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 #include "diag/fs_tracepoint.h"
  40 #include "accel/ipsec.h"
  41 #include "fpga/ipsec.h"
  42
  43 #define INIT_TREE_NODE_ARRAY_SIZE(...)  (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
  44                                          sizeof(struct init_tree_node))
  45
  46 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
  47                  ...) {.type = FS_TYPE_PRIO,\
  48         .min_ft_level = min_level_val,\
  49         .num_levels = num_levels_val,\
  50         .num_leaf_prios = num_prios_val,\
  51         .caps = caps_val,\
  52         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  53         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  54 }
  55
  56 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
  57         ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
  58                  __VA_ARGS__)\
  59
  60 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
  61         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  62         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  63 }
  64
  65 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  66                                    sizeof(long))
  67
  68 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  69
  70 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  71                                .caps = (long[]) {__VA_ARGS__} }
  72
  73 #define FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  74                                            FS_CAP(flow_table_properties_nic_receive.modify_root), \
  75                                            FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  76                                            FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  77
  78 #define LEFTOVERS_NUM_LEVELS 1
  79 #define LEFTOVERS_NUM_PRIOS 1
  80
  81 #define BY_PASS_PRIO_NUM_LEVELS 1
  82 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  83                            LEFTOVERS_NUM_PRIOS)
  84
  85 #define ETHTOOL_PRIO_NUM_LEVELS 1
  86 #define ETHTOOL_NUM_PRIOS 11
  87 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
  88 /* Vlan, mac, ttc, inner ttc, aRFS */
  89 #define KERNEL_NIC_PRIO_NUM_LEVELS 5
  90 #define KERNEL_NIC_NUM_PRIOS 1
  91 /* One more level for tc */
  92 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
  93
  94 #define KERNEL_NIC_TC_NUM_PRIOS  1
  95 #define KERNEL_NIC_TC_NUM_LEVELS 2
  96
  97 #define ANCHOR_NUM_LEVELS 1
  98 #define ANCHOR_NUM_PRIOS 1
  99 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 100
 101 #define OFFLOADS_MAX_FT 1
 102 #define OFFLOADS_NUM_PRIOS 1
 103 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
 104
 105 #define LAG_PRIO_NUM_LEVELS 1
 106 #define LAG_NUM_PRIOS 1
 107 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 108
 109 struct node_caps {
 110         size_t  arr_sz;
 111         long    *caps;
 112 };
 113
 114 static struct init_tree_node {
 115         enum fs_node_type       type;
 116         struct init_tree_node *children;
 117         int ar_size;
 118         struct node_caps caps;
 119         int min_ft_level;
 120         int num_leaf_prios;
 121         int prio;
 122         int num_levels;
 123 } root_fs = {
 124         .type = FS_TYPE_NAMESPACE,
 125         .ar_size = 7,
 126         .children = (struct init_tree_node[]) {
 127                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 128                          FS_CHAINING_CAPS,
 129                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 130                                                   BY_PASS_PRIO_NUM_LEVELS))),
 131                 ADD_PRIO(0, LAG_MIN_LEVEL, 0,
 132                          FS_CHAINING_CAPS,
 133                          ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 134                                                   LAG_PRIO_NUM_LEVELS))),
 135                 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
 136                          ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
 137                 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
 138                          FS_CHAINING_CAPS,
 139                          ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 140                                                   ETHTOOL_PRIO_NUM_LEVELS))),
 141                 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 142                          ADD_NS(ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS, KERNEL_NIC_TC_NUM_LEVELS),
 143                                 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 144                                                   KERNEL_NIC_PRIO_NUM_LEVELS))),
 145                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 146                          FS_CHAINING_CAPS,
 147                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
 148                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 149                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
 150         }
 151 };
 152
 153 enum fs_i_lock_class {
 154         FS_LOCK_GRANDPARENT,
 155         FS_LOCK_PARENT,
 156         FS_LOCK_CHILD
 157 };
 158
 159 static const struct rhashtable_params rhash_fte = {
 160         .key_len = FIELD_SIZEOF(struct fs_fte, val),
 161         .key_offset = offsetof(struct fs_fte, val),
 162         .head_offset = offsetof(struct fs_fte, hash),
 163         .automatic_shrinking = true,
 164         .min_size = 1,
 165 };
 166
 167 static const struct rhashtable_params rhash_fg = {
 168         .key_len = FIELD_SIZEOF(struct mlx5_flow_group, mask),
 169         .key_offset = offsetof(struct mlx5_flow_group, mask),
 170         .head_offset = offsetof(struct mlx5_flow_group, hash),
 171         .automatic_shrinking = true,
 172         .min_size = 1,
 173
 174 };
 175
 176 static void del_hw_flow_table(struct fs_node *node);
 177 static void del_hw_flow_group(struct fs_node *node);
 178 static void del_hw_fte(struct fs_node *node);
 179 static void del_sw_flow_table(struct fs_node *node);
 180 static void del_sw_flow_group(struct fs_node *node);
 181 static void del_sw_fte(struct fs_node *node);
 182 static void del_sw_prio(struct fs_node *node);
 183 static void del_sw_ns(struct fs_node *node);
 184 /* Delete rule (destination) is special case that
 185  * requires to lock the FTE for all the deletion process.
 186  */
 187 static void del_sw_hw_rule(struct fs_node *node);
 188 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 189                                 struct mlx5_flow_destination *d2);
 190 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 191 static struct mlx5_flow_rule *
 192 find_flow_rule(struct fs_fte *fte,
 193                struct mlx5_flow_destination *dest);
 194
 195 static void tree_init_node(struct fs_node *node,
 196                            void (*del_hw_func)(struct fs_node *),
 197                            void (*del_sw_func)(struct fs_node *))
 198 {
 199         refcount_set(&node->refcount, 1);
 200         INIT_LIST_HEAD(&node->list);
 201         INIT_LIST_HEAD(&node->children);
 202         init_rwsem(&node->lock);
 203         node->del_hw_func = del_hw_func;
 204         node->del_sw_func = del_sw_func;
 205         node->active = false;
 206 }
 207
 208 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 209 {
 210         if (parent)
 211                 refcount_inc(&parent->refcount);
 212         node->parent = parent;
 213
 214         /* Parent is the root */
 215         if (!parent)
 216                 node->root = node;
 217         else
 218                 node->root = parent->root;
 219 }
 220
 221 static int tree_get_node(struct fs_node *node)
 222 {
 223         return refcount_inc_not_zero(&node->refcount);
 224 }
 225
 226 static void nested_down_read_ref_node(struct fs_node *node,
 227                                       enum fs_i_lock_class class)
 228 {
 229         if (node) {
 230                 down_read_nested(&node->lock, class);
 231                 refcount_inc(&node->refcount);
 232         }
 233 }
 234
 235 static void nested_down_write_ref_node(struct fs_node *node,
 236                                        enum fs_i_lock_class class)
 237 {
 238         if (node) {
 239                 down_write_nested(&node->lock, class);
 240                 refcount_inc(&node->refcount);
 241         }
 242 }
 243
 244 static void down_write_ref_node(struct fs_node *node)
 245 {
 246         if (node) {
 247                 down_write(&node->lock);
 248                 refcount_inc(&node->refcount);
 249         }
 250 }
 251
 252 static void up_read_ref_node(struct fs_node *node)
 253 {
 254         refcount_dec(&node->refcount);
 255         up_read(&node->lock);
 256 }
 257
 258 static void up_write_ref_node(struct fs_node *node)
 259 {
 260         refcount_dec(&node->refcount);
 261         up_write(&node->lock);
 262 }
 263
 264 static void tree_put_node(struct fs_node *node)
 265 {
 266         struct fs_node *parent_node = node->parent;
 267
 268         if (refcount_dec_and_test(&node->refcount)) {
 269                 if (node->del_hw_func)
 270                         node->del_hw_func(node);
 271                 if (parent_node) {
 272                         /* Only root namespace doesn't have parent and we just
 273                          * need to free its node.
 274                          */
 275                         down_write_ref_node(parent_node);
 276                         list_del_init(&node->list);
 277                         if (node->del_sw_func)
 278                                 node->del_sw_func(node);
 279                         up_write_ref_node(parent_node);
 280                 } else {
 281                         kfree(node);
 282                 }
 283                 node = NULL;
 284         }
 285         if (!node && parent_node)
 286                 tree_put_node(parent_node);
 287 }
 288
 289 static int tree_remove_node(struct fs_node *node)
 290 {
 291         if (refcount_read(&node->refcount) > 1) {
 292                 refcount_dec(&node->refcount);
 293                 return -EEXIST;
 294         }
 295         tree_put_node(node);
 296         return 0;
 297 }
 298
 299 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 300                                  unsigned int prio)
 301 {
 302         struct fs_prio *iter_prio;
 303
 304         fs_for_each_prio(iter_prio, ns) {
 305                 if (iter_prio->prio == prio)
 306                         return iter_prio;
 307         }
 308
 309         return NULL;
 310 }
 311
 312 static bool check_last_reserved(const u32 *match_criteria)
 313 {
 314         char *match_criteria_reserved =
 315                 MLX5_ADDR_OF(fte_match_param, match_criteria, MLX5_FTE_MATCH_PARAM_RESERVED);
 316
 317         return  !match_criteria_reserved[0] &&
 318                 !memcmp(match_criteria_reserved, match_criteria_reserved + 1,
 319                         MLX5_FLD_SZ_BYTES(fte_match_param,
 320                                           MLX5_FTE_MATCH_PARAM_RESERVED) - 1);
 321 }
 322
 323 static bool check_valid_mask(u8 match_criteria_enable, const u32 *match_criteria)
 324 {
 325         if (match_criteria_enable & ~(
 326                 (1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS)   |
 327                 (1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS) |
 328                 (1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS) |
 329                 (1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS_2)))
 330                 return false;
 331
 332         if (!(match_criteria_enable &
 333               1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_OUTER_HEADERS)) {
 334                 char *fg_type_mask = MLX5_ADDR_OF(fte_match_param,
 335                                                   match_criteria, outer_headers);
 336
 337                 if (fg_type_mask[0] ||
 338                     memcmp(fg_type_mask, fg_type_mask + 1,
 339                            MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4) - 1))
 340                         return false;
 341         }
 342
 343         if (!(match_criteria_enable &
 344               1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS)) {
 345                 char *fg_type_mask = MLX5_ADDR_OF(fte_match_param,
 346                                                   match_criteria, misc_parameters);
 347
 348                 if (fg_type_mask[0] ||
 349                     memcmp(fg_type_mask, fg_type_mask + 1,
 350                            MLX5_ST_SZ_BYTES(fte_match_set_misc) - 1))
 351                         return false;
 352         }
 353
 354         if (!(match_criteria_enable &
 355               1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_INNER_HEADERS)) {
 356                 char *fg_type_mask = MLX5_ADDR_OF(fte_match_param,
 357                                                   match_criteria, inner_headers);
 358
 359                 if (fg_type_mask[0] ||
 360                     memcmp(fg_type_mask, fg_type_mask + 1,
 361                            MLX5_ST_SZ_BYTES(fte_match_set_lyr_2_4) - 1))
 362                         return false;
 363         }
 364
 365         if (!(match_criteria_enable &
 366               1 << MLX5_CREATE_FLOW_GROUP_IN_MATCH_CRITERIA_ENABLE_MISC_PARAMETERS_2)) {
 367                 char *fg_type_mask = MLX5_ADDR_OF(fte_match_param,
 368                                                   match_criteria, misc_parameters_2);
 369
 370                 if (fg_type_mask[0] ||
 371                     memcmp(fg_type_mask, fg_type_mask + 1,
 372                            MLX5_ST_SZ_BYTES(fte_match_set_misc2) - 1))
 373                         return false;
 374         }
 375
 376         return check_last_reserved(match_criteria);
 377 }
 378
 379 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 380 {
 381         int i;
 382
 383         if (!check_valid_mask(spec->match_criteria_enable, spec->match_criteria)) {
 384                 pr_warn("mlx5_core: Match criteria given mismatches match_criteria_enable\n");
 385                 return false;
 386         }
 387
 388         for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 389                 if (spec->match_value[i] & ~spec->match_criteria[i]) {
 390                         pr_warn("mlx5_core: match_value differs from match_criteria\n");
 391                         return false;
 392                 }
 393
 394         return check_last_reserved(spec->match_value);
 395 }
 396
 397 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 398 {
 399         struct fs_node *root;
 400         struct mlx5_flow_namespace *ns;
 401
 402         root = node->root;
 403
 404         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 405                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 406                 return NULL;
 407         }
 408
 409         ns = container_of(root, struct mlx5_flow_namespace, node);
 410         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 411 }
 412
 413 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 414 {
 415         struct mlx5_flow_root_namespace *root = find_root(node);
 416
 417         if (root)
 418                 return root->dev->priv.steering;
 419         return NULL;
 420 }
 421
 422 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 423 {
 424         struct mlx5_flow_root_namespace *root = find_root(node);
 425
 426         if (root)
 427                 return root->dev;
 428         return NULL;
 429 }
 430
 431 static void del_sw_ns(struct fs_node *node)
 432 {
 433         kfree(node);
 434 }
 435
 436 static void del_sw_prio(struct fs_node *node)
 437 {
 438         kfree(node);
 439 }
 440
 441 static void del_hw_flow_table(struct fs_node *node)
 442 {
 443         struct mlx5_flow_root_namespace *root;
 444         struct mlx5_flow_table *ft;
 445         struct mlx5_core_dev *dev;
 446         int err;
 447
 448         fs_get_obj(ft, node);
 449         dev = get_dev(&ft->node);
 450         root = find_root(&ft->node);
 451
 452         if (node->active) {
 453                 err = root->cmds->destroy_flow_table(dev, ft);
 454                 if (err)
 455                         mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 456         }
 457 }
 458
 459 static void del_sw_flow_table(struct fs_node *node)
 460 {
 461         struct mlx5_flow_table *ft;
 462         struct fs_prio *prio;
 463
 464         fs_get_obj(ft, node);
 465
 466         rhltable_destroy(&ft->fgs_hash);
 467         fs_get_obj(prio, ft->node.parent);
 468         prio->num_ft--;
 469         kfree(ft);
 470 }
 471
 472 static void del_sw_hw_rule(struct fs_node *node)
 473 {
 474         struct mlx5_flow_root_namespace *root;
 475         struct mlx5_flow_rule *rule;
 476         struct mlx5_flow_table *ft;
 477         struct mlx5_flow_group *fg;
 478         struct fs_fte *fte;
 479         int modify_mask;
 480         struct mlx5_core_dev *dev = get_dev(node);
 481         int err;
 482         bool update_fte = false;
 483
 484         fs_get_obj(rule, node);
 485         fs_get_obj(fte, rule->node.parent);
 486         fs_get_obj(fg, fte->node.parent);
 487         fs_get_obj(ft, fg->node.parent);
 488         trace_mlx5_fs_del_rule(rule);
 489         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 490                 mutex_lock(&rule->dest_attr.ft->lock);
 491                 list_del(&rule->next_ft);
 492                 mutex_unlock(&rule->dest_attr.ft->lock);
 493         }
 494
 495         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 496             --fte->dests_size) {
 497                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 498                               BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 499                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 500                 update_fte = true;
 501                 goto out;
 502         }
 503
 504         if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 505             --fte->dests_size) {
 506                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST),
 507                 update_fte = true;
 508         }
 509 out:
 510         root = find_root(&ft->node);
 511         if (update_fte && fte->dests_size) {
 512                 err = root->cmds->update_fte(dev, ft, fg->id, modify_mask, fte);
 513                 if (err)
 514                         mlx5_core_warn(dev,
 515                                        "%s can't del rule fg id=%d fte_index=%d\n",
 516                                        __func__, fg->id, fte->index);
 517         }
 518         kfree(rule);
 519 }
 520
 521 static void del_hw_fte(struct fs_node *node)
 522 {
 523         struct mlx5_flow_root_namespace *root;
 524         struct mlx5_flow_table *ft;
 525         struct mlx5_flow_group *fg;
 526         struct mlx5_core_dev *dev;
 527         struct fs_fte *fte;
 528         int err;
 529
 530         fs_get_obj(fte, node);
 531         fs_get_obj(fg, fte->node.parent);
 532         fs_get_obj(ft, fg->node.parent);
 533
 534         trace_mlx5_fs_del_fte(fte);
 535         dev = get_dev(&ft->node);
 536         root = find_root(&ft->node);
 537         if (node->active) {
 538                 err = root->cmds->delete_fte(dev, ft, fte);
 539                 if (err)
 540                         mlx5_core_warn(dev,
 541                                        "flow steering can't delete fte in index %d of flow group id %d\n",
 542                                        fte->index, fg->id);
 543         }
 544 }
 545
 546 static void del_sw_fte(struct fs_node *node)
 547 {
 548         struct mlx5_flow_steering *steering = get_steering(node);
 549         struct mlx5_flow_group *fg;
 550         struct fs_fte *fte;
 551         int err;
 552
 553         fs_get_obj(fte, node);
 554         fs_get_obj(fg, fte->node.parent);
 555
 556         err = rhashtable_remove_fast(&fg->ftes_hash,
 557                                      &fte->hash,
 558                                      rhash_fte);
 559         WARN_ON(err);
 560         ida_simple_remove(&fg->fte_allocator, fte->index - fg->start_index);
 561         kmem_cache_free(steering->ftes_cache, fte);
 562 }
 563
 564 static void del_hw_flow_group(struct fs_node *node)
 565 {
 566         struct mlx5_flow_root_namespace *root;
 567         struct mlx5_flow_group *fg;
 568         struct mlx5_flow_table *ft;
 569         struct mlx5_core_dev *dev;
 570
 571         fs_get_obj(fg, node);
 572         fs_get_obj(ft, fg->node.parent);
 573         dev = get_dev(&ft->node);
 574         trace_mlx5_fs_del_fg(fg);
 575
 576         root = find_root(&ft->node);
 577         if (fg->node.active && root->cmds->destroy_flow_group(dev, ft, fg->id))
 578                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 579                                fg->id, ft->id);
 580 }
 581
 582 static void del_sw_flow_group(struct fs_node *node)
 583 {
 584         struct mlx5_flow_steering *steering = get_steering(node);
 585         struct mlx5_flow_group *fg;
 586         struct mlx5_flow_table *ft;
 587         int err;
 588
 589         fs_get_obj(fg, node);
 590         fs_get_obj(ft, fg->node.parent);
 591
 592         rhashtable_destroy(&fg->ftes_hash);
 593         ida_destroy(&fg->fte_allocator);
 594         if (ft->autogroup.active)
 595                 ft->autogroup.num_groups--;
 596         err = rhltable_remove(&ft->fgs_hash,
 597                               &fg->hash,
 598                               rhash_fg);
 599         WARN_ON(err);
 600         kmem_cache_free(steering->fgs_cache, fg);
 601 }
 602
 603 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 604 {
 605         int index;
 606         int ret;
 607
 608         index = ida_simple_get(&fg->fte_allocator, 0, fg->max_ftes, GFP_KERNEL);
 609         if (index < 0)
 610                 return index;
 611
 612         fte->index = index + fg->start_index;
 613         ret = rhashtable_insert_fast(&fg->ftes_hash,
 614                                      &fte->hash,
 615                                      rhash_fte);
 616         if (ret)
 617                 goto err_ida_remove;
 618
 619         tree_add_node(&fte->node, &fg->node);
 620         list_add_tail(&fte->node.list, &fg->node.children);
 621         return 0;
 622
 623 err_ida_remove:
 624         ida_simple_remove(&fg->fte_allocator, index);
 625         return ret;
 626 }
 627
 628 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 629                                 u32 *match_value,
 630                                 struct mlx5_flow_act *flow_act)
 631 {
 632         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 633         struct fs_fte *fte;
 634
 635         fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 636         if (!fte)
 637                 return ERR_PTR(-ENOMEM);
 638
 639         memcpy(fte->val, match_value, sizeof(fte->val));
 640         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 641         fte->action = *flow_act;
 642
 643         tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
 644
 645         return fte;
 646 }
 647
 648 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 649                                struct mlx5_flow_group *fg)
 650 {
 651         rhashtable_destroy(&fg->ftes_hash);
 652         kmem_cache_free(steering->fgs_cache, fg);
 653 }
 654
 655 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 656                                                 u8 match_criteria_enable,
 657                                                 void *match_criteria,
 658                                                 int start_index,
 659                                                 int end_index)
 660 {
 661         struct mlx5_flow_group *fg;
 662         int ret;
 663
 664         fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 665         if (!fg)
 666                 return ERR_PTR(-ENOMEM);
 667
 668         ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 669         if (ret) {
 670                 kmem_cache_free(steering->fgs_cache, fg);
 671                 return ERR_PTR(ret);
 672 }
 673         ida_init(&fg->fte_allocator);
 674         fg->mask.match_criteria_enable = match_criteria_enable;
 675         memcpy(&fg->mask.match_criteria, match_criteria,
 676                sizeof(fg->mask.match_criteria));
 677         fg->node.type =  FS_TYPE_FLOW_GROUP;
 678         fg->start_index = start_index;
 679         fg->max_ftes = end_index - start_index + 1;
 680
 681         return fg;
 682 }
 683
 684 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 685                                                        u8 match_criteria_enable,
 686                                                        void *match_criteria,
 687                                                        int start_index,
 688                                                        int end_index,
 689                                                        struct list_head *prev)
 690 {
 691         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 692         struct mlx5_flow_group *fg;
 693         int ret;
 694
 695         fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 696                               start_index, end_index);
 697         if (IS_ERR(fg))
 698                 return fg;
 699
 700         /* initialize refcnt, add to parent list */
 701         ret = rhltable_insert(&ft->fgs_hash,
 702                               &fg->hash,
 703                               rhash_fg);
 704         if (ret) {
 705                 dealloc_flow_group(steering, fg);
 706                 return ERR_PTR(ret);
 707         }
 708
 709         tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 710         tree_add_node(&fg->node, &ft->node);
 711         /* Add node to group list */
 712         list_add(&fg->node.list, prev);
 713         atomic_inc(&ft->node.version);
 714
 715         return fg;
 716 }
 717
 718 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 719                                                 enum fs_flow_table_type table_type,
 720                                                 enum fs_flow_table_op_mod op_mod,
 721                                                 u32 flags)
 722 {
 723         struct mlx5_flow_table *ft;
 724         int ret;
 725
 726         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 727         if (!ft)
 728                 return ERR_PTR(-ENOMEM);
 729
 730         ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 731         if (ret) {
 732                 kfree(ft);
 733                 return ERR_PTR(ret);
 734         }
 735
 736         ft->level = level;
 737         ft->node.type = FS_TYPE_FLOW_TABLE;
 738         ft->op_mod = op_mod;
 739         ft->type = table_type;
 740         ft->vport = vport;
 741         ft->max_fte = max_fte;
 742         ft->flags = flags;
 743         INIT_LIST_HEAD(&ft->fwd_rules);
 744         mutex_init(&ft->lock);
 745
 746         return ft;
 747 }
 748
 749 /* If reverse is false, then we search for the first flow table in the
 750  * root sub-tree from start(closest from right), else we search for the
 751  * last flow table in the root sub-tree till start(closest from left).
 752  */
 753 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 754                                                          struct list_head *start,
 755                                                          bool reverse)
 756 {
 757 #define list_advance_entry(pos, reverse)                \
 758         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 759
 760 #define list_for_each_advance_continue(pos, head, reverse)      \
 761         for (pos = list_advance_entry(pos, reverse);            \
 762              &pos->list != (head);                              \
 763              pos = list_advance_entry(pos, reverse))
 764
 765         struct fs_node *iter = list_entry(start, struct fs_node, list);
 766         struct mlx5_flow_table *ft = NULL;
 767
 768         if (!root)
 769                 return NULL;
 770
 771         list_for_each_advance_continue(iter, &root->children, reverse) {
 772                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 773                         fs_get_obj(ft, iter);
 774                         return ft;
 775                 }
 776                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 777                 if (ft)
 778                         return ft;
 779         }
 780
 781         return ft;
 782 }
 783
 784 /* If reverse if false then return the first flow table in next priority of
 785  * prio in the tree, else return the last flow table in the previous priority
 786  * of prio in the tree.
 787  */
 788 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 789 {
 790         struct mlx5_flow_table *ft = NULL;
 791         struct fs_node *curr_node;
 792         struct fs_node *parent;
 793
 794         parent = prio->node.parent;
 795         curr_node = &prio->node;
 796         while (!ft && parent) {
 797                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 798                 curr_node = parent;
 799                 parent = curr_node->parent;
 800         }
 801         return ft;
 802 }
 803
 804 /* Assuming all the tree is locked by mutex chain lock */
 805 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 806 {
 807         return find_closest_ft(prio, false);
 808 }
 809
 810 /* Assuming all the tree is locked by mutex chain lock */
 811 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 812 {
 813         return find_closest_ft(prio, true);
 814 }
 815
 816 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 817                                struct fs_prio *prio,
 818                                struct mlx5_flow_table *ft)
 819 {
 820         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 821         struct mlx5_flow_table *iter;
 822         int i = 0;
 823         int err;
 824
 825         fs_for_each_ft(iter, prio) {
 826                 i++;
 827                 err = root->cmds->modify_flow_table(dev, iter, ft);
 828                 if (err) {
 829                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 830                                        iter->id);
 831                         /* The driver is out of sync with the FW */
 832                         if (i > 1)
 833                                 WARN_ON(true);
 834                         return err;
 835                 }
 836         }
 837         return 0;
 838 }
 839
 840 /* Connect flow tables from previous priority of prio to ft */
 841 static int connect_prev_fts(struct mlx5_core_dev *dev,
 842                             struct mlx5_flow_table *ft,
 843                             struct fs_prio *prio)
 844 {
 845         struct mlx5_flow_table *prev_ft;
 846
 847         prev_ft = find_prev_chained_ft(prio);
 848         if (prev_ft) {
 849                 struct fs_prio *prev_prio;
 850
 851                 fs_get_obj(prev_prio, prev_ft->node.parent);
 852                 return connect_fts_in_prio(dev, prev_prio, ft);
 853         }
 854         return 0;
 855 }
 856
 857 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 858                                  *prio)
 859 {
 860         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 861         struct mlx5_ft_underlay_qp *uqp;
 862         int min_level = INT_MAX;
 863         int err;
 864         u32 qpn;
 865
 866         if (root->root_ft)
 867                 min_level = root->root_ft->level;
 868
 869         if (ft->level >= min_level)
 870                 return 0;
 871
 872         if (list_empty(&root->underlay_qpns)) {
 873                 /* Don't set any QPN (zero) in case QPN list is empty */
 874                 qpn = 0;
 875                 err = root->cmds->update_root_ft(root->dev, ft, qpn, false);
 876         } else {
 877                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
 878                         qpn = uqp->qpn;
 879                         err = root->cmds->update_root_ft(root->dev, ft,
 880                                                          qpn, false);
 881                         if (err)
 882                                 break;
 883                 }
 884         }
 885
 886         if (err)
 887                 mlx5_core_warn(root->dev,
 888                                "Update root flow table of id(%u) qpn(%d) failed\n",
 889                                ft->id, qpn);
 890         else
 891                 root->root_ft = ft;
 892
 893         return err;
 894 }
 895
 896 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 897                                          struct mlx5_flow_destination *dest)
 898 {
 899         struct mlx5_flow_root_namespace *root;
 900         struct mlx5_flow_table *ft;
 901         struct mlx5_flow_group *fg;
 902         struct fs_fte *fte;
 903         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 904         int err = 0;
 905
 906         fs_get_obj(fte, rule->node.parent);
 907         if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 908                 return -EINVAL;
 909         down_write_ref_node(&fte->node);
 910         fs_get_obj(fg, fte->node.parent);
 911         fs_get_obj(ft, fg->node.parent);
 912
 913         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 914         root = find_root(&ft->node);
 915         err = root->cmds->update_fte(get_dev(&ft->node), ft, fg->id,
 916                                      modify_mask, fte);
 917         up_write_ref_node(&fte->node);
 918
 919         return err;
 920 }
 921
 922 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
 923                                  struct mlx5_flow_destination *new_dest,
 924                                  struct mlx5_flow_destination *old_dest)
 925 {
 926         int i;
 927
 928         if (!old_dest) {
 929                 if (handle->num_rules != 1)
 930                         return -EINVAL;
 931                 return _mlx5_modify_rule_destination(handle->rule[0],
 932                                                      new_dest);
 933         }
 934
 935         for (i = 0; i < handle->num_rules; i++) {
 936                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
 937                         return _mlx5_modify_rule_destination(handle->rule[i],
 938                                                              new_dest);
 939         }
 940
 941         return -EINVAL;
 942 }
 943
 944 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 945 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 946                              struct mlx5_flow_table *new_next_ft,
 947                              struct mlx5_flow_table *old_next_ft)
 948 {
 949         struct mlx5_flow_destination dest = {};
 950         struct mlx5_flow_rule *iter;
 951         int err = 0;
 952
 953         /* new_next_ft and old_next_ft could be NULL only
 954          * when we create/destroy the anchor flow table.
 955          */
 956         if (!new_next_ft || !old_next_ft)
 957                 return 0;
 958
 959         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 960         dest.ft = new_next_ft;
 961
 962         mutex_lock(&old_next_ft->lock);
 963         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 964         mutex_unlock(&old_next_ft->lock);
 965         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 966                 err = _mlx5_modify_rule_destination(iter, &dest);
 967                 if (err)
 968                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 969                                new_next_ft->id);
 970         }
 971         return 0;
 972 }
 973
 974 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 975                               struct fs_prio *prio)
 976 {
 977         struct mlx5_flow_table *next_ft;
 978         int err = 0;
 979
 980         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 981
 982         if (list_empty(&prio->node.children)) {
 983                 err = connect_prev_fts(dev, ft, prio);
 984                 if (err)
 985                         return err;
 986
 987                 next_ft = find_next_chained_ft(prio);
 988                 err = connect_fwd_rules(dev, ft, next_ft);
 989                 if (err)
 990                         return err;
 991         }
 992
 993         if (MLX5_CAP_FLOWTABLE(dev,
 994                                flow_table_properties_nic_receive.modify_root))
 995                 err = update_root_ft_create(ft, prio);
 996         return err;
 997 }
 998
 999 static void list_add_flow_table(struct mlx5_flow_table *ft,
1000                                 struct fs_prio *prio)
1001 {
1002         struct list_head *prev = &prio->node.children;
1003         struct mlx5_flow_table *iter;
1004
1005         fs_for_each_ft(iter, prio) {
1006                 if (iter->level > ft->level)
1007                         break;
1008                 prev = &iter->node.list;
1009         }
1010         list_add(&ft->node.list, prev);
1011 }
1012
1013 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1014                                                         struct mlx5_flow_table_attr *ft_attr,
1015                                                         enum fs_flow_table_op_mod op_mod,
1016                                                         u16 vport)
1017 {
1018         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1019         struct mlx5_flow_table *next_ft = NULL;
1020         struct fs_prio *fs_prio = NULL;
1021         struct mlx5_flow_table *ft;
1022         int log_table_sz;
1023         int err;
1024
1025         if (!root) {
1026                 pr_err("mlx5: flow steering failed to find root of namespace\n");
1027                 return ERR_PTR(-ENODEV);
1028         }
1029
1030         mutex_lock(&root->chain_lock);
1031         fs_prio = find_prio(ns, ft_attr->prio);
1032         if (!fs_prio) {
1033                 err = -EINVAL;
1034                 goto unlock_root;
1035         }
1036         if (ft_attr->level >= fs_prio->num_levels) {
1037                 err = -ENOSPC;
1038                 goto unlock_root;
1039         }
1040         /* The level is related to the
1041          * priority level range.
1042          */
1043         ft_attr->level += fs_prio->start_level;
1044         ft = alloc_flow_table(ft_attr->level,
1045                               vport,
1046                               ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
1047                               root->table_type,
1048                               op_mod, ft_attr->flags);
1049         if (IS_ERR(ft)) {
1050                 err = PTR_ERR(ft);
1051                 goto unlock_root;
1052         }
1053
1054         tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1055         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
1056         next_ft = find_next_chained_ft(fs_prio);
1057         err = root->cmds->create_flow_table(root->dev, ft->vport, ft->op_mod,
1058                                             ft->type, ft->level, log_table_sz,
1059                                             next_ft, &ft->id, ft->flags);
1060         if (err)
1061                 goto free_ft;
1062
1063         err = connect_flow_table(root->dev, ft, fs_prio);
1064         if (err)
1065                 goto destroy_ft;
1066         ft->node.active = true;
1067         down_write_ref_node(&fs_prio->node);
1068         tree_add_node(&ft->node, &fs_prio->node);
1069         list_add_flow_table(ft, fs_prio);
1070         fs_prio->num_ft++;
1071         up_write_ref_node(&fs_prio->node);
1072         mutex_unlock(&root->chain_lock);
1073         return ft;
1074 destroy_ft:
1075         root->cmds->destroy_flow_table(root->dev, ft);
1076 free_ft:
1077         kfree(ft);
1078 unlock_root:
1079         mutex_unlock(&root->chain_lock);
1080         return ERR_PTR(err);
1081 }
1082
1083 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1084                                                struct mlx5_flow_table_attr *ft_attr)
1085 {
1086         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1087 }
1088
1089 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1090                                                      int prio, int max_fte,
1091                                                      u32 level, u16 vport)
1092 {
1093         struct mlx5_flow_table_attr ft_attr = {};
1094
1095         ft_attr.max_fte = max_fte;
1096         ft_attr.level   = level;
1097         ft_attr.prio    = prio;
1098
1099         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1100 }
1101
1102 struct mlx5_flow_table*
1103 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1104                                  int prio, u32 level)
1105 {
1106         struct mlx5_flow_table_attr ft_attr = {};
1107
1108         ft_attr.level = level;
1109         ft_attr.prio  = prio;
1110         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1111 }
1112 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1113
1114 struct mlx5_flow_table*
1115 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1116                                     int prio,
1117                                     int num_flow_table_entries,
1118                                     int max_num_groups,
1119                                     u32 level,
1120                                     u32 flags)
1121 {
1122         struct mlx5_flow_table_attr ft_attr = {};
1123         struct mlx5_flow_table *ft;
1124
1125         if (max_num_groups > num_flow_table_entries)
1126                 return ERR_PTR(-EINVAL);
1127
1128         ft_attr.max_fte = num_flow_table_entries;
1129         ft_attr.prio    = prio;
1130         ft_attr.level   = level;
1131         ft_attr.flags   = flags;
1132
1133         ft = mlx5_create_flow_table(ns, &ft_attr);
1134         if (IS_ERR(ft))
1135                 return ft;
1136
1137         ft->autogroup.active = true;
1138         ft->autogroup.required_groups = max_num_groups;
1139
1140         return ft;
1141 }
1142 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1143
1144 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1145                                                u32 *fg_in)
1146 {
1147         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1148         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1149                                             fg_in, match_criteria);
1150         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1151                                             fg_in,
1152                                             match_criteria_enable);
1153         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1154                                    start_flow_index);
1155         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1156                                  end_flow_index);
1157         struct mlx5_core_dev *dev = get_dev(&ft->node);
1158         struct mlx5_flow_group *fg;
1159         int err;
1160
1161         if (!check_valid_mask(match_criteria_enable, match_criteria))
1162                 return ERR_PTR(-EINVAL);
1163
1164         if (ft->autogroup.active)
1165                 return ERR_PTR(-EPERM);
1166
1167         down_write_ref_node(&ft->node);
1168         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1169                                      start_index, end_index,
1170                                      ft->node.children.prev);
1171         up_write_ref_node(&ft->node);
1172         if (IS_ERR(fg))
1173                 return fg;
1174
1175         err = root->cmds->create_flow_group(dev, ft, fg_in, &fg->id);
1176         if (err) {
1177                 tree_put_node(&fg->node);
1178                 return ERR_PTR(err);
1179         }
1180         trace_mlx5_fs_add_fg(fg);
1181         fg->node.active = true;
1182
1183         return fg;
1184 }
1185
1186 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1187 {
1188         struct mlx5_flow_rule *rule;
1189
1190         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1191         if (!rule)
1192                 return NULL;
1193
1194         INIT_LIST_HEAD(&rule->next_ft);
1195         rule->node.type = FS_TYPE_FLOW_DEST;
1196         if (dest)
1197                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1198
1199         return rule;
1200 }
1201
1202 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1203 {
1204         struct mlx5_flow_handle *handle;
1205
1206         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1207         if (!handle)
1208                 return NULL;
1209
1210         handle->num_rules = num_rules;
1211
1212         return handle;
1213 }
1214
1215 static void destroy_flow_handle(struct fs_fte *fte,
1216                                 struct mlx5_flow_handle *handle,
1217                                 struct mlx5_flow_destination *dest,
1218                                 int i)
1219 {
1220         for (; --i >= 0;) {
1221                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1222                         fte->dests_size--;
1223                         list_del(&handle->rule[i]->node.list);
1224                         kfree(handle->rule[i]);
1225                 }
1226         }
1227         kfree(handle);
1228 }
1229
1230 static struct mlx5_flow_handle *
1231 create_flow_handle(struct fs_fte *fte,
1232                    struct mlx5_flow_destination *dest,
1233                    int dest_num,
1234                    int *modify_mask,
1235                    bool *new_rule)
1236 {
1237         struct mlx5_flow_handle *handle;
1238         struct mlx5_flow_rule *rule = NULL;
1239         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1240         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1241         int type;
1242         int i = 0;
1243
1244         handle = alloc_handle((dest_num) ? dest_num : 1);
1245         if (!handle)
1246                 return ERR_PTR(-ENOMEM);
1247
1248         do {
1249                 if (dest) {
1250                         rule = find_flow_rule(fte, dest + i);
1251                         if (rule) {
1252                                 refcount_inc(&rule->node.refcount);
1253                                 goto rule_found;
1254                         }
1255                 }
1256
1257                 *new_rule = true;
1258                 rule = alloc_rule(dest + i);
1259                 if (!rule)
1260                         goto free_rules;
1261
1262                 /* Add dest to dests list- we need flow tables to be in the
1263                  * end of the list for forward to next prio rules.
1264                  */
1265                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1266                 if (dest &&
1267                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1268                         list_add(&rule->node.list, &fte->node.children);
1269                 else
1270                         list_add_tail(&rule->node.list, &fte->node.children);
1271                 if (dest) {
1272                         fte->dests_size++;
1273
1274                         type = dest[i].type ==
1275                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1276                         *modify_mask |= type ? count : dst;
1277                 }
1278 rule_found:
1279                 handle->rule[i] = rule;
1280         } while (++i < dest_num);
1281
1282         return handle;
1283
1284 free_rules:
1285         destroy_flow_handle(fte, handle, dest, i);
1286         return ERR_PTR(-ENOMEM);
1287 }
1288
1289 /* fte should not be deleted while calling this function */
1290 static struct mlx5_flow_handle *
1291 add_rule_fte(struct fs_fte *fte,
1292              struct mlx5_flow_group *fg,
1293              struct mlx5_flow_destination *dest,
1294              int dest_num,
1295              bool update_action)
1296 {
1297         struct mlx5_flow_root_namespace *root;
1298         struct mlx5_flow_handle *handle;
1299         struct mlx5_flow_table *ft;
1300         int modify_mask = 0;
1301         int err;
1302         bool new_rule = false;
1303
1304         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1305                                     &new_rule);
1306         if (IS_ERR(handle) || !new_rule)
1307                 goto out;
1308
1309         if (update_action)
1310                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1311
1312         fs_get_obj(ft, fg->node.parent);
1313         root = find_root(&fg->node);
1314         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1315                 err = root->cmds->create_fte(get_dev(&ft->node),
1316                                              ft, fg, fte);
1317         else
1318                 err = root->cmds->update_fte(get_dev(&ft->node), ft, fg->id,
1319                                                      modify_mask, fte);
1320         if (err)
1321                 goto free_handle;
1322
1323         fte->node.active = true;
1324         fte->status |= FS_FTE_STATUS_EXISTING;
1325         atomic_inc(&fte->node.version);
1326
1327 out:
1328         return handle;
1329
1330 free_handle:
1331         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1332         return ERR_PTR(err);
1333 }
1334
1335 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1336                                                      struct mlx5_flow_spec *spec)
1337 {
1338         struct list_head *prev = &ft->node.children;
1339         struct mlx5_flow_group *fg;
1340         unsigned int candidate_index = 0;
1341         unsigned int group_size = 0;
1342
1343         if (!ft->autogroup.active)
1344                 return ERR_PTR(-ENOENT);
1345
1346         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1347                 /* We save place for flow groups in addition to max types */
1348                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1349
1350         /*  ft->max_fte == ft->autogroup.max_types */
1351         if (group_size == 0)
1352                 group_size = 1;
1353
1354         /* sorted by start_index */
1355         fs_for_each_fg(fg, ft) {
1356                 if (candidate_index + group_size > fg->start_index)
1357                         candidate_index = fg->start_index + fg->max_ftes;
1358                 else
1359                         break;
1360                 prev = &fg->node.list;
1361         }
1362
1363         if (candidate_index + group_size > ft->max_fte)
1364                 return ERR_PTR(-ENOSPC);
1365
1366         fg = alloc_insert_flow_group(ft,
1367                                      spec->match_criteria_enable,
1368                                      spec->match_criteria,
1369                                      candidate_index,
1370                                      candidate_index + group_size - 1,
1371                                      prev);
1372         if (IS_ERR(fg))
1373                 goto out;
1374
1375         ft->autogroup.num_groups++;
1376
1377 out:
1378         return fg;
1379 }
1380
1381 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1382                                   struct mlx5_flow_group *fg)
1383 {
1384         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1385         struct mlx5_core_dev *dev = get_dev(&ft->node);
1386         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1387         void *match_criteria_addr;
1388         u8 src_esw_owner_mask_on;
1389         void *misc;
1390         int err;
1391         u32 *in;
1392
1393         in = kvzalloc(inlen, GFP_KERNEL);
1394         if (!in)
1395                 return -ENOMEM;
1396
1397         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1398                  fg->mask.match_criteria_enable);
1399         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1400         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1401                  fg->max_ftes - 1);
1402
1403         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1404                             misc_parameters);
1405         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1406                                          source_eswitch_owner_vhca_id);
1407         MLX5_SET(create_flow_group_in, in,
1408                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1409
1410         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1411                                            in, match_criteria);
1412         memcpy(match_criteria_addr, fg->mask.match_criteria,
1413                sizeof(fg->mask.match_criteria));
1414
1415         err = root->cmds->create_flow_group(dev, ft, in, &fg->id);
1416         if (!err) {
1417                 fg->node.active = true;
1418                 trace_mlx5_fs_add_fg(fg);
1419         }
1420
1421         kvfree(in);
1422         return err;
1423 }
1424
1425 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1426                                 struct mlx5_flow_destination *d2)
1427 {
1428         if (d1->type == d2->type) {
1429                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1430                      d1->vport.num == d2->vport.num) ||
1431                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1432                      d1->ft == d2->ft) ||
1433                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1434                      d1->tir_num == d2->tir_num))
1435                         return true;
1436         }
1437
1438         return false;
1439 }
1440
1441 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1442                                              struct mlx5_flow_destination *dest)
1443 {
1444         struct mlx5_flow_rule *rule;
1445
1446         list_for_each_entry(rule, &fte->node.children, node.list) {
1447                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1448                         return rule;
1449         }
1450         return NULL;
1451 }
1452
1453 static bool check_conflicting_actions(u32 action1, u32 action2)
1454 {
1455         u32 xored_actions = action1 ^ action2;
1456
1457         /* if one rule only wants to count, it's ok */
1458         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1459             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1460                 return false;
1461
1462         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1463                              MLX5_FLOW_CONTEXT_ACTION_ENCAP |
1464                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1465                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1466                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1467                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH))
1468                 return true;
1469
1470         return false;
1471 }
1472
1473 static int check_conflicting_ftes(struct fs_fte *fte, const struct mlx5_flow_act *flow_act)
1474 {
1475         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1476                 mlx5_core_warn(get_dev(&fte->node),
1477                                "Found two FTEs with conflicting actions\n");
1478                 return -EEXIST;
1479         }
1480
1481         if (flow_act->has_flow_tag &&
1482             fte->action.flow_tag != flow_act->flow_tag) {
1483                 mlx5_core_warn(get_dev(&fte->node),
1484                                "FTE flow tag %u already exists with different flow tag %u\n",
1485                                fte->action.flow_tag,
1486                                flow_act->flow_tag);
1487                 return -EEXIST;
1488         }
1489
1490         return 0;
1491 }
1492
1493 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1494                                             u32 *match_value,
1495                                             struct mlx5_flow_act *flow_act,
1496                                             struct mlx5_flow_destination *dest,
1497                                             int dest_num,
1498                                             struct fs_fte *fte)
1499 {
1500         struct mlx5_flow_handle *handle;
1501         int old_action;
1502         int i;
1503         int ret;
1504
1505         ret = check_conflicting_ftes(fte, flow_act);
1506         if (ret)
1507                 return ERR_PTR(ret);
1508
1509         old_action = fte->action.action;
1510         fte->action.action |= flow_act->action;
1511         handle = add_rule_fte(fte, fg, dest, dest_num,
1512                               old_action != flow_act->action);
1513         if (IS_ERR(handle)) {
1514                 fte->action.action = old_action;
1515                 return handle;
1516         }
1517         trace_mlx5_fs_set_fte(fte, false);
1518
1519         for (i = 0; i < handle->num_rules; i++) {
1520                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1521                         tree_add_node(&handle->rule[i]->node, &fte->node);
1522                         trace_mlx5_fs_add_rule(handle->rule[i]);
1523                 }
1524         }
1525         return handle;
1526 }
1527
1528 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_handle *handle)
1529 {
1530         struct mlx5_flow_rule *dst;
1531         struct fs_fte *fte;
1532
1533         fs_get_obj(fte, handle->rule[0]->node.parent);
1534
1535         fs_for_each_dst(dst, fte) {
1536                 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1537                         return dst->dest_attr.counter;
1538         }
1539
1540         return NULL;
1541 }
1542
1543 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1544 {
1545         if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1546                 return !counter;
1547
1548         if (!counter)
1549                 return false;
1550
1551         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1552                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1553 }
1554
1555 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1556                           u32 action,
1557                           struct mlx5_flow_table *ft)
1558 {
1559         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1560                 return counter_is_valid(dest->counter, action);
1561
1562         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1563                 return true;
1564
1565         if (!dest || ((dest->type ==
1566             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1567             (dest->ft->level <= ft->level)))
1568                 return false;
1569         return true;
1570 }
1571
1572 struct match_list {
1573         struct list_head        list;
1574         struct mlx5_flow_group *g;
1575 };
1576
1577 struct match_list_head {
1578         struct list_head  list;
1579         struct match_list first;
1580 };
1581
1582 static void free_match_list(struct match_list_head *head)
1583 {
1584         if (!list_empty(&head->list)) {
1585                 struct match_list *iter, *match_tmp;
1586
1587                 list_del(&head->first.list);
1588                 tree_put_node(&head->first.g->node);
1589                 list_for_each_entry_safe(iter, match_tmp, &head->list,
1590                                          list) {
1591                         tree_put_node(&iter->g->node);
1592                         list_del(&iter->list);
1593                         kfree(iter);
1594                 }
1595         }
1596 }
1597
1598 static int build_match_list(struct match_list_head *match_head,
1599                             struct mlx5_flow_table *ft,
1600                             struct mlx5_flow_spec *spec)
1601 {
1602         struct rhlist_head *tmp, *list;
1603         struct mlx5_flow_group *g;
1604         int err = 0;
1605
1606         rcu_read_lock();
1607         INIT_LIST_HEAD(&match_head->list);
1608         /* Collect all fgs which has a matching match_criteria */
1609         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1610         /* RCU is atomic, we can't execute FW commands here */
1611         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1612                 struct match_list *curr_match;
1613
1614                 if (likely(list_empty(&match_head->list))) {
1615                         if (!tree_get_node(&g->node))
1616                                 continue;
1617                         match_head->first.g = g;
1618                         list_add_tail(&match_head->first.list,
1619                                       &match_head->list);
1620                         continue;
1621                 }
1622
1623                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1624                 if (!curr_match) {
1625                         free_match_list(match_head);
1626                         err = -ENOMEM;
1627                         goto out;
1628                 }
1629                 if (!tree_get_node(&g->node)) {
1630                         kfree(curr_match);
1631                         continue;
1632                 }
1633                 curr_match->g = g;
1634                 list_add_tail(&curr_match->list, &match_head->list);
1635         }
1636 out:
1637         rcu_read_unlock();
1638         return err;
1639 }
1640
1641 static u64 matched_fgs_get_version(struct list_head *match_head)
1642 {
1643         struct match_list *iter;
1644         u64 version = 0;
1645
1646         list_for_each_entry(iter, match_head, list)
1647                 version += (u64)atomic_read(&iter->g->node.version);
1648         return version;
1649 }
1650
1651 static struct mlx5_flow_handle *
1652 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1653                        struct list_head *match_head,
1654                        struct mlx5_flow_spec *spec,
1655                        struct mlx5_flow_act *flow_act,
1656                        struct mlx5_flow_destination *dest,
1657                        int dest_num,
1658                        int ft_version)
1659 {
1660         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1661         struct mlx5_flow_group *g;
1662         struct mlx5_flow_handle *rule;
1663         struct match_list *iter;
1664         bool take_write = false;
1665         struct fs_fte *fte;
1666         u64  version;
1667         int err;
1668
1669         fte = alloc_fte(ft, spec->match_value, flow_act);
1670         if (IS_ERR(fte))
1671                 return  ERR_PTR(-ENOMEM);
1672
1673         list_for_each_entry(iter, match_head, list) {
1674                 nested_down_read_ref_node(&iter->g->node, FS_LOCK_PARENT);
1675         }
1676
1677 search_again_locked:
1678         version = matched_fgs_get_version(match_head);
1679         /* Try to find a fg that already contains a matching fte */
1680         list_for_each_entry(iter, match_head, list) {
1681                 struct fs_fte *fte_tmp;
1682
1683                 g = iter->g;
1684                 fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, spec->match_value,
1685                                                  rhash_fte);
1686                 if (!fte_tmp || !tree_get_node(&fte_tmp->node))
1687                         continue;
1688
1689                 nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1690                 if (!take_write) {
1691                         list_for_each_entry(iter, match_head, list)
1692                                 up_read_ref_node(&iter->g->node);
1693                 } else {
1694                         list_for_each_entry(iter, match_head, list)
1695                                 up_write_ref_node(&iter->g->node);
1696                 }
1697
1698                 rule = add_rule_fg(g, spec->match_value,
1699                                    flow_act, dest, dest_num, fte_tmp);
1700                 up_write_ref_node(&fte_tmp->node);
1701                 tree_put_node(&fte_tmp->node);
1702                 kmem_cache_free(steering->ftes_cache, fte);
1703                 return rule;
1704         }
1705
1706         /* No group with matching fte found. Try to add a new fte to any
1707          * matching fg.
1708          */
1709
1710         if (!take_write) {
1711                 list_for_each_entry(iter, match_head, list)
1712                         up_read_ref_node(&iter->g->node);
1713                 list_for_each_entry(iter, match_head, list)
1714                         nested_down_write_ref_node(&iter->g->node,
1715                                                    FS_LOCK_PARENT);
1716                 take_write = true;
1717         }
1718
1719         /* Check the ft version, for case that new flow group
1720          * was added while the fgs weren't locked
1721          */
1722         if (atomic_read(&ft->node.version) != ft_version) {
1723                 rule = ERR_PTR(-EAGAIN);
1724                 goto out;
1725         }
1726
1727         /* Check the fgs version, for case the new FTE with the
1728          * same values was added while the fgs weren't locked
1729          */
1730         if (version != matched_fgs_get_version(match_head))
1731                 goto search_again_locked;
1732
1733         list_for_each_entry(iter, match_head, list) {
1734                 g = iter->g;
1735
1736                 if (!g->node.active)
1737                         continue;
1738                 err = insert_fte(g, fte);
1739                 if (err) {
1740                         if (err == -ENOSPC)
1741                                 continue;
1742                         list_for_each_entry(iter, match_head, list)
1743                                 up_write_ref_node(&iter->g->node);
1744                         kmem_cache_free(steering->ftes_cache, fte);
1745                         return ERR_PTR(err);
1746                 }
1747
1748                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1749                 list_for_each_entry(iter, match_head, list)
1750                         up_write_ref_node(&iter->g->node);
1751                 rule = add_rule_fg(g, spec->match_value,
1752                                    flow_act, dest, dest_num, fte);
1753                 up_write_ref_node(&fte->node);
1754                 tree_put_node(&fte->node);
1755                 return rule;
1756         }
1757         rule = ERR_PTR(-ENOENT);
1758 out:
1759         list_for_each_entry(iter, match_head, list)
1760                 up_write_ref_node(&iter->g->node);
1761         kmem_cache_free(steering->ftes_cache, fte);
1762         return rule;
1763 }
1764
1765 static struct mlx5_flow_handle *
1766 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1767                      struct mlx5_flow_spec *spec,
1768                      struct mlx5_flow_act *flow_act,
1769                      struct mlx5_flow_destination *dest,
1770                      int dest_num)
1771
1772 {
1773         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1774         struct mlx5_flow_group *g;
1775         struct mlx5_flow_handle *rule;
1776         struct match_list_head match_head;
1777         bool take_write = false;
1778         struct fs_fte *fte;
1779         int version;
1780         int err;
1781         int i;
1782
1783         if (!check_valid_spec(spec))
1784                 return ERR_PTR(-EINVAL);
1785
1786         for (i = 0; i < dest_num; i++) {
1787                 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1788                         return ERR_PTR(-EINVAL);
1789         }
1790         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1791 search_again_locked:
1792         version = atomic_read(&ft->node.version);
1793
1794         /* Collect all fgs which has a matching match_criteria */
1795         err = build_match_list(&match_head, ft, spec);
1796         if (err) {
1797                 if (take_write)
1798                         up_write_ref_node(&ft->node);
1799                 return ERR_PTR(err);
1800         }
1801
1802         if (!take_write)
1803                 up_read_ref_node(&ft->node);
1804
1805         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1806                                       dest_num, version);
1807         free_match_list(&match_head);
1808         if (!IS_ERR(rule) ||
1809             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1810                 if (take_write)
1811                         up_write_ref_node(&ft->node);
1812                 return rule;
1813         }
1814
1815         if (!take_write) {
1816                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1817                 take_write = true;
1818         }
1819
1820         if (PTR_ERR(rule) == -EAGAIN ||
1821             version != atomic_read(&ft->node.version))
1822                 goto search_again_locked;
1823
1824         g = alloc_auto_flow_group(ft, spec);
1825         if (IS_ERR(g)) {
1826                 rule = (void *)g;
1827                 up_write_ref_node(&ft->node);
1828                 return rule;
1829         }
1830
1831         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1832         up_write_ref_node(&ft->node);
1833
1834         err = create_auto_flow_group(ft, g);
1835         if (err)
1836                 goto err_release_fg;
1837
1838         fte = alloc_fte(ft, spec->match_value, flow_act);
1839         if (IS_ERR(fte)) {
1840                 err = PTR_ERR(fte);
1841                 goto err_release_fg;
1842         }
1843
1844         err = insert_fte(g, fte);
1845         if (err) {
1846                 kmem_cache_free(steering->ftes_cache, fte);
1847                 goto err_release_fg;
1848         }
1849
1850         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1851         up_write_ref_node(&g->node);
1852         rule = add_rule_fg(g, spec->match_value, flow_act, dest,
1853                            dest_num, fte);
1854         up_write_ref_node(&fte->node);
1855         tree_put_node(&fte->node);
1856         tree_put_node(&g->node);
1857         return rule;
1858
1859 err_release_fg:
1860         up_write_ref_node(&g->node);
1861         tree_put_node(&g->node);
1862         return ERR_PTR(err);
1863 }
1864
1865 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1866 {
1867         return ((ft->type == FS_FT_NIC_RX) &&
1868                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1869 }
1870
1871 struct mlx5_flow_handle *
1872 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1873                     struct mlx5_flow_spec *spec,
1874                     struct mlx5_flow_act *flow_act,
1875                     struct mlx5_flow_destination *dest,
1876                     int dest_num)
1877 {
1878         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1879         struct mlx5_flow_destination gen_dest = {};
1880         struct mlx5_flow_table *next_ft = NULL;
1881         struct mlx5_flow_handle *handle = NULL;
1882         u32 sw_action = flow_act->action;
1883         struct fs_prio *prio;
1884
1885         fs_get_obj(prio, ft->node.parent);
1886         if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1887                 if (!fwd_next_prio_supported(ft))
1888                         return ERR_PTR(-EOPNOTSUPP);
1889                 if (dest)
1890                         return ERR_PTR(-EINVAL);
1891                 mutex_lock(&root->chain_lock);
1892                 next_ft = find_next_chained_ft(prio);
1893                 if (next_ft) {
1894                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1895                         gen_dest.ft = next_ft;
1896                         dest = &gen_dest;
1897                         dest_num = 1;
1898                         flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1899                 } else {
1900                         mutex_unlock(&root->chain_lock);
1901                         return ERR_PTR(-EOPNOTSUPP);
1902                 }
1903         }
1904
1905         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, dest_num);
1906
1907         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1908                 if (!IS_ERR_OR_NULL(handle) &&
1909                     (list_empty(&handle->rule[0]->next_ft))) {
1910                         mutex_lock(&next_ft->lock);
1911                         list_add(&handle->rule[0]->next_ft,
1912                                  &next_ft->fwd_rules);
1913                         mutex_unlock(&next_ft->lock);
1914                         handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1915                 }
1916                 mutex_unlock(&root->chain_lock);
1917         }
1918         return handle;
1919 }
1920 EXPORT_SYMBOL(mlx5_add_flow_rules);
1921
1922 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1923 {
1924         int i;
1925
1926         for (i = handle->num_rules - 1; i >= 0; i--)
1927                 tree_remove_node(&handle->rule[i]->node);
1928         kfree(handle);
1929 }
1930 EXPORT_SYMBOL(mlx5_del_flow_rules);
1931
1932 /* Assuming prio->node.children(flow tables) is sorted by level */
1933 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1934 {
1935         struct fs_prio *prio;
1936
1937         fs_get_obj(prio, ft->node.parent);
1938
1939         if (!list_is_last(&ft->node.list, &prio->node.children))
1940                 return list_next_entry(ft, node.list);
1941         return find_next_chained_ft(prio);
1942 }
1943
1944 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1945 {
1946         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1947         struct mlx5_ft_underlay_qp *uqp;
1948         struct mlx5_flow_table *new_root_ft = NULL;
1949         int err = 0;
1950         u32 qpn;
1951
1952         if (root->root_ft != ft)
1953                 return 0;
1954
1955         new_root_ft = find_next_ft(ft);
1956         if (!new_root_ft) {
1957                 root->root_ft = NULL;
1958                 return 0;
1959         }
1960
1961         if (list_empty(&root->underlay_qpns)) {
1962                 /* Don't set any QPN (zero) in case QPN list is empty */
1963                 qpn = 0;
1964                 err = root->cmds->update_root_ft(root->dev, new_root_ft,
1965                                                  qpn, false);
1966         } else {
1967                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
1968                         qpn = uqp->qpn;
1969                         err = root->cmds->update_root_ft(root->dev,
1970                                                          new_root_ft, qpn,
1971                                                          false);
1972                         if (err)
1973                                 break;
1974                 }
1975         }
1976
1977         if (err)
1978                 mlx5_core_warn(root->dev,
1979                                "Update root flow table of id(%u) qpn(%d) failed\n",
1980                                ft->id, qpn);
1981         else
1982                 root->root_ft = new_root_ft;
1983
1984         return 0;
1985 }
1986
1987 /* Connect flow table from previous priority to
1988  * the next flow table.
1989  */
1990 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1991 {
1992         struct mlx5_core_dev *dev = get_dev(&ft->node);
1993         struct mlx5_flow_table *next_ft;
1994         struct fs_prio *prio;
1995         int err = 0;
1996
1997         err = update_root_ft_destroy(ft);
1998         if (err)
1999                 return err;
2000
2001         fs_get_obj(prio, ft->node.parent);
2002         if  (!(list_first_entry(&prio->node.children,
2003                                 struct mlx5_flow_table,
2004                                 node.list) == ft))
2005                 return 0;
2006
2007         next_ft = find_next_chained_ft(prio);
2008         err = connect_fwd_rules(dev, next_ft, ft);
2009         if (err)
2010                 return err;
2011
2012         err = connect_prev_fts(dev, next_ft, prio);
2013         if (err)
2014                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2015                                ft->id);
2016         return err;
2017 }
2018
2019 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2020 {
2021         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2022         int err = 0;
2023
2024         mutex_lock(&root->chain_lock);
2025         err = disconnect_flow_table(ft);
2026         if (err) {
2027                 mutex_unlock(&root->chain_lock);
2028                 return err;
2029         }
2030         if (tree_remove_node(&ft->node))
2031                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2032                                ft->id);
2033         mutex_unlock(&root->chain_lock);
2034
2035         return err;
2036 }
2037 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2038
2039 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2040 {
2041         if (tree_remove_node(&fg->node))
2042                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2043                                fg->id);
2044 }
2045
2046 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2047                                                     enum mlx5_flow_namespace_type type)
2048 {
2049         struct mlx5_flow_steering *steering = dev->priv.steering;
2050         struct mlx5_flow_root_namespace *root_ns;
2051         int prio;
2052         struct fs_prio *fs_prio;
2053         struct mlx5_flow_namespace *ns;
2054
2055         if (!steering)
2056                 return NULL;
2057
2058         switch (type) {
2059         case MLX5_FLOW_NAMESPACE_BYPASS:
2060         case MLX5_FLOW_NAMESPACE_LAG:
2061         case MLX5_FLOW_NAMESPACE_OFFLOADS:
2062         case MLX5_FLOW_NAMESPACE_ETHTOOL:
2063         case MLX5_FLOW_NAMESPACE_KERNEL:
2064         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
2065         case MLX5_FLOW_NAMESPACE_ANCHOR:
2066                 prio = type;
2067                 break;
2068         case MLX5_FLOW_NAMESPACE_FDB:
2069                 if (steering->fdb_root_ns)
2070                         return &steering->fdb_root_ns->ns;
2071                 else
2072                         return NULL;
2073         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2074                 if (steering->sniffer_rx_root_ns)
2075                         return &steering->sniffer_rx_root_ns->ns;
2076                 else
2077                         return NULL;
2078         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2079                 if (steering->sniffer_tx_root_ns)
2080                         return &steering->sniffer_tx_root_ns->ns;
2081                 else
2082                         return NULL;
2083         case MLX5_FLOW_NAMESPACE_EGRESS:
2084                 if (steering->egress_root_ns)
2085                         return &steering->egress_root_ns->ns;
2086                 else
2087                         return NULL;
2088         default:
2089                 return NULL;
2090         }
2091
2092         root_ns = steering->root_ns;
2093         if (!root_ns)
2094                 return NULL;
2095
2096         fs_prio = find_prio(&root_ns->ns, prio);
2097         if (!fs_prio)
2098                 return NULL;
2099
2100         ns = list_first_entry(&fs_prio->node.children,
2101                               typeof(*ns),
2102                               node.list);
2103
2104         return ns;
2105 }
2106 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2107
2108 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2109                                                               enum mlx5_flow_namespace_type type,
2110                                                               int vport)
2111 {
2112         struct mlx5_flow_steering *steering = dev->priv.steering;
2113
2114         if (!steering || vport >= MLX5_TOTAL_VPORTS(dev))
2115                 return NULL;
2116
2117         switch (type) {
2118         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2119                 if (steering->esw_egress_root_ns &&
2120                     steering->esw_egress_root_ns[vport])
2121                         return &steering->esw_egress_root_ns[vport]->ns;
2122                 else
2123                         return NULL;
2124         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2125                 if (steering->esw_ingress_root_ns &&
2126                     steering->esw_ingress_root_ns[vport])
2127                         return &steering->esw_ingress_root_ns[vport]->ns;
2128                 else
2129                         return NULL;
2130         default:
2131                 return NULL;
2132         }
2133 }
2134
2135 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2136                                       unsigned int prio, int num_levels)
2137 {
2138         struct fs_prio *fs_prio;
2139
2140         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2141         if (!fs_prio)
2142                 return ERR_PTR(-ENOMEM);
2143
2144         fs_prio->node.type = FS_TYPE_PRIO;
2145         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2146         tree_add_node(&fs_prio->node, &ns->node);
2147         fs_prio->num_levels = num_levels;
2148         fs_prio->prio = prio;
2149         list_add_tail(&fs_prio->node.list, &ns->node.children);
2150
2151         return fs_prio;
2152 }
2153
2154 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2155                                                      *ns)
2156 {
2157         ns->node.type = FS_TYPE_NAMESPACE;
2158
2159         return ns;
2160 }
2161
2162 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
2163 {
2164         struct mlx5_flow_namespace      *ns;
2165
2166         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2167         if (!ns)
2168                 return ERR_PTR(-ENOMEM);
2169
2170         fs_init_namespace(ns);
2171         tree_init_node(&ns->node, NULL, del_sw_ns);
2172         tree_add_node(&ns->node, &prio->node);
2173         list_add_tail(&ns->node.list, &prio->node.children);
2174
2175         return ns;
2176 }
2177
2178 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2179                              struct init_tree_node *prio_metadata)
2180 {
2181         struct fs_prio *fs_prio;
2182         int i;
2183
2184         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2185                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2186                 if (IS_ERR(fs_prio))
2187                         return PTR_ERR(fs_prio);
2188         }
2189         return 0;
2190 }
2191
2192 #define FLOW_TABLE_BIT_SZ 1
2193 #define GET_FLOW_TABLE_CAP(dev, offset) \
2194         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2195                         offset / 32)) >>                                        \
2196           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2197 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2198 {
2199         int i;
2200
2201         for (i = 0; i < caps->arr_sz; i++) {
2202                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2203                         return false;
2204         }
2205         return true;
2206 }
2207
2208 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2209                                     struct init_tree_node *init_node,
2210                                     struct fs_node *fs_parent_node,
2211                                     struct init_tree_node *init_parent_node,
2212                                     int prio)
2213 {
2214         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2215                                               flow_table_properties_nic_receive.
2216                                               max_ft_level);
2217         struct mlx5_flow_namespace *fs_ns;
2218         struct fs_prio *fs_prio;
2219         struct fs_node *base;
2220         int i;
2221         int err;
2222
2223         if (init_node->type == FS_TYPE_PRIO) {
2224                 if ((init_node->min_ft_level > max_ft_level) ||
2225                     !has_required_caps(steering->dev, &init_node->caps))
2226                         return 0;
2227
2228                 fs_get_obj(fs_ns, fs_parent_node);
2229                 if (init_node->num_leaf_prios)
2230                         return create_leaf_prios(fs_ns, prio, init_node);
2231                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2232                 if (IS_ERR(fs_prio))
2233                         return PTR_ERR(fs_prio);
2234                 base = &fs_prio->node;
2235         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2236                 fs_get_obj(fs_prio, fs_parent_node);
2237                 fs_ns = fs_create_namespace(fs_prio);
2238                 if (IS_ERR(fs_ns))
2239                         return PTR_ERR(fs_ns);
2240                 base = &fs_ns->node;
2241         } else {
2242                 return -EINVAL;
2243         }
2244         prio = 0;
2245         for (i = 0; i < init_node->ar_size; i++) {
2246                 err = init_root_tree_recursive(steering, &init_node->children[i],
2247                                                base, init_node, prio);
2248                 if (err)
2249                         return err;
2250                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2251                     init_node->children[i].num_leaf_prios) {
2252                         prio += init_node->children[i].num_leaf_prios;
2253                 }
2254         }
2255
2256         return 0;
2257 }
2258
2259 static int init_root_tree(struct mlx5_flow_steering *steering,
2260                           struct init_tree_node *init_node,
2261                           struct fs_node *fs_parent_node)
2262 {
2263         int i;
2264         struct mlx5_flow_namespace *fs_ns;
2265         int err;
2266
2267         fs_get_obj(fs_ns, fs_parent_node);
2268         for (i = 0; i < init_node->ar_size; i++) {
2269                 err = init_root_tree_recursive(steering, &init_node->children[i],
2270                                                &fs_ns->node,
2271                                                init_node, i);
2272                 if (err)
2273                         return err;
2274         }
2275         return 0;
2276 }
2277
2278 static struct mlx5_flow_root_namespace
2279 *create_root_ns(struct mlx5_flow_steering *steering,
2280                 enum fs_flow_table_type table_type)
2281 {
2282         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2283         struct mlx5_flow_root_namespace *root_ns;
2284         struct mlx5_flow_namespace *ns;
2285
2286         if (mlx5_accel_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2287             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2288                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2289
2290         /* Create the root namespace */
2291         root_ns = kvzalloc(sizeof(*root_ns), GFP_KERNEL);
2292         if (!root_ns)
2293                 return NULL;
2294
2295         root_ns->dev = steering->dev;
2296         root_ns->table_type = table_type;
2297         root_ns->cmds = cmds;
2298
2299         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2300
2301         ns = &root_ns->ns;
2302         fs_init_namespace(ns);
2303         mutex_init(&root_ns->chain_lock);
2304         tree_init_node(&ns->node, NULL, NULL);
2305         tree_add_node(&ns->node, NULL);
2306
2307         return root_ns;
2308 }
2309
2310 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2311
2312 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2313 {
2314         struct fs_prio *prio;
2315
2316         fs_for_each_prio(prio, ns) {
2317                  /* This updates prio start_level and num_levels */
2318                 set_prio_attrs_in_prio(prio, acc_level);
2319                 acc_level += prio->num_levels;
2320         }
2321         return acc_level;
2322 }
2323
2324 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2325 {
2326         struct mlx5_flow_namespace *ns;
2327         int acc_level_ns = acc_level;
2328
2329         prio->start_level = acc_level;
2330         fs_for_each_ns(ns, prio)
2331                 /* This updates start_level and num_levels of ns's priority descendants */
2332                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2333         if (!prio->num_levels)
2334                 prio->num_levels = acc_level_ns - prio->start_level;
2335         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2336 }
2337
2338 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2339 {
2340         struct mlx5_flow_namespace *ns = &root_ns->ns;
2341         struct fs_prio *prio;
2342         int start_level = 0;
2343
2344         fs_for_each_prio(prio, ns) {
2345                 set_prio_attrs_in_prio(prio, start_level);
2346                 start_level += prio->num_levels;
2347         }
2348 }
2349
2350 #define ANCHOR_PRIO 0
2351 #define ANCHOR_SIZE 1
2352 #define ANCHOR_LEVEL 0
2353 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2354 {
2355         struct mlx5_flow_namespace *ns = NULL;
2356         struct mlx5_flow_table_attr ft_attr = {};
2357         struct mlx5_flow_table *ft;
2358
2359         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2360         if (WARN_ON(!ns))
2361                 return -EINVAL;
2362
2363         ft_attr.max_fte = ANCHOR_SIZE;
2364         ft_attr.level   = ANCHOR_LEVEL;
2365         ft_attr.prio    = ANCHOR_PRIO;
2366
2367         ft = mlx5_create_flow_table(ns, &ft_attr);
2368         if (IS_ERR(ft)) {
2369                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2370                 return PTR_ERR(ft);
2371         }
2372         return 0;
2373 }
2374
2375 static int init_root_ns(struct mlx5_flow_steering *steering)
2376 {
2377         int err;
2378
2379         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2380         if (!steering->root_ns)
2381                 return -ENOMEM;
2382
2383         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2384         if (err)
2385                 goto out_err;
2386
2387         set_prio_attrs(steering->root_ns);
2388         err = create_anchor_flow_table(steering);
2389         if (err)
2390                 goto out_err;
2391
2392         return 0;
2393
2394 out_err:
2395         cleanup_root_ns(steering->root_ns);
2396         steering->root_ns = NULL;
2397         return err;
2398 }
2399
2400 static void clean_tree(struct fs_node *node)
2401 {
2402         if (node) {
2403                 struct fs_node *iter;
2404                 struct fs_node *temp;
2405
2406                 tree_get_node(node);
2407                 list_for_each_entry_safe(iter, temp, &node->children, list)
2408                         clean_tree(iter);
2409                 tree_put_node(node);
2410                 tree_remove_node(node);
2411         }
2412 }
2413
2414 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2415 {
2416         if (!root_ns)
2417                 return;
2418
2419         clean_tree(&root_ns->ns.node);
2420 }
2421
2422 static void cleanup_egress_acls_root_ns(struct mlx5_core_dev *dev)
2423 {
2424         struct mlx5_flow_steering *steering = dev->priv.steering;
2425         int i;
2426
2427         if (!steering->esw_egress_root_ns)
2428                 return;
2429
2430         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2431                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2432
2433         kfree(steering->esw_egress_root_ns);
2434 }
2435
2436 static void cleanup_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2437 {
2438         struct mlx5_flow_steering *steering = dev->priv.steering;
2439         int i;
2440
2441         if (!steering->esw_ingress_root_ns)
2442                 return;
2443
2444         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2445                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2446
2447         kfree(steering->esw_ingress_root_ns);
2448 }
2449
2450 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2451 {
2452         struct mlx5_flow_steering *steering = dev->priv.steering;
2453
2454         cleanup_root_ns(steering->root_ns);
2455         cleanup_egress_acls_root_ns(dev);
2456         cleanup_ingress_acls_root_ns(dev);
2457         cleanup_root_ns(steering->fdb_root_ns);
2458         cleanup_root_ns(steering->sniffer_rx_root_ns);
2459         cleanup_root_ns(steering->sniffer_tx_root_ns);
2460         cleanup_root_ns(steering->egress_root_ns);
2461         mlx5_cleanup_fc_stats(dev);
2462         kmem_cache_destroy(steering->ftes_cache);
2463         kmem_cache_destroy(steering->fgs_cache);
2464         kfree(steering);
2465 }
2466
2467 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2468 {
2469         struct fs_prio *prio;
2470
2471         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2472         if (!steering->sniffer_tx_root_ns)
2473                 return -ENOMEM;
2474
2475         /* Create single prio */
2476         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2477         if (IS_ERR(prio)) {
2478                 cleanup_root_ns(steering->sniffer_tx_root_ns);
2479                 return PTR_ERR(prio);
2480         }
2481         return 0;
2482 }
2483
2484 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2485 {
2486         struct fs_prio *prio;
2487
2488         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2489         if (!steering->sniffer_rx_root_ns)
2490                 return -ENOMEM;
2491
2492         /* Create single prio */
2493         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2494         if (IS_ERR(prio)) {
2495                 cleanup_root_ns(steering->sniffer_rx_root_ns);
2496                 return PTR_ERR(prio);
2497         }
2498         return 0;
2499 }
2500
2501 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2502 {
2503         struct fs_prio *prio;
2504
2505         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2506         if (!steering->fdb_root_ns)
2507                 return -ENOMEM;
2508
2509         prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 2);
2510         if (IS_ERR(prio))
2511                 goto out_err;
2512
2513         prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
2514         if (IS_ERR(prio))
2515                 goto out_err;
2516
2517         set_prio_attrs(steering->fdb_root_ns);
2518         return 0;
2519
2520 out_err:
2521         cleanup_root_ns(steering->fdb_root_ns);
2522         steering->fdb_root_ns = NULL;
2523         return PTR_ERR(prio);
2524 }
2525
2526 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2527 {
2528         struct fs_prio *prio;
2529
2530         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2531         if (!steering->esw_egress_root_ns[vport])
2532                 return -ENOMEM;
2533
2534         /* create 1 prio*/
2535         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2536         return PTR_ERR_OR_ZERO(prio);
2537 }
2538
2539 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2540 {
2541         struct fs_prio *prio;
2542
2543         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2544         if (!steering->esw_ingress_root_ns[vport])
2545                 return -ENOMEM;
2546
2547         /* create 1 prio*/
2548         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2549         return PTR_ERR_OR_ZERO(prio);
2550 }
2551
2552 static int init_egress_acls_root_ns(struct mlx5_core_dev *dev)
2553 {
2554         struct mlx5_flow_steering *steering = dev->priv.steering;
2555         int err;
2556         int i;
2557
2558         steering->esw_egress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2559                                                sizeof(*steering->esw_egress_root_ns),
2560                                                GFP_KERNEL);
2561         if (!steering->esw_egress_root_ns)
2562                 return -ENOMEM;
2563
2564         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2565                 err = init_egress_acl_root_ns(steering, i);
2566                 if (err)
2567                         goto cleanup_root_ns;
2568         }
2569
2570         return 0;
2571
2572 cleanup_root_ns:
2573         for (i--; i >= 0; i--)
2574                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2575         kfree(steering->esw_egress_root_ns);
2576         return err;
2577 }
2578
2579 static int init_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2580 {
2581         struct mlx5_flow_steering *steering = dev->priv.steering;
2582         int err;
2583         int i;
2584
2585         steering->esw_ingress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2586                                                 sizeof(*steering->esw_ingress_root_ns),
2587                                                 GFP_KERNEL);
2588         if (!steering->esw_ingress_root_ns)
2589                 return -ENOMEM;
2590
2591         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2592                 err = init_ingress_acl_root_ns(steering, i);
2593                 if (err)
2594                         goto cleanup_root_ns;
2595         }
2596
2597         return 0;
2598
2599 cleanup_root_ns:
2600         for (i--; i >= 0; i--)
2601                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2602         kfree(steering->esw_ingress_root_ns);
2603         return err;
2604 }
2605
2606 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2607 {
2608         struct fs_prio *prio;
2609
2610         steering->egress_root_ns = create_root_ns(steering,
2611                                                   FS_FT_NIC_TX);
2612         if (!steering->egress_root_ns)
2613                 return -ENOMEM;
2614
2615         /* create 1 prio*/
2616         prio = fs_create_prio(&steering->egress_root_ns->ns, 0, 1);
2617         return PTR_ERR_OR_ZERO(prio);
2618 }
2619
2620 int mlx5_init_fs(struct mlx5_core_dev *dev)
2621 {
2622         struct mlx5_flow_steering *steering;
2623         int err = 0;
2624
2625         err = mlx5_init_fc_stats(dev);
2626         if (err)
2627                 return err;
2628
2629         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2630         if (!steering)
2631                 return -ENOMEM;
2632         steering->dev = dev;
2633         dev->priv.steering = steering;
2634
2635         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2636                                                 sizeof(struct mlx5_flow_group), 0,
2637                                                 0, NULL);
2638         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2639                                                  0, NULL);
2640         if (!steering->ftes_cache || !steering->fgs_cache) {
2641                 err = -ENOMEM;
2642                 goto err;
2643         }
2644
2645         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2646               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2647              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2648               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2649             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2650                 err = init_root_ns(steering);
2651                 if (err)
2652                         goto err;
2653         }
2654
2655         if (MLX5_CAP_GEN(dev, eswitch_flow_table)) {
2656                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2657                         err = init_fdb_root_ns(steering);
2658                         if (err)
2659                                 goto err;
2660                 }
2661                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2662                         err = init_egress_acls_root_ns(dev);
2663                         if (err)
2664                                 goto err;
2665                 }
2666                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2667                         err = init_ingress_acls_root_ns(dev);
2668                         if (err)
2669                                 goto err;
2670                 }
2671         }
2672
2673         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2674                 err = init_sniffer_rx_root_ns(steering);
2675                 if (err)
2676                         goto err;
2677         }
2678
2679         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2680                 err = init_sniffer_tx_root_ns(steering);
2681                 if (err)
2682                         goto err;
2683         }
2684
2685         if (MLX5_IPSEC_DEV(dev)) {
2686                 err = init_egress_root_ns(steering);
2687                 if (err)
2688                         goto err;
2689         }
2690
2691         return 0;
2692 err:
2693         mlx5_cleanup_fs(dev);
2694         return err;
2695 }
2696
2697 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2698 {
2699         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2700         struct mlx5_ft_underlay_qp *new_uqp;
2701         int err = 0;
2702
2703         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
2704         if (!new_uqp)
2705                 return -ENOMEM;
2706
2707         mutex_lock(&root->chain_lock);
2708
2709         if (!root->root_ft) {
2710                 err = -EINVAL;
2711                 goto update_ft_fail;
2712         }
2713
2714         err = root->cmds->update_root_ft(dev, root->root_ft, underlay_qpn,
2715                                          false);
2716         if (err) {
2717                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
2718                                underlay_qpn, err);
2719                 goto update_ft_fail;
2720         }
2721
2722         new_uqp->qpn = underlay_qpn;
2723         list_add_tail(&new_uqp->list, &root->underlay_qpns);
2724
2725         mutex_unlock(&root->chain_lock);
2726
2727         return 0;
2728
2729 update_ft_fail:
2730         mutex_unlock(&root->chain_lock);
2731         kfree(new_uqp);
2732         return err;
2733 }
2734 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
2735
2736 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2737 {
2738         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2739         struct mlx5_ft_underlay_qp *uqp;
2740         bool found = false;
2741         int err = 0;
2742
2743         mutex_lock(&root->chain_lock);
2744         list_for_each_entry(uqp, &root->underlay_qpns, list) {
2745                 if (uqp->qpn == underlay_qpn) {
2746                         found = true;
2747                         break;
2748                 }
2749         }
2750
2751         if (!found) {
2752                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
2753                                underlay_qpn);
2754                 err = -EINVAL;
2755                 goto out;
2756         }
2757
2758         err = root->cmds->update_root_ft(dev, root->root_ft, underlay_qpn,
2759                                          true);
2760         if (err)
2761                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
2762                                underlay_qpn, err);
2763
2764         list_del(&uqp->list);
2765         mutex_unlock(&root->chain_lock);
2766         kfree(uqp);
2767
2768         return 0;
2769
2770 out:
2771         mutex_unlock(&root->chain_lock);
2772         return err;
2773 }
2774 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);