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