drivers/net/ethernet/mellanox/mlx5/core/fs_core.c

   1 /*
   2  * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #include <linux/mutex.h>
  34 #include <linux/mlx5/driver.h>
  35
  36 #include "mlx5_core.h"
  37 #include "fs_core.h"
  38 #include "fs_cmd.h"
  39
  40 #define INIT_TREE_NODE_ARRAY_SIZE(...)  (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
  41                                          sizeof(struct init_tree_node))
  42
  43 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
  44                  ...) {.type = FS_TYPE_PRIO,\
  45         .min_ft_level = min_level_val,\
  46         .num_levels = num_levels_val,\
  47         .num_leaf_prios = num_prios_val,\
  48         .caps = caps_val,\
  49         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  50         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  51 }
  52
  53 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
  54         ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
  55                  __VA_ARGS__)\
  56
  57 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
  58         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  59         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  60 }
  61
  62 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  63                                    sizeof(long))
  64
  65 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  66
  67 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  68                                .caps = (long[]) {__VA_ARGS__} }
  69
  70 #define LEFTOVERS_NUM_LEVELS 1
  71 #define LEFTOVERS_NUM_PRIOS 1
  72
  73 #define BY_PASS_PRIO_NUM_LEVELS 1
  74 #define BY_PASS_MIN_LEVEL (KERNEL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  75                            LEFTOVERS_NUM_PRIOS)
  76
  77 /* Vlan, mac, ttc, aRFS */
  78 #define KERNEL_NIC_PRIO_NUM_LEVELS 4
  79 #define KERNEL_NIC_NUM_PRIOS 1
  80 /* One more level for tc */
  81 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
  82
  83 #define ANCHOR_NUM_LEVELS 1
  84 #define ANCHOR_NUM_PRIOS 1
  85 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
  86 struct node_caps {
  87         size_t  arr_sz;
  88         long    *caps;
  89 };
  90 static struct init_tree_node {
  91         enum fs_node_type       type;
  92         struct init_tree_node *children;
  93         int ar_size;
  94         struct node_caps caps;
  95         int min_ft_level;
  96         int num_leaf_prios;
  97         int prio;
  98         int num_levels;
  99 } root_fs = {
 100         .type = FS_TYPE_NAMESPACE,
 101         .ar_size = 4,
 102         .children = (struct init_tree_node[]) {
 103                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 104                          FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en),
 105                                           FS_CAP(flow_table_properties_nic_receive.modify_root),
 106                                           FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode),
 107                                           FS_CAP(flow_table_properties_nic_receive.flow_table_modify)),
 108                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 109                                                   BY_PASS_PRIO_NUM_LEVELS))),
 110                 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 111                          ADD_NS(ADD_MULTIPLE_PRIO(1, 1),
 112                                 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 113                                                   KERNEL_NIC_PRIO_NUM_LEVELS))),
 114                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 115                          FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en),
 116                                           FS_CAP(flow_table_properties_nic_receive.modify_root),
 117                                           FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode),
 118                                           FS_CAP(flow_table_properties_nic_receive.flow_table_modify)),
 119                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
 120                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 121                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
 122         }
 123 };
 124
 125 enum fs_i_mutex_lock_class {
 126         FS_MUTEX_GRANDPARENT,
 127         FS_MUTEX_PARENT,
 128         FS_MUTEX_CHILD
 129 };
 130
 131 static void del_rule(struct fs_node *node);
 132 static void del_flow_table(struct fs_node *node);
 133 static void del_flow_group(struct fs_node *node);
 134 static void del_fte(struct fs_node *node);
 135
 136 static void tree_init_node(struct fs_node *node,
 137                            unsigned int refcount,
 138                            void (*remove_func)(struct fs_node *))
 139 {
 140         atomic_set(&node->refcount, refcount);
 141         INIT_LIST_HEAD(&node->list);
 142         INIT_LIST_HEAD(&node->children);
 143         mutex_init(&node->lock);
 144         node->remove_func = remove_func;
 145 }
 146
 147 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 148 {
 149         if (parent)
 150                 atomic_inc(&parent->refcount);
 151         node->parent = parent;
 152
 153         /* Parent is the root */
 154         if (!parent)
 155                 node->root = node;
 156         else
 157                 node->root = parent->root;
 158 }
 159
 160 static void tree_get_node(struct fs_node *node)
 161 {
 162         atomic_inc(&node->refcount);
 163 }
 164
 165 static void nested_lock_ref_node(struct fs_node *node,
 166                                  enum fs_i_mutex_lock_class class)
 167 {
 168         if (node) {
 169                 mutex_lock_nested(&node->lock, class);
 170                 atomic_inc(&node->refcount);
 171         }
 172 }
 173
 174 static void lock_ref_node(struct fs_node *node)
 175 {
 176         if (node) {
 177                 mutex_lock(&node->lock);
 178                 atomic_inc(&node->refcount);
 179         }
 180 }
 181
 182 static void unlock_ref_node(struct fs_node *node)
 183 {
 184         if (node) {
 185                 atomic_dec(&node->refcount);
 186                 mutex_unlock(&node->lock);
 187         }
 188 }
 189
 190 static void tree_put_node(struct fs_node *node)
 191 {
 192         struct fs_node *parent_node = node->parent;
 193
 194         lock_ref_node(parent_node);
 195         if (atomic_dec_and_test(&node->refcount)) {
 196                 if (parent_node)
 197                         list_del_init(&node->list);
 198                 if (node->remove_func)
 199                         node->remove_func(node);
 200                 kfree(node);
 201                 node = NULL;
 202         }
 203         unlock_ref_node(parent_node);
 204         if (!node && parent_node)
 205                 tree_put_node(parent_node);
 206 }
 207
 208 static int tree_remove_node(struct fs_node *node)
 209 {
 210         if (atomic_read(&node->refcount) > 1) {
 211                 atomic_dec(&node->refcount);
 212                 return -EEXIST;
 213         }
 214         tree_put_node(node);
 215         return 0;
 216 }
 217
 218 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 219                                  unsigned int prio)
 220 {
 221         struct fs_prio *iter_prio;
 222
 223         fs_for_each_prio(iter_prio, ns) {
 224                 if (iter_prio->prio == prio)
 225                         return iter_prio;
 226         }
 227
 228         return NULL;
 229 }
 230
 231 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 232 {
 233         unsigned int i;
 234
 235         for (i = 0; i < size; i++, mask++, val1++, val2++)
 236                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 237                     ((*(u8 *)val2) & (*(u8 *)mask)))
 238                         return false;
 239
 240         return true;
 241 }
 242
 243 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 244                                 void *fte_param1, void *fte_param2)
 245 {
 246         if (mask->match_criteria_enable &
 247             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 248                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 249                                                 fte_param1, outer_headers);
 250                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 251                                                 fte_param2, outer_headers);
 252                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 253                                               mask->match_criteria, outer_headers);
 254
 255                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 256                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 257                         return false;
 258         }
 259
 260         if (mask->match_criteria_enable &
 261             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 262                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 263                                                 fte_param1, misc_parameters);
 264                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 265                                                 fte_param2, misc_parameters);
 266                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 267                                           mask->match_criteria, misc_parameters);
 268
 269                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 270                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 271                         return false;
 272         }
 273
 274         if (mask->match_criteria_enable &
 275             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 276                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 277                                                 fte_param1, inner_headers);
 278                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 279                                                 fte_param2, inner_headers);
 280                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 281                                           mask->match_criteria, inner_headers);
 282
 283                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 284                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 285                         return false;
 286         }
 287         return true;
 288 }
 289
 290 static bool compare_match_criteria(u8 match_criteria_enable1,
 291                                    u8 match_criteria_enable2,
 292                                    void *mask1, void *mask2)
 293 {
 294         return match_criteria_enable1 == match_criteria_enable2 &&
 295                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 296 }
 297
 298 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 299 {
 300         struct fs_node *root;
 301         struct mlx5_flow_namespace *ns;
 302
 303         root = node->root;
 304
 305         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 306                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 307                 return NULL;
 308         }
 309
 310         ns = container_of(root, struct mlx5_flow_namespace, node);
 311         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 312 }
 313
 314 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 315 {
 316         struct mlx5_flow_root_namespace *root = find_root(node);
 317
 318         if (root)
 319                 return root->dev;
 320         return NULL;
 321 }
 322
 323 static void del_flow_table(struct fs_node *node)
 324 {
 325         struct mlx5_flow_table *ft;
 326         struct mlx5_core_dev *dev;
 327         struct fs_prio *prio;
 328         int err;
 329
 330         fs_get_obj(ft, node);
 331         dev = get_dev(&ft->node);
 332
 333         err = mlx5_cmd_destroy_flow_table(dev, ft);
 334         if (err)
 335                 pr_warn("flow steering can't destroy ft\n");
 336         fs_get_obj(prio, ft->node.parent);
 337         prio->num_ft--;
 338 }
 339
 340 static void del_rule(struct fs_node *node)
 341 {
 342         struct mlx5_flow_rule *rule;
 343         struct mlx5_flow_table *ft;
 344         struct mlx5_flow_group *fg;
 345         struct fs_fte *fte;
 346         u32     *match_value;
 347         struct mlx5_core_dev *dev = get_dev(node);
 348         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 349         int err;
 350
 351         match_value = mlx5_vzalloc(match_len);
 352         if (!match_value) {
 353                 pr_warn("failed to allocate inbox\n");
 354                 return;
 355         }
 356
 357         fs_get_obj(rule, node);
 358         fs_get_obj(fte, rule->node.parent);
 359         fs_get_obj(fg, fte->node.parent);
 360         memcpy(match_value, fte->val, sizeof(fte->val));
 361         fs_get_obj(ft, fg->node.parent);
 362         list_del(&rule->node.list);
 363         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 364                 mutex_lock(&rule->dest_attr.ft->lock);
 365                 list_del(&rule->next_ft);
 366                 mutex_unlock(&rule->dest_attr.ft->lock);
 367         }
 368         if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 369             --fte->dests_size) {
 370                 err = mlx5_cmd_update_fte(dev, ft,
 371                                           fg->id, fte);
 372                 if (err)
 373                         pr_warn("%s can't del rule fg id=%d fte_index=%d\n",
 374                                 __func__, fg->id, fte->index);
 375         }
 376         kvfree(match_value);
 377 }
 378
 379 static void del_fte(struct fs_node *node)
 380 {
 381         struct mlx5_flow_table *ft;
 382         struct mlx5_flow_group *fg;
 383         struct mlx5_core_dev *dev;
 384         struct fs_fte *fte;
 385         int err;
 386
 387         fs_get_obj(fte, node);
 388         fs_get_obj(fg, fte->node.parent);
 389         fs_get_obj(ft, fg->node.parent);
 390
 391         dev = get_dev(&ft->node);
 392         err = mlx5_cmd_delete_fte(dev, ft,
 393                                   fte->index);
 394         if (err)
 395                 pr_warn("flow steering can't delete fte in index %d of flow group id %d\n",
 396                         fte->index, fg->id);
 397
 398         fte->status = 0;
 399         fg->num_ftes--;
 400 }
 401
 402 static void del_flow_group(struct fs_node *node)
 403 {
 404         struct mlx5_flow_group *fg;
 405         struct mlx5_flow_table *ft;
 406         struct mlx5_core_dev *dev;
 407
 408         fs_get_obj(fg, node);
 409         fs_get_obj(ft, fg->node.parent);
 410         dev = get_dev(&ft->node);
 411
 412         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 413                 pr_warn("flow steering can't destroy fg %d of ft %d\n",
 414                         fg->id, ft->id);
 415 }
 416
 417 static struct fs_fte *alloc_fte(u8 action,
 418                                 u32 flow_tag,
 419                                 u32 *match_value,
 420                                 unsigned int index)
 421 {
 422         struct fs_fte *fte;
 423
 424         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 425         if (!fte)
 426                 return ERR_PTR(-ENOMEM);
 427
 428         memcpy(fte->val, match_value, sizeof(fte->val));
 429         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 430         fte->flow_tag = flow_tag;
 431         fte->index = index;
 432         fte->action = action;
 433
 434         return fte;
 435 }
 436
 437 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 438 {
 439         struct mlx5_flow_group *fg;
 440         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 441                                             create_fg_in, match_criteria);
 442         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 443                                             create_fg_in,
 444                                             match_criteria_enable);
 445         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 446         if (!fg)
 447                 return ERR_PTR(-ENOMEM);
 448
 449         fg->mask.match_criteria_enable = match_criteria_enable;
 450         memcpy(&fg->mask.match_criteria, match_criteria,
 451                sizeof(fg->mask.match_criteria));
 452         fg->node.type =  FS_TYPE_FLOW_GROUP;
 453         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 454                                    start_flow_index);
 455         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 456                                 end_flow_index) - fg->start_index + 1;
 457         return fg;
 458 }
 459
 460 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 461                                                 enum fs_flow_table_type table_type)
 462 {
 463         struct mlx5_flow_table *ft;
 464
 465         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 466         if (!ft)
 467                 return NULL;
 468
 469         ft->level = level;
 470         ft->node.type = FS_TYPE_FLOW_TABLE;
 471         ft->type = table_type;
 472         ft->vport = vport;
 473         ft->max_fte = max_fte;
 474         INIT_LIST_HEAD(&ft->fwd_rules);
 475         mutex_init(&ft->lock);
 476
 477         return ft;
 478 }
 479
 480 /* If reverse is false, then we search for the first flow table in the
 481  * root sub-tree from start(closest from right), else we search for the
 482  * last flow table in the root sub-tree till start(closest from left).
 483  */
 484 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 485                                                          struct list_head *start,
 486                                                          bool reverse)
 487 {
 488 #define list_advance_entry(pos, reverse)                \
 489         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 490
 491 #define list_for_each_advance_continue(pos, head, reverse)      \
 492         for (pos = list_advance_entry(pos, reverse);            \
 493              &pos->list != (head);                              \
 494              pos = list_advance_entry(pos, reverse))
 495
 496         struct fs_node *iter = list_entry(start, struct fs_node, list);
 497         struct mlx5_flow_table *ft = NULL;
 498
 499         if (!root)
 500                 return NULL;
 501
 502         list_for_each_advance_continue(iter, &root->children, reverse) {
 503                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 504                         fs_get_obj(ft, iter);
 505                         return ft;
 506                 }
 507                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 508                 if (ft)
 509                         return ft;
 510         }
 511
 512         return ft;
 513 }
 514
 515 /* If reverse if false then return the first flow table in next priority of
 516  * prio in the tree, else return the last flow table in the previous priority
 517  * of prio in the tree.
 518  */
 519 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 520 {
 521         struct mlx5_flow_table *ft = NULL;
 522         struct fs_node *curr_node;
 523         struct fs_node *parent;
 524
 525         parent = prio->node.parent;
 526         curr_node = &prio->node;
 527         while (!ft && parent) {
 528                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 529                 curr_node = parent;
 530                 parent = curr_node->parent;
 531         }
 532         return ft;
 533 }
 534
 535 /* Assuming all the tree is locked by mutex chain lock */
 536 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 537 {
 538         return find_closest_ft(prio, false);
 539 }
 540
 541 /* Assuming all the tree is locked by mutex chain lock */
 542 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 543 {
 544         return find_closest_ft(prio, true);
 545 }
 546
 547 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 548                                struct fs_prio *prio,
 549                                struct mlx5_flow_table *ft)
 550 {
 551         struct mlx5_flow_table *iter;
 552         int i = 0;
 553         int err;
 554
 555         fs_for_each_ft(iter, prio) {
 556                 i++;
 557                 err = mlx5_cmd_modify_flow_table(dev,
 558                                                  iter,
 559                                                  ft);
 560                 if (err) {
 561                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 562                                        iter->id);
 563                         /* The driver is out of sync with the FW */
 564                         if (i > 1)
 565                                 WARN_ON(true);
 566                         return err;
 567                 }
 568         }
 569         return 0;
 570 }
 571
 572 /* Connect flow tables from previous priority of prio to ft */
 573 static int connect_prev_fts(struct mlx5_core_dev *dev,
 574                             struct mlx5_flow_table *ft,
 575                             struct fs_prio *prio)
 576 {
 577         struct mlx5_flow_table *prev_ft;
 578
 579         prev_ft = find_prev_chained_ft(prio);
 580         if (prev_ft) {
 581                 struct fs_prio *prev_prio;
 582
 583                 fs_get_obj(prev_prio, prev_ft->node.parent);
 584                 return connect_fts_in_prio(dev, prev_prio, ft);
 585         }
 586         return 0;
 587 }
 588
 589 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 590                                  *prio)
 591 {
 592         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 593         int min_level = INT_MAX;
 594         int err;
 595
 596         if (root->root_ft)
 597                 min_level = root->root_ft->level;
 598
 599         if (ft->level >= min_level)
 600                 return 0;
 601
 602         err = mlx5_cmd_update_root_ft(root->dev, ft);
 603         if (err)
 604                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 605                                ft->id);
 606         else
 607                 root->root_ft = ft;
 608
 609         return err;
 610 }
 611
 612 int mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 613                                  struct mlx5_flow_destination *dest)
 614 {
 615         struct mlx5_flow_table *ft;
 616         struct mlx5_flow_group *fg;
 617         struct fs_fte *fte;
 618         int err = 0;
 619
 620         fs_get_obj(fte, rule->node.parent);
 621         if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 622                 return -EINVAL;
 623         lock_ref_node(&fte->node);
 624         fs_get_obj(fg, fte->node.parent);
 625         fs_get_obj(ft, fg->node.parent);
 626
 627         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 628         err = mlx5_cmd_update_fte(get_dev(&ft->node),
 629                                   ft, fg->id, fte);
 630         unlock_ref_node(&fte->node);
 631
 632         return err;
 633 }
 634
 635 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 636 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 637                              struct mlx5_flow_table *new_next_ft,
 638                              struct mlx5_flow_table *old_next_ft)
 639 {
 640         struct mlx5_flow_destination dest;
 641         struct mlx5_flow_rule *iter;
 642         int err = 0;
 643
 644         /* new_next_ft and old_next_ft could be NULL only
 645          * when we create/destroy the anchor flow table.
 646          */
 647         if (!new_next_ft || !old_next_ft)
 648                 return 0;
 649
 650         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 651         dest.ft = new_next_ft;
 652
 653         mutex_lock(&old_next_ft->lock);
 654         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 655         mutex_unlock(&old_next_ft->lock);
 656         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 657                 err = mlx5_modify_rule_destination(iter, &dest);
 658                 if (err)
 659                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 660                                new_next_ft->id);
 661         }
 662         return 0;
 663 }
 664
 665 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 666                               struct fs_prio *prio)
 667 {
 668         struct mlx5_flow_table *next_ft;
 669         int err = 0;
 670
 671         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 672
 673         if (list_empty(&prio->node.children)) {
 674                 err = connect_prev_fts(dev, ft, prio);
 675                 if (err)
 676                         return err;
 677
 678                 next_ft = find_next_chained_ft(prio);
 679                 err = connect_fwd_rules(dev, ft, next_ft);
 680                 if (err)
 681                         return err;
 682         }
 683
 684         if (MLX5_CAP_FLOWTABLE(dev,
 685                                flow_table_properties_nic_receive.modify_root))
 686                 err = update_root_ft_create(ft, prio);
 687         return err;
 688 }
 689
 690 static void list_add_flow_table(struct mlx5_flow_table *ft,
 691                                 struct fs_prio *prio)
 692 {
 693         struct list_head *prev = &prio->node.children;
 694         struct mlx5_flow_table *iter;
 695
 696         fs_for_each_ft(iter, prio) {
 697                 if (iter->level > ft->level)
 698                         break;
 699                 prev = &iter->node.list;
 700         }
 701         list_add(&ft->node.list, prev);
 702 }
 703
 704 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 705                                                         u16 vport, int prio,
 706                                                         int max_fte, u32 level)
 707 {
 708         struct mlx5_flow_table *next_ft = NULL;
 709         struct mlx5_flow_table *ft;
 710         int err;
 711         int log_table_sz;
 712         struct mlx5_flow_root_namespace *root =
 713                 find_root(&ns->node);
 714         struct fs_prio *fs_prio = NULL;
 715
 716         if (!root) {
 717                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 718                 return ERR_PTR(-ENODEV);
 719         }
 720
 721         mutex_lock(&root->chain_lock);
 722         fs_prio = find_prio(ns, prio);
 723         if (!fs_prio) {
 724                 err = -EINVAL;
 725                 goto unlock_root;
 726         }
 727         if (level >= fs_prio->num_levels) {
 728                 err = -ENOSPC;
 729                 goto unlock_root;
 730         }
 731         /* The level is related to the
 732          * priority level range.
 733          */
 734         level += fs_prio->start_level;
 735         ft = alloc_flow_table(level,
 736                               vport,
 737                               roundup_pow_of_two(max_fte),
 738                               root->table_type);
 739         if (!ft) {
 740                 err = -ENOMEM;
 741                 goto unlock_root;
 742         }
 743
 744         tree_init_node(&ft->node, 1, del_flow_table);
 745         log_table_sz = ilog2(ft->max_fte);
 746         next_ft = find_next_chained_ft(fs_prio);
 747         err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->type, ft->level,
 748                                          log_table_sz, next_ft, &ft->id);
 749         if (err)
 750                 goto free_ft;
 751
 752         err = connect_flow_table(root->dev, ft, fs_prio);
 753         if (err)
 754                 goto destroy_ft;
 755         lock_ref_node(&fs_prio->node);
 756         tree_add_node(&ft->node, &fs_prio->node);
 757         list_add_flow_table(ft, fs_prio);
 758         fs_prio->num_ft++;
 759         unlock_ref_node(&fs_prio->node);
 760         mutex_unlock(&root->chain_lock);
 761         return ft;
 762 destroy_ft:
 763         mlx5_cmd_destroy_flow_table(root->dev, ft);
 764 free_ft:
 765         kfree(ft);
 766 unlock_root:
 767         mutex_unlock(&root->chain_lock);
 768         return ERR_PTR(err);
 769 }
 770
 771 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 772                                                int prio, int max_fte,
 773                                                u32 level)
 774 {
 775         return __mlx5_create_flow_table(ns, 0, prio, max_fte, level);
 776 }
 777
 778 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
 779                                                      int prio, int max_fte,
 780                                                      u32 level, u16 vport)
 781 {
 782         return __mlx5_create_flow_table(ns, vport, prio, max_fte, level);
 783 }
 784
 785 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 786                                                             int prio,
 787                                                             int num_flow_table_entries,
 788                                                             int max_num_groups,
 789                                                             u32 level)
 790 {
 791         struct mlx5_flow_table *ft;
 792
 793         if (max_num_groups > num_flow_table_entries)
 794                 return ERR_PTR(-EINVAL);
 795
 796         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries, level);
 797         if (IS_ERR(ft))
 798                 return ft;
 799
 800         ft->autogroup.active = true;
 801         ft->autogroup.required_groups = max_num_groups;
 802
 803         return ft;
 804 }
 805 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 806
 807 /* Flow table should be locked */
 808 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 809                                                         u32 *fg_in,
 810                                                         struct list_head
 811                                                         *prev_fg,
 812                                                         bool is_auto_fg)
 813 {
 814         struct mlx5_flow_group *fg;
 815         struct mlx5_core_dev *dev = get_dev(&ft->node);
 816         int err;
 817
 818         if (!dev)
 819                 return ERR_PTR(-ENODEV);
 820
 821         fg = alloc_flow_group(fg_in);
 822         if (IS_ERR(fg))
 823                 return fg;
 824
 825         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 826         if (err) {
 827                 kfree(fg);
 828                 return ERR_PTR(err);
 829         }
 830
 831         if (ft->autogroup.active)
 832                 ft->autogroup.num_groups++;
 833         /* Add node to tree */
 834         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 835         tree_add_node(&fg->node, &ft->node);
 836         /* Add node to group list */
 837         list_add(&fg->node.list, ft->node.children.prev);
 838
 839         return fg;
 840 }
 841
 842 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 843                                                u32 *fg_in)
 844 {
 845         struct mlx5_flow_group *fg;
 846
 847         if (ft->autogroup.active)
 848                 return ERR_PTR(-EPERM);
 849
 850         lock_ref_node(&ft->node);
 851         fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
 852         unlock_ref_node(&ft->node);
 853
 854         return fg;
 855 }
 856
 857 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 858 {
 859         struct mlx5_flow_rule *rule;
 860
 861         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 862         if (!rule)
 863                 return NULL;
 864
 865         INIT_LIST_HEAD(&rule->next_ft);
 866         rule->node.type = FS_TYPE_FLOW_DEST;
 867         if (dest)
 868                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 869
 870         return rule;
 871 }
 872
 873 /* fte should not be deleted while calling this function */
 874 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
 875                                            struct mlx5_flow_group *fg,
 876                                            struct mlx5_flow_destination *dest)
 877 {
 878         struct mlx5_flow_table *ft;
 879         struct mlx5_flow_rule *rule;
 880         int err;
 881
 882         rule = alloc_rule(dest);
 883         if (!rule)
 884                 return ERR_PTR(-ENOMEM);
 885
 886         fs_get_obj(ft, fg->node.parent);
 887         /* Add dest to dests list- we need flow tables to be in the
 888          * end of the list for forward to next prio rules.
 889          */
 890         tree_init_node(&rule->node, 1, del_rule);
 891         if (dest && dest->type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
 892                 list_add(&rule->node.list, &fte->node.children);
 893         else
 894                 list_add_tail(&rule->node.list, &fte->node.children);
 895         if (dest)
 896                 fte->dests_size++;
 897         if (fte->dests_size == 1 || !dest)
 898                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 899                                           ft, fg->id, fte);
 900         else
 901                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 902                                           ft, fg->id, fte);
 903         if (err)
 904                 goto free_rule;
 905
 906         fte->status |= FS_FTE_STATUS_EXISTING;
 907
 908         return rule;
 909
 910 free_rule:
 911         list_del(&rule->node.list);
 912         kfree(rule);
 913         if (dest)
 914                 fte->dests_size--;
 915         return ERR_PTR(err);
 916 }
 917
 918 /* Assumed fg is locked */
 919 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 920                                        struct list_head **prev)
 921 {
 922         struct fs_fte *fte;
 923         unsigned int start = fg->start_index;
 924
 925         if (prev)
 926                 *prev = &fg->node.children;
 927
 928         /* assumed list is sorted by index */
 929         fs_for_each_fte(fte, fg) {
 930                 if (fte->index != start)
 931                         return start;
 932                 start++;
 933                 if (prev)
 934                         *prev = &fte->node.list;
 935         }
 936
 937         return start;
 938 }
 939
 940 /* prev is output, prev->next = new_fte */
 941 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
 942                                  u32 *match_value,
 943                                  u8 action,
 944                                  u32 flow_tag,
 945                                  struct list_head **prev)
 946 {
 947         struct fs_fte *fte;
 948         int index;
 949
 950         index = get_free_fte_index(fg, prev);
 951         fte = alloc_fte(action, flow_tag, match_value, index);
 952         if (IS_ERR(fte))
 953                 return fte;
 954
 955         return fte;
 956 }
 957
 958 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
 959                                                 u8 match_criteria_enable,
 960                                                 u32 *match_criteria)
 961 {
 962         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 963         struct list_head *prev = &ft->node.children;
 964         unsigned int candidate_index = 0;
 965         struct mlx5_flow_group *fg;
 966         void *match_criteria_addr;
 967         unsigned int group_size = 0;
 968         u32 *in;
 969
 970         if (!ft->autogroup.active)
 971                 return ERR_PTR(-ENOENT);
 972
 973         in = mlx5_vzalloc(inlen);
 974         if (!in)
 975                 return ERR_PTR(-ENOMEM);
 976
 977         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
 978                 /* We save place for flow groups in addition to max types */
 979                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
 980
 981         /*  ft->max_fte == ft->autogroup.max_types */
 982         if (group_size == 0)
 983                 group_size = 1;
 984
 985         /* sorted by start_index */
 986         fs_for_each_fg(fg, ft) {
 987                 if (candidate_index + group_size > fg->start_index)
 988                         candidate_index = fg->start_index + fg->max_ftes;
 989                 else
 990                         break;
 991                 prev = &fg->node.list;
 992         }
 993
 994         if (candidate_index + group_size > ft->max_fte) {
 995                 fg = ERR_PTR(-ENOSPC);
 996                 goto out;
 997         }
 998
 999         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1000                  match_criteria_enable);
1001         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1002         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
1003                  group_size - 1);
1004         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1005                                            in, match_criteria);
1006         memcpy(match_criteria_addr, match_criteria,
1007                MLX5_ST_SZ_BYTES(fte_match_param));
1008
1009         fg = create_flow_group_common(ft, in, prev, true);
1010 out:
1011         kvfree(in);
1012         return fg;
1013 }
1014
1015 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1016                                              struct mlx5_flow_destination *dest)
1017 {
1018         struct mlx5_flow_rule *rule;
1019
1020         list_for_each_entry(rule, &fte->node.children, node.list) {
1021                 if (rule->dest_attr.type == dest->type) {
1022                         if ((dest->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1023                              dest->vport_num == rule->dest_attr.vport_num) ||
1024                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1025                              dest->ft == rule->dest_attr.ft) ||
1026                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1027                              dest->tir_num == rule->dest_attr.tir_num))
1028                                 return rule;
1029                 }
1030         }
1031         return NULL;
1032 }
1033
1034 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
1035                                           u32 *match_value,
1036                                           u8 action,
1037                                           u32 flow_tag,
1038                                           struct mlx5_flow_destination *dest)
1039 {
1040         struct fs_fte *fte;
1041         struct mlx5_flow_rule *rule;
1042         struct mlx5_flow_table *ft;
1043         struct list_head *prev;
1044
1045         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1046         fs_for_each_fte(fte, fg) {
1047                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1048                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1049                     action == fte->action && flow_tag == fte->flow_tag) {
1050                         rule = find_flow_rule(fte, dest);
1051                         if (rule) {
1052                                 atomic_inc(&rule->node.refcount);
1053                                 unlock_ref_node(&fte->node);
1054                                 unlock_ref_node(&fg->node);
1055                                 return rule;
1056                         }
1057                         rule = add_rule_fte(fte, fg, dest);
1058                         unlock_ref_node(&fte->node);
1059                         if (IS_ERR(rule))
1060                                 goto unlock_fg;
1061                         else
1062                                 goto add_rule;
1063                 }
1064                 unlock_ref_node(&fte->node);
1065         }
1066         fs_get_obj(ft, fg->node.parent);
1067         if (fg->num_ftes >= fg->max_ftes) {
1068                 rule = ERR_PTR(-ENOSPC);
1069                 goto unlock_fg;
1070         }
1071
1072         fte = create_fte(fg, match_value, action, flow_tag, &prev);
1073         if (IS_ERR(fte)) {
1074                 rule = (void *)fte;
1075                 goto unlock_fg;
1076         }
1077         tree_init_node(&fte->node, 0, del_fte);
1078         rule = add_rule_fte(fte, fg, dest);
1079         if (IS_ERR(rule)) {
1080                 kfree(fte);
1081                 goto unlock_fg;
1082         }
1083
1084         fg->num_ftes++;
1085
1086         tree_add_node(&fte->node, &fg->node);
1087         list_add(&fte->node.list, prev);
1088 add_rule:
1089         tree_add_node(&rule->node, &fte->node);
1090 unlock_fg:
1091         unlock_ref_node(&fg->node);
1092         return rule;
1093 }
1094
1095 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1096                           u32 action,
1097                           struct mlx5_flow_table *ft)
1098 {
1099         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1100                 return true;
1101
1102         if (!dest || ((dest->type ==
1103             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1104             (dest->ft->level <= ft->level)))
1105                 return false;
1106         return true;
1107 }
1108
1109 static struct mlx5_flow_rule *
1110 _mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1111                     u8 match_criteria_enable,
1112                     u32 *match_criteria,
1113                     u32 *match_value,
1114                     u32 action,
1115                     u32 flow_tag,
1116                     struct mlx5_flow_destination *dest)
1117 {
1118         struct mlx5_flow_group *g;
1119         struct mlx5_flow_rule *rule;
1120
1121         if (!dest_is_valid(dest, action, ft))
1122                 return ERR_PTR(-EINVAL);
1123
1124         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1125         fs_for_each_fg(g, ft)
1126                 if (compare_match_criteria(g->mask.match_criteria_enable,
1127                                            match_criteria_enable,
1128                                            g->mask.match_criteria,
1129                                            match_criteria)) {
1130                         rule = add_rule_fg(g, match_value,
1131                                            action, flow_tag, dest);
1132                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1133                                 goto unlock;
1134                 }
1135
1136         g = create_autogroup(ft, match_criteria_enable, match_criteria);
1137         if (IS_ERR(g)) {
1138                 rule = (void *)g;
1139                 goto unlock;
1140         }
1141
1142         rule = add_rule_fg(g, match_value,
1143                            action, flow_tag, dest);
1144         if (IS_ERR(rule)) {
1145                 /* Remove assumes refcount > 0 and autogroup creates a group
1146                  * with a refcount = 0.
1147                  */
1148                 unlock_ref_node(&ft->node);
1149                 tree_get_node(&g->node);
1150                 tree_remove_node(&g->node);
1151                 return rule;
1152         }
1153 unlock:
1154         unlock_ref_node(&ft->node);
1155         return rule;
1156 }
1157
1158 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1159 {
1160         return ((ft->type == FS_FT_NIC_RX) &&
1161                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1162 }
1163
1164 struct mlx5_flow_rule *
1165 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1166                    u8 match_criteria_enable,
1167                    u32 *match_criteria,
1168                    u32 *match_value,
1169                    u32 action,
1170                    u32 flow_tag,
1171                    struct mlx5_flow_destination *dest)
1172 {
1173         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1174         struct mlx5_flow_destination gen_dest;
1175         struct mlx5_flow_table *next_ft = NULL;
1176         struct mlx5_flow_rule *rule = NULL;
1177         u32 sw_action = action;
1178         struct fs_prio *prio;
1179
1180         fs_get_obj(prio, ft->node.parent);
1181         if (action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1182                 if (!fwd_next_prio_supported(ft))
1183                         return ERR_PTR(-EOPNOTSUPP);
1184                 if (dest)
1185                         return ERR_PTR(-EINVAL);
1186                 mutex_lock(&root->chain_lock);
1187                 next_ft = find_next_chained_ft(prio);
1188                 if (next_ft) {
1189                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1190                         gen_dest.ft = next_ft;
1191                         dest = &gen_dest;
1192                         action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1193                 } else {
1194                         mutex_unlock(&root->chain_lock);
1195                         return ERR_PTR(-EOPNOTSUPP);
1196                 }
1197         }
1198
1199         rule =  _mlx5_add_flow_rule(ft, match_criteria_enable, match_criteria,
1200                                     match_value, action, flow_tag, dest);
1201
1202         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1203                 if (!IS_ERR_OR_NULL(rule) &&
1204                     (list_empty(&rule->next_ft))) {
1205                         mutex_lock(&next_ft->lock);
1206                         list_add(&rule->next_ft, &next_ft->fwd_rules);
1207                         mutex_unlock(&next_ft->lock);
1208                         rule->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1209                 }
1210                 mutex_unlock(&root->chain_lock);
1211         }
1212         return rule;
1213 }
1214 EXPORT_SYMBOL(mlx5_add_flow_rule);
1215
1216 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
1217 {
1218         tree_remove_node(&rule->node);
1219 }
1220 EXPORT_SYMBOL(mlx5_del_flow_rule);
1221
1222 /* Assuming prio->node.children(flow tables) is sorted by level */
1223 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1224 {
1225         struct fs_prio *prio;
1226
1227         fs_get_obj(prio, ft->node.parent);
1228
1229         if (!list_is_last(&ft->node.list, &prio->node.children))
1230                 return list_next_entry(ft, node.list);
1231         return find_next_chained_ft(prio);
1232 }
1233
1234 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1235 {
1236         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1237         struct mlx5_flow_table *new_root_ft = NULL;
1238
1239         if (root->root_ft != ft)
1240                 return 0;
1241
1242         new_root_ft = find_next_ft(ft);
1243         if (new_root_ft) {
1244                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1245
1246                 if (err) {
1247                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1248                                        ft->id);
1249                         return err;
1250                 }
1251                 root->root_ft = new_root_ft;
1252         }
1253         return 0;
1254 }
1255
1256 /* Connect flow table from previous priority to
1257  * the next flow table.
1258  */
1259 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1260 {
1261         struct mlx5_core_dev *dev = get_dev(&ft->node);
1262         struct mlx5_flow_table *next_ft;
1263         struct fs_prio *prio;
1264         int err = 0;
1265
1266         err = update_root_ft_destroy(ft);
1267         if (err)
1268                 return err;
1269
1270         fs_get_obj(prio, ft->node.parent);
1271         if  (!(list_first_entry(&prio->node.children,
1272                                 struct mlx5_flow_table,
1273                                 node.list) == ft))
1274                 return 0;
1275
1276         next_ft = find_next_chained_ft(prio);
1277         err = connect_fwd_rules(dev, next_ft, ft);
1278         if (err)
1279                 return err;
1280
1281         err = connect_prev_fts(dev, next_ft, prio);
1282         if (err)
1283                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1284                                ft->id);
1285         return err;
1286 }
1287
1288 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1289 {
1290         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1291         int err = 0;
1292
1293         mutex_lock(&root->chain_lock);
1294         err = disconnect_flow_table(ft);
1295         if (err) {
1296                 mutex_unlock(&root->chain_lock);
1297                 return err;
1298         }
1299         if (tree_remove_node(&ft->node))
1300                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1301                                ft->id);
1302         mutex_unlock(&root->chain_lock);
1303
1304         return err;
1305 }
1306 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1307
1308 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1309 {
1310         if (tree_remove_node(&fg->node))
1311                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1312                                fg->id);
1313 }
1314
1315 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1316                                                     enum mlx5_flow_namespace_type type)
1317 {
1318         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1319         int prio;
1320         struct fs_prio *fs_prio;
1321         struct mlx5_flow_namespace *ns;
1322
1323         if (!root_ns)
1324                 return NULL;
1325
1326         switch (type) {
1327         case MLX5_FLOW_NAMESPACE_BYPASS:
1328         case MLX5_FLOW_NAMESPACE_KERNEL:
1329         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1330         case MLX5_FLOW_NAMESPACE_ANCHOR:
1331                 prio = type;
1332                 break;
1333         case MLX5_FLOW_NAMESPACE_FDB:
1334                 if (dev->priv.fdb_root_ns)
1335                         return &dev->priv.fdb_root_ns->ns;
1336                 else
1337                         return NULL;
1338         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1339                 if (dev->priv.esw_egress_root_ns)
1340                         return &dev->priv.esw_egress_root_ns->ns;
1341                 else
1342                         return NULL;
1343         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1344                 if (dev->priv.esw_ingress_root_ns)
1345                         return &dev->priv.esw_ingress_root_ns->ns;
1346                 else
1347                         return NULL;
1348         default:
1349                 return NULL;
1350         }
1351
1352         fs_prio = find_prio(&root_ns->ns, prio);
1353         if (!fs_prio)
1354                 return NULL;
1355
1356         ns = list_first_entry(&fs_prio->node.children,
1357                               typeof(*ns),
1358                               node.list);
1359
1360         return ns;
1361 }
1362 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1363
1364 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1365                                       unsigned int prio, int num_levels)
1366 {
1367         struct fs_prio *fs_prio;
1368
1369         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1370         if (!fs_prio)
1371                 return ERR_PTR(-ENOMEM);
1372
1373         fs_prio->node.type = FS_TYPE_PRIO;
1374         tree_init_node(&fs_prio->node, 1, NULL);
1375         tree_add_node(&fs_prio->node, &ns->node);
1376         fs_prio->num_levels = num_levels;
1377         fs_prio->prio = prio;
1378         list_add_tail(&fs_prio->node.list, &ns->node.children);
1379
1380         return fs_prio;
1381 }
1382
1383 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1384                                                      *ns)
1385 {
1386         ns->node.type = FS_TYPE_NAMESPACE;
1387
1388         return ns;
1389 }
1390
1391 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1392 {
1393         struct mlx5_flow_namespace      *ns;
1394
1395         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1396         if (!ns)
1397                 return ERR_PTR(-ENOMEM);
1398
1399         fs_init_namespace(ns);
1400         tree_init_node(&ns->node, 1, NULL);
1401         tree_add_node(&ns->node, &prio->node);
1402         list_add_tail(&ns->node.list, &prio->node.children);
1403
1404         return ns;
1405 }
1406
1407 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1408                              struct init_tree_node *prio_metadata)
1409 {
1410         struct fs_prio *fs_prio;
1411         int i;
1412
1413         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1414                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1415                 if (IS_ERR(fs_prio))
1416                         return PTR_ERR(fs_prio);
1417         }
1418         return 0;
1419 }
1420
1421 #define FLOW_TABLE_BIT_SZ 1
1422 #define GET_FLOW_TABLE_CAP(dev, offset) \
1423         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1424                         offset / 32)) >>                                        \
1425           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1426 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1427 {
1428         int i;
1429
1430         for (i = 0; i < caps->arr_sz; i++) {
1431                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1432                         return false;
1433         }
1434         return true;
1435 }
1436
1437 static int init_root_tree_recursive(struct mlx5_core_dev *dev,
1438                                     struct init_tree_node *init_node,
1439                                     struct fs_node *fs_parent_node,
1440                                     struct init_tree_node *init_parent_node,
1441                                     int prio)
1442 {
1443         int max_ft_level = MLX5_CAP_FLOWTABLE(dev,
1444                                               flow_table_properties_nic_receive.
1445                                               max_ft_level);
1446         struct mlx5_flow_namespace *fs_ns;
1447         struct fs_prio *fs_prio;
1448         struct fs_node *base;
1449         int i;
1450         int err;
1451
1452         if (init_node->type == FS_TYPE_PRIO) {
1453                 if ((init_node->min_ft_level > max_ft_level) ||
1454                     !has_required_caps(dev, &init_node->caps))
1455                         return 0;
1456
1457                 fs_get_obj(fs_ns, fs_parent_node);
1458                 if (init_node->num_leaf_prios)
1459                         return create_leaf_prios(fs_ns, prio, init_node);
1460                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1461                 if (IS_ERR(fs_prio))
1462                         return PTR_ERR(fs_prio);
1463                 base = &fs_prio->node;
1464         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1465                 fs_get_obj(fs_prio, fs_parent_node);
1466                 fs_ns = fs_create_namespace(fs_prio);
1467                 if (IS_ERR(fs_ns))
1468                         return PTR_ERR(fs_ns);
1469                 base = &fs_ns->node;
1470         } else {
1471                 return -EINVAL;
1472         }
1473         prio = 0;
1474         for (i = 0; i < init_node->ar_size; i++) {
1475                 err = init_root_tree_recursive(dev, &init_node->children[i],
1476                                                base, init_node, prio);
1477                 if (err)
1478                         return err;
1479                 if (init_node->children[i].type == FS_TYPE_PRIO &&
1480                     init_node->children[i].num_leaf_prios) {
1481                         prio += init_node->children[i].num_leaf_prios;
1482                 }
1483         }
1484
1485         return 0;
1486 }
1487
1488 static int init_root_tree(struct mlx5_core_dev *dev,
1489                           struct init_tree_node *init_node,
1490                           struct fs_node *fs_parent_node)
1491 {
1492         int i;
1493         struct mlx5_flow_namespace *fs_ns;
1494         int err;
1495
1496         fs_get_obj(fs_ns, fs_parent_node);
1497         for (i = 0; i < init_node->ar_size; i++) {
1498                 err = init_root_tree_recursive(dev, &init_node->children[i],
1499                                                &fs_ns->node,
1500                                                init_node, i);
1501                 if (err)
1502                         return err;
1503         }
1504         return 0;
1505 }
1506
1507 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_core_dev *dev,
1508                                                        enum fs_flow_table_type
1509                                                        table_type)
1510 {
1511         struct mlx5_flow_root_namespace *root_ns;
1512         struct mlx5_flow_namespace *ns;
1513
1514         /* Create the root namespace */
1515         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1516         if (!root_ns)
1517                 return NULL;
1518
1519         root_ns->dev = dev;
1520         root_ns->table_type = table_type;
1521
1522         ns = &root_ns->ns;
1523         fs_init_namespace(ns);
1524         mutex_init(&root_ns->chain_lock);
1525         tree_init_node(&ns->node, 1, NULL);
1526         tree_add_node(&ns->node, NULL);
1527
1528         return root_ns;
1529 }
1530
1531 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1532
1533 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1534 {
1535         struct fs_prio *prio;
1536
1537         fs_for_each_prio(prio, ns) {
1538                  /* This updates prio start_level and num_levels */
1539                 set_prio_attrs_in_prio(prio, acc_level);
1540                 acc_level += prio->num_levels;
1541         }
1542         return acc_level;
1543 }
1544
1545 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1546 {
1547         struct mlx5_flow_namespace *ns;
1548         int acc_level_ns = acc_level;
1549
1550         prio->start_level = acc_level;
1551         fs_for_each_ns(ns, prio)
1552                 /* This updates start_level and num_levels of ns's priority descendants */
1553                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1554         if (!prio->num_levels)
1555                 prio->num_levels = acc_level_ns - prio->start_level;
1556         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1557 }
1558
1559 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1560 {
1561         struct mlx5_flow_namespace *ns = &root_ns->ns;
1562         struct fs_prio *prio;
1563         int start_level = 0;
1564
1565         fs_for_each_prio(prio, ns) {
1566                 set_prio_attrs_in_prio(prio, start_level);
1567                 start_level += prio->num_levels;
1568         }
1569 }
1570
1571 #define ANCHOR_PRIO 0
1572 #define ANCHOR_SIZE 1
1573 #define ANCHOR_LEVEL 0
1574 static int create_anchor_flow_table(struct mlx5_core_dev
1575                                                         *dev)
1576 {
1577         struct mlx5_flow_namespace *ns = NULL;
1578         struct mlx5_flow_table *ft;
1579
1580         ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1581         if (!ns)
1582                 return -EINVAL;
1583         ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL);
1584         if (IS_ERR(ft)) {
1585                 mlx5_core_err(dev, "Failed to create last anchor flow table");
1586                 return PTR_ERR(ft);
1587         }
1588         return 0;
1589 }
1590
1591 static int init_root_ns(struct mlx5_core_dev *dev)
1592 {
1593
1594         dev->priv.root_ns = create_root_ns(dev, FS_FT_NIC_RX);
1595         if (IS_ERR_OR_NULL(dev->priv.root_ns))
1596                 goto cleanup;
1597
1598         if (init_root_tree(dev, &root_fs, &dev->priv.root_ns->ns.node))
1599                 goto cleanup;
1600
1601         set_prio_attrs(dev->priv.root_ns);
1602
1603         if (create_anchor_flow_table(dev))
1604                 goto cleanup;
1605
1606         return 0;
1607
1608 cleanup:
1609         mlx5_cleanup_fs(dev);
1610         return -ENOMEM;
1611 }
1612
1613 static void cleanup_single_prio_root_ns(struct mlx5_core_dev *dev,
1614                                         struct mlx5_flow_root_namespace *root_ns)
1615 {
1616         struct fs_node *prio;
1617
1618         if (!root_ns)
1619                 return;
1620
1621         if (!list_empty(&root_ns->ns.node.children)) {
1622                 prio = list_first_entry(&root_ns->ns.node.children,
1623                                         struct fs_node,
1624                                  list);
1625                 if (tree_remove_node(prio))
1626                         mlx5_core_warn(dev,
1627                                        "Flow steering priority wasn't destroyed, refcount > 1\n");
1628         }
1629         if (tree_remove_node(&root_ns->ns.node))
1630                 mlx5_core_warn(dev,
1631                                "Flow steering namespace wasn't destroyed, refcount > 1\n");
1632         root_ns = NULL;
1633 }
1634
1635 static void destroy_flow_tables(struct fs_prio *prio)
1636 {
1637         struct mlx5_flow_table *iter;
1638         struct mlx5_flow_table *tmp;
1639
1640         fs_for_each_ft_safe(iter, tmp, prio)
1641                 mlx5_destroy_flow_table(iter);
1642 }
1643
1644 static void cleanup_root_ns(struct mlx5_core_dev *dev)
1645 {
1646         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1647         struct fs_prio *iter_prio;
1648
1649         if (!MLX5_CAP_GEN(dev, nic_flow_table))
1650                 return;
1651
1652         if (!root_ns)
1653                 return;
1654
1655         /* stage 1 */
1656         fs_for_each_prio(iter_prio, &root_ns->ns) {
1657                 struct fs_node *node;
1658                 struct mlx5_flow_namespace *iter_ns;
1659
1660                 fs_for_each_ns_or_ft(node, iter_prio) {
1661                         if (node->type == FS_TYPE_FLOW_TABLE)
1662                                 continue;
1663                         fs_get_obj(iter_ns, node);
1664                         while (!list_empty(&iter_ns->node.children)) {
1665                                 struct fs_prio *obj_iter_prio2;
1666                                 struct fs_node *iter_prio2 =
1667                                         list_first_entry(&iter_ns->node.children,
1668                                                          struct fs_node,
1669                                                          list);
1670
1671                                 fs_get_obj(obj_iter_prio2, iter_prio2);
1672                                 destroy_flow_tables(obj_iter_prio2);
1673                                 if (tree_remove_node(iter_prio2)) {
1674                                         mlx5_core_warn(dev,
1675                                                        "Priority %d wasn't destroyed, refcount > 1\n",
1676                                                        obj_iter_prio2->prio);
1677                                         return;
1678                                 }
1679                         }
1680                 }
1681         }
1682
1683         /* stage 2 */
1684         fs_for_each_prio(iter_prio, &root_ns->ns) {
1685                 while (!list_empty(&iter_prio->node.children)) {
1686                         struct fs_node *iter_ns =
1687                                 list_first_entry(&iter_prio->node.children,
1688                                                  struct fs_node,
1689                                                  list);
1690                         if (tree_remove_node(iter_ns)) {
1691                                 mlx5_core_warn(dev,
1692                                                "Namespace wasn't destroyed, refcount > 1\n");
1693                                 return;
1694                         }
1695                 }
1696         }
1697
1698         /* stage 3 */
1699         while (!list_empty(&root_ns->ns.node.children)) {
1700                 struct fs_prio *obj_prio_node;
1701                 struct fs_node *prio_node =
1702                         list_first_entry(&root_ns->ns.node.children,
1703                                          struct fs_node,
1704                                          list);
1705
1706                 fs_get_obj(obj_prio_node, prio_node);
1707                 if (tree_remove_node(prio_node)) {
1708                         mlx5_core_warn(dev,
1709                                        "Priority %d wasn't destroyed, refcount > 1\n",
1710                                        obj_prio_node->prio);
1711                         return;
1712                 }
1713         }
1714
1715         if (tree_remove_node(&root_ns->ns.node)) {
1716                 mlx5_core_warn(dev,
1717                                "root namespace wasn't destroyed, refcount > 1\n");
1718                 return;
1719         }
1720
1721         dev->priv.root_ns = NULL;
1722 }
1723
1724 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1725 {
1726         cleanup_root_ns(dev);
1727         cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1728         cleanup_single_prio_root_ns(dev, dev->priv.esw_egress_root_ns);
1729         cleanup_single_prio_root_ns(dev, dev->priv.esw_ingress_root_ns);
1730 }
1731
1732 static int init_fdb_root_ns(struct mlx5_core_dev *dev)
1733 {
1734         struct fs_prio *prio;
1735
1736         dev->priv.fdb_root_ns = create_root_ns(dev, FS_FT_FDB);
1737         if (!dev->priv.fdb_root_ns)
1738                 return -ENOMEM;
1739
1740         /* Create single prio */
1741         prio = fs_create_prio(&dev->priv.fdb_root_ns->ns, 0, 1);
1742         if (IS_ERR(prio)) {
1743                 cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1744                 return PTR_ERR(prio);
1745         } else {
1746                 return 0;
1747         }
1748 }
1749
1750 static int init_egress_acl_root_ns(struct mlx5_core_dev *dev)
1751 {
1752         struct fs_prio *prio;
1753
1754         dev->priv.esw_egress_root_ns = create_root_ns(dev, FS_FT_ESW_EGRESS_ACL);
1755         if (!dev->priv.esw_egress_root_ns)
1756                 return -ENOMEM;
1757
1758         /* create 1 prio*/
1759         prio = fs_create_prio(&dev->priv.esw_egress_root_ns->ns, 0, MLX5_TOTAL_VPORTS(dev));
1760         if (IS_ERR(prio))
1761                 return PTR_ERR(prio);
1762         else
1763                 return 0;
1764 }
1765
1766 static int init_ingress_acl_root_ns(struct mlx5_core_dev *dev)
1767 {
1768         struct fs_prio *prio;
1769
1770         dev->priv.esw_ingress_root_ns = create_root_ns(dev, FS_FT_ESW_INGRESS_ACL);
1771         if (!dev->priv.esw_ingress_root_ns)
1772                 return -ENOMEM;
1773
1774         /* create 1 prio*/
1775         prio = fs_create_prio(&dev->priv.esw_ingress_root_ns->ns, 0, MLX5_TOTAL_VPORTS(dev));
1776         if (IS_ERR(prio))
1777                 return PTR_ERR(prio);
1778         else
1779                 return 0;
1780 }
1781
1782 int mlx5_init_fs(struct mlx5_core_dev *dev)
1783 {
1784         int err = 0;
1785
1786         if (MLX5_CAP_GEN(dev, nic_flow_table)) {
1787                 err = init_root_ns(dev);
1788                 if (err)
1789                         return err;
1790         }
1791         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1792                 err = init_fdb_root_ns(dev);
1793                 if (err)
1794                         goto err;
1795         }
1796         if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
1797                 err = init_egress_acl_root_ns(dev);
1798                 if (err)
1799                         goto err;
1800         }
1801         if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
1802                 err = init_ingress_acl_root_ns(dev);
1803                 if (err)
1804                         goto err;
1805         }
1806
1807         return 0;
1808 err:
1809         mlx5_cleanup_fs(dev);
1810         return err;
1811 }