_index = 0;
1151         struct mlx5_flow_group *fg;
1152         void *match_criteria_addr;
1153         unsigned int group_size = 0;
1154         u32 *in;
1155
1156         if (!ft->autogroup.active)
1157                 return ERR_PTR(-ENOENT);
1158
1159         in = kvzalloc(inlen, GFP_KERNEL);
1160         if (!in)
1161                 return ERR_PTR(-ENOMEM);
1162
1163         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1164                 /* We save place for flow groups in addition to max types */
1165                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1166
1167         /*  ft->max_fte == ft->autogroup.max_types */
1168         if (group_size == 0)
1169                 group_size = 1;
1170
1171         /* sorted by start_index */
1172         fs_for_each_fg(fg, ft) {
1173                 if (candidate_index + group_size > fg->start_index)
1174                         candidate_index = fg->start_index + fg->max_ftes;
1175                 else
1176                         break;
1177                 prev = &fg->node.list;
1178         }
1179
1180         if (candidate_index + group_size > ft->max_fte) {
1181                 fg = ERR_PTR(-ENOSPC);
1182                 goto out;
1183         }
1184
1185         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1186                  match_criteria_enable);
1187         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1188         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
1189                  group_size - 1);
1190         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1191                                            in, match_criteria);
1192         memcpy(match_criteria_addr, match_criteria,
1193                MLX5_ST_SZ_BYTES(fte_match_param));
1194
1195         fg = create_flow_group_common(ft, in, prev, true);
1196 out:
1197         kvfree(in);
1198         return fg;
1199 }
1200
1201 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1202                                 struct mlx5_flow_destination *d2)
1203 {
1204         if (d1->type == d2->type) {
1205                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1206                      d1->vport_num == d2->vport_num) ||
1207                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1208                      d1->ft == d2->ft) ||
1209                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1210                      d1->tir_num == d2->tir_num))
1211                         return true;
1212         }
1213
1214         return false;
1215 }
1216
1217 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1218                                              struct mlx5_flow_destination *dest)
1219 {
1220         struct mlx5_flow_rule *rule;
1221
1222         list_for_each_entry(rule, &fte->node.children, node.list) {
1223                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1224                         return rule;
1225         }
1226         return NULL;
1227 }
1228
1229 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1230                                             u32 *match_value,
1231                                             struct mlx5_flow_act *flow_act,
1232                                             struct mlx5_flow_destination *dest,
1233                                             int dest_num)
1234 {
1235         struct mlx5_flow_handle *handle;
1236         struct mlx5_flow_table *ft;
1237         struct list_head *prev;
1238         struct fs_fte *fte;
1239         int i;
1240
1241         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1242         fs_for_each_fte(fte, fg) {
1243                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1244                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1245                     (flow_act->action & fte->action)) {
1246                         int old_action = fte->action;
1247
1248                         if (fte->flow_tag != flow_act->flow_tag) {
1249                                 mlx5_core_warn(get_dev(&fte->node),
1250                                                "FTE flow tag %u already exists with different flow tag %u\n",
1251                                                fte->flow_tag,
1252                                                flow_act->flow_tag);
1253                                 handle = ERR_PTR(-EEXIST);
1254                                 goto unlock_fte;
1255                         }
1256
1257                         fte->action |= flow_act->action;
1258                         handle = add_rule_fte(fte, fg, dest, dest_num,
1259                                               old_action != flow_act->action);
1260                         if (IS_ERR(handle)) {
1261                                 fte->action = old_action;
1262                                 goto unlock_fte;
1263                         } else {
1264                                 goto add_rules;
1265                         }
1266                 }
1267                 unlock_ref_node(&fte->node);
1268         }
1269         fs_get_obj(ft, fg->node.parent);
1270         if (fg->num_ftes >= fg->max_ftes) {
1271                 handle = ERR_PTR(-ENOSPC);
1272                 goto unlock_fg;
1273         }
1274
1275         fte = create_fte(fg, match_value, flow_act, &prev);
1276         if (IS_ERR(fte)) {
1277                 handle = (void *)fte;
1278                 goto unlock_fg;
1279         }
1280         tree_init_node(&fte->node, 0, del_fte);
1281         nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1282         handle = add_rule_fte(fte, fg, dest, dest_num, false);
1283         if (IS_ERR(handle)) {
1284                 unlock_ref_node(&fte->node);
1285                 kfree(fte);
1286                 goto unlock_fg;
1287         }
1288
1289         fg->num_ftes++;
1290
1291         tree_add_node(&fte->node, &fg->node);
1292         list_add(&fte->node.list, prev);
1293 add_rules:
1294         for (i = 0; i < handle->num_rules; i++) {
1295                 if (atomic_read(&handle->rule[i]->node.refcount) == 1)
1296                         tree_add_node(&handle->rule[i]->node, &fte->node);
1297         }
1298 unlock_fte:
1299         unlock_ref_node(&fte->node);
1300 unlock_fg:
1301         unlock_ref_node(&fg->node);
1302         return handle;
1303 }
1304
1305 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_handle *handle)
1306 {
1307         struct mlx5_flow_rule *dst;
1308         struct fs_fte *fte;
1309
1310         fs_get_obj(fte, handle->rule[0]->node.parent);
1311
1312         fs_for_each_dst(dst, fte) {
1313                 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1314                         return dst->dest_attr.counter;
1315         }
1316
1317         return NULL;
1318 }
1319
1320 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1321 {
1322         if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1323                 return !counter;
1324
1325         if (!counter)
1326                 return false;
1327
1328         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1329                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1330 }
1331
1332 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1333                           u32 action,
1334                           struct mlx5_flow_table *ft)
1335 {
1336         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1337                 return counter_is_valid(dest->counter, action);
1338
1339         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1340                 return true;
1341
1342         if (!dest || ((dest->type ==
1343             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1344             (dest->ft->level <= ft->level)))
1345                 return false;
1346         return true;
1347 }
1348
1349 static struct mlx5_flow_handle *
1350 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1351                      struct mlx5_flow_spec *spec,
1352                      struct mlx5_flow_act *flow_act,
1353                      struct mlx5_flow_destination *dest,
1354                      int dest_num)
1355
1356 {
1357         struct mlx5_flow_group *g;
1358         struct mlx5_flow_handle *rule;
1359         int i;
1360
1361         for (i = 0; i < dest_num; i++) {
1362                 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1363                         return ERR_PTR(-EINVAL);
1364         }
1365
1366         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1367         fs_for_each_fg(g, ft)
1368                 if (compare_match_criteria(g->mask.match_criteria_enable,
1369                                            spec->match_criteria_enable,
1370                                            g->mask.match_criteria,
1371                                            spec->match_criteria)) {
1372                         rule = add_rule_fg(g, spec->match_value,
1373                                            flow_act, dest, dest_num);
1374                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1375                                 goto unlock;
1376                 }
1377
1378         g = create_autogroup(ft, spec->match_criteria_enable,
1379                              spec->match_criteria);
1380         if (IS_ERR(g)) {
1381                 rule = (void *)g;
1382                 goto unlock;
1383         }
1384
1385         rule = add_rule_fg(g, spec->match_value, flow_act, dest, dest_num);
1386         if (IS_ERR(rule)) {
1387                 /* Remove assumes refcount > 0 and autogroup creates a group
1388                  * with a refcount = 0.
1389                  */
1390                 unlock_ref_node(&ft->node);
1391                 tree_get_node(&g->node);
1392                 tree_remove_node(&g->node);
1393                 return rule;
1394         }
1395 unlock:
1396         unlock_ref_node(&ft->node);
1397         return rule;
1398 }
1399
1400 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1401 {
1402         return ((ft->type == FS_FT_NIC_RX) &&
1403                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1404 }
1405
1406 struct mlx5_flow_handle *
1407 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1408                     struct mlx5_flow_spec *spec,
1409                     struct mlx5_flow_act *flow_act,
1410                     struct mlx5_flow_destination *dest,
1411                     int dest_num)
1412 {
1413         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1414         struct mlx5_flow_destination gen_dest;
1415         struct mlx5_flow_table *next_ft = NULL;
1416         struct mlx5_flow_handle *handle = NULL;
1417         u32 sw_action = flow_act->action;
1418         struct fs_prio *prio;
1419
1420         fs_get_obj(prio, ft->node.parent);
1421         if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1422                 if (!fwd_next_prio_supported(ft))
1423                         return ERR_PTR(-EOPNOTSUPP);
1424                 if (dest)
1425                         return ERR_PTR(-EINVAL);
1426                 mutex_lock(&root->chain_lock);
1427                 next_ft = find_next_chained_ft(prio);
1428                 if (next_ft) {
1429                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1430                         gen_dest.ft = next_ft;
1431                         dest = &gen_dest;
1432                         dest_num = 1;
1433                         flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1434                 } else {
1435                         mutex_unlock(&root->chain_lock);
1436                         return ERR_PTR(-EOPNOTSUPP);
1437                 }
1438         }
1439
1440         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, dest_num);
1441
1442         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1443                 if (!IS_ERR_OR_NULL(handle) &&
1444                     (list_empty(&handle->rule[0]->next_ft))) {
1445                         mutex_lock(&next_ft->lock);
1446                         list_add(&handle->rule[0]->next_ft,
1447                                  &next_ft->fwd_rules);
1448                         mutex_unlock(&next_ft->lock);
1449                         handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1450                 }
1451                 mutex_unlock(&root->chain_lock);
1452         }
1453         return handle;
1454 }
1455 EXPORT_SYMBOL(mlx5_add_flow_rules);
1456
1457 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1458 {
1459         int i;
1460
1461         for (i = handle->num_rules - 1; i >= 0; i--)
1462                 tree_remove_node(&handle->rule[i]->node);
1463         kfree(handle);
1464 }
1465 EXPORT_SYMBOL(mlx5_del_flow_rules);
1466
1467 /* Assuming prio->node.children(flow tables) is sorted by level */
1468 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1469 {
1470         struct fs_prio *prio;
1471
1472         fs_get_obj(prio, ft->node.parent);
1473
1474         if (!list_is_last(&ft->node.list, &prio->node.children))
1475                 return list_next_entry(ft, node.list);
1476         return find_next_chained_ft(prio);
1477 }
1478
1479 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1480 {
1481         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1482         struct mlx5_flow_table *new_root_ft = NULL;
1483
1484         if (root->root_ft != ft)
1485                 return 0;
1486
1487         new_root_ft = find_next_ft(ft);
1488         if (new_root_ft) {
1489                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft,
1490                                                   root->underlay_qpn);
1491
1492                 if (err) {
1493                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1494                                        ft->id);
1495                         return err;
1496                 }
1497         }
1498         root->root_ft = new_root_ft;
1499         return 0;
1500 }
1501
1502 /* Connect flow table from previous priority to
1503  * the next flow table.
1504  */
1505 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1506 {
1507         struct mlx5_core_dev *dev = get_dev(&ft->node);
1508         struct mlx5_flow_table *next_ft;
1509         struct fs_prio *prio;
1510         int err = 0;
1511
1512         err = update_root_ft_destroy(ft);
1513         if (err)
1514                 return err;
1515
1516         fs_get_obj(prio, ft->node.parent);
1517         if  (!(list_first_entry(&prio->node.children,
1518                                 struct mlx5_flow_table,
1519                                 node.list) == ft))
1520                 return 0;
1521
1522         next_ft = find_next_chained_ft(prio);
1523         err = connect_fwd_rules(dev, next_ft, ft);
1524         if (err)
1525                 return err;
1526
1527         err = connect_prev_fts(dev, next_ft, prio);
1528         if (err)
1529                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1530                                ft->id);
1531         return err;
1532 }
1533
1534 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1535 {
1536         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1537         int err = 0;
1538
1539         mutex_lock(&root->chain_lock);
1540         err = disconnect_flow_table(ft);
1541         if (err) {
1542                 mutex_unlock(&root->chain_lock);
1543                 return err;
1544         }
1545         if (tree_remove_node(&ft->node))
1546                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1547                                ft->id);
1548         mutex_unlock(&root->chain_lock);
1549
1550         return err;
1551 }
1552 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1553
1554 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1555 {
1556         if (tree_remove_node(&fg->node))
1557                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1558                                fg->id);
1559 }
1560
1561 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1562                                                     enum mlx5_flow_namespace_type type)
1563 {
1564         struct mlx5_flow_steering *steering = dev->priv.steering;
1565         struct mlx5_flow_root_namespace *root_ns;
1566         int prio;
1567         struct fs_prio *fs_prio;
1568         struct mlx5_flow_namespace *ns;
1569
1570         if (!steering)
1571                 return NULL;
1572
1573         switch (type) {
1574         case MLX5_FLOW_NAMESPACE_BYPASS:
1575         case MLX5_FLOW_NAMESPACE_LAG:
1576         case MLX5_FLOW_NAMESPACE_OFFLOADS:
1577         case MLX5_FLOW_NAMESPACE_ETHTOOL:
1578         case MLX5_FLOW_NAMESPACE_KERNEL:
1579         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1580         case MLX5_FLOW_NAMESPACE_ANCHOR:
1581                 prio = type;
1582                 break;
1583         case MLX5_FLOW_NAMESPACE_FDB:
1584                 if (steering->fdb_root_ns)
1585                         return &steering->fdb_root_ns->ns;
1586                 else
1587                         return NULL;
1588         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1589                 if (steering->esw_egress_root_ns)
1590                         return &steering->esw_egress_root_ns->ns;
1591                 else
1592                         return NULL;
1593         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1594                 if (steering->esw_ingress_root_ns)
1595                         return &steering->esw_ingress_root_ns->ns;
1596                 else
1597                         return NULL;
1598         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
1599                 if (steering->sniffer_rx_root_ns)
1600                         return &steering->sniffer_rx_root_ns->ns;
1601                 else
1602                         return NULL;
1603         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
1604                 if (steering->sniffer_tx_root_ns)
1605                         return &steering->sniffer_tx_root_ns->ns;
1606                 else
1607                         return NULL;
1608         default:
1609                 return NULL;
1610         }
1611
1612         root_ns = steering->root_ns;
1613         if (!root_ns)
1614                 return NULL;
1615
1616         fs_prio = find_prio(&root_ns->ns, prio);
1617         if (!fs_prio)
1618                 return NULL;
1619
1620         ns = list_first_entry(&fs_prio->node.children,
1621                               typeof(*ns),
1622                               node.list);
1623
1624         return ns;
1625 }
1626 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1627
1628 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1629                                       unsigned int prio, int num_levels)
1630 {
1631         struct fs_prio *fs_prio;
1632
1633         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1634         if (!fs_prio)
1635                 return ERR_PTR(-ENOMEM);
1636
1637         fs_prio->node.type = FS_TYPE_PRIO;
1638         tree_init_node(&fs_prio->node, 1, NULL);
1639         tree_add_node(&fs_prio->node, &ns->node);
1640         fs_prio->num_levels = num_levels;
1641         fs_prio->prio = prio;
1642         list_add_tail(&fs_prio->node.list, &ns->node.children);
1643
1644         return fs_prio;
1645 }
1646
1647 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1648                                                      *ns)
1649 {
1650         ns->node.type = FS_TYPE_NAMESPACE;
1651
1652         return ns;
1653 }
1654
1655 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1656 {
1657         struct mlx5_flow_namespace      *ns;
1658
1659         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1660         if (!ns)
1661                 return ERR_PTR(-ENOMEM);
1662
1663         fs_init_namespace(ns);
1664         tree_init_node(&ns->node, 1, NULL);
1665         tree_add_node(&ns->node, &prio->node);
1666         list_add_tail(&ns->node.list, &prio->node.children);
1667
1668         return ns;
1669 }
1670
1671 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1672                              struct init_tree_node *prio_metadata)
1673 {
1674         struct fs_prio *fs_prio;
1675         int i;
1676
1677         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1678                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1679                 if (IS_ERR(fs_prio))
1680                         return PTR_ERR(fs_prio);
1681         }
1682         return 0;
1683 }
1684
1685 #define FLOW_TABLE_BIT_SZ 1
1686 #define GET_FLOW_TABLE_CAP(dev, offset) \
1687         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
1688                         offset / 32)) >>                                        \
1689           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1690 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1691 {
1692         int i;
1693
1694         for (i = 0; i < caps->arr_sz; i++) {
1695                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1696                         return false;
1697         }
1698         return true;
1699 }
1700
1701 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
1702                                     struct init_tree_node *init_node,
1703                                     struct fs_node *fs_parent_node,
1704                                     struct init_tree_node *init_parent_node,
1705                                     int prio)
1706 {
1707         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
1708                                               flow_table_properties_nic_receive.
1709                                               max_ft_level);
1710         struct mlx5_flow_namespace *fs_ns;
1711         struct fs_prio *fs_prio;
1712         struct fs_node *base;
1713         int i;
1714         int err;
1715
1716         if (init_node->type == FS_TYPE_PRIO) {
1717                 if ((init_node->min_ft_level > max_ft_level) ||
1718                     !has_required_caps(steering->dev, &init_node->caps))
1719                         return 0;
1720
1721                 fs_get_obj(fs_ns, fs_parent_node);
1722                 if (init_node->num_leaf_prios)
1723                         return create_leaf_prios(fs_ns, prio, init_node);
1724                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1725                 if (IS_ERR(fs_prio))
1726                         return PTR_ERR(fs_prio);
1727                 base = &fs_prio->node;
1728         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1729                 fs_get_obj(fs_prio, fs_parent_node);
1730                 fs_ns = fs_create_namespace(fs_prio);
1731                 if (IS_ERR(fs_ns))
1732                         return PTR_ERR(fs_ns);
1733                 base = &fs_ns->node;
1734         } else {
1735                 return -EINVAL;
1736         }
1737         prio = 0;
1738         for (i = 0; i < init_node->ar_size; i++) {
1739                 err = init_root_tree_recursive(steering, &init_node->children[i],
1740                                                base, init_node, prio);
1741                 if (err)
1742                         return err;
1743                 if (init_node->children[i].type == FS_TYPE_PRIO &&
1744                     init_node->children[i].num_leaf_prios) {
1745                         prio += init_node->children[i].num_leaf_prios;
1746                 }
1747         }
1748
1749         return 0;
1750 }
1751
1752 static int init_root_tree(struct mlx5_flow_steering *steering,
1753                           struct init_tree_node *init_node,
1754                           struct fs_node *fs_parent_node)
1755 {
1756         int i;
1757         struct mlx5_flow_namespace *fs_ns;
1758         int err;
1759
1760         fs_get_obj(fs_ns, fs_parent_node);
1761         for (i = 0; i < init_node->ar_size; i++) {
1762                 err = init_root_tree_recursive(steering, &init_node->children[i],
1763                                                &fs_ns->node,
1764                                                init_node, i);
1765                 if (err)
1766                         return err;
1767         }
1768         return 0;
1769 }
1770
1771 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
1772                                                        enum fs_flow_table_type
1773                                                        table_type)
1774 {
1775         struct mlx5_flow_root_namespace *root_ns;
1776         struct mlx5_flow_namespace *ns;
1777
1778         /* Create the root namespace */
1779         root_ns = kvzalloc(sizeof(*root_ns), GFP_KERNEL);
1780         if (!root_ns)
1781                 return NULL;
1782
1783         root_ns->dev = steering->dev;
1784         root_ns->table_type = table_type;
1785
1786         ns = &root_ns->ns;
1787         fs_init_namespace(ns);
1788         mutex_init(&root_ns->chain_lock);
1789         tree_init_node(&ns->node, 1, NULL);
1790         tree_add_node(&ns->node, NULL);
1791
1792         return root_ns;
1793 }
1794
1795 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1796
1797 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1798 {
1799         struct fs_prio *prio;
1800
1801         fs_for_each_prio(prio, ns) {
1802                  /* This updates prio start_level and num_levels */
1803                 set_prio_attrs_in_prio(prio, acc_level);
1804                 acc_level += prio->num_levels;
1805         }
1806         return acc_level;
1807 }
1808
1809 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1810 {
1811         struct mlx5_flow_namespace *ns;
1812         int acc_level_ns = acc_level;
1813
1814         prio->start_level = acc_level;
1815         fs_for_each_ns(ns, prio)
1816                 /* This updates start_level and num_levels of ns's priority descendants */
1817                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1818         if (!prio->num_levels)
1819                 prio->num_levels = acc_level_ns - prio->start_level;
1820         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1821 }
1822
1823 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1824 {
1825         struct mlx5_flow_namespace *ns = &root_ns->ns;
1826         struct fs_prio *prio;
1827         int start_level = 0;
1828
1829         fs_for_each_prio(prio, ns) {
1830                 set_prio_attrs_in_prio(prio, start_level);
1831                 start_level += prio->num_levels;
1832         }
1833 }
1834
1835 #define ANCHOR_PRIO 0
1836 #define ANCHOR_SIZE 1
1837 #define ANCHOR_LEVEL 0
1838 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
1839 {
1840         struct mlx5_flow_namespace *ns = NULL;
1841         struct mlx5_flow_table_attr ft_attr = {};
1842         struct mlx5_flow_table *ft;
1843
1844         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1845         if (WARN_ON(!ns))
1846                 return -EINVAL;
1847
1848         ft_attr.max_fte = ANCHOR_SIZE;
1849         ft_attr.level   = ANCHOR_LEVEL;
1850         ft_attr.prio    = ANCHOR_PRIO;
1851
1852         ft = mlx5_create_flow_table(ns, &ft_attr);
1853         if (IS_ERR(ft)) {
1854                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
1855                 return PTR_ERR(ft);
1856         }
1857         return 0;
1858 }
1859
1860 static int init_root_ns(struct mlx5_flow_steering *steering)
1861 {
1862         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
1863         if (!steering->root_ns)
1864                 goto cleanup;
1865
1866         if (init_root_tree(steering, &root_fs, &steering->root_ns->ns.node))
1867                 goto cleanup;
1868
1869         set_prio_attrs(steering->root_ns);
1870
1871         if (create_anchor_flow_table(steering))
1872                 goto cleanup;
1873
1874         return 0;
1875
1876 cleanup:
1877         mlx5_cleanup_fs(steering->dev);
1878         return -ENOMEM;
1879 }
1880
1881 static void clean_tree(struct fs_node *node)
1882 {
1883         if (node) {
1884                 struct fs_node *iter;
1885                 struct fs_node *temp;
1886
1887                 list_for_each_entry_safe(iter, temp, &node->children, list)
1888                         clean_tree(iter);
1889                 tree_remove_node(node);
1890         }
1891 }
1892
1893 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
1894 {
1895         if (!root_ns)
1896                 return;
1897
1898         clean_tree(&root_ns->ns.node);
1899 }
1900
1901 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1902 {
1903         struct mlx5_flow_steering *steering = dev->priv.steering;
1904
1905         cleanup_root_ns(steering->root_ns);
1906         cleanup_root_ns(steering->esw_egress_root_ns);
1907         cleanup_root_ns(steering->esw_ingress_root_ns);
1908         cleanup_root_ns(steering->fdb_root_ns);
1909         cleanup_root_ns(steering->sniffer_rx_root_ns);
1910         cleanup_root_ns(steering->sniffer_tx_root_ns);
1911         mlx5_cleanup_fc_stats(dev);
1912         kfree(steering);
1913 }
1914
1915 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
1916 {
1917         struct fs_prio *prio;
1918
1919         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
1920         if (!steering->sniffer_tx_root_ns)
1921                 return -ENOMEM;
1922
1923         /* Create single prio */
1924         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
1925         if (IS_ERR(prio)) {
1926                 cleanup_root_ns(steering->sniffer_tx_root_ns);
1927                 return PTR_ERR(prio);
1928         }
1929         return 0;
1930 }
1931
1932 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
1933 {
1934         struct fs_prio *prio;
1935
1936         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
1937         if (!steering->sniffer_rx_root_ns)
1938                 return -ENOMEM;
1939
1940         /* Create single prio */
1941         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
1942         if (IS_ERR(prio)) {
1943                 cleanup_root_ns(steering->sniffer_rx_root_ns);
1944                 return PTR_ERR(prio);
1945         }
1946         return 0;
1947 }
1948
1949 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
1950 {
1951         struct fs_prio *prio;
1952
1953         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
1954         if (!steering->fdb_root_ns)
1955                 return -ENOMEM;
1956
1957         prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
1958         if (IS_ERR(prio))
1959                 goto out_err;
1960
1961         prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
1962         if (IS_ERR(prio))
1963                 goto out_err;
1964
1965         set_prio_attrs(steering->fdb_root_ns);
1966         return 0;
1967
1968 out_err:
1969         cleanup_root_ns(steering->fdb_root_ns);
1970         steering->fdb_root_ns = NULL;
1971         return PTR_ERR(prio);
1972 }
1973
1974 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
1975 {
1976         struct fs_prio *prio;
1977
1978         steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
1979         if (!steering->esw_egress_root_ns)
1980                 return -ENOMEM;
1981
1982         /* create 1 prio*/
1983         prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
1984                               MLX5_TOTAL_VPORTS(steering->dev));
1985         return PTR_ERR_OR_ZERO(prio);
1986 }
1987
1988 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
1989 {
1990         struct fs_prio *prio;
1991
1992         steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
1993         if (!steering->esw_ingress_root_ns)
1994                 return -ENOMEM;
1995
1996         /* create 1 prio*/
1997         prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
1998                               MLX5_TOTAL_VPORTS(steering->dev));
1999         return PTR_ERR_OR_ZERO(prio);
2000 }
2001
2002 int mlx5_init_fs(struct mlx5_core_dev *dev)
2003 {
2004         struct mlx5_flow_steering *steering;
2005         int err = 0;
2006
2007         err = mlx5_init_fc_stats(dev);
2008         if (err)
2009                 return err;
2010
2011         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2012         if (!steering)
2013                 return -ENOMEM;
2014         steering->dev = dev;
2015         dev->priv.steering = steering;
2016
2017         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2018               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2019              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2020               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2021             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2022                 err = init_root_ns(steering);
2023                 if (err)
2024                         goto err;
2025         }
2026
2027         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
2028                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2029                         err = init_fdb_root_ns(steering);
2030                         if (err)
2031                                 goto err;
2032                 }
2033                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2034                         err = init_egress_acl_root_ns(steering);
2035                         if (err)
2036                                 goto err;
2037                 }
2038                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2039                         err = init_ingress_acl_root_ns(steering);
2040                         if (err)
2041                                 goto err;
2042                 }
2043         }
2044
2045         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2046                 err = init_sniffer_rx_root_ns(steering);
2047                 if (err)
2048                         goto err;
2049         }
2050
2051         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2052                 err = init_sniffer_tx_root_ns(steering);
2053                 if (err)
2054                         goto err;
2055         }
2056
2057         return 0;
2058 err:
2059         mlx5_cleanup_fs(dev);
2060         return err;
2061 }
2062
2063 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2064 {
2065         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2066
2067         root->underlay_qpn = underlay_qpn;
2068         return 0;
2069 }
2070 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
2071
2072 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2073 {
2074         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2075
2076         root->underlay_qpn = 0;
2077         return 0;
2078 }
2079 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);