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, 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->max_fte = max_fte;
 473         INIT_LIST_HEAD(&ft->fwd_rules);
 474         mutex_init(&ft->lock);
 475
 476         return ft;
 477 }
 478
 479 /* If reverse is false, then we search for the first flow table in the
 480  * root sub-tree from start(closest from right), else we search for the
 481  * last flow table in the root sub-tree till start(closest from left).
 482  */
 483 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 484                                                          struct list_head *start,
 485                                                          bool reverse)
 486 {
 487 #define list_advance_entry(pos, reverse)                \
 488         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 489
 490 #define list_for_each_advance_continue(pos, head, reverse)      \
 491         for (pos = list_advance_entry(pos, reverse);            \
 492              &pos->list != (head);                              \
 493              pos = list_advance_entry(pos, reverse))
 494
 495         struct fs_node *iter = list_entry(start, struct fs_node, list);
 496         struct mlx5_flow_table *ft = NULL;
 497
 498         if (!root)
 499                 return NULL;
 500
 501         list_for_each_advance_continue(iter, &root->children, reverse) {
 502                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 503                         fs_get_obj(ft, iter);
 504                         return ft;
 505                 }
 506                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 507                 if (ft)
 508                         return ft;
 509         }
 510
 511         return ft;
 512 }
 513
 514 /* If reverse if false then return the first flow table in next priority of
 515  * prio in the tree, else return the last flow table in the previous priority
 516  * of prio in the tree.
 517  */
 518 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 519 {
 520         struct mlx5_flow_table *ft = NULL;
 521         struct fs_node *curr_node;
 522         struct fs_node *parent;
 523
 524         parent = prio->node.parent;
 525         curr_node = &prio->node;
 526         while (!ft && parent) {
 527                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 528                 curr_node = parent;
 529                 parent = curr_node->parent;
 530         }
 531         return ft;
 532 }
 533
 534 /* Assuming all the tree is locked by mutex chain lock */
 535 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 536 {
 537         return find_closest_ft(prio, false);
 538 }
 539
 540 /* Assuming all the tree is locked by mutex chain lock */
 541 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 542 {
 543         return find_closest_ft(prio, true);
 544 }
 545
 546 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 547                                struct fs_prio *prio,
 548                                struct mlx5_flow_table *ft)
 549 {
 550         struct mlx5_flow_table *iter;
 551         int i = 0;
 552         int err;
 553
 554         fs_for_each_ft(iter, prio) {
 555                 i++;
 556                 err = mlx5_cmd_modify_flow_table(dev,
 557                                                  iter,
 558                                                  ft);
 559                 if (err) {
 560                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 561                                        iter->id);
 562                         /* The driver is out of sync with the FW */
 563                         if (i > 1)
 564                                 WARN_ON(true);
 565                         return err;
 566                 }
 567         }
 568         return 0;
 569 }
 570
 571 /* Connect flow tables from previous priority of prio to ft */
 572 static int connect_prev_fts(struct mlx5_core_dev *dev,
 573                             struct mlx5_flow_table *ft,
 574                             struct fs_prio *prio)
 575 {
 576         struct mlx5_flow_table *prev_ft;
 577
 578         prev_ft = find_prev_chained_ft(prio);
 579         if (prev_ft) {
 580                 struct fs_prio *prev_prio;
 581
 582                 fs_get_obj(prev_prio, prev_ft->node.parent);
 583                 return connect_fts_in_prio(dev, prev_prio, ft);
 584         }
 585         return 0;
 586 }
 587
 588 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 589                                  *prio)
 590 {
 591         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 592         int min_level = INT_MAX;
 593         int err;
 594
 595         if (root->root_ft)
 596                 min_level = root->root_ft->level;
 597
 598         if (ft->level >= min_level)
 599                 return 0;
 600
 601         err = mlx5_cmd_update_root_ft(root->dev, ft);
 602         if (err)
 603                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 604                                ft->id);
 605         else
 606                 root->root_ft = ft;
 607
 608         return err;
 609 }
 610
 611 int mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 612                                  struct mlx5_flow_destination *dest)
 613 {
 614         struct mlx5_flow_table *ft;
 615         struct mlx5_flow_group *fg;
 616         struct fs_fte *fte;
 617         int err = 0;
 618
 619         fs_get_obj(fte, rule->node.parent);
 620         if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 621                 return -EINVAL;
 622         lock_ref_node(&fte->node);
 623         fs_get_obj(fg, fte->node.parent);
 624         fs_get_obj(ft, fg->node.parent);
 625
 626         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 627         err = mlx5_cmd_update_fte(get_dev(&ft->node),
 628                                   ft, fg->id, fte);
 629         unlock_ref_node(&fte->node);
 630
 631         return err;
 632 }
 633
 634 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 635 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 636                              struct mlx5_flow_table *new_next_ft,
 637                              struct mlx5_flow_table *old_next_ft)
 638 {
 639         struct mlx5_flow_destination dest;
 640         struct mlx5_flow_rule *iter;
 641         int err = 0;
 642
 643         /* new_next_ft and old_next_ft could be NULL only
 644          * when we create/destroy the anchor flow table.
 645          */
 646         if (!new_next_ft || !old_next_ft)
 647                 return 0;
 648
 649         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 650         dest.ft = new_next_ft;
 651
 652         mutex_lock(&old_next_ft->lock);
 653         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 654         mutex_unlock(&old_next_ft->lock);
 655         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 656                 err = mlx5_modify_rule_destination(iter, &dest);
 657                 if (err)
 658                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 659                                new_next_ft->id);
 660         }
 661         return 0;
 662 }
 663
 664 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 665                               struct fs_prio *prio)
 666 {
 667         struct mlx5_flow_table *next_ft;
 668         int err = 0;
 669
 670         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 671
 672         if (list_empty(&prio->node.children)) {
 673                 err = connect_prev_fts(dev, ft, prio);
 674                 if (err)
 675                         return err;
 676
 677                 next_ft = find_next_chained_ft(prio);
 678                 err = connect_fwd_rules(dev, ft, next_ft);
 679                 if (err)
 680                         return err;
 681         }
 682
 683         if (MLX5_CAP_FLOWTABLE(dev,
 684                                flow_table_properties_nic_receive.modify_root))
 685                 err = update_root_ft_create(ft, prio);
 686         return err;
 687 }
 688
 689 static void list_add_flow_table(struct mlx5_flow_table *ft,
 690                                 struct fs_prio *prio)
 691 {
 692         struct list_head *prev = &prio->node.children;
 693         struct mlx5_flow_table *iter;
 694
 695         fs_for_each_ft(iter, prio) {
 696                 if (iter->level > ft->level)
 697                         break;
 698                 prev = &iter->node.list;
 699         }
 700         list_add(&ft->node.list, prev);
 701 }
 702
 703 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 704                                                int prio, int max_fte,
 705                                                u32 level)
 706 {
 707         struct mlx5_flow_table *next_ft = NULL;
 708         struct mlx5_flow_table *ft;
 709         int err;
 710         int log_table_sz;
 711         struct mlx5_flow_root_namespace *root =
 712                 find_root(&ns->node);
 713         struct fs_prio *fs_prio = NULL;
 714
 715         if (!root) {
 716                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 717                 return ERR_PTR(-ENODEV);
 718         }
 719
 720         mutex_lock(&root->chain_lock);
 721         fs_prio = find_prio(ns, prio);
 722         if (!fs_prio) {
 723                 err = -EINVAL;
 724                 goto unlock_root;
 725         }
 726         if (level >= fs_prio->num_levels) {
 727                 err = -ENOSPC;
 728                 goto unlock_root;
 729         }
 730         /* The level is related to the
 731          * priority level range.
 732          */
 733         level += fs_prio->start_level;
 734         ft = alloc_flow_table(level,
 735                               roundup_pow_of_two(max_fte),
 736                               root->table_type);
 737         if (!ft) {
 738                 err = -ENOMEM;
 739                 goto unlock_root;
 740         }
 741
 742         tree_init_node(&ft->node, 1, del_flow_table);
 743         log_table_sz = ilog2(ft->max_fte);
 744         next_ft = find_next_chained_ft(fs_prio);
 745         err = mlx5_cmd_create_flow_table(root->dev, ft->type, ft->level,
 746                                          log_table_sz, next_ft, &ft->id);
 747         if (err)
 748                 goto free_ft;
 749
 750         err = connect_flow_table(root->dev, ft, fs_prio);
 751         if (err)
 752                 goto destroy_ft;
 753         lock_ref_node(&fs_prio->node);
 754         tree_add_node(&ft->node, &fs_prio->node);
 755         list_add_flow_table(ft, fs_prio);
 756         fs_prio->num_ft++;
 757         unlock_ref_node(&fs_prio->node);
 758         mutex_unlock(&root->chain_lock);
 759         return ft;
 760 destroy_ft:
 761         mlx5_cmd_destroy_flow_table(root->dev, ft);
 762 free_ft:
 763         kfree(ft);
 764 unlock_root:
 765         mutex_unlock(&root->chain_lock);
 766         return ERR_PTR(err);
 767 }
 768
 769 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 770                                                             int prio,
 771                                                             int num_flow_table_entries,
 772                                                             int max_num_groups,
 773                                                             u32 level)
 774 {
 775         struct mlx5_flow_table *ft;
 776
 777         if (max_num_groups > num_flow_table_entries)
 778                 return ERR_PTR(-EINVAL);
 779
 780         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries, level);
 781         if (IS_ERR(ft))
 782                 return ft;
 783
 784         ft->autogroup.active = true;
 785         ft->autogroup.required_groups = max_num_groups;
 786
 787         return ft;
 788 }
 789 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 790
 791 /* Flow table should be locked */
 792 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 793                                                         u32 *fg_in,
 794                                                         struct list_head
 795                                                         *prev_fg,
 796                                                         bool is_auto_fg)
 797 {
 798         struct mlx5_flow_group *fg;
 799         struct mlx5_core_dev *dev = get_dev(&ft->node);
 800         int err;
 801
 802         if (!dev)
 803                 return ERR_PTR(-ENODEV);
 804
 805         fg = alloc_flow_group(fg_in);
 806         if (IS_ERR(fg))
 807                 return fg;
 808
 809         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 810         if (err) {
 811                 kfree(fg);
 812                 return ERR_PTR(err);
 813         }
 814
 815         if (ft->autogroup.active)
 816                 ft->autogroup.num_groups++;
 817         /* Add node to tree */
 818         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 819         tree_add_node(&fg->node, &ft->node);
 820         /* Add node to group list */
 821         list_add(&fg->node.list, ft->node.children.prev);
 822
 823         return fg;
 824 }
 825
 826 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 827                                                u32 *fg_in)
 828 {
 829         struct mlx5_flow_group *fg;
 830
 831         if (ft->autogroup.active)
 832                 return ERR_PTR(-EPERM);
 833
 834         lock_ref_node(&ft->node);
 835         fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
 836         unlock_ref_node(&ft->node);
 837
 838         return fg;
 839 }
 840
 841 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 842 {
 843         struct mlx5_flow_rule *rule;
 844
 845         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 846         if (!rule)
 847                 return NULL;
 848
 849         INIT_LIST_HEAD(&rule->next_ft);
 850         rule->node.type = FS_TYPE_FLOW_DEST;
 851         if (dest)
 852                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 853
 854         return rule;
 855 }
 856
 857 /* fte should not be deleted while calling this function */
 858 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
 859                                            struct mlx5_flow_group *fg,
 860                                            struct mlx5_flow_destination *dest)
 861 {
 862         struct mlx5_flow_table *ft;
 863         struct mlx5_flow_rule *rule;
 864         int err;
 865
 866         rule = alloc_rule(dest);
 867         if (!rule)
 868                 return ERR_PTR(-ENOMEM);
 869
 870         fs_get_obj(ft, fg->node.parent);
 871         /* Add dest to dests list- we need flow tables to be in the
 872          * end of the list for forward to next prio rules.
 873          */
 874         tree_init_node(&rule->node, 1, del_rule);
 875         if (dest && dest->type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
 876                 list_add(&rule->node.list, &fte->node.children);
 877         else
 878                 list_add_tail(&rule->node.list, &fte->node.children);
 879         if (dest)
 880                 fte->dests_size++;
 881         if (fte->dests_size == 1 || !dest)
 882                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 883                                           ft, fg->id, fte);
 884         else
 885                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 886                                           ft, fg->id, fte);
 887         if (err)
 888                 goto free_rule;
 889
 890         fte->status |= FS_FTE_STATUS_EXISTING;
 891
 892         return rule;
 893
 894 free_rule:
 895         list_del(&rule->node.list);
 896         kfree(rule);
 897         if (dest)
 898                 fte->dests_size--;
 899         return ERR_PTR(err);
 900 }
 901
 902 /* Assumed fg is locked */
 903 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 904                                        struct list_head **prev)
 905 {
 906         struct fs_fte *fte;
 907         unsigned int start = fg->start_index;
 908
 909         if (prev)
 910                 *prev = &fg->node.children;
 911
 912         /* assumed list is sorted by index */
 913         fs_for_each_fte(fte, fg) {
 914                 if (fte->index != start)
 915                         return start;
 916                 start++;
 917                 if (prev)
 918                         *prev = &fte->node.list;
 919         }
 920
 921         return start;
 922 }
 923
 924 /* prev is output, prev->next = new_fte */
 925 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
 926                                  u32 *match_value,
 927                                  u8 action,
 928                                  u32 flow_tag,
 929                                  struct list_head **prev)
 930 {
 931         struct fs_fte *fte;
 932         int index;
 933
 934         index = get_free_fte_index(fg, prev);
 935         fte = alloc_fte(action, flow_tag, match_value, index);
 936         if (IS_ERR(fte))
 937                 return fte;
 938
 939         return fte;
 940 }
 941
 942 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
 943                                                 u8 match_criteria_enable,
 944                                                 u32 *match_criteria)
 945 {
 946         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 947         struct list_head *prev = &ft->node.children;
 948         unsigned int candidate_index = 0;
 949         struct mlx5_flow_group *fg;
 950         void *match_criteria_addr;
 951         unsigned int group_size = 0;
 952         u32 *in;
 953
 954         if (!ft->autogroup.active)
 955                 return ERR_PTR(-ENOENT);
 956
 957         in = mlx5_vzalloc(inlen);
 958         if (!in)
 959                 return ERR_PTR(-ENOMEM);
 960
 961         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
 962                 /* We save place for flow groups in addition to max types */
 963                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
 964
 965         /*  ft->max_fte == ft->autogroup.max_types */
 966         if (group_size == 0)
 967                 group_size = 1;
 968
 969         /* sorted by start_index */
 970         fs_for_each_fg(fg, ft) {
 971                 if (candidate_index + group_size > fg->start_index)
 972                         candidate_index = fg->start_index + fg->max_ftes;
 973                 else
 974                         break;
 975                 prev = &fg->node.list;
 976         }
 977
 978         if (candidate_index + group_size > ft->max_fte) {
 979                 fg = ERR_PTR(-ENOSPC);
 980                 goto out;
 981         }
 982
 983         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
 984                  match_criteria_enable);
 985         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
 986         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
 987                  group_size - 1);
 988         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
 989                                            in, match_criteria);
 990         memcpy(match_criteria_addr, match_criteria,
 991                MLX5_ST_SZ_BYTES(fte_match_param));
 992
 993         fg = create_flow_group_common(ft, in, prev, true);
 994 out:
 995         kvfree(in);
 996         return fg;
 997 }
 998
 999 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1000                                              struct mlx5_flow_destination *dest)
1001 {
1002         struct mlx5_flow_rule *rule;
1003
1004         list_for_each_entry(rule, &fte->node.children, node.list) {
1005                 if (rule->dest_attr.type == dest->type) {
1006                         if ((dest->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1007                              dest->vport_num == rule->dest_attr.vport_num) ||
1008                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1009                              dest->ft == rule->dest_attr.ft) ||
1010                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1011                              dest->tir_num == rule->dest_attr.tir_num))
1012                                 return rule;
1013                 }
1014         }
1015         return NULL;
1016 }
1017
1018 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
1019                                           u32 *match_value,
1020                                           u8 action,
1021                                           u32 flow_tag,
1022                                           struct mlx5_flow_destination *dest)
1023 {
1024         struct fs_fte *fte;
1025         struct mlx5_flow_rule *rule;
1026         struct mlx5_flow_table *ft;
1027         struct list_head *prev;
1028
1029         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1030         fs_for_each_fte(fte, fg) {
1031                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1032                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1033                     action == fte->action && flow_tag == fte->flow_tag) {
1034                         rule = find_flow_rule(fte, dest);
1035                         if (rule) {
1036                                 atomic_inc(&rule->node.refcount);
1037                                 unlock_ref_node(&fte->node);
1038                                 unlock_ref_node(&fg->node);
1039                                 return rule;
1040                         }
1041                         rule = add_rule_fte(fte, fg, dest);
1042                         unlock_ref_node(&fte->node);
1043                         if (IS_ERR(rule))
1044                                 goto unlock_fg;
1045                         else
1046                                 goto add_rule;
1047                 }
1048                 unlock_ref_node(&fte->node);
1049         }
1050         fs_get_obj(ft, fg->node.parent);
1051         if (fg->num_ftes >= fg->max_ftes) {
1052                 rule = ERR_PTR(-ENOSPC);
1053                 goto unlock_fg;
1054         }
1055
1056         fte = create_fte(fg, match_value, action, flow_tag, &prev);
1057         if (IS_ERR(fte)) {
1058                 rule = (void *)fte;
1059                 goto unlock_fg;
1060         }
1061         tree_init_node(&fte->node, 0, del_fte);
1062         rule = add_rule_fte(fte, fg, dest);
1063         if (IS_ERR(rule)) {
1064                 kfree(fte);
1065                 goto unlock_fg;
1066         }
1067
1068         fg->num_ftes++;
1069
1070         tree_add_node(&fte->node, &fg->node);
1071         list_add(&fte->node.list, prev);
1072 add_rule:
1073         tree_add_node(&rule->node, &fte->node);
1074 unlock_fg:
1075         unlock_ref_node(&fg->node);
1076         return rule;
1077 }
1078
1079 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1080                           u32 action,
1081                           struct mlx5_flow_table *ft)
1082 {
1083         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1084                 return true;
1085
1086         if (!dest || ((dest->type ==
1087             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1088             (dest->ft->level <= ft->level)))
1089                 return false;
1090         return true;
1091 }
1092
1093 static struct mlx5_flow_rule *
1094 _mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1095                     u8 match_criteria_enable,
1096                     u32 *match_criteria,
1097                     u32 *match_value,
1098                     u32 action,
1099                     u32 flow_tag,
1100                     struct mlx5_flow_destination *dest)
1101 {
1102         struct mlx5_flow_group *g;
1103         struct mlx5_flow_rule *rule;
1104
1105         if (!dest_is_valid(dest, action, ft))
1106                 return ERR_PTR(-EINVAL);
1107
1108         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1109         fs_for_each_fg(g, ft)
1110                 if (compare_match_criteria(g->mask.match_criteria_enable,
1111                                            match_criteria_enable,
1112                                            g->mask.match_criteria,
1113                                            match_criteria)) {
1114                         rule = add_rule_fg(g, match_value,
1115                                            action, flow_tag, dest);
1116                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1117                                 goto unlock;
1118                 }
1119
1120         g = create_autogroup(ft, match_criteria_enable, match_criteria);
1121         if (IS_ERR(g)) {
1122                 rule = (void *)g;
1123                 goto unlock;
1124         }
1125
1126         rule = add_rule_fg(g, match_value,
1127                            action, flow_tag, dest);
1128         if (IS_ERR(rule)) {
1129                 /* Remove assumes refcount > 0 and autogroup creates a group
1130                  * with a refcount = 0.
1131                  */
1132                 unlock_ref_node(&ft->node);
1133                 tree_get_node(&g->node);
1134                 tree_remove_node(&g->node);
1135                 return rule;
1136         }
1137 unlock:
1138         unlock_ref_node(&ft->node);
1139         return rule;
1140 }
1141
1142 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1143 {
1144         return ((ft->type == FS_FT_NIC_RX) &&
1145                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1146 }
1147
1148 struct mlx5_flow_rule *
1149 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1150                    u8 match_criteria_enable,
1151                    u32 *match_criteria,
1152                    u32 *match_value,
1153                    u32 action,
1154                    u32 flow_tag,
1155                    struct mlx5_flow_destination *dest)
1156 {
1157         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1158         struct mlx5_flow_destination gen_dest;
1159         struct mlx5_flow_table *next_ft = NULL;
1160         struct mlx5_flow_rule *rule = NULL;
1161         u32 sw_action = action;
1162         struct fs_prio *prio;
1163
1164         fs_get_obj(prio, ft->node.parent);
1165         if (action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1166                 if (!fwd_next_prio_supported(ft))
1167                         return ERR_PTR(-EOPNOTSUPP);
1168                 if (dest)
1169                         return ERR_PTR(-EINVAL);
1170                 mutex_lock(&root->chain_lock);
1171                 next_ft = find_next_chained_ft(prio);
1172                 if (next_ft) {
1173                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1174                         gen_dest.ft = next_ft;
1175                         dest = &gen_dest;
1176                         action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1177                 } else {
1178                         mutex_unlock(&root->chain_lock);
1179                         return ERR_PTR(-EOPNOTSUPP);
1180                 }
1181         }
1182
1183         rule =  _mlx5_add_flow_rule(ft, match_criteria_enable, match_criteria,
1184                                     match_value, action, flow_tag, dest);
1185
1186         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1187                 if (!IS_ERR_OR_NULL(rule) &&
1188                     (list_empty(&rule->next_ft))) {
1189                         mutex_lock(&next_ft->lock);
1190                         list_add(&rule->next_ft, &next_ft->fwd_rules);
1191                         mutex_unlock(&next_ft->lock);
1192                         rule->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1193                 }
1194                 mutex_unlock(&root->chain_lock);
1195         }
1196         return rule;
1197 }
1198 EXPORT_SYMBOL(mlx5_add_flow_rule);
1199
1200 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
1201 {
1202         tree_remove_node(&rule->node);
1203 }
1204 EXPORT_SYMBOL(mlx5_del_flow_rule);
1205
1206 /* Assuming prio->node.children(flow tables) is sorted by level */
1207 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1208 {
1209         struct fs_prio *prio;
1210
1211         fs_get_obj(prio, ft->node.parent);
1212
1213         if (!list_is_last(&ft->node.list, &prio->node.children))
1214                 return list_next_entry(ft, node.list);
1215         return find_next_chained_ft(prio);
1216 }
1217
1218 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1219 {
1220         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1221         struct mlx5_flow_table *new_root_ft = NULL;
1222
1223         if (root->root_ft != ft)
1224                 return 0;
1225
1226         new_root_ft = find_next_ft(ft);
1227         if (new_root_ft) {
1228                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1229
1230                 if (err) {
1231                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1232                                        ft->id);
1233                         return err;
1234                 }
1235                 root->root_ft = new_root_ft;
1236         }
1237         return 0;
1238 }
1239
1240 /* Connect flow table from previous priority to
1241  * the next flow table.
1242  */
1243 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1244 {
1245         struct mlx5_core_dev *dev = get_dev(&ft->node);
1246         struct mlx5_flow_table *next_ft;
1247         struct fs_prio *prio;
1248         int err = 0;
1249
1250         err = update_root_ft_destroy(ft);
1251         if (err)
1252                 return err;
1253
1254         fs_get_obj(prio, ft->node.parent);
1255         if  (!(list_first_entry(&prio->node.children,
1256                                 struct mlx5_flow_table,
1257                                 node.list) == ft))
1258                 return 0;
1259
1260         next_ft = find_next_chained_ft(prio);
1261         err = connect_fwd_rules(dev, next_ft, ft);
1262         if (err)
1263                 return err;
1264
1265         err = connect_prev_fts(dev, next_ft, prio);
1266         if (err)
1267                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1268                                ft->id);
1269         return err;
1270 }
1271
1272 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1273 {
1274         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1275         int err = 0;
1276
1277         mutex_lock(&root->chain_lock);
1278         err = disconnect_flow_table(ft);
1279         if (err) {
1280                 mutex_unlock(&root->chain_lock);
1281                 return err;
1282         }
1283         if (tree_remove_node(&ft->node))
1284                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1285                                ft->id);
1286         mutex_unlock(&root->chain_lock);
1287
1288         return err;
1289 }
1290 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1291
1292 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1293 {
1294         if (tree_remove_node(&fg->node))
1295                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1296                                fg->id);
1297 }
1298
1299 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1300                                                     enum mlx5_flow_namespace_type type)
1301 {
1302         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1303         int prio;
1304         struct fs_prio *fs_prio;
1305         struct mlx5_flow_namespace *ns;
1306
1307         if (!root_ns)
1308                 return NULL;
1309
1310         switch (type) {
1311         case MLX5_FLOW_NAMESPACE_BYPASS:
1312         case MLX5_FLOW_NAMESPACE_KERNEL:
1313         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1314         case MLX5_FLOW_NAMESPACE_ANCHOR:
1315                 prio = type;
1316                 break;
1317         case MLX5_FLOW_NAMESPACE_FDB:
1318                 if (dev->priv.fdb_root_ns)
1319                         return &dev->priv.fdb_root_ns->ns;
1320                 else
1321                         return NULL;
1322         default:
1323                 return NULL;
1324         }
1325
1326         fs_prio = find_prio(&root_ns->ns, prio);
1327         if (!fs_prio)
1328                 return NULL;
1329
1330         ns = list_first_entry(&fs_prio->node.children,
1331                               typeof(*ns),
1332                               node.list);
1333
1334         return ns;
1335 }
1336 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1337
1338 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1339                                       unsigned int prio, int num_levels)
1340 {
1341         struct fs_prio *fs_prio;
1342
1343         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1344         if (!fs_prio)
1345                 return ERR_PTR(-ENOMEM);
1346
1347         fs_prio->node.type = FS_TYPE_PRIO;
1348         tree_init_node(&fs_prio->node, 1, NULL);
1349         tree_add_node(&fs_prio->node, &ns->node);
1350         fs_prio->num_levels = num_levels;
1351         fs_prio->prio = prio;
1352         list_add_tail(&fs_prio->node.list, &ns->node.children);
1353
1354         return fs_prio;
1355 }
1356
1357 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1358                                                      *ns)
1359 {
1360         ns->node.type = FS_TYPE_NAMESPACE;
1361
1362         return ns;
1363 }
1364
1365 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1366 {
1367         struct mlx5_flow_namespace      *ns;
1368
1369         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1370         if (!ns)
1371                 return ERR_PTR(-ENOMEM);
1372
1373         fs_init_namespace(ns);
1374         tree_init_node(&ns->node, 1, NULL);
1375         tree_add_node(&ns->node, &prio->node);
1376         list_add_tail(&ns->node.list, &prio->node.children);
1377
1378         return ns;
1379 }
1380
1381 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1382                              struct init_tree_node *prio_metadata)
1383 {
1384         struct fs_prio *fs_prio;
1385         int i;
1386
1387         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1388                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1389                 if (IS_ERR(fs_prio))
1390                         return PTR_ERR(fs_prio);
1391         }
1392         return 0;
1393 }
1394
1395 #define FLOW_TABLE_BIT_SZ 1
1396 #define GET_FLOW_TABLE_CAP(dev, offset) \
1397         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1398                         offset / 32)) >>                                        \
1399           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1400 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1401 {
1402         int i;
1403
1404         for (i = 0; i < caps->arr_sz; i++) {
1405                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1406                         return false;
1407         }
1408         return true;
1409 }
1410
1411 static int init_root_tree_recursive(struct mlx5_core_dev *dev,
1412                                     struct init_tree_node *init_node,
1413                                     struct fs_node *fs_parent_node,
1414                                     struct init_tree_node *init_parent_node,
1415                                     int prio)
1416 {
1417         int max_ft_level = MLX5_CAP_FLOWTABLE(dev,
1418                                               flow_table_properties_nic_receive.
1419                                               max_ft_level);
1420         struct mlx5_flow_namespace *fs_ns;
1421         struct fs_prio *fs_prio;
1422         struct fs_node *base;
1423         int i;
1424         int err;
1425
1426         if (init_node->type == FS_TYPE_PRIO) {
1427                 if ((init_node->min_ft_level > max_ft_level) ||
1428                     !has_required_caps(dev, &init_node->caps))
1429                         return 0;
1430
1431                 fs_get_obj(fs_ns, fs_parent_node);
1432                 if (init_node->num_leaf_prios)
1433                         return create_leaf_prios(fs_ns, prio, init_node);
1434                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1435                 if (IS_ERR(fs_prio))
1436                         return PTR_ERR(fs_prio);
1437                 base = &fs_prio->node;
1438         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1439                 fs_get_obj(fs_prio, fs_parent_node);
1440                 fs_ns = fs_create_namespace(fs_prio);
1441                 if (IS_ERR(fs_ns))
1442                         return PTR_ERR(fs_ns);
1443                 base = &fs_ns->node;
1444         } else {
1445                 return -EINVAL;
1446         }
1447         prio = 0;
1448         for (i = 0; i < init_node->ar_size; i++) {
1449                 err = init_root_tree_recursive(dev, &init_node->children[i],
1450                                                base, init_node, prio);
1451                 if (err)
1452                         return err;
1453                 if (init_node->children[i].type == FS_TYPE_PRIO &&
1454                     init_node->children[i].num_leaf_prios) {
1455                         prio += init_node->children[i].num_leaf_prios;
1456                 }
1457         }
1458
1459         return 0;
1460 }
1461
1462 static int init_root_tree(struct mlx5_core_dev *dev,
1463                           struct init_tree_node *init_node,
1464                           struct fs_node *fs_parent_node)
1465 {
1466         int i;
1467         struct mlx5_flow_namespace *fs_ns;
1468         int err;
1469
1470         fs_get_obj(fs_ns, fs_parent_node);
1471         for (i = 0; i < init_node->ar_size; i++) {
1472                 err = init_root_tree_recursive(dev, &init_node->children[i],
1473                                                &fs_ns->node,
1474                                                init_node, i);
1475                 if (err)
1476                         return err;
1477         }
1478         return 0;
1479 }
1480
1481 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_core_dev *dev,
1482                                                        enum fs_flow_table_type
1483                                                        table_type)
1484 {
1485         struct mlx5_flow_root_namespace *root_ns;
1486         struct mlx5_flow_namespace *ns;
1487
1488         /* Create the root namespace */
1489         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1490         if (!root_ns)
1491                 return NULL;
1492
1493         root_ns->dev = dev;
1494         root_ns->table_type = table_type;
1495
1496         ns = &root_ns->ns;
1497         fs_init_namespace(ns);
1498         mutex_init(&root_ns->chain_lock);
1499         tree_init_node(&ns->node, 1, NULL);
1500         tree_add_node(&ns->node, NULL);
1501
1502         return root_ns;
1503 }
1504
1505 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1506
1507 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1508 {
1509         struct fs_prio *prio;
1510
1511         fs_for_each_prio(prio, ns) {
1512                  /* This updates prio start_level and num_levels */
1513                 set_prio_attrs_in_prio(prio, acc_level);
1514                 acc_level += prio->num_levels;
1515         }
1516         return acc_level;
1517 }
1518
1519 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1520 {
1521         struct mlx5_flow_namespace *ns;
1522         int acc_level_ns = acc_level;
1523
1524         prio->start_level = acc_level;
1525         fs_for_each_ns(ns, prio)
1526                 /* This updates start_level and num_levels of ns's priority descendants */
1527                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1528         if (!prio->num_levels)
1529                 prio->num_levels = acc_level_ns - prio->start_level;
1530         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1531 }
1532
1533 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1534 {
1535         struct mlx5_flow_namespace *ns = &root_ns->ns;
1536         struct fs_prio *prio;
1537         int start_level = 0;
1538
1539         fs_for_each_prio(prio, ns) {
1540                 set_prio_attrs_in_prio(prio, start_level);
1541                 start_level += prio->num_levels;
1542         }
1543 }
1544
1545 #define ANCHOR_PRIO 0
1546 #define ANCHOR_SIZE 1
1547 #define ANCHOR_LEVEL 0
1548 static int create_anchor_flow_table(struct mlx5_core_dev
1549                                                         *dev)
1550 {
1551         struct mlx5_flow_namespace *ns = NULL;
1552         struct mlx5_flow_table *ft;
1553
1554         ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1555         if (!ns)
1556                 return -EINVAL;
1557         ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL);
1558         if (IS_ERR(ft)) {
1559                 mlx5_core_err(dev, "Failed to create last anchor flow table");
1560                 return PTR_ERR(ft);
1561         }
1562         return 0;
1563 }
1564
1565 static int init_root_ns(struct mlx5_core_dev *dev)
1566 {
1567
1568         dev->priv.root_ns = create_root_ns(dev, FS_FT_NIC_RX);
1569         if (IS_ERR_OR_NULL(dev->priv.root_ns))
1570                 goto cleanup;
1571
1572         if (init_root_tree(dev, &root_fs, &dev->priv.root_ns->ns.node))
1573                 goto cleanup;
1574
1575         set_prio_attrs(dev->priv.root_ns);
1576
1577         if (create_anchor_flow_table(dev))
1578                 goto cleanup;
1579
1580         return 0;
1581
1582 cleanup:
1583         mlx5_cleanup_fs(dev);
1584         return -ENOMEM;
1585 }
1586
1587 static void cleanup_single_prio_root_ns(struct mlx5_core_dev *dev,
1588                                         struct mlx5_flow_root_namespace *root_ns)
1589 {
1590         struct fs_node *prio;
1591
1592         if (!root_ns)
1593                 return;
1594
1595         if (!list_empty(&root_ns->ns.node.children)) {
1596                 prio = list_first_entry(&root_ns->ns.node.children,
1597                                         struct fs_node,
1598                                  list);
1599                 if (tree_remove_node(prio))
1600                         mlx5_core_warn(dev,
1601                                        "Flow steering priority wasn't destroyed, refcount > 1\n");
1602         }
1603         if (tree_remove_node(&root_ns->ns.node))
1604                 mlx5_core_warn(dev,
1605                                "Flow steering namespace wasn't destroyed, refcount > 1\n");
1606         root_ns = NULL;
1607 }
1608
1609 static void destroy_flow_tables(struct fs_prio *prio)
1610 {
1611         struct mlx5_flow_table *iter;
1612         struct mlx5_flow_table *tmp;
1613
1614         fs_for_each_ft_safe(iter, tmp, prio)
1615                 mlx5_destroy_flow_table(iter);
1616 }
1617
1618 static void cleanup_root_ns(struct mlx5_core_dev *dev)
1619 {
1620         struct mlx5_flow_root_namespace *root_ns = dev->priv.root_ns;
1621         struct fs_prio *iter_prio;
1622
1623         if (!MLX5_CAP_GEN(dev, nic_flow_table))
1624                 return;
1625
1626         if (!root_ns)
1627                 return;
1628
1629         /* stage 1 */
1630         fs_for_each_prio(iter_prio, &root_ns->ns) {
1631                 struct fs_node *node;
1632                 struct mlx5_flow_namespace *iter_ns;
1633
1634                 fs_for_each_ns_or_ft(node, iter_prio) {
1635                         if (node->type == FS_TYPE_FLOW_TABLE)
1636                                 continue;
1637                         fs_get_obj(iter_ns, node);
1638                         while (!list_empty(&iter_ns->node.children)) {
1639                                 struct fs_prio *obj_iter_prio2;
1640                                 struct fs_node *iter_prio2 =
1641                                         list_first_entry(&iter_ns->node.children,
1642                                                          struct fs_node,
1643                                                          list);
1644
1645                                 fs_get_obj(obj_iter_prio2, iter_prio2);
1646                                 destroy_flow_tables(obj_iter_prio2);
1647                                 if (tree_remove_node(iter_prio2)) {
1648                                         mlx5_core_warn(dev,
1649                                                        "Priority %d wasn't destroyed, refcount > 1\n",
1650                                                        obj_iter_prio2->prio);
1651                                         return;
1652                                 }
1653                         }
1654                 }
1655         }
1656
1657         /* stage 2 */
1658         fs_for_each_prio(iter_prio, &root_ns->ns) {
1659                 while (!list_empty(&iter_prio->node.children)) {
1660                         struct fs_node *iter_ns =
1661                                 list_first_entry(&iter_prio->node.children,
1662                                                  struct fs_node,
1663                                                  list);
1664                         if (tree_remove_node(iter_ns)) {
1665                                 mlx5_core_warn(dev,
1666                                                "Namespace wasn't destroyed, refcount > 1\n");
1667                                 return;
1668                         }
1669                 }
1670         }
1671
1672         /* stage 3 */
1673         while (!list_empty(&root_ns->ns.node.children)) {
1674                 struct fs_prio *obj_prio_node;
1675                 struct fs_node *prio_node =
1676                         list_first_entry(&root_ns->ns.node.children,
1677                                          struct fs_node,
1678                                          list);
1679
1680                 fs_get_obj(obj_prio_node, prio_node);
1681                 if (tree_remove_node(prio_node)) {
1682                         mlx5_core_warn(dev,
1683                                        "Priority %d wasn't destroyed, refcount > 1\n",
1684                                        obj_prio_node->prio);
1685                         return;
1686                 }
1687         }
1688
1689         if (tree_remove_node(&root_ns->ns.node)) {
1690                 mlx5_core_warn(dev,
1691                                "root namespace wasn't destroyed, refcount > 1\n");
1692                 return;
1693         }
1694
1695         dev->priv.root_ns = NULL;
1696 }
1697
1698 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1699 {
1700         cleanup_root_ns(dev);
1701         cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1702 }
1703
1704 static int init_fdb_root_ns(struct mlx5_core_dev *dev)
1705 {
1706         struct fs_prio *prio;
1707
1708         dev->priv.fdb_root_ns = create_root_ns(dev, FS_FT_FDB);
1709         if (!dev->priv.fdb_root_ns)
1710                 return -ENOMEM;
1711
1712         /* Create single prio */
1713         prio = fs_create_prio(&dev->priv.fdb_root_ns->ns, 0, 1);
1714         if (IS_ERR(prio)) {
1715                 cleanup_single_prio_root_ns(dev, dev->priv.fdb_root_ns);
1716                 return PTR_ERR(prio);
1717         } else {
1718                 return 0;
1719         }
1720 }
1721
1722 int mlx5_init_fs(struct mlx5_core_dev *dev)
1723 {
1724         int err = 0;
1725
1726         if (MLX5_CAP_GEN(dev, nic_flow_table)) {
1727                 err = init_root_ns(dev);
1728                 if (err)
1729                         return err;
1730         }
1731         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1732                 err = init_fdb_root_ns(dev);
1733                 if (err)
1734                         cleanup_root_ns(dev);
1735         }
1736
1737         return err;
1738 }