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 11
  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 #define LAG_PRIO_NUM_LEVELS 1
 100 #define LAG_NUM_PRIOS 1
 101 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 102
 103 struct node_caps {
 104         size_t  arr_sz;
 105         long    *caps;
 106 };
 107 static struct init_tree_node {
 108         enum fs_node_type       type;
 109         struct init_tree_node *children;
 110         int ar_size;
 111         struct node_caps caps;
 112         int min_ft_level;
 113         int num_leaf_prios;
 114         int prio;
 115         int num_levels;
 116 } root_fs = {
 117         .type = FS_TYPE_NAMESPACE,
 118         .ar_size = 7,
 119         .children = (struct init_tree_node[]) {
 120                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 121                          FS_CHAINING_CAPS,
 122                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 123                                                   BY_PASS_PRIO_NUM_LEVELS))),
 124                 ADD_PRIO(0, LAG_MIN_LEVEL, 0,
 125                          FS_CHAINING_CAPS,
 126                          ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 127                                                   LAG_PRIO_NUM_LEVELS))),
 128                 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
 129                          ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
 130                 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
 131                          FS_CHAINING_CAPS,
 132                          ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 133                                                   ETHTOOL_PRIO_NUM_LEVELS))),
 134                 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 135                          ADD_NS(ADD_MULTIPLE_PRIO(1, 1),
 136                                 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 137                                                   KERNEL_NIC_PRIO_NUM_LEVELS))),
 138                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 139                          FS_CHAINING_CAPS,
 140                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
 141                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 142                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
 143         }
 144 };
 145
 146 enum fs_i_mutex_lock_class {
 147         FS_MUTEX_GRANDPARENT,
 148         FS_MUTEX_PARENT,
 149         FS_MUTEX_CHILD
 150 };
 151
 152 static void del_rule(struct fs_node *node);
 153 static void del_flow_table(struct fs_node *node);
 154 static void del_flow_group(struct fs_node *node);
 155 static void del_fte(struct fs_node *node);
 156 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 157                                 struct mlx5_flow_destination *d2);
 158 static struct mlx5_flow_rule *
 159 find_flow_rule(struct fs_fte *fte,
 160                struct mlx5_flow_destination *dest);
 161
 162 static void tree_init_node(struct fs_node *node,
 163                            unsigned int refcount,
 164                            void (*remove_func)(struct fs_node *))
 165 {
 166         atomic_set(&node->refcount, refcount);
 167         INIT_LIST_HEAD(&node->list);
 168         INIT_LIST_HEAD(&node->children);
 169         mutex_init(&node->lock);
 170         node->remove_func = remove_func;
 171 }
 172
 173 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 174 {
 175         if (parent)
 176                 atomic_inc(&parent->refcount);
 177         node->parent = parent;
 178
 179         /* Parent is the root */
 180         if (!parent)
 181                 node->root = node;
 182         else
 183                 node->root = parent->root;
 184 }
 185
 186 static void tree_get_node(struct fs_node *node)
 187 {
 188         atomic_inc(&node->refcount);
 189 }
 190
 191 static void nested_lock_ref_node(struct fs_node *node,
 192                                  enum fs_i_mutex_lock_class class)
 193 {
 194         if (node) {
 195                 mutex_lock_nested(&node->lock, class);
 196                 atomic_inc(&node->refcount);
 197         }
 198 }
 199
 200 static void lock_ref_node(struct fs_node *node)
 201 {
 202         if (node) {
 203                 mutex_lock(&node->lock);
 204                 atomic_inc(&node->refcount);
 205         }
 206 }
 207
 208 static void unlock_ref_node(struct fs_node *node)
 209 {
 210         if (node) {
 211                 atomic_dec(&node->refcount);
 212                 mutex_unlock(&node->lock);
 213         }
 214 }
 215
 216 static void tree_put_node(struct fs_node *node)
 217 {
 218         struct fs_node *parent_node = node->parent;
 219
 220         lock_ref_node(parent_node);
 221         if (atomic_dec_and_test(&node->refcount)) {
 222                 if (parent_node)
 223                         list_del_init(&node->list);
 224                 if (node->remove_func)
 225                         node->remove_func(node);
 226                 kfree(node);
 227                 node = NULL;
 228         }
 229         unlock_ref_node(parent_node);
 230         if (!node && parent_node)
 231                 tree_put_node(parent_node);
 232 }
 233
 234 static int tree_remove_node(struct fs_node *node)
 235 {
 236         if (atomic_read(&node->refcount) > 1) {
 237                 atomic_dec(&node->refcount);
 238                 return -EEXIST;
 239         }
 240         tree_put_node(node);
 241         return 0;
 242 }
 243
 244 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 245                                  unsigned int prio)
 246 {
 247         struct fs_prio *iter_prio;
 248
 249         fs_for_each_prio(iter_prio, ns) {
 250                 if (iter_prio->prio == prio)
 251                         return iter_prio;
 252         }
 253
 254         return NULL;
 255 }
 256
 257 static bool masked_memcmp(void *mask, void *val1, void *val2, size_t size)
 258 {
 259         unsigned int i;
 260
 261         for (i = 0; i < size; i++, mask++, val1++, val2++)
 262                 if ((*((u8 *)val1) & (*(u8 *)mask)) !=
 263                     ((*(u8 *)val2) & (*(u8 *)mask)))
 264                         return false;
 265
 266         return true;
 267 }
 268
 269 static bool compare_match_value(struct mlx5_flow_group_mask *mask,
 270                                 void *fte_param1, void *fte_param2)
 271 {
 272         if (mask->match_criteria_enable &
 273             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS) {
 274                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 275                                                 fte_param1, outer_headers);
 276                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 277                                                 fte_param2, outer_headers);
 278                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 279                                               mask->match_criteria, outer_headers);
 280
 281                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 282                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 283                         return false;
 284         }
 285
 286         if (mask->match_criteria_enable &
 287             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) {
 288                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 289                                                 fte_param1, misc_parameters);
 290                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 291                                                 fte_param2, misc_parameters);
 292                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 293                                           mask->match_criteria, misc_parameters);
 294
 295                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 296                                    MLX5_ST_SZ_BYTES(fte_match_set_misc)))
 297                         return false;
 298         }
 299
 300         if (mask->match_criteria_enable &
 301             1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) {
 302                 void *fte_match1 = MLX5_ADDR_OF(fte_match_param,
 303                                                 fte_param1, inner_headers);
 304                 void *fte_match2 = MLX5_ADDR_OF(fte_match_param,
 305                                                 fte_param2, inner_headers);
 306                 void *fte_mask = MLX5_ADDR_OF(fte_match_param,
 307                                           mask->match_criteria, inner_headers);
 308
 309                 if (!masked_memcmp(fte_mask, fte_match1, fte_match2,
 310                                    MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4)))
 311                         return false;
 312         }
 313         return true;
 314 }
 315
 316 static bool compare_match_criteria(u8 match_criteria_enable1,
 317                                    u8 match_criteria_enable2,
 318                                    void *mask1, void *mask2)
 319 {
 320         return match_criteria_enable1 == match_criteria_enable2 &&
 321                 !memcmp(mask1, mask2, MLX5_ST_SZ_BYTES(fte_match_param));
 322 }
 323
 324 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 325 {
 326         struct fs_node *root;
 327         struct mlx5_flow_namespace *ns;
 328
 329         root = node->root;
 330
 331         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 332                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 333                 return NULL;
 334         }
 335
 336         ns = container_of(root, struct mlx5_flow_namespace, node);
 337         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 338 }
 339
 340 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 341 {
 342         struct mlx5_flow_root_namespace *root = find_root(node);
 343
 344         if (root)
 345                 return root->dev;
 346         return NULL;
 347 }
 348
 349 static void del_flow_table(struct fs_node *node)
 350 {
 351         struct mlx5_flow_table *ft;
 352         struct mlx5_core_dev *dev;
 353         struct fs_prio *prio;
 354         int err;
 355
 356         fs_get_obj(ft, node);
 357         dev = get_dev(&ft->node);
 358
 359         err = mlx5_cmd_destroy_flow_table(dev, ft);
 360         if (err)
 361                 mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 362         fs_get_obj(prio, ft->node.parent);
 363         prio->num_ft--;
 364 }
 365
 366 static void del_rule(struct fs_node *node)
 367 {
 368         struct mlx5_flow_rule *rule;
 369         struct mlx5_flow_table *ft;
 370         struct mlx5_flow_group *fg;
 371         struct fs_fte *fte;
 372         u32     *match_value;
 373         int modify_mask;
 374         struct mlx5_core_dev *dev = get_dev(node);
 375         int match_len = MLX5_ST_SZ_BYTES(fte_match_param);
 376         int err;
 377         bool update_fte = false;
 378
 379         match_value = mlx5_vzalloc(match_len);
 380         if (!match_value) {
 381                 mlx5_core_warn(dev, "failed to allocate inbox\n");
 382                 return;
 383         }
 384
 385         fs_get_obj(rule, node);
 386         fs_get_obj(fte, rule->node.parent);
 387         fs_get_obj(fg, fte->node.parent);
 388         memcpy(match_value, fte->val, sizeof(fte->val));
 389         fs_get_obj(ft, fg->node.parent);
 390         list_del(&rule->node.list);
 391         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 392                 mutex_lock(&rule->dest_attr.ft->lock);
 393                 list_del(&rule->next_ft);
 394                 mutex_unlock(&rule->dest_attr.ft->lock);
 395         }
 396
 397         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 398             --fte->dests_size) {
 399                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
 400                 fte->action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 401                 update_fte = true;
 402                 goto out;
 403         }
 404
 405         if ((fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 406             --fte->dests_size) {
 407                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST),
 408                 update_fte = true;
 409         }
 410 out:
 411         if (update_fte && fte->dests_size) {
 412                 err = mlx5_cmd_update_fte(dev, ft, fg->id, modify_mask, fte);
 413                 if (err)
 414                         mlx5_core_warn(dev,
 415                                        "%s can't del rule fg id=%d fte_index=%d\n",
 416                                        __func__, fg->id, fte->index);
 417         }
 418         kvfree(match_value);
 419 }
 420
 421 static void del_fte(struct fs_node *node)
 422 {
 423         struct mlx5_flow_table *ft;
 424         struct mlx5_flow_group *fg;
 425         struct mlx5_core_dev *dev;
 426         struct fs_fte *fte;
 427         int err;
 428
 429         fs_get_obj(fte, node);
 430         fs_get_obj(fg, fte->node.parent);
 431         fs_get_obj(ft, fg->node.parent);
 432
 433         dev = get_dev(&ft->node);
 434         err = mlx5_cmd_delete_fte(dev, ft,
 435                                   fte->index);
 436         if (err)
 437                 mlx5_core_warn(dev,
 438                                "flow steering can't delete fte in index %d of flow group id %d\n",
 439                                fte->index, fg->id);
 440
 441         fte->status = 0;
 442         fg->num_ftes--;
 443 }
 444
 445 static void del_flow_group(struct fs_node *node)
 446 {
 447         struct mlx5_flow_group *fg;
 448         struct mlx5_flow_table *ft;
 449         struct mlx5_core_dev *dev;
 450
 451         fs_get_obj(fg, node);
 452         fs_get_obj(ft, fg->node.parent);
 453         dev = get_dev(&ft->node);
 454
 455         if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id))
 456                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 457                                fg->id, ft->id);
 458 }
 459
 460 static struct fs_fte *alloc_fte(u8 action,
 461                                 u32 flow_tag,
 462                                 u32 *match_value,
 463                                 unsigned int index)
 464 {
 465         struct fs_fte *fte;
 466
 467         fte = kzalloc(sizeof(*fte), GFP_KERNEL);
 468         if (!fte)
 469                 return ERR_PTR(-ENOMEM);
 470
 471         memcpy(fte->val, match_value, sizeof(fte->val));
 472         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 473         fte->flow_tag = flow_tag;
 474         fte->index = index;
 475         fte->action = action;
 476
 477         return fte;
 478 }
 479
 480 static struct mlx5_flow_group *alloc_flow_group(u32 *create_fg_in)
 481 {
 482         struct mlx5_flow_group *fg;
 483         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
 484                                             create_fg_in, match_criteria);
 485         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
 486                                             create_fg_in,
 487                                             match_criteria_enable);
 488         fg = kzalloc(sizeof(*fg), GFP_KERNEL);
 489         if (!fg)
 490                 return ERR_PTR(-ENOMEM);
 491
 492         fg->mask.match_criteria_enable = match_criteria_enable;
 493         memcpy(&fg->mask.match_criteria, match_criteria,
 494                sizeof(fg->mask.match_criteria));
 495         fg->node.type =  FS_TYPE_FLOW_GROUP;
 496         fg->start_index = MLX5_GET(create_flow_group_in, create_fg_in,
 497                                    start_flow_index);
 498         fg->max_ftes = MLX5_GET(create_flow_group_in, create_fg_in,
 499                                 end_flow_index) - fg->start_index + 1;
 500         return fg;
 501 }
 502
 503 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 504                                                 enum fs_flow_table_type table_type,
 505                                                 enum fs_flow_table_op_mod op_mod)
 506 {
 507         struct mlx5_flow_table *ft;
 508
 509         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 510         if (!ft)
 511                 return NULL;
 512
 513         ft->level = level;
 514         ft->node.type = FS_TYPE_FLOW_TABLE;
 515         ft->op_mod = op_mod;
 516         ft->type = table_type;
 517         ft->vport = vport;
 518         ft->max_fte = max_fte;
 519         INIT_LIST_HEAD(&ft->fwd_rules);
 520         mutex_init(&ft->lock);
 521
 522         return ft;
 523 }
 524
 525 /* If reverse is false, then we search for the first flow table in the
 526  * root sub-tree from start(closest from right), else we search for the
 527  * last flow table in the root sub-tree till start(closest from left).
 528  */
 529 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 530                                                          struct list_head *start,
 531                                                          bool reverse)
 532 {
 533 #define list_advance_entry(pos, reverse)                \
 534         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 535
 536 #define list_for_each_advance_continue(pos, head, reverse)      \
 537         for (pos = list_advance_entry(pos, reverse);            \
 538              &pos->list != (head);                              \
 539              pos = list_advance_entry(pos, reverse))
 540
 541         struct fs_node *iter = list_entry(start, struct fs_node, list);
 542         struct mlx5_flow_table *ft = NULL;
 543
 544         if (!root)
 545                 return NULL;
 546
 547         list_for_each_advance_continue(iter, &root->children, reverse) {
 548                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 549                         fs_get_obj(ft, iter);
 550                         return ft;
 551                 }
 552                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 553                 if (ft)
 554                         return ft;
 555         }
 556
 557         return ft;
 558 }
 559
 560 /* If reverse if false then return the first flow table in next priority of
 561  * prio in the tree, else return the last flow table in the previous priority
 562  * of prio in the tree.
 563  */
 564 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 565 {
 566         struct mlx5_flow_table *ft = NULL;
 567         struct fs_node *curr_node;
 568         struct fs_node *parent;
 569
 570         parent = prio->node.parent;
 571         curr_node = &prio->node;
 572         while (!ft && parent) {
 573                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 574                 curr_node = parent;
 575                 parent = curr_node->parent;
 576         }
 577         return ft;
 578 }
 579
 580 /* Assuming all the tree is locked by mutex chain lock */
 581 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 582 {
 583         return find_closest_ft(prio, false);
 584 }
 585
 586 /* Assuming all the tree is locked by mutex chain lock */
 587 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 588 {
 589         return find_closest_ft(prio, true);
 590 }
 591
 592 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 593                                struct fs_prio *prio,
 594                                struct mlx5_flow_table *ft)
 595 {
 596         struct mlx5_flow_table *iter;
 597         int i = 0;
 598         int err;
 599
 600         fs_for_each_ft(iter, prio) {
 601                 i++;
 602                 err = mlx5_cmd_modify_flow_table(dev,
 603                                                  iter,
 604                                                  ft);
 605                 if (err) {
 606                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 607                                        iter->id);
 608                         /* The driver is out of sync with the FW */
 609                         if (i > 1)
 610                                 WARN_ON(true);
 611                         return err;
 612                 }
 613         }
 614         return 0;
 615 }
 616
 617 /* Connect flow tables from previous priority of prio to ft */
 618 static int connect_prev_fts(struct mlx5_core_dev *dev,
 619                             struct mlx5_flow_table *ft,
 620                             struct fs_prio *prio)
 621 {
 622         struct mlx5_flow_table *prev_ft;
 623
 624         prev_ft = find_prev_chained_ft(prio);
 625         if (prev_ft) {
 626                 struct fs_prio *prev_prio;
 627
 628                 fs_get_obj(prev_prio, prev_ft->node.parent);
 629                 return connect_fts_in_prio(dev, prev_prio, ft);
 630         }
 631         return 0;
 632 }
 633
 634 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 635                                  *prio)
 636 {
 637         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 638         int min_level = INT_MAX;
 639         int err;
 640
 641         if (root->root_ft)
 642                 min_level = root->root_ft->level;
 643
 644         if (ft->level >= min_level)
 645                 return 0;
 646
 647         err = mlx5_cmd_update_root_ft(root->dev, ft);
 648         if (err)
 649                 mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
 650                                ft->id);
 651         else
 652                 root->root_ft = ft;
 653
 654         return err;
 655 }
 656
 657 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 658                                          struct mlx5_flow_destination *dest)
 659 {
 660         struct mlx5_flow_table *ft;
 661         struct mlx5_flow_group *fg;
 662         struct fs_fte *fte;
 663         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 664         int err = 0;
 665
 666         fs_get_obj(fte, rule->node.parent);
 667         if (!(fte->action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 668                 return -EINVAL;
 669         lock_ref_node(&fte->node);
 670         fs_get_obj(fg, fte->node.parent);
 671         fs_get_obj(ft, fg->node.parent);
 672
 673         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 674         err = mlx5_cmd_update_fte(get_dev(&ft->node),
 675                                   ft, fg->id,
 676                                   modify_mask,
 677                                   fte);
 678         unlock_ref_node(&fte->node);
 679
 680         return err;
 681 }
 682
 683 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
 684                                  struct mlx5_flow_destination *new_dest,
 685                                  struct mlx5_flow_destination *old_dest)
 686 {
 687         int i;
 688
 689         if (!old_dest) {
 690                 if (handle->num_rules != 1)
 691                         return -EINVAL;
 692                 return _mlx5_modify_rule_destination(handle->rule[0],
 693                                                      new_dest);
 694         }
 695
 696         for (i = 0; i < handle->num_rules; i++) {
 697                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
 698                         return _mlx5_modify_rule_destination(handle->rule[i],
 699                                                              new_dest);
 700         }
 701
 702         return -EINVAL;
 703 }
 704
 705 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 706 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 707                              struct mlx5_flow_table *new_next_ft,
 708                              struct mlx5_flow_table *old_next_ft)
 709 {
 710         struct mlx5_flow_destination dest;
 711         struct mlx5_flow_rule *iter;
 712         int err = 0;
 713
 714         /* new_next_ft and old_next_ft could be NULL only
 715          * when we create/destroy the anchor flow table.
 716          */
 717         if (!new_next_ft || !old_next_ft)
 718                 return 0;
 719
 720         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 721         dest.ft = new_next_ft;
 722
 723         mutex_lock(&old_next_ft->lock);
 724         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 725         mutex_unlock(&old_next_ft->lock);
 726         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 727                 err = _mlx5_modify_rule_destination(iter, &dest);
 728                 if (err)
 729                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 730                                new_next_ft->id);
 731         }
 732         return 0;
 733 }
 734
 735 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 736                               struct fs_prio *prio)
 737 {
 738         struct mlx5_flow_table *next_ft;
 739         int err = 0;
 740
 741         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 742
 743         if (list_empty(&prio->node.children)) {
 744                 err = connect_prev_fts(dev, ft, prio);
 745                 if (err)
 746                         return err;
 747
 748                 next_ft = find_next_chained_ft(prio);
 749                 err = connect_fwd_rules(dev, ft, next_ft);
 750                 if (err)
 751                         return err;
 752         }
 753
 754         if (MLX5_CAP_FLOWTABLE(dev,
 755                                flow_table_properties_nic_receive.modify_root))
 756                 err = update_root_ft_create(ft, prio);
 757         return err;
 758 }
 759
 760 static void list_add_flow_table(struct mlx5_flow_table *ft,
 761                                 struct fs_prio *prio)
 762 {
 763         struct list_head *prev = &prio->node.children;
 764         struct mlx5_flow_table *iter;
 765
 766         fs_for_each_ft(iter, prio) {
 767                 if (iter->level > ft->level)
 768                         break;
 769                 prev = &iter->node.list;
 770         }
 771         list_add(&ft->node.list, prev);
 772 }
 773
 774 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 775                                                         enum fs_flow_table_op_mod op_mod,
 776                                                         u16 vport, int prio,
 777                                                         int max_fte, u32 level)
 778 {
 779         struct mlx5_flow_table *next_ft = NULL;
 780         struct mlx5_flow_table *ft;
 781         int err;
 782         int log_table_sz;
 783         struct mlx5_flow_root_namespace *root =
 784                 find_root(&ns->node);
 785         struct fs_prio *fs_prio = NULL;
 786
 787         if (!root) {
 788                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 789                 return ERR_PTR(-ENODEV);
 790         }
 791
 792         mutex_lock(&root->chain_lock);
 793         fs_prio = find_prio(ns, prio);
 794         if (!fs_prio) {
 795                 err = -EINVAL;
 796                 goto unlock_root;
 797         }
 798         if (level >= fs_prio->num_levels) {
 799                 err = -ENOSPC;
 800                 goto unlock_root;
 801         }
 802         /* The level is related to the
 803          * priority level range.
 804          */
 805         level += fs_prio->start_level;
 806         ft = alloc_flow_table(level,
 807                               vport,
 808                               max_fte ? roundup_pow_of_two(max_fte) : 0,
 809                               root->table_type,
 810                               op_mod);
 811         if (!ft) {
 812                 err = -ENOMEM;
 813                 goto unlock_root;
 814         }
 815
 816         tree_init_node(&ft->node, 1, del_flow_table);
 817         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
 818         next_ft = find_next_chained_ft(fs_prio);
 819         err = mlx5_cmd_create_flow_table(root->dev, ft->vport, ft->op_mod, ft->type,
 820                                          ft->level, log_table_sz, next_ft, &ft->id);
 821         if (err)
 822                 goto free_ft;
 823
 824         err = connect_flow_table(root->dev, ft, fs_prio);
 825         if (err)
 826                 goto destroy_ft;
 827         lock_ref_node(&fs_prio->node);
 828         tree_add_node(&ft->node, &fs_prio->node);
 829         list_add_flow_table(ft, fs_prio);
 830         fs_prio->num_ft++;
 831         unlock_ref_node(&fs_prio->node);
 832         mutex_unlock(&root->chain_lock);
 833         return ft;
 834 destroy_ft:
 835         mlx5_cmd_destroy_flow_table(root->dev, ft);
 836 free_ft:
 837         kfree(ft);
 838 unlock_root:
 839         mutex_unlock(&root->chain_lock);
 840         return ERR_PTR(err);
 841 }
 842
 843 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 844                                                int prio, int max_fte,
 845                                                u32 level)
 846 {
 847         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, 0, prio,
 848                                         max_fte, level);
 849 }
 850
 851 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
 852                                                      int prio, int max_fte,
 853                                                      u32 level, u16 vport)
 854 {
 855         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_NORMAL, vport, prio,
 856                                         max_fte, level);
 857 }
 858
 859 struct mlx5_flow_table *mlx5_create_lag_demux_flow_table(
 860                                                struct mlx5_flow_namespace *ns,
 861                                                int prio, u32 level)
 862 {
 863         return __mlx5_create_flow_table(ns, FS_FT_OP_MOD_LAG_DEMUX, 0, prio, 0,
 864                                         level);
 865 }
 866 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
 867
 868 struct mlx5_flow_table *mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
 869                                                             int prio,
 870                                                             int num_flow_table_entries,
 871                                                             int max_num_groups,
 872                                                             u32 level)
 873 {
 874         struct mlx5_flow_table *ft;
 875
 876         if (max_num_groups > num_flow_table_entries)
 877                 return ERR_PTR(-EINVAL);
 878
 879         ft = mlx5_create_flow_table(ns, prio, num_flow_table_entries, level);
 880         if (IS_ERR(ft))
 881                 return ft;
 882
 883         ft->autogroup.active = true;
 884         ft->autogroup.required_groups = max_num_groups;
 885
 886         return ft;
 887 }
 888 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
 889
 890 /* Flow table should be locked */
 891 static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table *ft,
 892                                                         u32 *fg_in,
 893                                                         struct list_head
 894                                                         *prev_fg,
 895                                                         bool is_auto_fg)
 896 {
 897         struct mlx5_flow_group *fg;
 898         struct mlx5_core_dev *dev = get_dev(&ft->node);
 899         int err;
 900
 901         if (!dev)
 902                 return ERR_PTR(-ENODEV);
 903
 904         fg = alloc_flow_group(fg_in);
 905         if (IS_ERR(fg))
 906                 return fg;
 907
 908         err = mlx5_cmd_create_flow_group(dev, ft, fg_in, &fg->id);
 909         if (err) {
 910                 kfree(fg);
 911                 return ERR_PTR(err);
 912         }
 913
 914         if (ft->autogroup.active)
 915                 ft->autogroup.num_groups++;
 916         /* Add node to tree */
 917         tree_init_node(&fg->node, !is_auto_fg, del_flow_group);
 918         tree_add_node(&fg->node, &ft->node);
 919         /* Add node to group list */
 920         list_add(&fg->node.list, ft->node.children.prev);
 921
 922         return fg;
 923 }
 924
 925 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
 926                                                u32 *fg_in)
 927 {
 928         struct mlx5_flow_group *fg;
 929
 930         if (ft->autogroup.active)
 931                 return ERR_PTR(-EPERM);
 932
 933         lock_ref_node(&ft->node);
 934         fg = create_flow_group_common(ft, fg_in, &ft->node.children, false);
 935         unlock_ref_node(&ft->node);
 936
 937         return fg;
 938 }
 939
 940 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
 941 {
 942         struct mlx5_flow_rule *rule;
 943
 944         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
 945         if (!rule)
 946                 return NULL;
 947
 948         INIT_LIST_HEAD(&rule->next_ft);
 949         rule->node.type = FS_TYPE_FLOW_DEST;
 950         if (dest)
 951                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
 952
 953         return rule;
 954 }
 955
 956 static struct mlx5_flow_handle *alloc_handle(int num_rules)
 957 {
 958         struct mlx5_flow_handle *handle;
 959
 960         handle = kzalloc(sizeof(*handle) + sizeof(handle->rule[0]) *
 961                           num_rules, GFP_KERNEL);
 962         if (!handle)
 963                 return NULL;
 964
 965         handle->num_rules = num_rules;
 966
 967         return handle;
 968 }
 969
 970 static void destroy_flow_handle(struct fs_fte *fte,
 971                                 struct mlx5_flow_handle *handle,
 972                                 struct mlx5_flow_destination *dest,
 973                                 int i)
 974 {
 975         for (; --i >= 0;) {
 976                 if (atomic_dec_and_test(&handle->rule[i]->node.refcount)) {
 977                         fte->dests_size--;
 978                         list_del(&handle->rule[i]->node.list);
 979                         kfree(handle->rule[i]);
 980                 }
 981         }
 982         kfree(handle);
 983 }
 984
 985 static struct mlx5_flow_handle *
 986 create_flow_handle(struct fs_fte *fte,
 987                    struct mlx5_flow_destination *dest,
 988                    int dest_num,
 989                    int *modify_mask,
 990                    bool *new_rule)
 991 {
 992         struct mlx5_flow_handle *handle;
 993         struct mlx5_flow_rule *rule = NULL;
 994         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 995         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 996         int type;
 997         int i = 0;
 998
 999         handle = alloc_handle((dest_num) ? dest_num : 1);
1000         if (!handle)
1001                 return ERR_PTR(-ENOMEM);
1002
1003         do {
1004                 if (dest) {
1005                         rule = find_flow_rule(fte, dest + i);
1006                         if (rule) {
1007                                 atomic_inc(&rule->node.refcount);
1008                                 goto rule_found;
1009                         }
1010                 }
1011
1012                 *new_rule = true;
1013                 rule = alloc_rule(dest + i);
1014                 if (!rule)
1015                         goto free_rules;
1016
1017                 /* Add dest to dests list- we need flow tables to be in the
1018                  * end of the list for forward to next prio rules.
1019                  */
1020                 tree_init_node(&rule->node, 1, del_rule);
1021                 if (dest &&
1022                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1023                         list_add(&rule->node.list, &fte->node.children);
1024                 else
1025                         list_add_tail(&rule->node.list, &fte->node.children);
1026                 if (dest) {
1027                         fte->dests_size++;
1028
1029                         type = dest[i].type ==
1030                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1031                         *modify_mask |= type ? count : dst;
1032                 }
1033 rule_found:
1034                 handle->rule[i] = rule;
1035         } while (++i < dest_num);
1036
1037         return handle;
1038
1039 free_rules:
1040         destroy_flow_handle(fte, handle, dest, i);
1041         return ERR_PTR(-ENOMEM);
1042 }
1043
1044 /* fte should not be deleted while calling this function */
1045 static struct mlx5_flow_handle *
1046 add_rule_fte(struct fs_fte *fte,
1047              struct mlx5_flow_group *fg,
1048              struct mlx5_flow_destination *dest,
1049              int dest_num,
1050              bool update_action)
1051 {
1052         struct mlx5_flow_handle *handle;
1053         struct mlx5_flow_table *ft;
1054         int modify_mask = 0;
1055         int err;
1056         bool new_rule = false;
1057
1058         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1059                                     &new_rule);
1060         if (IS_ERR(handle) || !new_rule)
1061                 goto out;
1062
1063         if (update_action)
1064                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1065
1066         fs_get_obj(ft, fg->node.parent);
1067         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1068                 err = mlx5_cmd_create_fte(get_dev(&ft->node),
1069                                           ft, fg->id, fte);
1070         else
1071                 err = mlx5_cmd_update_fte(get_dev(&ft->node),
1072                                           ft, fg->id, modify_mask, fte);
1073         if (err)
1074                 goto free_handle;
1075
1076         fte->status |= FS_FTE_STATUS_EXISTING;
1077
1078 out:
1079         return handle;
1080
1081 free_handle:
1082         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1083         return ERR_PTR(err);
1084 }
1085
1086 /* Assumed fg is locked */
1087 static unsigned int get_free_fte_index(struct mlx5_flow_group *fg,
1088                                        struct list_head **prev)
1089 {
1090         struct fs_fte *fte;
1091         unsigned int start = fg->start_index;
1092
1093         if (prev)
1094                 *prev = &fg->node.children;
1095
1096         /* assumed list is sorted by index */
1097         fs_for_each_fte(fte, fg) {
1098                 if (fte->index != start)
1099                         return start;
1100                 start++;
1101                 if (prev)
1102                         *prev = &fte->node.list;
1103         }
1104
1105         return start;
1106 }
1107
1108 /* prev is output, prev->next = new_fte */
1109 static struct fs_fte *create_fte(struct mlx5_flow_group *fg,
1110                                  u32 *match_value,
1111                                  u8 action,
1112                                  u32 flow_tag,
1113                                  struct list_head **prev)
1114 {
1115         struct fs_fte *fte;
1116         int index;
1117
1118         index = get_free_fte_index(fg, prev);
1119         fte = alloc_fte(action, flow_tag, match_value, index);
1120         if (IS_ERR(fte))
1121                 return fte;
1122
1123         return fte;
1124 }
1125
1126 static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft,
1127                                                 u8 match_criteria_enable,
1128                                                 u32 *match_criteria)
1129 {
1130         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1131         struct list_head *prev = &ft->node.children;
1132         unsigned int candidate_index = 0;
1133         struct mlx5_flow_group *fg;
1134         void *match_criteria_addr;
1135         unsigned int group_size = 0;
1136         u32 *in;
1137
1138         if (!ft->autogroup.active)
1139                 return ERR_PTR(-ENOENT);
1140
1141         in = mlx5_vzalloc(inlen);
1142         if (!in)
1143                 return ERR_PTR(-ENOMEM);
1144
1145         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1146                 /* We save place for flow groups in addition to max types */
1147                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1148
1149         /*  ft->max_fte == ft->autogroup.max_types */
1150         if (group_size == 0)
1151                 group_size = 1;
1152
1153         /* sorted by start_index */
1154         fs_for_each_fg(fg, ft) {
1155                 if (candidate_index + group_size > fg->start_index)
1156                         candidate_index = fg->start_index + fg->max_ftes;
1157                 else
1158                         break;
1159                 prev = &fg->node.list;
1160         }
1161
1162         if (candidate_index + group_size > ft->max_fte) {
1163                 fg = ERR_PTR(-ENOSPC);
1164                 goto out;
1165         }
1166
1167         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1168                  match_criteria_enable);
1169         MLX5_SET(create_flow_group_in, in, start_flow_index, candidate_index);
1170         MLX5_SET(create_flow_group_in, in, end_flow_index,   candidate_index +
1171                  group_size - 1);
1172         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1173                                            in, match_criteria);
1174         memcpy(match_criteria_addr, match_criteria,
1175                MLX5_ST_SZ_BYTES(fte_match_param));
1176
1177         fg = create_flow_group_common(ft, in, prev, true);
1178 out:
1179         kvfree(in);
1180         return fg;
1181 }
1182
1183 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1184                                 struct mlx5_flow_destination *d2)
1185 {
1186         if (d1->type == d2->type) {
1187                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1188                      d1->vport_num == d2->vport_num) ||
1189                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1190                      d1->ft == d2->ft) ||
1191                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1192                      d1->tir_num == d2->tir_num))
1193                         return true;
1194         }
1195
1196         return false;
1197 }
1198
1199 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1200                                              struct mlx5_flow_destination *dest)
1201 {
1202         struct mlx5_flow_rule *rule;
1203
1204         list_for_each_entry(rule, &fte->node.children, node.list) {
1205                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1206                         return rule;
1207         }
1208         return NULL;
1209 }
1210
1211 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1212                                             u32 *match_value,
1213                                             u8 action,
1214                                             u32 flow_tag,
1215                                             struct mlx5_flow_destination *dest,
1216                                             int dest_num)
1217 {
1218         struct mlx5_flow_handle *handle;
1219         struct mlx5_flow_table *ft;
1220         struct list_head *prev;
1221         struct fs_fte *fte;
1222         int i;
1223
1224         nested_lock_ref_node(&fg->node, FS_MUTEX_PARENT);
1225         fs_for_each_fte(fte, fg) {
1226                 nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1227                 if (compare_match_value(&fg->mask, match_value, &fte->val) &&
1228                     (action & fte->action) && flow_tag == fte->flow_tag) {
1229                         int old_action = fte->action;
1230
1231                         fte->action |= action;
1232                         handle = add_rule_fte(fte, fg, dest, dest_num,
1233                                               old_action != action);
1234                         if (IS_ERR(handle)) {
1235                                 fte->action = old_action;
1236                                 goto unlock_fte;
1237                         } else {
1238                                 goto add_rules;
1239                         }
1240                 }
1241                 unlock_ref_node(&fte->node);
1242         }
1243         fs_get_obj(ft, fg->node.parent);
1244         if (fg->num_ftes >= fg->max_ftes) {
1245                 handle = ERR_PTR(-ENOSPC);
1246                 goto unlock_fg;
1247         }
1248
1249         fte = create_fte(fg, match_value, action, flow_tag, &prev);
1250         if (IS_ERR(fte)) {
1251                 handle = (void *)fte;
1252                 goto unlock_fg;
1253         }
1254         tree_init_node(&fte->node, 0, del_fte);
1255         nested_lock_ref_node(&fte->node, FS_MUTEX_CHILD);
1256         handle = add_rule_fte(fte, fg, dest, dest_num, false);
1257         if (IS_ERR(handle)) {
1258                 kfree(fte);
1259                 goto unlock_fg;
1260         }
1261
1262         fg->num_ftes++;
1263
1264         tree_add_node(&fte->node, &fg->node);
1265         list_add(&fte->node.list, prev);
1266 add_rules:
1267         for (i = 0; i < handle->num_rules; i++) {
1268                 if (atomic_read(&handle->rule[i]->node.refcount) == 1)
1269                         tree_add_node(&handle->rule[i]->node, &fte->node);
1270         }
1271 unlock_fte:
1272         unlock_ref_node(&fte->node);
1273 unlock_fg:
1274         unlock_ref_node(&fg->node);
1275         return handle;
1276 }
1277
1278 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_handle *handle)
1279 {
1280         struct mlx5_flow_rule *dst;
1281         struct fs_fte *fte;
1282
1283         fs_get_obj(fte, handle->rule[0]->node.parent);
1284
1285         fs_for_each_dst(dst, fte) {
1286                 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1287                         return dst->dest_attr.counter;
1288         }
1289
1290         return NULL;
1291 }
1292
1293 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1294 {
1295         if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1296                 return !counter;
1297
1298         if (!counter)
1299                 return false;
1300
1301         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1302                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST)) &&
1303                 (action & MLX5_FLOW_CONTEXT_ACTION_COUNT);
1304 }
1305
1306 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1307                           u32 action,
1308                           struct mlx5_flow_table *ft)
1309 {
1310         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1311                 return counter_is_valid(dest->counter, action);
1312
1313         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1314                 return true;
1315
1316         if (!dest || ((dest->type ==
1317             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1318             (dest->ft->level <= ft->level)))
1319                 return false;
1320         return true;
1321 }
1322
1323 static struct mlx5_flow_handle *
1324 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1325                      struct mlx5_flow_spec *spec,
1326                      u32 action,
1327                      u32 flow_tag,
1328                      struct mlx5_flow_destination *dest,
1329                      int dest_num)
1330 {
1331         struct mlx5_flow_group *g;
1332         struct mlx5_flow_handle *rule;
1333         int i;
1334
1335         for (i = 0; i < dest_num; i++) {
1336                 if (!dest_is_valid(&dest[i], action, ft))
1337                         return ERR_PTR(-EINVAL);
1338         }
1339
1340         nested_lock_ref_node(&ft->node, FS_MUTEX_GRANDPARENT);
1341         fs_for_each_fg(g, ft)
1342                 if (compare_match_criteria(g->mask.match_criteria_enable,
1343                                            spec->match_criteria_enable,
1344                                            g->mask.match_criteria,
1345                                            spec->match_criteria)) {
1346                         rule = add_rule_fg(g, spec->match_value,
1347                                            action, flow_tag, dest, dest_num);
1348                         if (!IS_ERR(rule) || PTR_ERR(rule) != -ENOSPC)
1349                                 goto unlock;
1350                 }
1351
1352         g = create_autogroup(ft, spec->match_criteria_enable,
1353                              spec->match_criteria);
1354         if (IS_ERR(g)) {
1355                 rule = (void *)g;
1356                 goto unlock;
1357         }
1358
1359         rule = add_rule_fg(g, spec->match_value,
1360                            action, flow_tag, dest, dest_num);
1361         if (IS_ERR(rule)) {
1362                 /* Remove assumes refcount > 0 and autogroup creates a group
1363                  * with a refcount = 0.
1364                  */
1365                 unlock_ref_node(&ft->node);
1366                 tree_get_node(&g->node);
1367                 tree_remove_node(&g->node);
1368                 return rule;
1369         }
1370 unlock:
1371         unlock_ref_node(&ft->node);
1372         return rule;
1373 }
1374
1375 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1376 {
1377         return ((ft->type == FS_FT_NIC_RX) &&
1378                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1379 }
1380
1381 struct mlx5_flow_handle *
1382 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1383                     struct mlx5_flow_spec *spec,
1384                     u32 action,
1385                     u32 flow_tag,
1386                     struct mlx5_flow_destination *dest,
1387                     int dest_num)
1388 {
1389         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1390         struct mlx5_flow_destination gen_dest;
1391         struct mlx5_flow_table *next_ft = NULL;
1392         struct mlx5_flow_handle *handle = NULL;
1393         u32 sw_action = action;
1394         struct fs_prio *prio;
1395
1396         fs_get_obj(prio, ft->node.parent);
1397         if (action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1398                 if (!fwd_next_prio_supported(ft))
1399                         return ERR_PTR(-EOPNOTSUPP);
1400                 if (dest)
1401                         return ERR_PTR(-EINVAL);
1402                 mutex_lock(&root->chain_lock);
1403                 next_ft = find_next_chained_ft(prio);
1404                 if (next_ft) {
1405                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1406                         gen_dest.ft = next_ft;
1407                         dest = &gen_dest;
1408                         dest_num = 1;
1409                         action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1410                 } else {
1411                         mutex_unlock(&root->chain_lock);
1412                         return ERR_PTR(-EOPNOTSUPP);
1413                 }
1414         }
1415
1416         handle = _mlx5_add_flow_rules(ft, spec, action, flow_tag, dest,
1417                                       dest_num);
1418
1419         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1420                 if (!IS_ERR_OR_NULL(handle) &&
1421                     (list_empty(&handle->rule[0]->next_ft))) {
1422                         mutex_lock(&next_ft->lock);
1423                         list_add(&handle->rule[0]->next_ft,
1424                                  &next_ft->fwd_rules);
1425                         mutex_unlock(&next_ft->lock);
1426                         handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1427                 }
1428                 mutex_unlock(&root->chain_lock);
1429         }
1430         return handle;
1431 }
1432 EXPORT_SYMBOL(mlx5_add_flow_rules);
1433
1434 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1435 {
1436         int i;
1437
1438         for (i = handle->num_rules - 1; i >= 0; i--)
1439                 tree_remove_node(&handle->rule[i]->node);
1440         kfree(handle);
1441 }
1442 EXPORT_SYMBOL(mlx5_del_flow_rules);
1443
1444 /* Assuming prio->node.children(flow tables) is sorted by level */
1445 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1446 {
1447         struct fs_prio *prio;
1448
1449         fs_get_obj(prio, ft->node.parent);
1450
1451         if (!list_is_last(&ft->node.list, &prio->node.children))
1452                 return list_next_entry(ft, node.list);
1453         return find_next_chained_ft(prio);
1454 }
1455
1456 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1457 {
1458         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1459         struct mlx5_flow_table *new_root_ft = NULL;
1460
1461         if (root->root_ft != ft)
1462                 return 0;
1463
1464         new_root_ft = find_next_ft(ft);
1465         if (new_root_ft) {
1466                 int err = mlx5_cmd_update_root_ft(root->dev, new_root_ft);
1467
1468                 if (err) {
1469                         mlx5_core_warn(root->dev, "Update root flow table of id=%u failed\n",
1470                                        ft->id);
1471                         return err;
1472                 }
1473         }
1474         root->root_ft = new_root_ft;
1475         return 0;
1476 }
1477
1478 /* Connect flow table from previous priority to
1479  * the next flow table.
1480  */
1481 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1482 {
1483         struct mlx5_core_dev *dev = get_dev(&ft->node);
1484         struct mlx5_flow_table *next_ft;
1485         struct fs_prio *prio;
1486         int err = 0;
1487
1488         err = update_root_ft_destroy(ft);
1489         if (err)
1490                 return err;
1491
1492         fs_get_obj(prio, ft->node.parent);
1493         if  (!(list_first_entry(&prio->node.children,
1494                                 struct mlx5_flow_table,
1495                                 node.list) == ft))
1496                 return 0;
1497
1498         next_ft = find_next_chained_ft(prio);
1499         err = connect_fwd_rules(dev, next_ft, ft);
1500         if (err)
1501                 return err;
1502
1503         err = connect_prev_fts(dev, next_ft, prio);
1504         if (err)
1505                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1506                                ft->id);
1507         return err;
1508 }
1509
1510 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1511 {
1512         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1513         int err = 0;
1514
1515         mutex_lock(&root->chain_lock);
1516         err = disconnect_flow_table(ft);
1517         if (err) {
1518                 mutex_unlock(&root->chain_lock);
1519                 return err;
1520         }
1521         if (tree_remove_node(&ft->node))
1522                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1523                                ft->id);
1524         mutex_unlock(&root->chain_lock);
1525
1526         return err;
1527 }
1528 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1529
1530 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1531 {
1532         if (tree_remove_node(&fg->node))
1533                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1534                                fg->id);
1535 }
1536
1537 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1538                                                     enum mlx5_flow_namespace_type type)
1539 {
1540         struct mlx5_flow_steering *steering = dev->priv.steering;
1541         struct mlx5_flow_root_namespace *root_ns;
1542         int prio;
1543         struct fs_prio *fs_prio;
1544         struct mlx5_flow_namespace *ns;
1545
1546         if (!steering)
1547                 return NULL;
1548
1549         switch (type) {
1550         case MLX5_FLOW_NAMESPACE_BYPASS:
1551         case MLX5_FLOW_NAMESPACE_LAG:
1552         case MLX5_FLOW_NAMESPACE_OFFLOADS:
1553         case MLX5_FLOW_NAMESPACE_ETHTOOL:
1554         case MLX5_FLOW_NAMESPACE_KERNEL:
1555         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1556         case MLX5_FLOW_NAMESPACE_ANCHOR:
1557                 prio = type;
1558                 break;
1559         case MLX5_FLOW_NAMESPACE_FDB:
1560                 if (steering->fdb_root_ns)
1561                         return &steering->fdb_root_ns->ns;
1562                 else
1563                         return NULL;
1564         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
1565                 if (steering->esw_egress_root_ns)
1566                         return &steering->esw_egress_root_ns->ns;
1567                 else
1568                         return NULL;
1569         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
1570                 if (steering->esw_ingress_root_ns)
1571                         return &steering->esw_ingress_root_ns->ns;
1572                 else
1573                         return NULL;
1574         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
1575                 if (steering->sniffer_rx_root_ns)
1576                         return &steering->sniffer_rx_root_ns->ns;
1577                 else
1578                         return NULL;
1579         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
1580                 if (steering->sniffer_tx_root_ns)
1581                         return &steering->sniffer_tx_root_ns->ns;
1582                 else
1583                         return NULL;
1584         default:
1585                 return NULL;
1586         }
1587
1588         root_ns = steering->root_ns;
1589         if (!root_ns)
1590                 return NULL;
1591
1592         fs_prio = find_prio(&root_ns->ns, prio);
1593         if (!fs_prio)
1594                 return NULL;
1595
1596         ns = list_first_entry(&fs_prio->node.children,
1597                               typeof(*ns),
1598                               node.list);
1599
1600         return ns;
1601 }
1602 EXPORT_SYMBOL(mlx5_get_flow_namespace);
1603
1604 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
1605                                       unsigned int prio, int num_levels)
1606 {
1607         struct fs_prio *fs_prio;
1608
1609         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
1610         if (!fs_prio)
1611                 return ERR_PTR(-ENOMEM);
1612
1613         fs_prio->node.type = FS_TYPE_PRIO;
1614         tree_init_node(&fs_prio->node, 1, NULL);
1615         tree_add_node(&fs_prio->node, &ns->node);
1616         fs_prio->num_levels = num_levels;
1617         fs_prio->prio = prio;
1618         list_add_tail(&fs_prio->node.list, &ns->node.children);
1619
1620         return fs_prio;
1621 }
1622
1623 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
1624                                                      *ns)
1625 {
1626         ns->node.type = FS_TYPE_NAMESPACE;
1627
1628         return ns;
1629 }
1630
1631 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
1632 {
1633         struct mlx5_flow_namespace      *ns;
1634
1635         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
1636         if (!ns)
1637                 return ERR_PTR(-ENOMEM);
1638
1639         fs_init_namespace(ns);
1640         tree_init_node(&ns->node, 1, NULL);
1641         tree_add_node(&ns->node, &prio->node);
1642         list_add_tail(&ns->node.list, &prio->node.children);
1643
1644         return ns;
1645 }
1646
1647 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
1648                              struct init_tree_node *prio_metadata)
1649 {
1650         struct fs_prio *fs_prio;
1651         int i;
1652
1653         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
1654                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
1655                 if (IS_ERR(fs_prio))
1656                         return PTR_ERR(fs_prio);
1657         }
1658         return 0;
1659 }
1660
1661 #define FLOW_TABLE_BIT_SZ 1
1662 #define GET_FLOW_TABLE_CAP(dev, offset) \
1663         ((be32_to_cpu(*((__be32 *)(dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE]) +    \
1664                         offset / 32)) >>                                        \
1665           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
1666 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
1667 {
1668         int i;
1669
1670         for (i = 0; i < caps->arr_sz; i++) {
1671                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
1672                         return false;
1673         }
1674         return true;
1675 }
1676
1677 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
1678                                     struct init_tree_node *init_node,
1679                                     struct fs_node *fs_parent_node,
1680                                     struct init_tree_node *init_parent_node,
1681                                     int prio)
1682 {
1683         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
1684                                               flow_table_properties_nic_receive.
1685                                               max_ft_level);
1686         struct mlx5_flow_namespace *fs_ns;
1687         struct fs_prio *fs_prio;
1688         struct fs_node *base;
1689         int i;
1690         int err;
1691
1692         if (init_node->type == FS_TYPE_PRIO) {
1693                 if ((init_node->min_ft_level > max_ft_level) ||
1694                     !has_required_caps(steering->dev, &init_node->caps))
1695                         return 0;
1696
1697                 fs_get_obj(fs_ns, fs_parent_node);
1698                 if (init_node->num_leaf_prios)
1699                         return create_leaf_prios(fs_ns, prio, init_node);
1700                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
1701                 if (IS_ERR(fs_prio))
1702                         return PTR_ERR(fs_prio);
1703                 base = &fs_prio->node;
1704         } else if (init_node->type == FS_TYPE_NAMESPACE) {
1705                 fs_get_obj(fs_prio, fs_parent_node);
1706                 fs_ns = fs_create_namespace(fs_prio);
1707                 if (IS_ERR(fs_ns))
1708                         return PTR_ERR(fs_ns);
1709                 base = &fs_ns->node;
1710         } else {
1711                 return -EINVAL;
1712         }
1713         prio = 0;
1714         for (i = 0; i < init_node->ar_size; i++) {
1715                 err = init_root_tree_recursive(steering, &init_node->children[i],
1716                                                base, init_node, prio);
1717                 if (err)
1718                         return err;
1719                 if (init_node->children[i].type == FS_TYPE_PRIO &&
1720                     init_node->children[i].num_leaf_prios) {
1721                         prio += init_node->children[i].num_leaf_prios;
1722                 }
1723         }
1724
1725         return 0;
1726 }
1727
1728 static int init_root_tree(struct mlx5_flow_steering *steering,
1729                           struct init_tree_node *init_node,
1730                           struct fs_node *fs_parent_node)
1731 {
1732         int i;
1733         struct mlx5_flow_namespace *fs_ns;
1734         int err;
1735
1736         fs_get_obj(fs_ns, fs_parent_node);
1737         for (i = 0; i < init_node->ar_size; i++) {
1738                 err = init_root_tree_recursive(steering, &init_node->children[i],
1739                                                &fs_ns->node,
1740                                                init_node, i);
1741                 if (err)
1742                         return err;
1743         }
1744         return 0;
1745 }
1746
1747 static struct mlx5_flow_root_namespace *create_root_ns(struct mlx5_flow_steering *steering,
1748                                                        enum fs_flow_table_type
1749                                                        table_type)
1750 {
1751         struct mlx5_flow_root_namespace *root_ns;
1752         struct mlx5_flow_namespace *ns;
1753
1754         /* Create the root namespace */
1755         root_ns = mlx5_vzalloc(sizeof(*root_ns));
1756         if (!root_ns)
1757                 return NULL;
1758
1759         root_ns->dev = steering->dev;
1760         root_ns->table_type = table_type;
1761
1762         ns = &root_ns->ns;
1763         fs_init_namespace(ns);
1764         mutex_init(&root_ns->chain_lock);
1765         tree_init_node(&ns->node, 1, NULL);
1766         tree_add_node(&ns->node, NULL);
1767
1768         return root_ns;
1769 }
1770
1771 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
1772
1773 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
1774 {
1775         struct fs_prio *prio;
1776
1777         fs_for_each_prio(prio, ns) {
1778                  /* This updates prio start_level and num_levels */
1779                 set_prio_attrs_in_prio(prio, acc_level);
1780                 acc_level += prio->num_levels;
1781         }
1782         return acc_level;
1783 }
1784
1785 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
1786 {
1787         struct mlx5_flow_namespace *ns;
1788         int acc_level_ns = acc_level;
1789
1790         prio->start_level = acc_level;
1791         fs_for_each_ns(ns, prio)
1792                 /* This updates start_level and num_levels of ns's priority descendants */
1793                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
1794         if (!prio->num_levels)
1795                 prio->num_levels = acc_level_ns - prio->start_level;
1796         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
1797 }
1798
1799 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
1800 {
1801         struct mlx5_flow_namespace *ns = &root_ns->ns;
1802         struct fs_prio *prio;
1803         int start_level = 0;
1804
1805         fs_for_each_prio(prio, ns) {
1806                 set_prio_attrs_in_prio(prio, start_level);
1807                 start_level += prio->num_levels;
1808         }
1809 }
1810
1811 #define ANCHOR_PRIO 0
1812 #define ANCHOR_SIZE 1
1813 #define ANCHOR_LEVEL 0
1814 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
1815 {
1816         struct mlx5_flow_namespace *ns = NULL;
1817         struct mlx5_flow_table *ft;
1818
1819         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
1820         if (!ns)
1821                 return -EINVAL;
1822         ft = mlx5_create_flow_table(ns, ANCHOR_PRIO, ANCHOR_SIZE, ANCHOR_LEVEL);
1823         if (IS_ERR(ft)) {
1824                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
1825                 return PTR_ERR(ft);
1826         }
1827         return 0;
1828 }
1829
1830 static int init_root_ns(struct mlx5_flow_steering *steering)
1831 {
1832
1833         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
1834         if (IS_ERR_OR_NULL(steering->root_ns))
1835                 goto cleanup;
1836
1837         if (init_root_tree(steering, &root_fs, &steering->root_ns->ns.node))
1838                 goto cleanup;
1839
1840         set_prio_attrs(steering->root_ns);
1841
1842         if (create_anchor_flow_table(steering))
1843                 goto cleanup;
1844
1845         return 0;
1846
1847 cleanup:
1848         mlx5_cleanup_fs(steering->dev);
1849         return -ENOMEM;
1850 }
1851
1852 static void clean_tree(struct fs_node *node)
1853 {
1854         if (node) {
1855                 struct fs_node *iter;
1856                 struct fs_node *temp;
1857
1858                 list_for_each_entry_safe(iter, temp, &node->children, list)
1859                         clean_tree(iter);
1860                 tree_remove_node(node);
1861         }
1862 }
1863
1864 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
1865 {
1866         if (!root_ns)
1867                 return;
1868
1869         clean_tree(&root_ns->ns.node);
1870 }
1871
1872 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
1873 {
1874         struct mlx5_flow_steering *steering = dev->priv.steering;
1875
1876         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1877                 return;
1878
1879         cleanup_root_ns(steering->root_ns);
1880         cleanup_root_ns(steering->esw_egress_root_ns);
1881         cleanup_root_ns(steering->esw_ingress_root_ns);
1882         cleanup_root_ns(steering->fdb_root_ns);
1883         cleanup_root_ns(steering->sniffer_rx_root_ns);
1884         cleanup_root_ns(steering->sniffer_tx_root_ns);
1885         mlx5_cleanup_fc_stats(dev);
1886         kfree(steering);
1887 }
1888
1889 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
1890 {
1891         struct fs_prio *prio;
1892
1893         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
1894         if (!steering->sniffer_tx_root_ns)
1895                 return -ENOMEM;
1896
1897         /* Create single prio */
1898         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
1899         if (IS_ERR(prio)) {
1900                 cleanup_root_ns(steering->sniffer_tx_root_ns);
1901                 return PTR_ERR(prio);
1902         }
1903         return 0;
1904 }
1905
1906 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
1907 {
1908         struct fs_prio *prio;
1909
1910         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
1911         if (!steering->sniffer_rx_root_ns)
1912                 return -ENOMEM;
1913
1914         /* Create single prio */
1915         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
1916         if (IS_ERR(prio)) {
1917                 cleanup_root_ns(steering->sniffer_rx_root_ns);
1918                 return PTR_ERR(prio);
1919         }
1920         return 0;
1921 }
1922
1923 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
1924 {
1925         struct fs_prio *prio;
1926
1927         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
1928         if (!steering->fdb_root_ns)
1929                 return -ENOMEM;
1930
1931         prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 1);
1932         if (IS_ERR(prio))
1933                 goto out_err;
1934
1935         prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
1936         if (IS_ERR(prio))
1937                 goto out_err;
1938
1939         set_prio_attrs(steering->fdb_root_ns);
1940         return 0;
1941
1942 out_err:
1943         cleanup_root_ns(steering->fdb_root_ns);
1944         steering->fdb_root_ns = NULL;
1945         return PTR_ERR(prio);
1946 }
1947
1948 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering)
1949 {
1950         struct fs_prio *prio;
1951
1952         steering->esw_egress_root_ns = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
1953         if (!steering->esw_egress_root_ns)
1954                 return -ENOMEM;
1955
1956         /* create 1 prio*/
1957         prio = fs_create_prio(&steering->esw_egress_root_ns->ns, 0,
1958                               MLX5_TOTAL_VPORTS(steering->dev));
1959         return PTR_ERR_OR_ZERO(prio);
1960 }
1961
1962 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering)
1963 {
1964         struct fs_prio *prio;
1965
1966         steering->esw_ingress_root_ns = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
1967         if (!steering->esw_ingress_root_ns)
1968                 return -ENOMEM;
1969
1970         /* create 1 prio*/
1971         prio = fs_create_prio(&steering->esw_ingress_root_ns->ns, 0,
1972                               MLX5_TOTAL_VPORTS(steering->dev));
1973         return PTR_ERR_OR_ZERO(prio);
1974 }
1975
1976 int mlx5_init_fs(struct mlx5_core_dev *dev)
1977 {
1978         struct mlx5_flow_steering *steering;
1979         int err = 0;
1980
1981         if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
1982                 return 0;
1983
1984         err = mlx5_init_fc_stats(dev);
1985         if (err)
1986                 return err;
1987
1988         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
1989         if (!steering)
1990                 return -ENOMEM;
1991         steering->dev = dev;
1992         dev->priv.steering = steering;
1993
1994         if (MLX5_CAP_GEN(dev, nic_flow_table) &&
1995             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
1996                 err = init_root_ns(steering);
1997                 if (err)
1998                         goto err;
1999         }
2000
2001         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
2002                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2003                         err = init_fdb_root_ns(steering);
2004                         if (err)
2005                                 goto err;
2006                 }
2007                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2008                         err = init_egress_acl_root_ns(steering);
2009                         if (err)
2010                                 goto err;
2011                 }
2012                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2013                         err = init_ingress_acl_root_ns(steering);
2014                         if (err)
2015                                 goto err;
2016                 }
2017         }
2018
2019         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2020                 err = init_sniffer_rx_root_ns(steering);
2021                 if (err)
2022                         goto err;
2023         }
2024
2025         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2026                 err = init_sniffer_tx_root_ns(steering);
2027                 if (err)
2028                         goto err;
2029         }
2030
2031         return 0;
2032 err:
2033         mlx5_cleanup_fs(dev);
2034         return err;
2035 }