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 INIT_PRIO(min_level_val, max_ft_val,\
  44                   start_level_val, ...) {.type = FS_TYPE_PRIO,\
  45         .min_ft_level = min_level_val,\
  46         .start_level = start_level_val,\
  47         .max_ft = max_ft_val,\
  48         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  49         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  50 }
  51
  52 #define ADD_PRIO(min_level_val, max_ft_val, start_level_val, ...)\
  53         INIT_PRIO(min_level_val, max_ft_val, start_level_val,\
  54                   __VA_ARGS__)\
  55
  56 #define ADD_FT_PRIO(max_ft_val, start_level_val, ...)\
  57         INIT_PRIO(0, max_ft_val, start_level_val,\
  58                   __VA_ARGS__)\
  59
  60 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
  61         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  62         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  63 }
  64
  65 #define KERNEL_START_LEVEL 0
  66 #define KERNEL_P0_START_LEVEL KERNEL_START_LEVEL
  67 #define KERNEL_MAX_FT 2
  68 #define KENREL_MIN_LEVEL 2
  69 static struct init_tree_node {
  70         enum fs_node_type       type;
  71         struct init_tree_node *children;
  72         int ar_size;
  73         int min_ft_level;
  74         int prio;
  75         int max_ft;
  76         int start_level;
  77 } root_fs = {
  78         .type = FS_TYPE_NAMESPACE,
  79         .ar_size = 1,
  80         .children = (struct init_tree_node[]) {
  81                 ADD_PRIO(KENREL_MIN_LEVEL, KERNEL_MAX_FT,
  82                          KERNEL_START_LEVEL,
  83                          ADD_NS(ADD_FT_PRIO(KERNEL_MAX_FT,
  84                                             KERNEL_P0_START_LEVEL))),
  85         }
  86 };
  87
  88 enum fs_i_mutex_lock_class {
  89         FS_MUTEX_GRANDPARENT,
  90         FS_MUTEX_PARENT,
  91         FS_MUTEX_CHILD
  92 };
  93
  94 static void del_rule(struct fs_node *node);
  95 static void del_flow_table(struct fs_node *node);
  96 static void del_flow_group(struct fs_node *node);
  97 static void del_fte(struct fs_node *node);
  98
  99 static void tree_init_node(struct fs_node *node,
 100                            unsigned int refcount,
 101                            void (*remove_func)(struct fs_node *))
 102 {
 103         atomic_set(&node->refcount, refcount);
 104         INIT_LIST_HEAD(&node->list);
 105         INIT_LIST_HEAD(&node->children);
 106         mutex_init(&node->lock);
 107         node->remove_func = remove_func;
 108 }
 109
 110 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 111 {
 112         if (parent)
 113                 atomic_inc(&parent->refcount);
 114         node->parent = parent;
 115
 116         /* Parent is the root */
 117         if (!parent)
 118                 node->root = node;
 119         else
 120                 node->root = parent->root;
 121 }
 122
 123 static void tree_get_node(struct fs_node *node)
 124 {
 125         atomic_inc(&node->refcount);
 126 }
 127
 128 static void nested_lock_ref_node(struct fs_node *node,
 129                                  enum fs_i_mutex_lock_class class)
 130 {
 131         if (node) {
 132                 mutex_lock_nested(&node->lock, class);
 133                 atomic_inc(&node->refcount);
 134         }
 135 }
 136
 137 static void lock_ref_node(struct fs_node *node)
 138 {
 139         if (node) {
 140                 mutex_lock(&node->lock);
 141                 atomic_inc(&node->refcount);
 142         }
 143 }
 144
 145 static void unlock_ref_node(struct fs_node *node)
 146 {
 147         if (node) {
 148                 atomic_dec(&node->refcount);
 149                 mutex_unlock(&node->lock);
 150         }
 151 }
 152
 153 static void tree_put_node(struct fs_node *node)
 154 {
 155         struct fs_node *parent_node = node->parent;
 156
 157         lock_ref_node(parent_node);
 158         if (atomic_dec_and_test(&node->refcount)) {
 159                 if (parent_node)
 160                         list_del_init(&node->list);
 161                 if (node->remove_func)
 162                         node->remove_func(node);
 163                 kfree(node);
 164                 node = NULL;
 165         }
 166         unlock_ref_node(parent_node);
 167         if (!node && parent_node)
 168                 tree_put_node(parent_node);
 169 }
 170
 171 static int tree_remove_node(struct fs_node *node)
 172 {
 173         if (atomic_read(&node->refcount) > 1)
 174                 return -EPERM;
 175         tree_put_node(node);
 176         return 0;
 177 }
 178
 179 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 180                                  unsigned int prio)
 181 {
 182         struct fs_prio *iter_prio;
 183
 184         fs_for_each_prio(iter_prio, ns) {
 185                 if (iter_prio->prio == prio)
 186                         return iter_prio;
 187         }
 188
 189         return NULL;
 190 }
 191
 192 static unsigned int find_next_free_level(struct fs_prio *prio)
 193 {
 194         if (!list_empty(&prio->node.children)) {
 195                 struct mlx5_flow_table *ft;
 196
 197                 ft = list_last_entry(&prio->node.children,
 198                                      struct mlx5_flow_table,
 199                                      node.list);
 200                 return ft->level + 1;
 201         }
 202         return prio->start_level;
 203 }
 204
 205 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 206 {
 207         unsigned int i;
 208
 209         for (i = 0; i < size; i++, mask++, val1++, val2++)
 210                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 211                     ((*(u8 *)val2) & (*(u8 *)mask)))
 212                         return false;
 213
 214         return true;
 215 }
 216
 217 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 218                                 void *fte_param1, void *fte_param2)
 219 {
 220         if (mask->match_criteria_enable &
 221             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 222                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 223                                                 fte_param1, outer_headers);
 224                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 225                                                 fte_param2, outer_headers);
 226                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 227                                               mask->match_criteria, outer_headers);
 228
 229                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 230                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 231                         return false;
 232         }
 233
 234         if (mask->match_criteria_enable &
 235             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 236                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 237                                                 fte_param1, misc_parameters);
 238                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 239                                                 fte_param2, misc_parameters);
 240                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 241                                           mask->match_criteria, misc_parameters);
 242
 243                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 244                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 245                         return false;
 246         }
 247
 248         if (mask->match_criteria_enable &
 249             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 250                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 251                                                 fte_param1, inner_headers);
 252                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 253                                                 fte_param2, inner_headers);
 254                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 255                                           mask->match_criteria, inner_headers);
 256
 257                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 258                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 259                         return false;
 260         }
 261         return true;
 262 }
 263
 264 static bool compare_match_criteria(u8 match_criteria_enable1,
 265                                    u8 match_criteria_enable2,
 266                                    void *mask1, void *mask2)
 267 {
 268         return match_criteria_enable1 == match_criteria_enable2 &&
 269                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 270 }
 271
 272 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 273 {
 274         struct fs_node *root;
 275         struct mlx5_flow_namespace *ns;
 276
 277         root = node->root;
 278
 279         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 280                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 281                 return NULL;
 282         }
 283
 284         ns = container_of(root, struct mlx5_flow_namespace, node);
 285         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 286 }
 287
 288 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 289 {
 290         struct mlx5_flow_root_namespace *root = find_root(node);
 291
 292         if (root)
 293                 return root->dev;
 294         return NULL;
 295 }
 296
 297 static void del_flow_table(struct fs_node *node)
 298 {
 299         struct mlx5_flow_table *ft;
 300         struct mlx5_core_dev *dev;
 301         struct fs_prio *prio;
 302         int err;
 303
 304         fs_get_obj(ft, node);
 305         dev = get_dev(&ft->node);
 306
 307         err = mlx5_cmd_destroy_flow_table(dev, ft);
 308         if (err)
 309                 pr_warn("flow steering can't destroy ft\n");
 310         fs_get_obj(prio, ft->node.parent);
 311         prio->num_ft--;
 312 }
 313
 314 static void del_rule(struct fs_node *node)
 315 {
 316         struct mlx5_flow_rule *rule;
 317         struct mlx5_flow_table *ft;
 318         struct mlx5_flow_group *fg;
 319         struct fs_fte *fte;
 320         u32     *match_value;
 321         struct mlx5_core_dev *dev = get_dev(node);
 322         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 323         int err;
 324
 325         match_value = mlx5_vzalloc(match_len);
 326         if (!match_value) {
 327                 pr_warn("failed to allocate inbox\n");
 328                 return;
 329         }
 330
 331         fs_get_obj(rule, node);
 332         fs_get_obj(fte, rule->node.parent);
 333         fs_get_obj(fg, fte->node.parent);
 334         memcpy(match_value, fte->val, sizeof(fte->val));
 335         fs_get_obj(ft, fg->node.parent);
 336         list_del(&rule->node.list);
 337         fte->dests_size--;
 338         if (fte->dests_size) {
 339                 err = mlx5_cmd_update_fte(dev, ft,
 340                                           fg->id, fte);
 341                 if (err)
 342                         pr_warn("%s can't del rule fg id=%d fte_index=%d\n",
 343                                 __func__, fg->id, fte->index);
 344         }
 345         kvfree(match_value);
 346 }
 347
 348 static void del_fte(struct fs_node *node)
 349 {
 350         struct mlx5_flow_table *ft;
 351         struct mlx5_flow_group *fg;
 352         struct mlx5_core_dev *dev;
 353         struct fs_fte *fte;
 354         int err;
 355
 356         fs_get_obj(fte, node);
 357         fs_get_obj(fg, fte->node.parent);
 358         fs_get_obj(ft, fg->node.parent);
 359
 360         dev = get_dev(&ft->node);
 361         err = mlx5_cmd_delete_fte(dev, ft,
 362                                   fte->index);
 363         if (err)
 364                 pr_warn("flow steering can't delete fte in index %d of flow group id %d\n",
 365                         fte->index, fg->id);
 366
 367         fte->status = 0;
 368         fg->num_ftes--;
 369 }
 370
 371 static void del_flow_group(struct fs_node *node)
 372 {
 373         struct mlx5_flow_group *fg;
 374         struct mlx5_flow_table *ft;
 375         struct mlx5_core_dev *dev;
 376
 377         fs_get_obj(fg, node);
 378         fs_get_obj(ft, fg->node.parent);
 379         dev = get_dev(&ft->node);
 380
 381         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 382                 pr_warn("flow steering can't destroy fg %d of ft %d\n",
 383                         fg->id, ft->id);
 384 }
 385
 386 static struct fs_fte *alloc_fte(u8 action,
 387                                 u32 flow_tag,
 388                                 u32 *match_value,
 389                                 unsigned int index)
 390 {
 391         struct fs_fte *fte;
 392
 393         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 394         if (!fte)
 395                 return ERR_PTR(-ENOMEM);
 396
 397         memcpy(fte->val, match_value, sizeof(fte->val));
 398         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 399         fte->flow_tag = flow_tag;
 400         fte->index = index;
 401         fte->action = action;
 402
 403         return fte;
 404 }
 405
 406 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 407 {
 408         struct mlx5_flow_group *fg;
 409         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 410                                             create_fg_in, match_criteria);
 411         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 412                                             create_fg_in,
 413                                             match_criteria_enable);
 414         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 415         if (!fg)
 416                 return ERR_PTR(-ENOMEM);
 417
 418         fg->mask.match_criteria_enable = match_criteria_enable;
 419         memcpy(&fg->mask.match_criteria, match_criteria,
 420                sizeof(fg->mask.match_criteria));
 421         fg->node.type =  FS_TYPE_FLOW_GROUP;
 422         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 423                                    start_flow_index);
 424         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 425                                 end_flow_index) - fg->start_index + 1;
 426         return fg;
 427 }
 428
 429 static struct mlx5_flow_table *alloc_flow_table(int level, int max_fte,
 430                                                 enum fs_flow_table_type table_type)
 431 {
 432         struct mlx5_flow_table *ft;
 433
 434         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 435         if (!ft)
 436                 return NULL;
 437
 438         ft->level = level;
 439         ft->node.type = FS_TYPE_FLOW_TABLE;
 440         ft->type = table_type;
 441         ft->max_fte = max_fte;
 442
 443         return ft;
 444 }
 445
 446 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 447                                                int prio,
 448                                                int max_fte)
 449 {
 450         struct mlx5_flow_table *ft;
 451         int err;
 452         int log_table_sz;
 453         struct mlx5_flow_root_namespace *root =
 454                 find_root(&ns->node);
 455         struct fs_prio *fs_prio = NULL;
 456
 457         if (!root) {
 458                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 459                 return ERR_PTR(-ENODEV);
 460         }
 461
 462         fs_prio = find_prio(ns, prio);
 463         if (!fs_prio)
 464                 return ERR_PTR(-EINVAL);
 465
 466         lock_ref_node(&fs_prio->node);
 467         if (fs_prio->num_ft == fs_prio->max_ft) {
 468                 err = -ENOSPC;
 469                 goto unlock_prio;
 470         }
 471
 472         ft = alloc_flow_table(find_next_free_level(fs_prio),
 473                               roundup_pow_of_two(max_fte),
 474                               root->table_type);
 475         if (!ft) {
 476                 err = -ENOMEM;
 477                 goto unlock_prio;
 478         }
 479
 480         tree_init_node(&ft->node, 1, del_flow_table);
 481         log_table_sz = ilog2(ft->max_fte);
 482         err = mlx5_cmd_create_flow_table(root->dev, ft->type, ft->level,
 483                                          log_table_sz, &ft->id);
 484         if (err)
 485                 goto free_ft;
 486
 487         tree_add_node(&ft->node, &fs_prio->node);
 488         list_add_tail(&ft->node.list, &fs_prio->node.children);
 489         fs_prio->num_ft++;
 490         unlock_ref_node(&fs_prio->node);
 491         return ft;
 492 free_ft:
 493         kfree(ft);
 494 unlock_prio:
 495         unlock_ref_node(&fs_prio->node);
 496         return ERR_PTR(err);
 497 }
 498
 499 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 500                                                             int prio,
 501                                                             int num_flow_table_entries,
 502                                                             int max_num_groups)
 503 {
 504         struct mlx5_flow_table *ft;
 505
 506         if (max_num_groups > num_flow_table_entries)
 507                 return ERR_PTR(-EINVAL);
 508
 509         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries);
 510         if (IS_ERR(ft))
 511                 return ft;
 512
 513         ft->autogroup.active = true;
 514         ft->autogroup.required_groups = max_num_groups;
 515
 516         return ft;
 517 }
 518
 519 /* Flow table should be locked */
 520 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 521                                                         u32 *fg_in,
 522                                                         struct list_head
 523                                                         *prev_fg,
 524                                                         bool is_auto_fg)
 525 {
 526         struct mlx5_flow_group *fg;
 527         struct mlx5_core_dev *dev = get_dev(&ft->node);
 528         int err;
 529
 530         if (!dev)
 531                 return ERR_PTR(-ENODEV);
 532
 533         fg = alloc_flow_group(fg_in);
 534         if (IS_ERR(fg))
 535                 return fg;
 536
 537         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 538         if (err) {
 539                 kfree(fg);
 540                 return ERR_PTR(err);
 541         }
 542
 543         if (ft->autogroup.active)
 544                 ft->autogroup.num_groups++;
 545         /* Add node to tree */
 546         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 547         tree_add_node(&fg->node, &ft->node);
 548         /* Add node to group list */
 549         list_add(&fg->node.list, ft->node.children.prev);
 550
 551         return fg;
 552 }
 553
 554 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 555                                                u32 *fg_in)
 556 {
 557         struct mlx5_flow_group *fg;
 558
 559         if (ft->autogroup.active)
 560                 return ERR_PTR(-EPERM);
 561
 562         lock_ref_node(&ft->node);
 563         fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
 564         unlock_ref_node(&ft->node);
 565
 566         return fg;
 567 }
 568
 569 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 570 {
 571         struct mlx5_flow_rule *rule;
 572
 573         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 574         if (!rule)
 575                 return NULL;
 576
 577         rule->node.type = FS_TYPE_FLOW_DEST;
 578         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 579
 580         return rule;
 581 }
 582
 583 /* fte should not be deleted while calling this function */
 584 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
 585                                            struct mlx5_flow_group *fg,
 586                                            struct mlx5_flow_destination *dest)
 587 {
 588         struct mlx5_flow_table *ft;
 589         struct mlx5_flow_rule *rule;
 590         int err;
 591
 592         rule = alloc_rule(dest);
 593         if (!rule)
 594                 return ERR_PTR(-ENOMEM);
 595
 596         fs_get_obj(ft, fg->node.parent);
 597         /* Add dest to dests list- added as first element after the head */
 598         tree_init_node(&rule->node, 1, del_rule);
 599         list_add_tail(&rule->node.list, &fte->node.children);
 600         fte->dests_size++;
 601         if (fte->dests_size == 1)
 602                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 603                                           ft, fg->id, fte);
 604         else
 605                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 606                                           ft, fg->id, fte);
 607         if (err)
 608                 goto free_rule;
 609
 610         fte->status |= FS_FTE_STATUS_EXISTING;
 611
 612         return rule;
 613
 614 free_rule:
 615         list_del(&rule->node.list);
 616         kfree(rule);
 617         fte->dests_size--;
 618         return ERR_PTR(err);
 619 }
 620
 621 /* Assumed fg is locked */
 622 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 623                                        struct list_head **prev)
 624 {
 625         struct fs_fte *fte;
 626         unsigned int start = fg->start_index;
 627
 628         if (prev)
 629                 *prev = &fg->node.children;
 630
 631         /* assumed list is sorted by index */
 632         fs_for_each_fte(fte, fg) {
 633                 if (fte->index != start)
 634                         return start;
 635                 start++;
 636                 if (prev)
 637                         *prev = &fte->node.list;
 638         }
 639
 640         return start;
 641 }
 642
 643 /* prev is output, prev->next = new_fte */
 644 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
 645                                  u32 *match_value,
 646                                  u8 action,
 647                                  u32 flow_tag,
 648                                  struct list_head **prev)
 649 {
 650         struct fs_fte *fte;
 651         int index;
 652
 653         index = get_free_fte_index(fg, prev);
 654         fte = alloc_fte(action, flow_tag, match_value, index);
 655         if (IS_ERR(fte))
 656                 return fte;
 657
 658         return fte;
 659 }
 660
 661 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
 662                                                 u8 match_criteria_enable,
 663                                                 u32 *match_criteria)
 664 {
 665         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 666         struct list_head *prev = &ft->node.children;
 667         unsigned int candidate_index = 0;
 668         struct mlx5_flow_group *fg;
 669         void *match_criteria_addr;
 670         unsigned int group_size = 0;
 671         u32 *in;
 672
 673         if (!ft->autogroup.active)
 674                 return ERR_PTR(-ENOENT);
 675
 676         in = mlx5_vzalloc(inlen);
 677         if (!in)
 678                 return ERR_PTR(-ENOMEM);
 679
 680         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
 681                 /* We save place for flow groups in addition to max types */
 682                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
 683
 684         /*  ft->max_fte == ft->autogroup.max_types */
 685         if (group_size == 0)
 686                 group_size = 1;
 687
 688         /* sorted by start_index */
 689         fs_for_each_fg(fg, ft) {
 690                 if (candidate_index + group_size > fg->start_index)
 691                         candidate_index = fg->start_index + fg->max_ftes;
 692                 else
 693                         break;
 694                 prev = &fg->node.list;
 695         }
 696
 697         if (candidate_index + group_size > ft->max_fte) {
 698                 fg = ERR_PTR(-ENOSPC);
 699                 goto out;
 700         }
 701
 702         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
 703                  match_criteria_enable);
 704         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
 705         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
 706                  group_size - 1);
 707         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
 708                                            in, match_criteria);
 709         memcpy(match_criteria_addr, match_criteria,
 710                MLX5_ST_SZ_BYTES(fte_match_param));
 711
 712         fg = create_flow_group_common(ft, in, prev, true);
 713 out:
 714         kvfree(in);
 715         return fg;
 716 }
 717
 718 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
 719                                           u32 *match_value,
 720                                           u8 action,
 721                                           u32 flow_tag,
 722                                           struct mlx5_flow_destination *dest)
 723 {
 724         struct fs_fte *fte;
 725         struct mlx5_flow_rule *rule;
 726         struct mlx5_flow_table *ft;
 727         struct list_head *prev;
 728
 729         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
 730         fs_for_each_fte(fte, fg) {
 731                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
 732                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
 733                     action == fte->action && flow_tag == fte->flow_tag) {
 734                         rule = add_rule_fte(fte, fg, dest);
 735                         unlock_ref_node(&fte->node);
 736                         if (IS_ERR(rule))
 737                                 goto unlock_fg;
 738                         else
 739                                 goto add_rule;
 740                 }
 741                 unlock_ref_node(&fte->node);
 742         }
 743         fs_get_obj(ft, fg->node.parent);
 744         if (fg->num_ftes >= fg->max_ftes) {
 745                 rule = ERR_PTR(-ENOSPC);
 746                 goto unlock_fg;
 747         }
 748
 749         fte = create_fte(fg, match_value, action, flow_tag, &prev);
 750         if (IS_ERR(fte)) {
 751                 rule = (void *)fte;
 752                 goto unlock_fg;
 753         }
 754         tree_init_node(&fte->node, 0, del_fte);
 755         rule = add_rule_fte(fte, fg, dest);
 756         if (IS_ERR(rule)) {
 757                 kfree(fte);
 758                 goto unlock_fg;
 759         }
 760
 761         fg->num_ftes++;
 762
 763         tree_add_node(&fte->node, &fg->node);
 764         list_add(&fte->node.list, prev);
 765 add_rule:
 766         tree_add_node(&rule->node, &fte->node);
 767 unlock_fg:
 768         unlock_ref_node(&fg->node);
 769         return rule;
 770 }
 771
 772 static struct mlx5_flow_rule *add_rule_to_auto_fg(struct mlx5_flow_table *ft,
 773                                                   u8 match_criteria_enable,
 774                                                   u32 *match_criteria,
 775                                                   u32 *match_value,
 776                                                   u8 action,
 777                                                   u32 flow_tag,
 778                                                   struct mlx5_flow_destination *dest)
 779 {
 780         struct mlx5_flow_rule *rule;
 781         struct mlx5_flow_group *g;
 782
 783         g = create_autogroup(ft, match_criteria_enable, match_criteria);
 784         if (IS_ERR(g))
 785                 return (void *)g;
 786
 787         rule = add_rule_fg(g, match_value,
 788                            action, flow_tag, dest);
 789         if (IS_ERR(rule)) {
 790                 /* Remove assumes refcount > 0 and autogroup creates a group
 791                  * with a refcount = 0.
 792                  */
 793                 tree_get_node(&g->node);
 794                 tree_remove_node(&g->node);
 795         }
 796         return rule;
 797 }
 798
 799 struct mlx5_flow_rule *
 800 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
 801                    u8 match_criteria_enable,
 802                    u32 *match_criteria,
 803                    u32 *match_value,
 804                    u32 action,
 805                    u32 flow_tag,
 806                    struct mlx5_flow_destination *dest)
 807 {
 808         struct mlx5_flow_group *g;
 809         struct mlx5_flow_rule *rule;
 810
 811         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
 812         fs_for_each_fg(g, ft)
 813                 if (compare_match_criteria(g->mask.match_criteria_enable,
 814                                            match_criteria_enable,
 815                                            g->mask.match_criteria,
 816                                            match_criteria)) {
 817                         rule = add_rule_fg(g, match_value,
 818                                            action, flow_tag, dest);
 819                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
 820                                 goto unlock;
 821                 }
 822
 823         rule = add_rule_to_auto_fg(ft, match_criteria_enable, match_criteria,
 824                                    match_value, action, flow_tag, dest);
 825 unlock:
 826         unlock_ref_node(&ft->node);
 827         return rule;
 828 }
 829
 830 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
 831 {
 832         tree_remove_node(&rule->node);
 833 }
 834
 835 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
 836 {
 837         if (tree_remove_node(&ft->node))
 838                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
 839                                ft->id);
 840
 841         return 0;
 842 }
 843
 844 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
 845 {
 846         if (tree_remove_node(&fg->node))
 847                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
 848                                fg->id);
 849 }
 850
 851 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
 852                                                     enum mlx5_flow_namespace_type type)
 853 {
 854         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
 855         int prio;
 856         static struct fs_prio *fs_prio;
 857         struct mlx5_flow_namespace *ns;
 858
 859         if (!root_ns)
 860                 return NULL;
 861
 862         switch (type) {
 863         case MLX5_FLOW_NAMESPACE_KERNEL:
 864                 prio = 0;
 865                 break;
 866         case MLX5_FLOW_NAMESPACE_FDB:
 867                 if (dev->priv.fdb_root_ns)
 868                         return &dev->priv.fdb_root_ns->ns;
 869                 else
 870                         return NULL;
 871         default:
 872                 return NULL;
 873         }
 874
 875         fs_prio = find_prio(&root_ns->ns, prio);
 876         if (!fs_prio)
 877                 return NULL;
 878
 879         ns = list_first_entry(&fs_prio->node.children,
 880                               typeof(*ns),
 881                               node.list);
 882
 883         return ns;
 884 }
 885
 886 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
 887                                       unsigned prio, int max_ft,
 888                                       int start_level)
 889 {
 890         struct fs_prio *fs_prio;
 891
 892         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
 893         if (!fs_prio)
 894                 return ERR_PTR(-ENOMEM);
 895
 896         fs_prio->node.type = FS_TYPE_PRIO;
 897         tree_init_node(&fs_prio->node, 1, NULL);
 898         tree_add_node(&fs_prio->node, &ns->node);
 899         fs_prio->max_ft = max_ft;
 900         fs_prio->prio = prio;
 901         fs_prio->start_level = start_level;
 902         list_add_tail(&fs_prio->node.list, &ns->node.children);
 903
 904         return fs_prio;
 905 }
 906
 907 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
 908                                                      *ns)
 909 {
 910         ns->node.type = FS_TYPE_NAMESPACE;
 911
 912         return ns;
 913 }
 914
 915 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
 916 {
 917         struct mlx5_flow_namespace      *ns;
 918
 919         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
 920         if (!ns)
 921                 return ERR_PTR(-ENOMEM);
 922
 923         fs_init_namespace(ns);
 924         tree_init_node(&ns->node, 1, NULL);
 925         tree_add_node(&ns->node, &prio->node);
 926         list_add_tail(&ns->node.list, &prio->node.children);
 927
 928         return ns;
 929 }
 930
 931 static int init_root_tree_recursive(int max_ft_level, struct init_tree_node *init_node,
 932                                     struct fs_node *fs_parent_node,
 933                                     struct init_tree_node *init_parent_node,
 934                                     int index)
 935 {
 936         struct mlx5_flow_namespace *fs_ns;
 937         struct fs_prio *fs_prio;
 938         struct fs_node *base;
 939         int i;
 940         int err;
 941
 942         if (init_node->type == FS_TYPE_PRIO) {
 943                 if (init_node->min_ft_level > max_ft_level)
 944                         return -ENOTSUPP;
 945
 946                 fs_get_obj(fs_ns, fs_parent_node);
 947                 fs_prio = fs_create_prio(fs_ns, index, init_node->max_ft,
 948                                          init_node->start_level);
 949                 if (IS_ERR(fs_prio))
 950                         return PTR_ERR(fs_prio);
 951                 base = &fs_prio->node;
 952         } else if (init_node->type == FS_TYPE_NAMESPACE) {
 953                 fs_get_obj(fs_prio, fs_parent_node);
 954                 fs_ns = fs_create_namespace(fs_prio);
 955                 if (IS_ERR(fs_ns))
 956                         return PTR_ERR(fs_ns);
 957                 base = &fs_ns->node;
 958         } else {
 959                 return -EINVAL;
 960         }
 961         for (i = 0; i < init_node->ar_size; i++) {
 962                 err = init_root_tree_recursive(max_ft_level,
 963                                                &init_node->children[i], base,
 964                                                init_node, i);
 965                 if (err)
 966                         return err;
 967         }
 968
 969         return 0;
 970 }
 971
 972 static int init_root_tree(int max_ft_level, struct init_tree_node *init_node,
 973                           struct fs_node *fs_parent_node)
 974 {
 975         int i;
 976         struct mlx5_flow_namespace *fs_ns;
 977         int err;
 978
 979         fs_get_obj(fs_ns, fs_parent_node);
 980         for (i = 0; i < init_node->ar_size; i++) {
 981                 err = init_root_tree_recursive(max_ft_level,
 982                                                &init_node->children[i],
 983                                                &fs_ns->node,
 984                                                init_node, i);
 985                 if (err)
 986                         return err;
 987         }
 988         return 0;
 989 }
 990
 991 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_core_dev *dev,
 992                                                        enum fs_flow_table_type
 993                                                        table_type)
 994 {
 995         struct mlx5_flow_root_namespace *root_ns;
 996         struct mlx5_flow_namespace *ns;
 997
 998         /* Create the root namespace */
 999         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1000         if (!root_ns)
1001                 return NULL;
1002
1003         root_ns->dev = dev;
1004         root_ns->table_type = table_type;
1005
1006         ns = &root_ns->ns;
1007         fs_init_namespace(ns);
1008         tree_init_node(&ns->node, 1, NULL);
1009         tree_add_node(&ns->node, NULL);
1010
1011         return root_ns;
1012 }
1013
1014 static int init_root_ns(struct mlx5_core_dev *dev)
1015 {
1016         int max_ft_level = MLX5_CAP_FLOWTABLE(dev,
1017                                               flow_table_properties_nic_receive.
1018                                               max_ft_level);
1019
1020         dev->priv.root_ns = create_root_ns(dev, FS_FT_NIC_RX);
1021         if (IS_ERR_OR_NULL(dev->priv.root_ns))
1022                 goto cleanup;
1023
1024         if (init_root_tree(max_ft_level, &root_fs, &dev->priv.root_ns->ns.node))
1025                 goto cleanup;
1026
1027         return 0;
1028
1029 cleanup:
1030         mlx5_cleanup_fs(dev);
1031         return -ENOMEM;
1032 }
1033
1034 static void cleanup_single_prio_root_ns(struct mlx5_core_dev *dev,
1035                                         struct mlx5_flow_root_namespace *root_ns)
1036 {
1037         struct fs_node *prio;
1038
1039         if (!root_ns)
1040                 return;
1041
1042         if (!list_empty(&root_ns->ns.node.children)) {
1043                 prio = list_first_entry(&root_ns->ns.node.children,
1044                                         struct fs_node,
1045                                  list);
1046                 if (tree_remove_node(prio))
1047                         mlx5_core_warn(dev,
1048                                        "Flow steering priority wasn't destroyed, refcount > 1\n");
1049         }
1050         if (tree_remove_node(&root_ns->ns.node))
1051                 mlx5_core_warn(dev,
1052                                "Flow steering namespace wasn't destroyed, refcount > 1\n");
1053         root_ns = NULL;
1054 }
1055
1056 static void cleanup_root_ns(struct mlx5_core_dev *dev)
1057 {
1058         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1059         struct fs_prio *iter_prio;
1060
1061         if (!MLX5_CAP_GEN(dev, nic_flow_table))
1062                 return;
1063
1064         if (!root_ns)
1065                 return;
1066
1067         /* stage 1 */
1068         fs_for_each_prio(iter_prio, &root_ns->ns) {
1069                 struct fs_node *node;
1070                 struct mlx5_flow_namespace *iter_ns;
1071
1072                 fs_for_each_ns_or_ft(node, iter_prio) {
1073                         if (node->type == FS_TYPE_FLOW_TABLE)
1074                                 continue;
1075                         fs_get_obj(iter_ns, node);
1076                         while (!list_empty(&iter_ns->node.children)) {
1077                                 struct fs_prio *obj_iter_prio2;
1078                                 struct fs_node *iter_prio2 =
1079                                         list_first_entry(&iter_ns->node.children,
1080                                                          struct fs_node,
1081                                                          list);
1082
1083                                 fs_get_obj(obj_iter_prio2, iter_prio2);
1084                                 if (tree_remove_node(iter_prio2)) {
1085                                         mlx5_core_warn(dev,
1086                                                        "Priority %d wasn't destroyed, refcount > 1\n",
1087                                                        obj_iter_prio2->prio);
1088                                         return;
1089                                 }
1090                         }
1091                 }
1092         }
1093
1094         /* stage 2 */
1095         fs_for_each_prio(iter_prio, &root_ns->ns) {
1096                 while (!list_empty(&iter_prio->node.children)) {
1097                         struct fs_node *iter_ns =
1098                                 list_first_entry(&iter_prio->node.children,
1099                                                  struct fs_node,
1100                                                  list);
1101                         if (tree_remove_node(iter_ns)) {
1102                                 mlx5_core_warn(dev,
1103                                                "Namespace wasn't destroyed, refcount > 1\n");
1104                                 return;
1105                         }
1106                 }
1107         }
1108
1109         /* stage 3 */
1110         while (!list_empty(&root_ns->ns.node.children)) {
1111                 struct fs_prio *obj_prio_node;
1112                 struct fs_node *prio_node =
1113                         list_first_entry(&root_ns->ns.node.children,
1114                                          struct fs_node,
1115                                          list);
1116
1117                 fs_get_obj(obj_prio_node, prio_node);
1118                 if (tree_remove_node(prio_node)) {
1119                         mlx5_core_warn(dev,
1120                                        "Priority %d wasn't destroyed, refcount > 1\n",
1121                                        obj_prio_node->prio);
1122                         return;
1123                 }
1124         }
1125
1126         if (tree_remove_node(&root_ns->ns.node)) {
1127                 mlx5_core_warn(dev,
1128                                "root namespace wasn't destroyed, refcount > 1\n");
1129                 return;
1130         }
1131
1132         dev->priv.root_ns = NULL;
1133 }
1134
1135 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1136 {
1137         cleanup_root_ns(dev);
1138         cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1139 }
1140
1141 static int init_fdb_root_ns(struct mlx5_core_dev *dev)
1142 {
1143         struct fs_prio *prio;
1144
1145         dev->priv.fdb_root_ns = create_root_ns(dev, FS_FT_FDB);
1146         if (!dev->priv.fdb_root_ns)
1147                 return -ENOMEM;
1148
1149         /* Create single prio */
1150         prio = fs_create_prio(&dev->priv.fdb_root_ns->ns, 0, 1, 0);
1151         if (IS_ERR(prio)) {
1152                 cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1153                 return PTR_ERR(prio);
1154         } else {
1155                 return 0;
1156         }
1157 }
1158
1159 int mlx5_init_fs(struct mlx5_core_dev *dev)
1160 {
1161         int err = 0;
1162
1163         if (MLX5_CAP_GEN(dev, nic_flow_table)) {
1164                 err = init_root_ns(dev);
1165                 if (err)
1166                         return err;
1167         }
1168         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1169                 err = init_fdb_root_ns(dev);
1170                 if (err)
1171                         cleanup_root_ns(dev);
1172         }
1173
1174         return err;
1175 }