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