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 FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  71                                            FS_CAP(flow_table_properties_nic_receive.modify_root), \
  72                                            FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  73                                            FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  74
  75 #define LEFTOVERS_NUM_LEVELS 1
  76 #define LEFTOVERS_NUM_PRIOS 1
  77
  78 #define BY_PASS_PRIO_NUM_LEVELS 1
  79 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  80                            LEFTOVERS_NUM_PRIOS)
  81
  82 #define ETHTOOL_PRIO_NUM_LEVELS 1
  83 #define ETHTOOL_NUM_PRIOS 10
  84 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
  85 /* Vlan, mac, ttc, aRFS */
  86 #define KERNEL_NIC_PRIO_NUM_LEVELS 4
  87 #define KERNEL_NIC_NUM_PRIOS 1
  88 /* One more level for tc */
  89 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
  90
  91 #define ANCHOR_NUM_LEVELS 1
  92 #define ANCHOR_NUM_PRIOS 1
  93 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
  94
  95 #define OFFLOADS_MAX_FT 1
  96 #define OFFLOADS_NUM_PRIOS 1
  97 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
  98
  99 struct node_caps {
 100         size_t  arr_sz;
 101         long    *caps;
 102 };
 103 static struct init_tree_node {
 104         enum fs_node_type       type;
 105         struct init_tree_node *children;
 106         int ar_size;
 107         struct node_caps caps;
 108         int min_ft_level;
 109         int num_leaf_prios;
 110         int prio;
 111         int num_levels;
 112 } root_fs = {
 113         .type = FS_TYPE_NAMESPACE,
 114         .ar_size = 6,
 115         .children = (struct init_tree_node[]) {
 116                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 117                          FS_CHAINING_CAPS,
 118                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 119                                                   BY_PASS_PRIO_NUM_LEVELS))),
 120                 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
 121                          ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
 122                 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
 123                          FS_CHAINING_CAPS,
 124                          ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 125                                                   ETHTOOL_PRIO_NUM_LEVELS))),
 126                 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 127                          ADD_NS(ADD_MULTIPLE_PRIO(1, 1),
 128                                 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 129                                                   KERNEL_NIC_PRIO_NUM_LEVELS))),
 130                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 131                          FS_CHAINING_CAPS,
 132                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
 133                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 134                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
 135         }
 136 };
 137
 138 enum fs_i_mutex_lock_class {
 139         FS_MUTEX_GRANDPARENT,
 140         FS_MUTEX_PARENT,
 141         FS_MUTEX_CHILD
 142 };
 143
 144 static void del_rule(struct fs_node *node);
 145 static void del_flow_table(struct fs_node *node);
 146 static void del_flow_group(struct fs_node *node);
 147 static void del_fte(struct fs_node *node);
 148
 149 static void tree_init_node(struct fs_node *node,
 150                            unsigned int refcount,
 151                            void (*remove_func)(struct fs_node *))
 152 {
 153         atomic_set(&node->refcount, refcount);
 154         INIT_LIST_HEAD(&node->list);
 155         INIT_LIST_HEAD(&node->children);
 156         mutex_init(&node->lock);
 157         node->remove_func = remove_func;
 158 }
 159
 160 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 161 {
 162         if (parent)
 163                 atomic_inc(&parent->refcount);
 164         node->parent = parent;
 165
 166         /* Parent is the root */
 167         if (!parent)
 168                 node->root = node;
 169         else
 170                 node->root = parent->root;
 171 }
 172
 173 static void tree_get_node(struct fs_node *node)
 174 {
 175         atomic_inc(&node->refcount);
 176 }
 177
 178 static void nested_lock_ref_node(struct fs_node *node,
 179                                  enum fs_i_mutex_lock_class class)
 180 {
 181         if (node) {
 182                 mutex_lock_nested(&node->lock, class);
 183                 atomic_inc(&node->refcount);
 184         }
 185 }
 186
 187 static void lock_ref_node(struct fs_node *node)
 188 {
 189         if (node) {
 190                 mutex_lock(&node->lock);
 191                 atomic_inc(&node->refcount);
 192         }
 193 }
 194
 195 static void unlock_ref_node(struct fs_node *node)
 196 {
 197         if (node) {
 198                 atomic_dec(&node->refcount);
 199                 mutex_unlock(&node->lock);
 200         }
 201 }
 202
 203 static void tree_put_node(struct fs_node *node)
 204 {
 205         struct fs_node *parent_node = node->parent;
 206
 207         lock_ref_node(parent_node);
 208         if (atomic_dec_and_test(&node->refcount)) {
 209                 if (parent_node)
 210                         list_del_init(&node->list);
 211                 if (node->remove_func)
 212                         node->remove_func(node);
 213                 kfree(node);
 214                 node = NULL;
 215         }
 216         unlock_ref_node(parent_node);
 217         if (!node && parent_node)
 218                 tree_put_node(parent_node);
 219 }
 220
 221 static int tree_remove_node(struct fs_node *node)
 222 {
 223         if (atomic_read(&node->refcount) > 1) {
 224                 atomic_dec(&node->refcount);
 225                 return -EEXIST;
 226         }
 227         tree_put_node(node);
 228         return 0;
 229 }
 230
 231 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 232                                  unsigned int prio)
 233 {
 234         struct fs_prio *iter_prio;
 235
 236         fs_for_each_prio(iter_prio, ns) {
 237                 if (iter_prio->prio == prio)
 238                         return iter_prio;
 239         }
 240
 241         return NULL;
 242 }
 243
 244 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 245 {
 246         unsigned int i;
 247
 248         for (i = 0; i < size; i++, mask++, val1++, val2++)
 249                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 250                     ((*(u8 *)val2) & (*(u8 *)mask)))
 251                         return false;
 252
 253         return true;
 254 }
 255
 256 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 257                                 void *fte_param1, void *fte_param2)
 258 {
 259         if (mask->match_criteria_enable &
 260             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 261                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 262                                                 fte_param1, outer_headers);
 263                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 264                                                 fte_param2, outer_headers);
 265                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 266                                               mask->match_criteria, outer_headers);
 267
 268                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 269                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 270                         return false;
 271         }
 272
 273         if (mask->match_criteria_enable &
 274             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 275                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 276                                                 fte_param1, misc_parameters);
 277                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 278                                                 fte_param2, misc_parameters);
 279                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 280                                           mask->match_criteria, misc_parameters);
 281
 282                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 283                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 284                         return false;
 285         }
 286
 287         if (mask->match_criteria_enable &
 288             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 289                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 290                                                 fte_param1, inner_headers);
 291                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 292                                                 fte_param2, inner_headers);
 293                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 294                                           mask->match_criteria, inner_headers);
 295
 296                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 297                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 298                         return false;
 299         }
 300         return true;
 301 }
 302
 303 static bool compare_match_criteria(u8 match_criteria_enable1,
 304                                    u8 match_criteria_enable2,
 305                                    void *mask1, void *mask2)
 306 {
 307         return match_criteria_enable1 == match_criteria_enable2 &&
 308                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 309 }
 310
 311 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 312 {
 313         struct fs_node *root;
 314         struct mlx5_flow_namespace *ns;
 315
 316         root = node->root;
 317
 318         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 319                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 320                 return NULL;
 321         }
 322
 323         ns = container_of(root, struct mlx5_flow_namespace, node);
 324         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 325 }
 326
 327 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 328 {
 329         struct mlx5_flow_root_namespace *root = find_root(node);
 330
 331         if (root)
 332                 return root->dev;
 333         return NULL;
 334 }
 335
 336 static void del_flow_table(struct fs_node *node)
 337 {
 338         struct mlx5_flow_table *ft;
 339         struct mlx5_core_dev *dev;
 340         struct fs_prio *prio;
 341         int err;
 342
 343         fs_get_obj(ft, node);
 344         dev = get_dev(&ft->node);
 345
 346         err = mlx5_cmd_destroy_flow_table(dev, ft);
 347         if (err)
 348                 pr_warn("flow steering can't destroy ft\n");
 349         fs_get_obj(prio, ft->node.parent);
 350         prio->num_ft--;
 351 }
 352
 353 static void del_rule(struct fs_node *node)
 354 {
 355         struct mlx5_flow_rule *rule;
 356         struct mlx5_flow_table *ft;
 357         struct mlx5_flow_group *fg;
 358         struct fs_fte *fte;
 359         u32     *match_value;
 360         int modify_mask;
 361         struct mlx5_core_dev *dev = get_dev(node);
 362         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 363         int err;
 364
 365         match_value = mlx5_vzalloc(match_len);
 366         if (!match_value) {
 367                 pr_warn("failed to allocate inbox\n");
 368                 return;
 369         }
 370
 371         fs_get_obj(rule, node);
 372         fs_get_obj(fte, rule->node.parent);
 373         fs_get_obj(fg, fte->node.parent);
 374         memcpy(match_value, fte->val, sizeof(fte->val));
 375         fs_get_obj(ft, fg->node.parent);
 376         list_del(&rule->node.list);
 377         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 378                 mutex_lock(&rule->dest_attr.ft->lock);
 379                 list_del(&rule->next_ft);
 380                 mutex_unlock(&rule->dest_attr.ft->lock);
 381         }
 382         if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 383             --fte->dests_size) {
 384                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST),
 385                 err = mlx5_cmd_update_fte(dev, ft,
 386                                           fg->id,
 387                                           modify_mask,
 388                                           fte);
 389                 if (err)
 390                         pr_warn("%s can't del rule fg id=%d fte_index=%d\n",
 391                                 __func__, fg->id, fte->index);
 392         }
 393         kvfree(match_value);
 394 }
 395
 396 static void del_fte(struct fs_node *node)
 397 {
 398         struct mlx5_flow_table *ft;
 399         struct mlx5_flow_group *fg;
 400         struct mlx5_core_dev *dev;
 401         struct fs_fte *fte;
 402         int err;
 403
 404         fs_get_obj(fte, node);
 405         fs_get_obj(fg, fte->node.parent);
 406         fs_get_obj(ft, fg->node.parent);
 407
 408         dev = get_dev(&ft->node);
 409         err = mlx5_cmd_delete_fte(dev, ft,
 410                                   fte->index);
 411         if (err)
 412                 pr_warn("flow steering can't delete fte in index %d of flow group id %d\n",
 413                         fte->index, fg->id);
 414
 415         fte->status = 0;
 416         fg->num_ftes--;
 417 }
 418
 419 static void del_flow_group(struct fs_node *node)
 420 {
 421         struct mlx5_flow_group *fg;
 422         struct mlx5_flow_table *ft;
 423         struct mlx5_core_dev *dev;
 424
 425         fs_get_obj(fg, node);
 426         fs_get_obj(ft, fg->node.parent);
 427         dev = get_dev(&ft->node);
 428
 429         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 430                 pr_warn("flow steering can't destroy fg %d of ft %d\n",
 431                         fg->id, ft->id);
 432 }
 433
 434 static struct fs_fte *alloc_fte(u8 action,
 435                                 u32 flow_tag,
 436                                 u32 *match_value,
 437                                 unsigned int index)
 438 {
 439         struct fs_fte *fte;
 440
 441         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 442         if (!fte)
 443                 return ERR_PTR(-ENOMEM);
 444
 445         memcpy(fte->val, match_value, sizeof(fte->val));
 446         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 447         fte->flow_tag = flow_tag;
 448         fte->index = index;
 449         fte->action = action;
 450
 451         return fte;
 452 }
 453
 454 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 455 {
 456         struct mlx5_flow_group *fg;
 457         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 458                                             create_fg_in, match_criteria);
 459         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 460                                             create_fg_in,
 461                                             match_criteria_enable);
 462         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 463         if (!fg)
 464                 return ERR_PTR(-ENOMEM);
 465
 466         fg->mask.match_criteria_enable = match_criteria_enable;
 467         memcpy(&fg->mask.match_criteria, match_criteria,
 468                sizeof(fg->mask.match_criteria));
 469         fg->node.type =  FS_TYPE_FLOW_GROUP;
 470         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 471                                    start_flow_index);
 472         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 473                                 end_flow_index) - fg->start_index + 1;
 474         return fg;
 475 }
 476
 477 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 478                                                 enum fs_flow_table_type table_type)
 479 {
 480         struct mlx5_flow_table *ft;
 481
 482         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 483         if (!ft)
 484                 return NULL;
 485
 486         ft->level = level;
 487         ft->node.type = FS_TYPE_FLOW_TABLE;
 488         ft->type = table_type;
 489         ft->vport = vport;
 490         ft->max_fte = max_fte;
 491         INIT_LIST_HEAD(&ft->fwd_rules);
 492         mutex_init(&ft->lock);
 493
 494         return ft;
 495 }
 496
 497 /* If reverse is false, then we search for the first flow table in the
 498  * root sub-tree from start(closest from right), else we search for the
 499  * last flow table in the root sub-tree till start(closest from left).
 500  */
 501 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 502                                                          struct list_head *start,
 503                                                          bool reverse)
 504 {
 505 #define list_advance_entry(pos, reverse)                \
 506         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 507
 508 #define list_for_each_advance_continue(pos, head, reverse)      \
 509         for (pos = list_advance_entry(pos, reverse);            \
 510              &pos->list != (head);                              \
 511              pos = list_advance_entry(pos, reverse))
 512
 513         struct fs_node *iter = list_entry(start, struct fs_node, list);
 514         struct mlx5_flow_table *ft = NULL;
 515
 516         if (!root)
 517                 return NULL;
 518
 519         list_for_each_advance_continue(iter, &root->children, reverse) {
 520                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 521                         fs_get_obj(ft, iter);
 522                         return ft;
 523                 }
 524                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 525                 if (ft)
 526                         return ft;
 527         }
 528
 529         return ft;
 530 }
 531
 532 /* If reverse if false then return the first flow table in next priority of
 533  * prio in the tree, else return the last flow table in the previous priority
 534  * of prio in the tree.
 535  */
 536 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 537 {
 538         struct mlx5_flow_table *ft = NULL;
 539         struct fs_node *curr_node;
 540         struct fs_node *parent;
 541
 542         parent = prio->node.parent;
 543         curr_node = &prio->node;
 544         while (!ft && parent) {
 545                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 546                 curr_node = parent;
 547                 parent = curr_node->parent;
 548         }
 549         return ft;
 550 }
 551
 552 /* Assuming all the tree is locked by mutex chain lock */
 553 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 554 {
 555         return find_closest_ft(prio, false);
 556 }
 557
 558 /* Assuming all the tree is locked by mutex chain lock */
 559 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 560 {
 561         return find_closest_ft(prio, true);
 562 }
 563
 564 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 565                                struct fs_prio *prio,
 566                                struct mlx5_flow_table *ft)
 567 {
 568         struct mlx5_flow_table *iter;
 569         int i = 0;
 570         int err;
 571
 572         fs_for_each_ft(iter, prio) {
 573                 i++;
 574                 err = mlx5_cmd_modify_flow_table(dev,
 575                                                  iter,
 576                                                  ft);
 577                 if (err) {
 578                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 579                                        iter->id);
 580                         /* The driver is out of sync with the FW */
 581                         if (i > 1)
 582                                 WARN_ON(true);
 583                         return err;
 584                 }
 585         }
 586         return 0;
 587 }
 588
 589 /* Connect flow tables from previous priority of prio to ft */
 590 static int connect_prev_fts(struct mlx5_core_dev *dev,
 591                             struct mlx5_flow_table *ft,
 592                             struct fs_prio *prio)
 593 {
 594         struct mlx5_flow_table *prev_ft;
 595
 596         prev_ft = find_prev_chained_ft(prio);
 597         if (prev_ft) {
 598                 struct fs_prio *prev_prio;
 599
 600                 fs_get_obj(prev_prio, prev_ft->node.parent);
 601                 return connect_fts_in_prio(dev, prev_prio, ft);
 602         }
 603         return 0;
 604 }
 605
 606 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 607                                  *prio)
 608 {
 609         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 610         int min_level = INT_MAX;
 611         int err;
 612
 613         if (root->root_ft)
 614                 min_level = root->root_ft->level;
 615
 616         if (ft->level >= min_level)
 617                 return 0;
 618
 619         err = mlx5_cmd_update_root_ft(root->dev, ft);
 620         if (err)
 621                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 622                                ft->id);
 623         else
 624                 root->root_ft = ft;
 625
 626         return err;
 627 }
 628
 629 int mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 630                                  struct mlx5_flow_destination *dest)
 631 {
 632         struct mlx5_flow_table *ft;
 633         struct mlx5_flow_group *fg;
 634         struct fs_fte *fte;
 635         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 636         int err = 0;
 637
 638         fs_get_obj(fte, rule->node.parent);
 639         if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 640                 return -EINVAL;
 641         lock_ref_node(&fte->node);
 642         fs_get_obj(fg, fte->node.parent);
 643         fs_get_obj(ft, fg->node.parent);
 644
 645         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 646         err = mlx5_cmd_update_fte(get_dev(&ft->node),
 647                                   ft, fg->id,
 648                                   modify_mask,
 649                                   fte);
 650         unlock_ref_node(&fte->node);
 651
 652         return err;
 653 }
 654
 655 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 656 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 657                              struct mlx5_flow_table *new_next_ft,
 658                              struct mlx5_flow_table *old_next_ft)
 659 {
 660         struct mlx5_flow_destination dest;
 661         struct mlx5_flow_rule *iter;
 662         int err = 0;
 663
 664         /* new_next_ft and old_next_ft could be NULL only
 665          * when we create/destroy the anchor flow table.
 666          */
 667         if (!new_next_ft || !old_next_ft)
 668                 return 0;
 669
 670         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 671         dest.ft = new_next_ft;
 672
 673         mutex_lock(&old_next_ft->lock);
 674         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 675         mutex_unlock(&old_next_ft->lock);
 676         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 677                 err = mlx5_modify_rule_destination(iter, &dest);
 678                 if (err)
 679                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 680                                new_next_ft->id);
 681         }
 682         return 0;
 683 }
 684
 685 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 686                               struct fs_prio *prio)
 687 {
 688         struct mlx5_flow_table *next_ft;
 689         int err = 0;
 690
 691         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 692
 693         if (list_empty(&prio->node.children)) {
 694                 err = connect_prev_fts(dev, ft, prio);
 695                 if (err)
 696                         return err;
 697
 698                 next_ft = find_next_chained_ft(prio);
 699                 err = connect_fwd_rules(dev, ft, next_ft);
 700                 if (err)
 701                         return err;
 702         }
 703
 704         if (MLX5_CAP_FLOWTABLE(dev,
 705                                flow_table_properties_nic_receive.modify_root))
 706                 err = update_root_ft_create(ft, prio);
 707         return err;
 708 }
 709
 710 static void list_add_flow_table(struct mlx5_flow_table *ft,
 711                                 struct fs_prio *prio)
 712 {
 713         struct list_head *prev = &prio->node.children;
 714         struct mlx5_flow_table *iter;
 715
 716         fs_for_each_ft(iter, prio) {
 717                 if (iter->level > ft->level)
 718                         break;
 719                 prev = &iter->node.list;
 720         }
 721         list_add(&ft->node.list, prev);
 722 }
 723
 724 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 725                                                         u16 vport, int prio,
 726                                                         int max_fte, u32 level)
 727 {
 728         struct mlx5_flow_table *next_ft = NULL;
 729         struct mlx5_flow_table *ft;
 730         int err;
 731         int log_table_sz;
 732         struct mlx5_flow_root_namespace *root =
 733                 find_root(&ns->node);
 734         struct fs_prio *fs_prio = NULL;
 735
 736         if (!root) {
 737                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 738                 return ERR_PTR(-ENODEV);
 739         }
 740
 741         mutex_lock(&root->chain_lock);
 742         fs_prio = find_prio(ns, prio);
 743         if (!fs_prio) {
 744                 err = -EINVAL;
 745                 goto unlock_root;
 746         }
 747         if (level >= fs_prio->num_levels) {
 748                 err = -ENOSPC;
 749                 goto unlock_root;
 750         }
 751         /* The level is related to the
 752          * priority level range.
 753          */
 754         level += fs_prio->start_level;
 755         ft = alloc_flow_table(level,
 756                               vport,
 757                               roundup_pow_of_two(max_fte),
 758                               root->table_type);
 759         if (!ft) {
 760                 err = -ENOMEM;
 761                 goto unlock_root;
 762         }
 763
 764         tree_init_node(&ft->node, 1, del_flow_table);
 765         log_table_sz = ilog2(ft->max_fte);
 766         next_ft = find_next_chained_ft(fs_prio);
 767         err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->type, ft->level,
 768                                          log_table_sz, next_ft, &ft->id);
 769         if (err)
 770                 goto free_ft;
 771
 772         err = connect_flow_table(root->dev, ft, fs_prio);
 773         if (err)
 774                 goto destroy_ft;
 775         lock_ref_node(&fs_prio->node);
 776         tree_add_node(&ft->node, &fs_prio->node);
 777         list_add_flow_table(ft, fs_prio);
 778         fs_prio->num_ft++;
 779         unlock_ref_node(&fs_prio->node);
 780         mutex_unlock(&root->chain_lock);
 781         return ft;
 782 destroy_ft:
 783         mlx5_cmd_destroy_flow_table(root->dev, ft);
 784 free_ft:
 785         kfree(ft);
 786 unlock_root:
 787         mutex_unlock(&root->chain_lock);
 788         return ERR_PTR(err);
 789 }
 790
 791 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 792                                                int prio, int max_fte,
 793                                                u32 level)
 794 {
 795         return __mlx5_create_flow_table(ns, 0, prio, max_fte, level);
 796 }
 797
 798 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
 799                                                      int prio, int max_fte,
 800                                                      u32 level, u16 vport)
 801 {
 802         return __mlx5_create_flow_table(ns, vport, prio, max_fte, level);
 803 }
 804
 805 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 806                                                             int prio,
 807                                                             int num_flow_table_entries,
 808                                                             int max_num_groups,
 809                                                             u32 level)
 810 {
 811         struct mlx5_flow_table *ft;
 812
 813         if (max_num_groups > num_flow_table_entries)
 814                 return ERR_PTR(-EINVAL);
 815
 816         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries, level);
 817         if (IS_ERR(ft))
 818                 return ft;
 819
 820         ft->autogroup.active = true;
 821         ft->autogroup.required_groups = max_num_groups;
 822
 823         return ft;
 824 }
 825 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 826
 827 /* Flow table should be locked */
 828 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 829                                                         u32 *fg_in,
 830                                                         struct list_head
 831                                                         *prev_fg,
 832                                                         bool is_auto_fg)
 833 {
 834         struct mlx5_flow_group *fg;
 835         struct mlx5_core_dev *dev = get_dev(&ft->node);
 836         int err;
 837
 838         if (!dev)
 839                 return ERR_PTR(-ENODEV);
 840
 841         fg = alloc_flow_group(fg_in);
 842         if (IS_ERR(fg))
 843                 return fg;
 844
 845         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 846         if (err) {
 847                 kfree(fg);
 848                 return ERR_PTR(err);
 849         }
 850
 851         if (ft->autogroup.active)
 852                 ft->autogroup.num_groups++;
 853         /* Add node to tree */
 854         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 855         tree_add_node(&fg->node, &ft->node);
 856         /* Add node to group list */
 857         list_add(&fg->node.list, ft->node.children.prev);
 858
 859         return fg;
 860 }
 861
 862 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 863                                                u32 *fg_in)
 864 {
 865         struct mlx5_flow_group *fg;
 866
 867         if (ft->autogroup.active)
 868                 return ERR_PTR(-EPERM);
 869
 870         lock_ref_node(&ft->node);
 871         fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
 872         unlock_ref_node(&ft->node);
 873
 874         return fg;
 875 }
 876
 877 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 878 {
 879         struct mlx5_flow_rule *rule;
 880
 881         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 882         if (!rule)
 883                 return NULL;
 884
 885         INIT_LIST_HEAD(&rule->next_ft);
 886         rule->node.type = FS_TYPE_FLOW_DEST;
 887         if (dest)
 888                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 889
 890         return rule;
 891 }
 892
 893 /* fte should not be deleted while calling this function */
 894 static struct mlx5_flow_rule *add_rule_fte(struct fs_fte *fte,
 895                                            struct mlx5_flow_group *fg,
 896                                            struct mlx5_flow_destination *dest)
 897 {
 898         struct mlx5_flow_table *ft;
 899         struct mlx5_flow_rule *rule;
 900         int modify_mask = 0;
 901         int err;
 902
 903         rule = alloc_rule(dest);
 904         if (!rule)
 905                 return ERR_PTR(-ENOMEM);
 906
 907         fs_get_obj(ft, fg->node.parent);
 908         /* Add dest to dests list- we need flow tables to be in the
 909          * end of the list for forward to next prio rules.
 910          */
 911         tree_init_node(&rule->node, 1, del_rule);
 912         if (dest && dest->type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
 913                 list_add(&rule->node.list, &fte->node.children);
 914         else
 915                 list_add_tail(&rule->node.list, &fte->node.children);
 916         if (dest) {
 917                 fte->dests_size++;
 918
 919                 modify_mask |= dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER ?
 920                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS) :
 921                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 922         }
 923
 924         if (fte->dests_size == 1 || !dest)
 925                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
 926                                           ft, fg->id, fte);
 927         else
 928                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
 929                                           ft, fg->id, modify_mask, fte);
 930         if (err)
 931                 goto free_rule;
 932
 933         fte->status |= FS_FTE_STATUS_EXISTING;
 934
 935         return rule;
 936
 937 free_rule:
 938         list_del(&rule->node.list);
 939         kfree(rule);
 940         if (dest)
 941                 fte->dests_size--;
 942         return ERR_PTR(err);
 943 }
 944
 945 /* Assumed fg is locked */
 946 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
 947                                        struct list_head **prev)
 948 {
 949         struct fs_fte *fte;
 950         unsigned int start = fg->start_index;
 951
 952         if (prev)
 953                 *prev = &fg->node.children;
 954
 955         /* assumed list is sorted by index */
 956         fs_for_each_fte(fte, fg) {
 957                 if (fte->index != start)
 958                         return start;
 959                 start++;
 960                 if (prev)
 961                         *prev = &fte->node.list;
 962         }
 963
 964         return start;
 965 }
 966
 967 /* prev is output, prev->next = new_fte */
 968 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
 969                                  u32 *match_value,
 970                                  u8 action,
 971                                  u32 flow_tag,
 972                                  struct list_head **prev)
 973 {
 974         struct fs_fte *fte;
 975         int index;
 976
 977         index = get_free_fte_index(fg, prev);
 978         fte = alloc_fte(action, flow_tag, match_value, index);
 979         if (IS_ERR(fte))
 980                 return fte;
 981
 982         return fte;
 983 }
 984
 985 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
 986                                                 u8 match_criteria_enable,
 987                                                 u32 *match_criteria)
 988 {
 989         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
 990         struct list_head *prev = &ft->node.children;
 991         unsigned int candidate_index = 0;
 992         struct mlx5_flow_group *fg;
 993         void *match_criteria_addr;
 994         unsigned int group_size = 0;
 995         u32 *in;
 996
 997         if (!ft->autogroup.active)
 998                 return ERR_PTR(-ENOENT);
 999
1000         in = mlx5_vzalloc(inlen);
1001         if (!in)
1002                 return ERR_PTR(-ENOMEM);
1003
1004         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1005                 /* We save place for flow groups in addition to max types */
1006                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1007
1008         /*  ft->max_fte == ft->autogroup.max_types */
1009         if (group_size == 0)
1010                 group_size = 1;
1011
1012         /* sorted by start_index */
1013         fs_for_each_fg(fg, ft) {
1014                 if (candidate_index + group_size > fg->start_index)
1015                         candidate_index = fg->start_index + fg->max_ftes;
1016                 else
1017                         break;
1018                 prev = &fg->node.list;
1019         }
1020
1021         if (candidate_index + group_size > ft->max_fte) {
1022                 fg = ERR_PTR(-ENOSPC);
1023                 goto out;
1024         }
1025
1026         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1027                  match_criteria_enable);
1028         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1029         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
1030                  group_size - 1);
1031         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1032                                            in, match_criteria);
1033         memcpy(match_criteria_addr, match_criteria,
1034                MLX5_ST_SZ_BYTES(fte_match_param));
1035
1036         fg = create_flow_group_common(ft, in, prev, true);
1037 out:
1038         kvfree(in);
1039         return fg;
1040 }
1041
1042 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1043                                              struct mlx5_flow_destination *dest)
1044 {
1045         struct mlx5_flow_rule *rule;
1046
1047         list_for_each_entry(rule, &fte->node.children, node.list) {
1048                 if (rule->dest_attr.type == dest->type) {
1049                         if ((dest->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1050                              dest->vport_num == rule->dest_attr.vport_num) ||
1051                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1052                              dest->ft == rule->dest_attr.ft) ||
1053                             (dest->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1054                              dest->tir_num == rule->dest_attr.tir_num))
1055                                 return rule;
1056                 }
1057         }
1058         return NULL;
1059 }
1060
1061 static struct mlx5_flow_rule *add_rule_fg(struct mlx5_flow_group *fg,
1062                                           u32 *match_value,
1063                                           u8 action,
1064                                           u32 flow_tag,
1065                                           struct mlx5_flow_destination *dest)
1066 {
1067         struct fs_fte *fte;
1068         struct mlx5_flow_rule *rule;
1069         struct mlx5_flow_table *ft;
1070         struct list_head *prev;
1071
1072         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1073         fs_for_each_fte(fte, fg) {
1074                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1075                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1076                     action == fte->action && flow_tag == fte->flow_tag) {
1077                         rule = find_flow_rule(fte, dest);
1078                         if (rule) {
1079                                 atomic_inc(&rule->node.refcount);
1080                                 unlock_ref_node(&fte->node);
1081                                 unlock_ref_node(&fg->node);
1082                                 return rule;
1083                         }
1084                         rule = add_rule_fte(fte, fg, dest);
1085                         unlock_ref_node(&fte->node);
1086                         if (IS_ERR(rule))
1087                                 goto unlock_fg;
1088                         else
1089                                 goto add_rule;
1090                 }
1091                 unlock_ref_node(&fte->node);
1092         }
1093         fs_get_obj(ft, fg->node.parent);
1094         if (fg->num_ftes >= fg->max_ftes) {
1095                 rule = ERR_PTR(-ENOSPC);
1096                 goto unlock_fg;
1097         }
1098
1099         fte = create_fte(fg, match_value, action, flow_tag, &prev);
1100         if (IS_ERR(fte)) {
1101                 rule = (void *)fte;
1102                 goto unlock_fg;
1103         }
1104         tree_init_node(&fte->node, 0, del_fte);
1105         rule = add_rule_fte(fte, fg, dest);
1106         if (IS_ERR(rule)) {
1107                 kfree(fte);
1108                 goto unlock_fg;
1109         }
1110
1111         fg->num_ftes++;
1112
1113         tree_add_node(&fte->node, &fg->node);
1114         list_add(&fte->node.list, prev);
1115 add_rule:
1116         tree_add_node(&rule->node, &fte->node);
1117 unlock_fg:
1118         unlock_ref_node(&fg->node);
1119         return rule;
1120 }
1121
1122 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_rule *rule)
1123 {
1124         struct mlx5_flow_rule *dst;
1125         struct fs_fte *fte;
1126
1127         fs_get_obj(fte, rule->node.parent);
1128
1129         fs_for_each_dst(dst, fte) {
1130                 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1131                         return dst->dest_attr.counter;
1132         }
1133
1134         return NULL;
1135 }
1136
1137 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1138 {
1139         if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1140                 return !counter;
1141
1142         if (!counter)
1143                 return false;
1144
1145         /* Hardware support counter for a drop action only */
1146         return action == (MLX5_FLOW_CONTEXT_ACTION_DROP | MLX5_FLOW_CONTEXT_ACTION_COUNT);
1147 }
1148
1149 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1150                           u32 action,
1151                           struct mlx5_flow_table *ft)
1152 {
1153         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1154                 return counter_is_valid(dest->counter, action);
1155
1156         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1157                 return true;
1158
1159         if (!dest || ((dest->type ==
1160             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1161             (dest->ft->level <= ft->level)))
1162                 return false;
1163         return true;
1164 }
1165
1166 static struct mlx5_flow_rule *
1167 _mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1168                    struct mlx5_flow_spec *spec,
1169                     u32 action,
1170                     u32 flow_tag,
1171                     struct mlx5_flow_destination *dest)
1172 {
1173         struct mlx5_flow_group *g;
1174         struct mlx5_flow_rule *rule;
1175
1176         if (!dest_is_valid(dest, action, ft))
1177                 return ERR_PTR(-EINVAL);
1178
1179         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1180         fs_for_each_fg(g, ft)
1181                 if (compare_match_criteria(g->mask.match_criteria_enable,
1182                                            spec->match_criteria_enable,
1183                                            g->mask.match_criteria,
1184                                            spec->match_criteria)) {
1185                         rule = add_rule_fg(g, spec->match_value,
1186                                            action, flow_tag, dest);
1187                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1188                                 goto unlock;
1189                 }
1190
1191         g = create_autogroup(ft, spec->match_criteria_enable,
1192                              spec->match_criteria);
1193         if (IS_ERR(g)) {
1194                 rule = (void *)g;
1195                 goto unlock;
1196         }
1197
1198         rule = add_rule_fg(g, spec->match_value,
1199                            action, flow_tag, dest);
1200         if (IS_ERR(rule)) {
1201                 /* Remove assumes refcount > 0 and autogroup creates a group
1202                  * with a refcount = 0.
1203                  */
1204                 unlock_ref_node(&ft->node);
1205                 tree_get_node(&g->node);
1206                 tree_remove_node(&g->node);
1207                 return rule;
1208         }
1209 unlock:
1210         unlock_ref_node(&ft->node);
1211         return rule;
1212 }
1213
1214 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1215 {
1216         return ((ft->type == FS_FT_NIC_RX) &&
1217                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1218 }
1219
1220 struct mlx5_flow_rule *
1221 mlx5_add_flow_rule(struct mlx5_flow_table *ft,
1222                    struct mlx5_flow_spec *spec,
1223                    u32 action,
1224                    u32 flow_tag,
1225                    struct mlx5_flow_destination *dest)
1226 {
1227         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1228         struct mlx5_flow_destination gen_dest;
1229         struct mlx5_flow_table *next_ft = NULL;
1230         struct mlx5_flow_rule *rule = NULL;
1231         u32 sw_action = action;
1232         struct fs_prio *prio;
1233
1234         fs_get_obj(prio, ft->node.parent);
1235         if (action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1236                 if (!fwd_next_prio_supported(ft))
1237                         return ERR_PTR(-EOPNOTSUPP);
1238                 if (dest)
1239                         return ERR_PTR(-EINVAL);
1240                 mutex_lock(&root->chain_lock);
1241                 next_ft = find_next_chained_ft(prio);
1242                 if (next_ft) {
1243                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1244                         gen_dest.ft = next_ft;
1245                         dest = &gen_dest;
1246                         action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1247                 } else {
1248                         mutex_unlock(&root->chain_lock);
1249                         return ERR_PTR(-EOPNOTSUPP);
1250                 }
1251         }
1252
1253         rule = _mlx5_add_flow_rule(ft, spec, action, flow_tag, dest);
1254
1255         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1256                 if (!IS_ERR_OR_NULL(rule) &&
1257                     (list_empty(&rule->next_ft))) {
1258                         mutex_lock(&next_ft->lock);
1259                         list_add(&rule->next_ft, &next_ft->fwd_rules);
1260                         mutex_unlock(&next_ft->lock);
1261                         rule->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1262                 }
1263                 mutex_unlock(&root->chain_lock);
1264         }
1265         return rule;
1266 }
1267 EXPORT_SYMBOL(mlx5_add_flow_rule);
1268
1269 void mlx5_del_flow_rule(struct mlx5_flow_rule *rule)
1270 {
1271         tree_remove_node(&rule->node);
1272 }
1273 EXPORT_SYMBOL(mlx5_del_flow_rule);
1274
1275 /* Assuming prio->node.children(flow tables) is sorted by level */
1276 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1277 {
1278         struct fs_prio *prio;
1279
1280         fs_get_obj(prio, ft->node.parent);
1281
1282         if (!list_is_last(&ft->node.list, &prio->node.children))
1283                 return list_next_entry(ft, node.list);
1284         return find_next_chained_ft(prio);
1285 }
1286
1287 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1288 {
1289         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1290         struct mlx5_flow_table *new_root_ft = NULL;
1291
1292         if (root->root_ft != ft)
1293                 return 0;
1294
1295         new_root_ft = find_next_ft(ft);
1296         if (new_root_ft) {
1297                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1298
1299                 if (err) {
1300                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1301                                        ft->id);
1302                         return err;
1303                 }
1304         }
1305         root->root_ft = new_root_ft;
1306         return 0;
1307 }
1308
1309 /* Connect flow table from previous priority to
1310  * the next flow table.
1311  */
1312 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1313 {
1314         struct mlx5_core_dev *dev = get_dev(&ft->node);
1315         struct mlx5_flow_table *next_ft;
1316         struct fs_prio *prio;
1317         int err = 0;
1318
1319         err = update_root_ft_destroy(ft);
1320         if (err)
1321                 return err;
1322
1323         fs_get_obj(prio, ft->node.parent);
1324         if  (!(list_first_entry(&prio->node.children,
1325                                 struct mlx5_flow_table,
1326                                 node.list) == ft))
1327                 return 0;
1328
1329         next_ft = find_next_chained_ft(prio);
1330         err = connect_fwd_rules(dev, next_ft, ft);
1331         if (err)
1332                 return err;
1333
1334         err = connect_prev_fts(dev, next_ft, prio);
1335         if (err)
1336                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1337                                ft->id);
1338         return err;
1339 }
1340
1341 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1342 {
1343         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1344         int err = 0;
1345
1346         mutex_lock(&root->chain_lock);
1347         err = disconnect_flow_table(ft);
1348         if (err) {
1349                 mutex_unlock(&root->chain_lock);
1350                 return err;
1351         }
1352         if (tree_remove_node(&ft->node))
1353                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1354                                ft->id);
1355         mutex_unlock(&root->chain_lock);
1356
1357         return err;
1358 }
1359 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1360
1361 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1362 {
1363         if (tree_remove_node(&fg->node))
1364                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1365                                fg->id);
1366 }
1367
1368 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1369                                                     enum mlx5_flow_namespace_type type)
1370 {
1371         struct mlx5_flow_steering *steering = dev->priv.steering;
1372         struct mlx5_flow_root_namespace *root_ns;
1373         int prio;
1374         struct fs_prio *fs_prio;
1375         struct mlx5_flow_namespace *ns;
1376
1377         if (!steering)
1378                 return NULL;
1379
1380         switch (type) {
1381         case MLX5_FLOW_NAMESPACE_BYPASS:
1382         case MLX5_FLOW_NAMESPACE_OFFLOADS:
1383         case MLX5_FLOW_NAMESPACE_ETHTOOL:
1384         case MLX5_FLOW_NAMESPACE_KERNEL:
1385         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1386         case MLX5_FLOW_NAMESPACE_ANCHOR:
1387                 prio = type;
1388                 break;
1389         case MLX5_FLOW_NAMESPACE_FDB:
1390                 if (steering->fdb_root_ns)
1391                         return &steering->fdb_root_ns->ns;
1392                 else
1393                         return NULL;
1394         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1395                 if (steering->esw_egress_root_ns)
1396                         return &steering->esw_egress_root_ns->ns;
1397                 else
1398                         return NULL;
1399         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1400                 if (steering->esw_ingress_root_ns)
1401                         return &steering->esw_ingress_root_ns->ns;
1402                 else
1403                         return NULL;
1404         default:
1405                 return NULL;
1406         }
1407
1408         root_ns = steering->root_ns;
1409         if (!root_ns)
1410                 return NULL;
1411
1412         fs_prio = find_prio(&root_ns->ns, prio);
1413         if (!fs_prio)
1414                 return NULL;
1415
1416         ns = list_first_entry(&fs_prio->node.children,
1417                               typeof(*ns),
1418                               node.list);
1419
1420         return ns;
1421 }
1422 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1423
1424 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1425                                       unsigned int prio, int num_levels)
1426 {
1427         struct fs_prio *fs_prio;
1428
1429         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1430         if (!fs_prio)
1431                 return ERR_PTR(-ENOMEM);
1432
1433         fs_prio->node.type = FS_TYPE_PRIO;
1434         tree_init_node(&fs_prio->node, 1, NULL);
1435         tree_add_node(&fs_prio->node, &ns->node);
1436         fs_prio->num_levels = num_levels;
1437         fs_prio->prio = prio;
1438         list_add_tail(&fs_prio->node.list, &ns->node.children);
1439
1440         return fs_prio;
1441 }
1442
1443 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1444                                                      *ns)
1445 {
1446         ns->node.type = FS_TYPE_NAMESPACE;
1447
1448         return ns;
1449 }
1450
1451 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1452 {
1453         struct mlx5_flow_namespace      *ns;
1454
1455         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1456         if (!ns)
1457                 return ERR_PTR(-ENOMEM);
1458
1459         fs_init_namespace(ns);
1460         tree_init_node(&ns->node, 1, NULL);
1461         tree_add_node(&ns->node, &prio->node);
1462         list_add_tail(&ns->node.list, &prio->node.children);
1463
1464         return ns;
1465 }
1466
1467 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1468                              struct init_tree_node *prio_metadata)
1469 {
1470         struct fs_prio *fs_prio;
1471         int i;
1472
1473         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1474                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1475                 if (IS_ERR(fs_prio))
1476                         return PTR_ERR(fs_prio);
1477         }
1478         return 0;
1479 }
1480
1481 #define FLOW_TABLE_BIT_SZ 1
1482 #define GET_FLOW_TABLE_CAP(dev, offset) \
1483         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1484                         offset / 32)) >>                                        \
1485           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1486 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1487 {
1488         int i;
1489
1490         for (i = 0; i < caps->arr_sz; i++) {
1491                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1492                         return false;
1493         }
1494         return true;
1495 }
1496
1497 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
1498                                     struct init_tree_node *init_node,
1499                                     struct fs_node *fs_parent_node,
1500                                     struct init_tree_node *init_parent_node,
1501                                     int prio)
1502 {
1503         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
1504                                               flow_table_properties_nic_receive.
1505                                               max_ft_level);
1506         struct mlx5_flow_namespace *fs_ns;
1507         struct fs_prio *fs_prio;
1508         struct fs_node *base;
1509         int i;
1510         int err;
1511
1512         if (init_node->type == FS_TYPE_PRIO) {
1513                 if ((init_node->min_ft_level > max_ft_level) ||
1514                     !has_required_caps(steering->dev, &init_node->caps))
1515                         return 0;
1516
1517                 fs_get_obj(fs_ns, fs_parent_node);
1518                 if (init_node->num_leaf_prios)
1519                         return create_leaf_prios(fs_ns, prio, init_node);
1520                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1521                 if (IS_ERR(fs_prio))
1522                         return PTR_ERR(fs_prio);
1523                 base = &fs_prio->node;
1524         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1525                 fs_get_obj(fs_prio, fs_parent_node);
1526                 fs_ns = fs_create_namespace(fs_prio);
1527                 if (IS_ERR(fs_ns))
1528                         return PTR_ERR(fs_ns);
1529                 base = &fs_ns->node;
1530         } else {
1531                 return -EINVAL;
1532         }
1533         prio = 0;
1534         for (i = 0; i < init_node->ar_size; i++) {
1535                 err = init_root_tree_recursive(steering, &init_node->children[i],
1536                                                base, init_node, prio);
1537                 if (err)
1538                         return err;
1539                 if (init_node->children[i].type == FS_TYPE_PRIO &&
1540                     init_node->children[i].num_leaf_prios) {
1541                         prio += init_node->children[i].num_leaf_prios;
1542                 }
1543         }
1544
1545         return 0;
1546 }
1547
1548 static int init_root_tree(struct mlx5_flow_steering *steering,
1549                           struct init_tree_node *init_node,
1550                           struct fs_node *fs_parent_node)
1551 {
1552         int i;
1553         struct mlx5_flow_namespace *fs_ns;
1554         int err;
1555
1556         fs_get_obj(fs_ns, fs_parent_node);
1557         for (i = 0; i < init_node->ar_size; i++) {
1558                 err = init_root_tree_recursive(steering, &init_node->children[i],
1559                                                &fs_ns->node,
1560                                                init_node, i);
1561                 if (err)
1562                         return err;
1563         }
1564         return 0;
1565 }
1566
1567 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
1568                                                        enum fs_flow_table_type
1569                                                        table_type)
1570 {
1571         struct mlx5_flow_root_namespace *root_ns;
1572         struct mlx5_flow_namespace *ns;
1573
1574         /* Create the root namespace */
1575         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1576         if (!root_ns)
1577                 return NULL;
1578
1579         root_ns->dev = steering->dev;
1580         root_ns->table_type = table_type;
1581
1582         ns = &root_ns->ns;
1583         fs_init_namespace(ns);
1584         mutex_init(&root_ns->chain_lock);
1585         tree_init_node(&ns->node, 1, NULL);
1586         tree_add_node(&ns->node, NULL);
1587
1588         return root_ns;
1589 }
1590
1591 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1592
1593 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1594 {
1595         struct fs_prio *prio;
1596
1597         fs_for_each_prio(prio, ns) {
1598                  /* This updates prio start_level and num_levels */
1599                 set_prio_attrs_in_prio(prio, acc_level);
1600                 acc_level += prio->num_levels;
1601         }
1602         return acc_level;
1603 }
1604
1605 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1606 {
1607         struct mlx5_flow_namespace *ns;
1608         int acc_level_ns = acc_level;
1609
1610         prio->start_level = acc_level;
1611         fs_for_each_ns(ns, prio)
1612                 /* This updates start_level and num_levels of ns's priority descendants */
1613                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1614         if (!prio->num_levels)
1615                 prio->num_levels = acc_level_ns - prio->start_level;
1616         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1617 }
1618
1619 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1620 {
1621         struct mlx5_flow_namespace *ns = &root_ns->ns;
1622         struct fs_prio *prio;
1623         int start_level = 0;
1624
1625         fs_for_each_prio(prio, ns) {
1626                 set_prio_attrs_in_prio(prio, start_level);
1627                 start_level += prio->num_levels;
1628         }
1629 }
1630
1631 #define ANCHOR_PRIO 0
1632 #define ANCHOR_SIZE 1
1633 #define ANCHOR_LEVEL 0
1634 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
1635 {
1636         struct mlx5_flow_namespace *ns = NULL;
1637         struct mlx5_flow_table *ft;
1638
1639         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1640         if (!ns)
1641                 return -EINVAL;
1642         ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL);
1643         if (IS_ERR(ft)) {
1644                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
1645                 return PTR_ERR(ft);
1646         }
1647         return 0;
1648 }
1649
1650 static int init_root_ns(struct mlx5_flow_steering *steering)
1651 {
1652
1653         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
1654         if (IS_ERR_OR_NULL(steering->root_ns))
1655                 goto cleanup;
1656
1657         if (init_root_tree(steering, &root_fs, &steering->root_ns->ns.node))
1658                 goto cleanup;
1659
1660         set_prio_attrs(steering->root_ns);
1661
1662         if (create_anchor_flow_table(steering))
1663                 goto cleanup;
1664
1665         return 0;
1666
1667 cleanup:
1668         mlx5_cleanup_fs(steering->dev);
1669         return -ENOMEM;
1670 }
1671
1672 static void clean_tree(struct fs_node *node)
1673 {
1674         if (node) {
1675                 struct fs_node *iter;
1676                 struct fs_node *temp;
1677
1678                 list_for_each_entry_safe(iter, temp, &node->children, list)
1679                         clean_tree(iter);
1680                 tree_remove_node(node);
1681         }
1682 }
1683
1684 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
1685 {
1686         if (!root_ns)
1687                 return;
1688
1689         clean_tree(&root_ns->ns.node);
1690 }
1691
1692 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1693 {
1694         struct mlx5_flow_steering *steering = dev->priv.steering;
1695
1696         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1697                 return;
1698
1699         cleanup_root_ns(steering->root_ns);
1700         cleanup_root_ns(steering->esw_egress_root_ns);
1701         cleanup_root_ns(steering->esw_ingress_root_ns);
1702         cleanup_root_ns(steering->fdb_root_ns);
1703         mlx5_cleanup_fc_stats(dev);
1704         kfree(steering);
1705 }
1706
1707 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
1708 {
1709         struct fs_prio *prio;
1710
1711         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
1712         if (!steering->fdb_root_ns)
1713                 return -ENOMEM;
1714
1715         /* Create single prio */
1716         prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
1717         if (IS_ERR(prio)) {
1718                 cleanup_root_ns(steering->fdb_root_ns);
1719                 steering->fdb_root_ns = NULL;
1720                 return PTR_ERR(prio);
1721         } else {
1722                 return 0;
1723         }
1724 }
1725
1726 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
1727 {
1728         struct fs_prio *prio;
1729
1730         steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
1731         if (!steering->esw_egress_root_ns)
1732                 return -ENOMEM;
1733
1734         /* create 1 prio*/
1735         prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
1736                               MLX5_TOTAL_VPORTS(steering->dev));
1737         if (IS_ERR(prio))
1738                 return PTR_ERR(prio);
1739         else
1740                 return 0;
1741 }
1742
1743 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
1744 {
1745         struct fs_prio *prio;
1746
1747         steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
1748         if (!steering->esw_ingress_root_ns)
1749                 return -ENOMEM;
1750
1751         /* create 1 prio*/
1752         prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
1753                               MLX5_TOTAL_VPORTS(steering->dev));
1754         if (IS_ERR(prio))
1755                 return PTR_ERR(prio);
1756         else
1757                 return 0;
1758 }
1759
1760 int mlx5_init_fs(struct mlx5_core_dev *dev)
1761 {
1762         struct mlx5_flow_steering *steering;
1763         int err = 0;
1764
1765         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1766                 return 0;
1767
1768         err = mlx5_init_fc_stats(dev);
1769         if (err)
1770                 return err;
1771
1772         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
1773         if (!steering)
1774                 return -ENOMEM;
1775         steering->dev = dev;
1776         dev->priv.steering = steering;
1777
1778         if (MLX5_CAP_GEN(dev, nic_flow_table) &&
1779             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
1780                 err = init_root_ns(steering);
1781                 if (err)
1782                         goto err;
1783         }
1784
1785         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
1786                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
1787                         err = init_fdb_root_ns(steering);
1788                         if (err)
1789                                 goto err;
1790                 }
1791                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
1792                         err = init_egress_acl_root_ns(steering);
1793                         if (err)
1794                                 goto err;
1795                 }
1796                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
1797                         err = init_ingress_acl_root_ns(steering);
1798                         if (err)
1799                                 goto err;
1800                 }
1801         }
1802
1803         return 0;
1804 err:
1805         mlx5_cleanup_fs(dev);
1806         return err;
1807 }