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, max_ft_val, caps_val,\
  44                  ...) {.type = FS_TYPE_PRIO,\
  45         .min_ft_level = min_level_val,\
  46         .max_ft = max_ft_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, max_ft_val, ...)\
  54         ADD_PRIO(num_prios_val, 0, max_ft_val, {},\
  55                  __VA_ARGS__)\
  56
  57 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
  58         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  59         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  60 }
  61
  62 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  63                                    sizeof(long))
  64
  65 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  66
  67 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  68                                .caps = (long[]) {__VA_ARGS__} }
  69
  70 #define LEFTOVERS_MAX_FT 1
  71 #define LEFTOVERS_NUM_PRIOS 1
  72 #define BY_PASS_PRIO_MAX_FT 1
  73 #define BY_PASS_MIN_LEVEL (KENREL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  74                            LEFTOVERS_MAX_FT)
  75
  76 #define KERNEL_MAX_FT 3
  77 #define KERNEL_NUM_PRIOS 2
  78 #define KENREL_MIN_LEVEL 2
  79
  80 #define ANCHOR_MAX_FT 1
  81 #define ANCHOR_NUM_PRIOS 1
  82 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
  83 struct node_caps {
  84         size_t  arr_sz;
  85         long    *caps;
  86 };
  87 static struct init_tree_node {
  88         enum fs_node_type       type;
  89         struct init_tree_node *children;
  90         int ar_size;
  91         struct node_caps caps;
  92         int min_ft_level;
  93         int num_leaf_prios;
  94         int prio;
  95         int max_ft;
  96 } root_fs = {
  97         .type = FS_TYPE_NAMESPACE,
  98         .ar_size = 4,
  99         .children = (struct init_tree_node[]) {
 100                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 101                          FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en),
 102                                           FS_CAP(flow_table_properties_nic_receive.modify_root),
 103                                           FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode),
 104                                           FS_CAP(flow_table_properties_nic_receive.flow_table_modify)),
 105                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS, BY_PASS_PRIO_MAX_FT))),
 106                 ADD_PRIO(0, KENREL_MIN_LEVEL, 0, {},
 107                          ADD_NS(ADD_MULTIPLE_PRIO(KERNEL_NUM_PRIOS, KERNEL_MAX_FT))),
 108                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 109                          FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en),
 110                                           FS_CAP(flow_table_properties_nic_receive.modify_root),
 111                                           FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode),
 112                                           FS_CAP(flow_table_properties_nic_receive.flow_table_modify)),
 113                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_MAX_FT))),
 114                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 115                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_MAX_FT))),
 116         }
 117 };
 118
 119 enum fs_i_mutex_lock_class {
 120         FS_MUTEX_GRANDPARENT,
 121         FS_MUTEX_PARENT,
 122         FS_MUTEX_CHILD
 123 };
 124
 125 static void del_rule(struct fs_node *node);
 126 static void del_flow_table(struct fs_node *node);
 127 static void del_flow_group(struct fs_node *node);
 128 static void del_fte(struct fs_node *node);
 129
 130 static void tree_init_node(struct fs_node *node,
 131                            unsigned int refcount,
 132                            void (*remove_func)(struct fs_node *))
 133 {
 134         atomic_set(&node->refcount, refcount);
 135         INIT_LIST_HEAD(&node->list);
 136         INIT_LIST_HEAD(&node->children);
 137         mutex_init(&node->lock);
 138         node->remove_func = remove_func;
 139 }
 140
 141 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 142 {
 143         if (parent)
 144                 atomic_inc(&parent->refcount);
 145         node->parent = parent;
 146
 147         /* Parent is the root */
 148         if (!parent)
 149                 node->root = node;
 150         else
 151                 node->root = parent->root;
 152 }
 153
 154 static void tree_get_node(struct fs_node *node)
 155 {
 156         atomic_inc(&node->refcount);
 157 }
 158
 159 static void nested_lock_ref_node(struct fs_node *node,
 160                                  enum fs_i_mutex_lock_class class)
 161 {
 162         if (node) {
 163                 mutex_lock_nested(&node->lock, class);
 164                 atomic_inc(&node->refcount);
 165         }
 166 }
 167
 168 static void lock_ref_node(struct fs_node *node)
 169 {
 170         if (node) {
 171                 mutex_lock(&node->lock);
 172                 atomic_inc(&node->refcount);
 173         }
 174 }
 175
 176 static void unlock_ref_node(struct fs_node *node)
 177 {
 178         if (node) {
 179                 atomic_dec(&node->refcount);
 180                 mutex_unlock(&node->lock);
 181         }
 182 }
 183
 184 static void tree_put_node(struct fs_node *node)
 185 {
 186         struct fs_node *parent_node = node->parent;
 187
 188         lock_ref_node(parent_node);
 189         if (atomic_dec_and_test(&node->refcount)) {
 190                 if (parent_node)
 191                         list_del_init(&node->list);
 192                 if (node->remove_func)
 193                         node->remove_func(node);
 194                 kfree(node);
 195                 node = NULL;
 196         }
 197         unlock_ref_node(parent_node);
 198         if (!node && parent_node)
 199                 tree_put_node(parent_node);
 200 }
 201
 202 static int tree_remove_node(struct fs_node *node)
 203 {
 204         if (atomic_read(&node->refcount) > 1) {
 205                 atomic_dec(&node->refcount);
 206                 return -EEXIST;
 207         }
 208         tree_put_node(node);
 209         return 0;
 210 }
 211
 212 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 213                                  unsigned int prio)
 214 {
 215         struct fs_prio *iter_prio;
 216
 217         fs_for_each_prio(iter_prio, ns) {
 218                 if (iter_prio->prio == prio)
 219                         return iter_prio;
 220         }
 221
 222         return NULL;
 223 }
 224
 225 static unsigned int find_next_free_level(struct fs_prio *prio)
 226 {
 227         if (!list_empty(&prio->node.children)) {
 228                 struct mlx5_flow_table *ft;
 229
 230                 ft = list_last_entry(&prio->node.children,
 231                                      struct mlx5_flow_table,
 232                                      node.list);
 233                 return ft->level + 1;
 234         }
 235         return prio->start_level;
 236 }
 237
 238 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 239 {
 240         unsigned int i;
 241
 242         for (i = 0; i < size; i++, mask++, val1++, val2++)
 243                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 244                     ((*(u8 *)val2) & (*(u8 *)mask)))
 245                         return false;
 246
 247         return true;
 248 }
 249
 250 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 251                                 void *fte_param1, void *fte_param2)
 252 {
 253         if (mask->match_criteria_enable &
 254             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 255                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 256                                                 fte_param1, outer_headers);
 257                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 258                                                 fte_param2, outer_headers);
 259                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 260                                               mask->match_criteria, outer_headers);
 261
 262                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 263                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 264                         return false;
 265         }
 266
 267         if (mask->match_criteria_enable &
 268             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 269                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 270                                                 fte_param1, misc_parameters);
 271                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 272                                                 fte_param2, misc_parameters);
 273                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 274                                           mask->match_criteria, misc_parameters);
 275
 276                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 277                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 278                         return false;
 279         }
 280
 281         if (mask->match_criteria_enable &
 282             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 283                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 284                                                 fte_param1, inner_headers);
 285                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 286                                                 fte_param2, inner_headers);
 287                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 288                                           mask->match_criteria, inner_headers);
 289
 290                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 291                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 292                         return false;
 293         }
 294         return true;
 295 }
 296
 297 static bool compare_match_criteria(u8 match_criteria_enable1,
 298                                    u8 match_criteria_enable2,
 299                                    void *mask1, void *mask2)
 300 {
 301         return match_criteria_enable1 == match_criteria_enable2 &&
 302                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 303 }
 304
 305 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 306 {
 307         struct fs_node *root;
 308         struct mlx5_flow_namespace *ns;
 309
 310         root = node->root;
 311
 312         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 313                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 314                 return NULL;
 315         }
 316
 317         ns = container_of(root, struct mlx5_flow_namespace, node);
 318         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 319 }
 320
 321 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 322 {
 323         struct mlx5_flow_root_namespace *root = find_root(node);
 324
 325         if (root)
 326                 return root->dev;
 327         return NULL;
 328 }
 329
 330 static void del_flow_table(struct fs_node *node)
 331 {
 332         struct mlx5_flow_table *ft;
 333         struct mlx5_core_dev *dev;
 334         struct fs_prio *prio;
 335         int err;
 336
 337         fs_get_obj(ft, node);
 338         dev = get_dev(&ft->node);
 339
 340         err = mlx5_cmd_destroy_flow_table(dev, ft);
 341         if (err)
 342                 pr_warn("flow steering can't destroy ft\n");
 343         fs_get_obj(prio, ft->node.parent);
 344         prio->num_ft--;
 345 }
 346
 347 static void del_rule(struct fs_node *node)
 348 {
 349         struct mlx5_flow_rule *rule;
 350         struct mlx5_flow_table *ft;
 351         struct mlx5_flow_group *fg;
 352         struct fs_fte *fte;
 353         u32     *match_value;
 354         struct mlx5_core_dev *dev = get_dev(node);
 355         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 356         int err;
 357
 358         match_value = mlx5_vzalloc(match_len);
 359         if (!match_value) {
 360                 pr_warn("failed to allocate inbox\n");
 361                 return;
 362         }
 363
 364         fs_get_obj(rule, node);
 365         fs_get_obj(fte, rule->node.parent);
 366         fs_get_obj(fg, fte->node.parent);
 367         memcpy(match_value, fte->val, sizeof(fte->val));
 368         fs_get_obj(ft, fg->node.parent);
 369         list_del(&rule->node.list);
 370         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 371                 mutex_lock(&rule->dest_attr.ft->lock);
 372                 list_del(&rule->next_ft);
 373                 mutex_unlock(&rule->dest_attr.ft->lock);
 374         }
 375         if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 376             --fte->dests_size) {
 377                 err = mlx5_cmd_update_fte(dev, ft,
 378                                           fg->id, fte);
 379                 if (err)
 380                         pr_warn("%s can't del rule fg id=%d fte_index=%d\n",
 381                                 __func__, fg->id, fte->index);
 382         }
 383         kvfree(match_value);
 384 }
 385
 386 static void del_fte(struct fs_node *node)
 387 {
 388         struct mlx5_flow_table *ft;
 389         struct mlx5_flow_group *fg;
 390         struct mlx5_core_dev *dev;
 391         struct fs_fte *fte;
 392         int err;
 393
 394         fs_get_obj(fte, node);
 395         fs_get_obj(fg, fte->node.parent);
 396         fs_get_obj(ft, fg->node.parent);
 397
 398         dev = get_dev(&ft->node);
 399         err = mlx5_cmd_delete_fte(dev, ft,
 400                                   fte->index);
 401         if (err)
 402                 pr_warn("flow steering can't delete fte in index %d of flow group id %d\n",
 403                         fte->index, fg->id);
 404
 405         fte->status = 0;
 406         fg->num_ftes--;
 407 }
 408
 409 static void del_flow_group(struct fs_node *node)
 410 {
 411         struct mlx5_flow_group *fg;
 412         struct mlx5_flow_table *ft;
 413         struct mlx5_core_dev *dev;
 414
 415         fs_get_obj(fg, node);
 416         fs_get_obj(ft, fg->node.parent);
 417         dev = get_dev(&ft->node);
 418
 419         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 420                 pr_warn("flow steering can't destroy fg %d of ft %d\n",
 421                         fg->id, ft->id);
 422 }
 423
 424 static struct fs_fte *alloc_fte(u8 action,
 425                                 u32 flow_tag,
 426                                 u32 *match_value,
 427                                 unsigned int index)
 428 {
 429         struct fs_fte *fte;
 430
 431         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 432         if (!fte)
 433                 return ERR_PTR(-ENOMEM);
 434
 435         memcpy(fte->val, match_value, sizeof(fte->val));
 436         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 437         fte->flow_tag = flow_tag;
 438         fte->index = index;
 439         fte->action = action;
 440
 441         return fte;
 442 }
 443
 444 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 445 {
 446         struct mlx5_flow_group *fg;
 447         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 448                                             create_fg_in, match_criteria);
 449         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 450                                             create_fg_in,
 451                                             match_criteria_enable);
 452         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 453         if (!fg)
 454                 return ERR_PTR(-ENOMEM);
 455
 456         fg->mask.match_criteria_enable = match_criteria_enable;
 457         memcpy(&fg->mask.match_criteria, match_criteria,
 458                sizeof(fg->mask.match_criteria));
 459         fg->node.type =  FS_TYPE_FLOW_GROUP;
 460         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 461                                    start_flow_index);
 462         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 463                                 end_flow_index) - fg->start_index + 1;
 464         return fg;
 465 }
 466
 467 static struct mlx5_flow_table *alloc_flow_table(int level, int max_fte,
 468                                                 enum fs_flow_table_type table_type)
 469 {
 470         struct mlx5_flow_table *ft;
 471
 472         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 473         if (!ft)
 474                 return NULL;
 475
 476         ft->level = level;
 477         ft->node.type = FS_TYPE_FLOW_TABLE;
 478         ft->type = table_type;
 479         ft->max_fte = max_fte;
 480         INIT_LIST_HEAD(&ft->fwd_rules);
 481         mutex_init(&ft->lock);
 482
 483         return ft;
 484 }
 485
 486 /* If reverse is false, then we search for the first flow table in the
 487  * root sub-tree from start(closest from right), else we search for the
 488  * last flow table in the root sub-tree till start(closest from left).
 489  */
 490 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 491                                                          struct list_head *start,
 492                                                          bool reverse)
 493 {
 494 #define list_advance_entry(pos, reverse)                \
 495         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 496
 497 #define list_for_each_advance_continue(pos, head, reverse)      \
 498         for (pos = list_advance_entry(pos, reverse);            \
 499              &pos->list != (head);                              \
 500              pos = list_advance_entry(pos, reverse))
 501
 502         struct fs_node *iter = list_entry(start, struct fs_node, list);
 503         struct mlx5_flow_table *ft = NULL;
 504
 505         if (!root)
 506                 return NULL;
 507
 508         list_for_each_advance_continue(iter, &root->children, reverse) {
 509                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 510                         fs_get_obj(ft, iter);
 511                         return ft;
 512                 }
 513                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 514                 if (ft)
 515                         return ft;
 516         }
 517
 518         return ft;
 519 }
 520
 521 /* If reverse if false then return the first flow table in next priority of
 522  * prio in the tree, else return the last flow table in the previous priority
 523  * of prio in the tree.
 524  */
 525 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 526 {
 527         struct mlx5_flow_table *ft = NULL;
 528         struct fs_node *curr_node;
 529         struct fs_node *parent;
 530
 531         parent = prio->node.parent;
 532         curr_node = &prio->node;
 533         while (!ft && parent) {
 534                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 535                 curr_node = parent;
 536                 parent = curr_node->parent;
 537         }
 538         return ft;
 539 }
 540
 541 /* Assuming all the tree is locked by mutex chain lock */
 542 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 543 {
 544         return find_closest_ft(prio, false);
 545 }
 546
 547 /* Assuming all the tree is locked by mutex chain lock */
 548 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 549 {
 550         return find_closest_ft(prio, true);
 551 }
 552
 553 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 554                                struct fs_prio *prio,
 555                                struct mlx5_flow_table *ft)
 556 {
 557         struct mlx5_flow_table *iter;
 558         int i = 0;
 559         int err;
 560
 561         fs_for_each_ft(iter, prio) {
 562                 i++;
 563                 err = mlx5_cmd_modify_flow_table(dev,
 564                                                  iter,
 565                                                  ft);
 566                 if (err) {
 567                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 568                                        iter->id);
 569                         /* The driver is out of sync with the FW */
 570                         if (i > 1)
 571                                 WARN_ON(true);
 572                         return err;
 573                 }
 574         }
 575         return 0;
 576 }
 577
 578 /* Connect flow tables from previous priority of prio to ft */
 579 static int connect_prev_fts(struct mlx5_core_dev *dev,
 580                             struct mlx5_flow_table *ft,
 581                             struct fs_prio *prio)
 582 {
 583         struct mlx5_flow_table *prev_ft;
 584
 585         prev_ft = find_prev_chained_ft(prio);
 586         if (prev_ft) {
 587                 struct fs_prio *prev_prio;
 588
 589                 fs_get_obj(prev_prio, prev_ft->node.parent);
 590                 return connect_fts_in_prio(dev, prev_prio, ft);
 591         }
 592         return 0;
 593 }
 594
 595 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 596                                  *prio)
 597 {
 598         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 599         int min_level = INT_MAX;
 600         int err;
 601
 602         if (root->root_ft)
 603                 min_level = root->root_ft->level;
 604
 605         if (ft->level >= min_level)
 606                 return 0;
 607
 608         err = mlx5_cmd_update_root_ft(root->dev, ft);
 609         if (err)
 610                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 611                                ft->id);
 612         else
 613                 root->root_ft = ft;
 614
 615         return err;
 616 }
 617
 618 static int mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 619                                         struct mlx5_flow_destination *dest)
 620 {
 621         struct mlx5_flow_table *ft;
 622         struct mlx5_flow_group *fg;
 623         struct fs_fte *fte;
 624         int err = 0;
 625
 626         fs_get_obj(fte, rule->node.parent);
 627         if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 628                 return -EINVAL;
 629         lock_ref_node(&fte->node);
 630         fs_get_obj(fg, fte->node.parent);
 631         fs_get_obj(ft, fg->node.parent);
 632
 633         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 634         err = mlx5_cmd_update_fte(get_dev(&ft->node),
 635                                   ft, fg->id, fte);
 636         unlock_ref_node(&fte->node);
 637
 638         return err;
 639 }
 640
 641 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 642 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 643                              struct mlx5_flow_table *new_next_ft,
 644                              struct mlx5_flow_table *old_next_ft)
 645 {
 646         struct mlx5_flow_destination dest;
 647         struct mlx5_flow_rule *iter;
 648         int err = 0;
 649
 650         /* new_next_ft and old_next_ft could be NULL only
 651          * when we create/destroy the anchor flow table.
 652          */
 653         if (!new_next_ft || !old_next_ft)
 654                 return 0;
 655
 656         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 657         dest.ft = new_next_ft;
 658
 659         mutex_lock(&old_next_ft->lock);
 660         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 661         mutex_unlock(&old_next_ft->lock);
 662         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 663                 err = mlx5_modify_rule_destination(iter, &dest);
 664                 if (err)
 665                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 666                                new_next_ft->id);
 667         }
 668         return 0;
 669 }
 670
 671 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 672                               struct fs_prio *prio)
 673 {
 674         struct mlx5_flow_table *next_ft;
 675         int err = 0;
 676
 677         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 678
 679         if (list_empty(&prio->node.children)) {
 680                 err = connect_prev_fts(dev, ft, prio);
 681                 if (err)
 682                         return err;
 683
 684                 next_ft = find_next_chained_ft(prio);
 685                 err = connect_fwd_rules(dev, ft, next_ft);
 686                 if (err)
 687                         return err;
 688         }
 689
 690         if (MLX5_CAP_FLOWTABLE(dev,
 691                                flow_table_properties_nic_receive.modify_root))
 692                 err = update_root_ft_create(ft, prio);
 693         return err;
 694 }
 695
 696 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 697                                                int prio,
 698                                                int max_fte)
 699 {
 700         struct mlx5_flow_table *next_ft = NULL;
 701         struct mlx5_flow_table *ft;
 702         int err;
 703         int log_table_sz;
 704         struct mlx5_flow_root_namespace *root =
 705                 find_root(&ns->node);
 706         struct fs_prio *fs_prio = NULL;
 707
 708         if (!root) {
 709                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 710                 return ERR_PTR(-ENODEV);
 711         }
 712
 713         mutex_lock(&root->chain_lock);
 714         fs_prio = find_prio(ns, prio);
 715         if (!fs_prio) {
 716                 err = -EINVAL;
 717                 goto unlock_root;
 718         }
 719         if (fs_prio->num_ft == fs_prio->max_ft) {
 720                 err = -ENOSPC;
 721                 goto unlock_root;
 722         }
 723
 724         ft = alloc_flow_table(find_next_free_level(fs_prio),
 725                               roundup_pow_of_two(max_fte),
 726                               root->table_type);
 727         if (!ft) {
 728                 err = -ENOMEM;
 729                 goto unlock_root;
 730         }
 731
 732         tree_init_node(&ft->node, 1, del_flow_table);
 733         log_table_sz = ilog2(ft->max_fte);
 734         next_ft = find_next_chained_ft(fs_prio);
 735         err = mlx5_cmd_create_flow_table(root->dev, ft->type, ft->level,
 736                                          log_table_sz, next_ft, &ft->id);
 737         if (err)
 738                 goto free_ft;
 739
 740         err = connect_flow_table(root->dev, ft, fs_prio);
 741         if (err)
 742                 goto destroy_ft;
 743         lock_ref_node(&fs_prio->node);
 744         tree_add_node(&ft->node, &fs_prio->node);
 745         list_add_tail(&ft->node.list, &fs_prio->node.children);
 746         fs_prio->num_ft++;
 747         unlock_ref_node(&fs_prio->node);
 748         mutex_unlock(&root->chain_lock);
 749         return ft;
 750 destroy_ft:
 751         mlx5_cmd_destroy_flow_table(root->dev, ft);
 752 free_ft:
 753         kfree(ft);
 754 unlock_root:
 755         mutex_unlock(&root->chain_lock);
 756         return ERR_PTR(err);
 757 }
 758
 759 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 760                                                             int prio,
 761                                                             int num_flow_table_entries,
 762                                                             int max_num_groups)
 763 {
 764         struct mlx5_flow_table *ft;
 765
 766         if (max_num_groups > num_flow_table_entries)
 767                 return ERR_PTR(-EINVAL);
 768
 769         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries);
 770         if (IS_ERR(ft))
 771                 return ft;
 772
 773         ft->autogroup.active = true;
 774         ft->autogroup.required_groups = max_num_groups;
 775
 776         return ft;
 777 }
 778 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 779
 780 /* Flow table should be locked */
 781 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 782                                                         u32 *fg_in,
 783                                                         struct list_head
 784                                                         *prev_fg,
 785                                                         bool is_auto_fg)
 786 {
 787         struct mlx5_flow_group *fg;
 788         struct mlx5_core_dev *dev = get_dev(&ft->node);
 789         int err;
 790
 791         if (!dev)
 792                 return ERR_PTR(-ENODEV);
 793
 794         fg = alloc_flow_group(fg_in);
 795         if (IS_ERR(fg))
 796                 return fg;
 797
 798         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 799         if (err) {
 800                 kfree(fg);
 801                 return ERR_PTR(err);
 802         }
 803
 804         if (ft->autogroup.active)
 805                 ft->autogroup.num_groups++;
 806         /* Add node to tree */
 807         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 808         tree_add_node(&fg->node, &ft->node);
 809         /* Add node to group list */
 810         list_add(&fg->node.list, ft->node.children.prev);
 811
 812         return fg;
 813 }
 814
 815 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 816                                                u32 *fg_in)
 817 {
 818         struct mlx5_flow_group *fg;
 819
 820         if (ft->autogroup.active)
 821                 return ERR_PTR(-EPERM);
 822
 823         lock_ref_node(&ft->node);
 824         fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
 825         unlock_ref_node(&ft->node);
 826
 827         return fg;
 828 }
 829
 830 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 831 {
 832         struct mlx5_flow_rule *rule;
 833
 834         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 835         if (!rule)
 836                 return NULL;
 837
 838         INIT_LIST_HEAD(&rule->next_ft);
 839         rule->node.type = FS_TYPE_FLOW_DEST;
 840         if (dest)
 841                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 842
 843         return rule;
 844 }
 845
 846 /* fte should not be deleted while calling this function */
 847 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
 848                                            struct mlx5_flow_group *fg,
 849                                            struct mlx5_flow_destination *dest)
 850 {
 851         struct mlx5_flow_table *ft;
 852         struct mlx5_flow_rule *rule;
 853         int err;
 854
 855         rule = alloc_rule(dest);
 856         if (!rule)
 857                 return ERR_PTR(-ENOMEM);
 858
 859         fs_get_obj(ft, fg->node.parent);
 860         /* Add dest to dests list- we need flow tables to be in the
 861          * end of the list for forward to next prio rules.
 862          */
 863         tree_init_node(&rule->node, 1, del_rule);
 864         if (dest && dest->type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
 865                 list_add(&rule->node.list, &fte->node.children);
 866         else
 867                 list_add_tail(&rule->node.list, &fte->node.children);
 868         if (dest)
 869                 fte->dests_size++;
 870         if (fte->dests_size == 1 || !dest)
 871                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 872                                           ft, fg->id, fte);
 873         else
 874                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 875                                           ft, fg->id, fte);
 876         if (err)
 877                 goto free_rule;
 878
 879         fte->status |= FS_FTE_STATUS_EXISTING;
 880
 881         return rule;
 882
 883 free_rule:
 884         list_del(&rule->node.list);
 885         kfree(rule);
 886         if (dest)
 887                 fte->dests_size--;
 888         return ERR_PTR(err);
 889 }
 890
 891 /* Assumed fg is locked */
 892 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 893                                        struct list_head **prev)
 894 {
 895         struct fs_fte *fte;
 896         unsigned int start = fg->start_index;
 897
 898         if (prev)
 899                 *prev = &fg->node.children;
 900
 901         /* assumed list is sorted by index */
 902         fs_for_each_fte(fte, fg) {
 903                 if (fte->index != start)
 904                         return start;
 905                 start++;
 906                 if (prev)
 907                         *prev = &fte->node.list;
 908         }
 909
 910         return start;
 911 }
 912
 913 /* prev is output, prev->next = new_fte */
 914 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
 915                                  u32 *match_value,
 916                                  u8 action,
 917                                  u32 flow_tag,
 918                                  struct list_head **prev)
 919 {
 920         struct fs_fte *fte;
 921         int index;
 922
 923         index = get_free_fte_index(fg, prev);
 924         fte = alloc_fte(action, flow_tag, match_value, index);
 925         if (IS_ERR(fte))
 926                 return fte;
 927
 928         return fte;
 929 }
 930
 931 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
 932                                                 u8 match_criteria_enable,
 933                                                 u32 *match_criteria)
 934 {
 935         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 936         struct list_head *prev = &ft->node.children;
 937         unsigned int candidate_index = 0;
 938         struct mlx5_flow_group *fg;
 939         void *match_criteria_addr;
 940         unsigned int group_size = 0;
 941         u32 *in;
 942
 943         if (!ft->autogroup.active)
 944                 return ERR_PTR(-ENOENT);
 945
 946         in = mlx5_vzalloc(inlen);
 947         if (!in)
 948                 return ERR_PTR(-ENOMEM);
 949
 950         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
 951                 /* We save place for flow groups in addition to max types */
 952                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
 953
 954         /*  ft->max_fte == ft->autogroup.max_types */
 955         if (group_size == 0)
 956                 group_size = 1;
 957
 958         /* sorted by start_index */
 959         fs_for_each_fg(fg, ft) {
 960                 if (candidate_index + group_size > fg->start_index)
 961                         candidate_index = fg->start_index + fg->max_ftes;
 962                 else
 963                         break;
 964                 prev = &fg->node.list;
 965         }
 966
 967         if (candidate_index + group_size > ft->max_fte) {
 968                 fg = ERR_PTR(-ENOSPC);
 969                 goto out;
 970         }
 971
 972         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
 973                  match_criteria_enable);
 974         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
 975         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
 976                  group_size - 1);
 977         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
 978                                            in, match_criteria);
 979         memcpy(match_criteria_addr, match_criteria,
 980                MLX5_ST_SZ_BYTES(fte_match_param));
 981
 982         fg = create_flow_group_common(ft, in, prev, true);
 983 out:
 984         kvfree(in);
 985         return fg;
 986 }
 987
 988 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
 989                                              struct mlx5_flow_destination *dest)
 990 {
 991         struct mlx5_flow_rule *rule;
 992
 993         list_for_each_entry(rule, &fte->node.children, node.list) {
 994                 if (rule->dest_attr.type == dest->type) {
 995                         if ((dest->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
 996                              dest->vport_num == rule->dest_attr.vport_num) ||
 997                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
 998                              dest->ft == rule->dest_attr.ft) ||
 999                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1000                              dest->tir_num == rule->dest_attr.tir_num))
1001                                 return rule;
1002                 }
1003         }
1004         return NULL;
1005 }
1006
1007 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
1008                                           u32 *match_value,
1009                                           u8 action,
1010                                           u32 flow_tag,
1011                                           struct mlx5_flow_destination *dest)
1012 {
1013         struct fs_fte *fte;
1014         struct mlx5_flow_rule *rule;
1015         struct mlx5_flow_table *ft;
1016         struct list_head *prev;
1017
1018         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1019         fs_for_each_fte(fte, fg) {
1020                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1021                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1022                     action == fte->action && flow_tag == fte->flow_tag) {
1023                         rule = find_flow_rule(fte, dest);
1024                         if (rule) {
1025                                 atomic_inc(&rule->node.refcount);
1026                                 unlock_ref_node(&fte->node);
1027                                 unlock_ref_node(&fg->node);
1028                                 return rule;
1029                         }
1030                         rule = add_rule_fte(fte, fg, dest);
1031                         unlock_ref_node(&fte->node);
1032                         if (IS_ERR(rule))
1033                                 goto unlock_fg;
1034                         else
1035                                 goto add_rule;
1036                 }
1037                 unlock_ref_node(&fte->node);
1038         }
1039         fs_get_obj(ft, fg->node.parent);
1040         if (fg->num_ftes >= fg->max_ftes) {
1041                 rule = ERR_PTR(-ENOSPC);
1042                 goto unlock_fg;
1043         }
1044
1045         fte = create_fte(fg, match_value, action, flow_tag, &prev);
1046         if (IS_ERR(fte)) {
1047                 rule = (void *)fte;
1048                 goto unlock_fg;
1049         }
1050         tree_init_node(&fte->node, 0, del_fte);
1051         rule = add_rule_fte(fte, fg, dest);
1052         if (IS_ERR(rule)) {
1053                 kfree(fte);
1054                 goto unlock_fg;
1055         }
1056
1057         fg->num_ftes++;
1058
1059         tree_add_node(&fte->node, &fg->node);
1060         list_add(&fte->node.list, prev);
1061 add_rule:
1062         tree_add_node(&rule->node, &fte->node);
1063 unlock_fg:
1064         unlock_ref_node(&fg->node);
1065         return rule;
1066 }
1067
1068 static struct mlx5_flow_rule *add_rule_to_auto_fg(struct mlx5_flow_table *ft,
1069                                                   u8 match_criteria_enable,
1070                                                   u32 *match_criteria,
1071                                                   u32 *match_value,
1072                                                   u8 action,
1073                                                   u32 flow_tag,
1074                                                   struct mlx5_flow_destination *dest)
1075 {
1076         struct mlx5_flow_rule *rule;
1077         struct mlx5_flow_group *g;
1078
1079         g = create_autogroup(ft, match_criteria_enable, match_criteria);
1080         if (IS_ERR(g))
1081                 return (void *)g;
1082
1083         rule = add_rule_fg(g, match_value,
1084                            action, flow_tag, dest);
1085         if (IS_ERR(rule)) {
1086                 /* Remove assumes refcount > 0 and autogroup creates a group
1087                  * with a refcount = 0.
1088                  */
1089                 tree_get_node(&g->node);
1090                 tree_remove_node(&g->node);
1091         }
1092         return rule;
1093 }
1094
1095 static struct mlx5_flow_rule *
1096 _mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1097                     u8 match_criteria_enable,
1098                     u32 *match_criteria,
1099                     u32 *match_value,
1100                     u32 action,
1101                     u32 flow_tag,
1102                     struct mlx5_flow_destination *dest)
1103 {
1104         struct mlx5_flow_group *g;
1105         struct mlx5_flow_rule *rule;
1106
1107         if ((action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) && !dest)
1108                 return ERR_PTR(-EINVAL);
1109
1110         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1111         fs_for_each_fg(g, ft)
1112                 if (compare_match_criteria(g->mask.match_criteria_enable,
1113                                            match_criteria_enable,
1114                                            g->mask.match_criteria,
1115                                            match_criteria)) {
1116                         rule = add_rule_fg(g, match_value,
1117                                            action, flow_tag, dest);
1118                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1119                                 goto unlock;
1120                 }
1121
1122         rule = add_rule_to_auto_fg(ft, match_criteria_enable, match_criteria,
1123                                    match_value, action, flow_tag, dest);
1124 unlock:
1125         unlock_ref_node(&ft->node);
1126         return rule;
1127 }
1128
1129 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1130 {
1131         return ((ft->type == FS_FT_NIC_RX) &&
1132                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1133 }
1134
1135 struct mlx5_flow_rule *
1136 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1137                    u8 match_criteria_enable,
1138                    u32 *match_criteria,
1139                    u32 *match_value,
1140                    u32 action,
1141                    u32 flow_tag,
1142                    struct mlx5_flow_destination *dest)
1143 {
1144         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1145         struct mlx5_flow_destination gen_dest;
1146         struct mlx5_flow_table *next_ft = NULL;
1147         struct mlx5_flow_rule *rule = NULL;
1148         u32 sw_action = action;
1149         struct fs_prio *prio;
1150
1151         fs_get_obj(prio, ft->node.parent);
1152         if (action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1153                 if (!fwd_next_prio_supported(ft))
1154                         return ERR_PTR(-EOPNOTSUPP);
1155                 if (dest)
1156                         return ERR_PTR(-EINVAL);
1157                 mutex_lock(&root->chain_lock);
1158                 next_ft = find_next_chained_ft(prio);
1159                 if (next_ft) {
1160                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1161                         gen_dest.ft = next_ft;
1162                         dest = &gen_dest;
1163                         action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1164                 } else {
1165                         mutex_unlock(&root->chain_lock);
1166                         return ERR_PTR(-EOPNOTSUPP);
1167                 }
1168         }
1169
1170         rule =  _mlx5_add_flow_rule(ft, match_criteria_enable, match_criteria,
1171                                     match_value, action, flow_tag, dest);
1172
1173         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1174                 if (!IS_ERR_OR_NULL(rule) &&
1175                     (list_empty(&rule->next_ft))) {
1176                         mutex_lock(&next_ft->lock);
1177                         list_add(&rule->next_ft, &next_ft->fwd_rules);
1178                         mutex_unlock(&next_ft->lock);
1179                         rule->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1180                 }
1181                 mutex_unlock(&root->chain_lock);
1182         }
1183         return rule;
1184 }
1185 EXPORT_SYMBOL(mlx5_add_flow_rule);
1186
1187 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
1188 {
1189         tree_remove_node(&rule->node);
1190 }
1191 EXPORT_SYMBOL(mlx5_del_flow_rule);
1192
1193 /* Assuming prio->node.children(flow tables) is sorted by level */
1194 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1195 {
1196         struct fs_prio *prio;
1197
1198         fs_get_obj(prio, ft->node.parent);
1199
1200         if (!list_is_last(&ft->node.list, &prio->node.children))
1201                 return list_next_entry(ft, node.list);
1202         return find_next_chained_ft(prio);
1203 }
1204
1205 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1206 {
1207         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1208         struct mlx5_flow_table *new_root_ft = NULL;
1209
1210         if (root->root_ft != ft)
1211                 return 0;
1212
1213         new_root_ft = find_next_ft(ft);
1214         if (new_root_ft) {
1215                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1216
1217                 if (err) {
1218                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1219                                        ft->id);
1220                         return err;
1221                 }
1222                 root->root_ft = new_root_ft;
1223         }
1224         return 0;
1225 }
1226
1227 /* Connect flow table from previous priority to
1228  * the next flow table.
1229  */
1230 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1231 {
1232         struct mlx5_core_dev *dev = get_dev(&ft->node);
1233         struct mlx5_flow_table *next_ft;
1234         struct fs_prio *prio;
1235         int err = 0;
1236
1237         err = update_root_ft_destroy(ft);
1238         if (err)
1239                 return err;
1240
1241         fs_get_obj(prio, ft->node.parent);
1242         if  (!(list_first_entry(&prio->node.children,
1243                                 struct mlx5_flow_table,
1244                                 node.list) == ft))
1245                 return 0;
1246
1247         next_ft = find_next_chained_ft(prio);
1248         err = connect_fwd_rules(dev, next_ft, ft);
1249         if (err)
1250                 return err;
1251
1252         err = connect_prev_fts(dev, next_ft, prio);
1253         if (err)
1254                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1255                                ft->id);
1256         return err;
1257 }
1258
1259 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1260 {
1261         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1262         int err = 0;
1263
1264         mutex_lock(&root->chain_lock);
1265         err = disconnect_flow_table(ft);
1266         if (err) {
1267                 mutex_unlock(&root->chain_lock);
1268                 return err;
1269         }
1270         if (tree_remove_node(&ft->node))
1271                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1272                                ft->id);
1273         mutex_unlock(&root->chain_lock);
1274
1275         return err;
1276 }
1277 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1278
1279 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1280 {
1281         if (tree_remove_node(&fg->node))
1282                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1283                                fg->id);
1284 }
1285
1286 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1287                                                     enum mlx5_flow_namespace_type type)
1288 {
1289         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1290         int prio;
1291         static struct fs_prio *fs_prio;
1292         struct mlx5_flow_namespace *ns;
1293
1294         if (!root_ns)
1295                 return NULL;
1296
1297         switch (type) {
1298         case MLX5_FLOW_NAMESPACE_BYPASS:
1299         case MLX5_FLOW_NAMESPACE_KERNEL:
1300         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1301         case MLX5_FLOW_NAMESPACE_ANCHOR:
1302                 prio = type;
1303                 break;
1304         case MLX5_FLOW_NAMESPACE_FDB:
1305                 if (dev->priv.fdb_root_ns)
1306                         return &dev->priv.fdb_root_ns->ns;
1307                 else
1308                         return NULL;
1309         default:
1310                 return NULL;
1311         }
1312
1313         fs_prio = find_prio(&root_ns->ns, prio);
1314         if (!fs_prio)
1315                 return NULL;
1316
1317         ns = list_first_entry(&fs_prio->node.children,
1318                               typeof(*ns),
1319                               node.list);
1320
1321         return ns;
1322 }
1323 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1324
1325 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1326                                       unsigned prio, int max_ft)
1327 {
1328         struct fs_prio *fs_prio;
1329
1330         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1331         if (!fs_prio)
1332                 return ERR_PTR(-ENOMEM);
1333
1334         fs_prio->node.type = FS_TYPE_PRIO;
1335         tree_init_node(&fs_prio->node, 1, NULL);
1336         tree_add_node(&fs_prio->node, &ns->node);
1337         fs_prio->max_ft = max_ft;
1338         fs_prio->prio = prio;
1339         list_add_tail(&fs_prio->node.list, &ns->node.children);
1340
1341         return fs_prio;
1342 }
1343
1344 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1345                                                      *ns)
1346 {
1347         ns->node.type = FS_TYPE_NAMESPACE;
1348
1349         return ns;
1350 }
1351
1352 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1353 {
1354         struct mlx5_flow_namespace      *ns;
1355
1356         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1357         if (!ns)
1358                 return ERR_PTR(-ENOMEM);
1359
1360         fs_init_namespace(ns);
1361         tree_init_node(&ns->node, 1, NULL);
1362         tree_add_node(&ns->node, &prio->node);
1363         list_add_tail(&ns->node.list, &prio->node.children);
1364
1365         return ns;
1366 }
1367
1368 static int create_leaf_prios(struct mlx5_flow_namespace *ns, struct init_tree_node
1369                              *prio_metadata)
1370 {
1371         struct fs_prio *fs_prio;
1372         int i;
1373
1374         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1375                 fs_prio = fs_create_prio(ns, i, prio_metadata->max_ft);
1376                 if (IS_ERR(fs_prio))
1377                         return PTR_ERR(fs_prio);
1378         }
1379         return 0;
1380 }
1381
1382 #define FLOW_TABLE_BIT_SZ 1
1383 #define GET_FLOW_TABLE_CAP(dev, offset) \
1384         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1385                         offset / 32)) >>                                        \
1386           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1387 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1388 {
1389         int i;
1390
1391         for (i = 0; i < caps->arr_sz; i++) {
1392                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1393                         return false;
1394         }
1395         return true;
1396 }
1397
1398 static int init_root_tree_recursive(struct mlx5_core_dev *dev,
1399                                     struct init_tree_node *init_node,
1400                                     struct fs_node *fs_parent_node,
1401                                     struct init_tree_node *init_parent_node,
1402                                     int index)
1403 {
1404         int max_ft_level = MLX5_CAP_FLOWTABLE(dev,
1405                                               flow_table_properties_nic_receive.
1406                                               max_ft_level);
1407         struct mlx5_flow_namespace *fs_ns;
1408         struct fs_prio *fs_prio;
1409         struct fs_node *base;
1410         int i;
1411         int err;
1412
1413         if (init_node->type == FS_TYPE_PRIO) {
1414                 if ((init_node->min_ft_level > max_ft_level) ||
1415                     !has_required_caps(dev, &init_node->caps))
1416                         return 0;
1417
1418                 fs_get_obj(fs_ns, fs_parent_node);
1419                 if (init_node->num_leaf_prios)
1420                         return create_leaf_prios(fs_ns, init_node);
1421                 fs_prio = fs_create_prio(fs_ns, index, init_node->max_ft);
1422                 if (IS_ERR(fs_prio))
1423                         return PTR_ERR(fs_prio);
1424                 base = &fs_prio->node;
1425         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1426                 fs_get_obj(fs_prio, fs_parent_node);
1427                 fs_ns = fs_create_namespace(fs_prio);
1428                 if (IS_ERR(fs_ns))
1429                         return PTR_ERR(fs_ns);
1430                 base = &fs_ns->node;
1431         } else {
1432                 return -EINVAL;
1433         }
1434         for (i = 0; i < init_node->ar_size; i++) {
1435                 err = init_root_tree_recursive(dev, &init_node->children[i],
1436                                                base, init_node, i);
1437                 if (err)
1438                         return err;
1439         }
1440
1441         return 0;
1442 }
1443
1444 static int init_root_tree(struct mlx5_core_dev *dev,
1445                           struct init_tree_node *init_node,
1446                           struct fs_node *fs_parent_node)
1447 {
1448         int i;
1449         struct mlx5_flow_namespace *fs_ns;
1450         int err;
1451
1452         fs_get_obj(fs_ns, fs_parent_node);
1453         for (i = 0; i < init_node->ar_size; i++) {
1454                 err = init_root_tree_recursive(dev, &init_node->children[i],
1455                                                &fs_ns->node,
1456                                                init_node, i);
1457                 if (err)
1458                         return err;
1459         }
1460         return 0;
1461 }
1462
1463 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_core_dev *dev,
1464                                                        enum fs_flow_table_type
1465                                                        table_type)
1466 {
1467         struct mlx5_flow_root_namespace *root_ns;
1468         struct mlx5_flow_namespace *ns;
1469
1470         /* Create the root namespace */
1471         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1472         if (!root_ns)
1473                 return NULL;
1474
1475         root_ns->dev = dev;
1476         root_ns->table_type = table_type;
1477
1478         ns = &root_ns->ns;
1479         fs_init_namespace(ns);
1480         mutex_init(&root_ns->chain_lock);
1481         tree_init_node(&ns->node, 1, NULL);
1482         tree_add_node(&ns->node, NULL);
1483
1484         return root_ns;
1485 }
1486
1487 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1488
1489 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1490 {
1491         struct fs_prio *prio;
1492
1493         fs_for_each_prio(prio, ns) {
1494                  /* This updates prio start_level and max_ft */
1495                 set_prio_attrs_in_prio(prio, acc_level);
1496                 acc_level += prio->max_ft;
1497         }
1498         return acc_level;
1499 }
1500
1501 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1502 {
1503         struct mlx5_flow_namespace *ns;
1504         int acc_level_ns = acc_level;
1505
1506         prio->start_level = acc_level;
1507         fs_for_each_ns(ns, prio)
1508                 /* This updates start_level and max_ft of ns's priority descendants */
1509                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1510         if (!prio->max_ft)
1511                 prio->max_ft = acc_level_ns - prio->start_level;
1512         WARN_ON(prio->max_ft < acc_level_ns - prio->start_level);
1513 }
1514
1515 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1516 {
1517         struct mlx5_flow_namespace *ns = &root_ns->ns;
1518         struct fs_prio *prio;
1519         int start_level = 0;
1520
1521         fs_for_each_prio(prio, ns) {
1522                 set_prio_attrs_in_prio(prio, start_level);
1523                 start_level += prio->max_ft;
1524         }
1525 }
1526
1527 #define ANCHOR_PRIO 0
1528 #define ANCHOR_SIZE 1
1529 static int create_anchor_flow_table(struct mlx5_core_dev
1530                                                         *dev)
1531 {
1532         struct mlx5_flow_namespace *ns = NULL;
1533         struct mlx5_flow_table *ft;
1534
1535         ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1536         if (!ns)
1537                 return -EINVAL;
1538         ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE);
1539         if (IS_ERR(ft)) {
1540                 mlx5_core_err(dev, "Failed to create last anchor flow table");
1541                 return PTR_ERR(ft);
1542         }
1543         return 0;
1544 }
1545
1546 static int init_root_ns(struct mlx5_core_dev *dev)
1547 {
1548
1549         dev->priv.root_ns = create_root_ns(dev, FS_FT_NIC_RX);
1550         if (IS_ERR_OR_NULL(dev->priv.root_ns))
1551                 goto cleanup;
1552
1553         if (init_root_tree(dev, &root_fs, &dev->priv.root_ns->ns.node))
1554                 goto cleanup;
1555
1556         set_prio_attrs(dev->priv.root_ns);
1557
1558         if (create_anchor_flow_table(dev))
1559                 goto cleanup;
1560
1561         return 0;
1562
1563 cleanup:
1564         mlx5_cleanup_fs(dev);
1565         return -ENOMEM;
1566 }
1567
1568 static void cleanup_single_prio_root_ns(struct mlx5_core_dev *dev,
1569                                         struct mlx5_flow_root_namespace *root_ns)
1570 {
1571         struct fs_node *prio;
1572
1573         if (!root_ns)
1574                 return;
1575
1576         if (!list_empty(&root_ns->ns.node.children)) {
1577                 prio = list_first_entry(&root_ns->ns.node.children,
1578                                         struct fs_node,
1579                                  list);
1580                 if (tree_remove_node(prio))
1581                         mlx5_core_warn(dev,
1582                                        "Flow steering priority wasn't destroyed, refcount > 1\n");
1583         }
1584         if (tree_remove_node(&root_ns->ns.node))
1585                 mlx5_core_warn(dev,
1586                                "Flow steering namespace wasn't destroyed, refcount > 1\n");
1587         root_ns = NULL;
1588 }
1589
1590 static void destroy_flow_tables(struct fs_prio *prio)
1591 {
1592         struct mlx5_flow_table *iter;
1593         struct mlx5_flow_table *tmp;
1594
1595         fs_for_each_ft_safe(iter, tmp, prio)
1596                 mlx5_destroy_flow_table(iter);
1597 }
1598
1599 static void cleanup_root_ns(struct mlx5_core_dev *dev)
1600 {
1601         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1602         struct fs_prio *iter_prio;
1603
1604         if (!MLX5_CAP_GEN(dev, nic_flow_table))
1605                 return;
1606
1607         if (!root_ns)
1608                 return;
1609
1610         /* stage 1 */
1611         fs_for_each_prio(iter_prio, &root_ns->ns) {
1612                 struct fs_node *node;
1613                 struct mlx5_flow_namespace *iter_ns;
1614
1615                 fs_for_each_ns_or_ft(node, iter_prio) {
1616                         if (node->type == FS_TYPE_FLOW_TABLE)
1617                                 continue;
1618                         fs_get_obj(iter_ns, node);
1619                         while (!list_empty(&iter_ns->node.children)) {
1620                                 struct fs_prio *obj_iter_prio2;
1621                                 struct fs_node *iter_prio2 =
1622                                         list_first_entry(&iter_ns->node.children,
1623                                                          struct fs_node,
1624                                                          list);
1625
1626                                 fs_get_obj(obj_iter_prio2, iter_prio2);
1627                                 destroy_flow_tables(obj_iter_prio2);
1628                                 if (tree_remove_node(iter_prio2)) {
1629                                         mlx5_core_warn(dev,
1630                                                        "Priority %d wasn't destroyed, refcount > 1\n",
1631                                                        obj_iter_prio2->prio);
1632                                         return;
1633                                 }
1634                         }
1635                 }
1636         }
1637
1638         /* stage 2 */
1639         fs_for_each_prio(iter_prio, &root_ns->ns) {
1640                 while (!list_empty(&iter_prio->node.children)) {
1641                         struct fs_node *iter_ns =
1642                                 list_first_entry(&iter_prio->node.children,
1643                                                  struct fs_node,
1644                                                  list);
1645                         if (tree_remove_node(iter_ns)) {
1646                                 mlx5_core_warn(dev,
1647                                                "Namespace wasn't destroyed, refcount > 1\n");
1648                                 return;
1649                         }
1650                 }
1651         }
1652
1653         /* stage 3 */
1654         while (!list_empty(&root_ns->ns.node.children)) {
1655                 struct fs_prio *obj_prio_node;
1656                 struct fs_node *prio_node =
1657                         list_first_entry(&root_ns->ns.node.children,
1658                                          struct fs_node,
1659                                          list);
1660
1661                 fs_get_obj(obj_prio_node, prio_node);
1662                 if (tree_remove_node(prio_node)) {
1663                         mlx5_core_warn(dev,
1664                                        "Priority %d wasn't destroyed, refcount > 1\n",
1665                                        obj_prio_node->prio);
1666                         return;
1667                 }
1668         }
1669
1670         if (tree_remove_node(&root_ns->ns.node)) {
1671                 mlx5_core_warn(dev,
1672                                "root namespace wasn't destroyed, refcount > 1\n");
1673                 return;
1674         }
1675
1676         dev->priv.root_ns = NULL;
1677 }
1678
1679 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1680 {
1681         cleanup_root_ns(dev);
1682         cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1683 }
1684
1685 static int init_fdb_root_ns(struct mlx5_core_dev *dev)
1686 {
1687         struct fs_prio *prio;
1688
1689         dev->priv.fdb_root_ns = create_root_ns(dev, FS_FT_FDB);
1690         if (!dev->priv.fdb_root_ns)
1691                 return -ENOMEM;
1692
1693         /* Create single prio */
1694         prio = fs_create_prio(&dev->priv.fdb_root_ns->ns, 0, 1);
1695         if (IS_ERR(prio)) {
1696                 cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1697                 return PTR_ERR(prio);
1698         } else {
1699                 return 0;
1700         }
1701 }
1702
1703 int mlx5_init_fs(struct mlx5_core_dev *dev)
1704 {
1705         int err = 0;
1706
1707         if (MLX5_CAP_GEN(dev, nic_flow_table)) {
1708                 err = init_root_ns(dev);
1709                 if (err)
1710                         return err;
1711         }
1712         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1713                 err = init_fdb_root_ns(dev);
1714                 if (err)
1715                         cleanup_root_ns(dev);
1716         }
1717
1718         return err;
1719 }