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 #include <linux/mlx5/vport.h>
  36 #include <linux/mlx5/eswitch.h>
  37
  38 #include "mlx5_core.h"
  39 #include "fs_core.h"
  40 #include "fs_cmd.h"
  41 #include "fs_ft_pool.h"
  42 #include "diag/fs_tracepoint.h"
  43 #include "accel/ipsec.h"
  44 #include "fpga/ipsec.h"
  45
  46 #define INIT_TREE_NODE_ARRAY_SIZE(...)  (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
  47                                          sizeof(struct init_tree_node))
  48
  49 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
  50                  ...) {.type = FS_TYPE_PRIO,\
  51         .min_ft_level = min_level_val,\
  52         .num_levels = num_levels_val,\
  53         .num_leaf_prios = num_prios_val,\
  54         .caps = caps_val,\
  55         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  56         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  57 }
  58
  59 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
  60         ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
  61                  __VA_ARGS__)\
  62
  63 #define ADD_NS(def_miss_act, ...) {.type = FS_TYPE_NAMESPACE,   \
  64         .def_miss_action = def_miss_act,\
  65         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  66         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  67 }
  68
  69 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  70                                    sizeof(long))
  71
  72 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  73
  74 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  75                                .caps = (long[]) {__VA_ARGS__} }
  76
  77 #define FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  78                                            FS_CAP(flow_table_properties_nic_receive.modify_root), \
  79                                            FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  80                                            FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  81
  82 #define FS_CHAINING_CAPS_EGRESS                                                \
  83         FS_REQUIRED_CAPS(                                                      \
  84                 FS_CAP(flow_table_properties_nic_transmit.flow_modify_en),     \
  85                 FS_CAP(flow_table_properties_nic_transmit.modify_root),        \
  86                 FS_CAP(flow_table_properties_nic_transmit                      \
  87                                .identified_miss_table_mode),                   \
  88                 FS_CAP(flow_table_properties_nic_transmit.flow_table_modify))
  89
  90 #define FS_CHAINING_CAPS_RDMA_TX                                                \
  91         FS_REQUIRED_CAPS(                                                       \
  92                 FS_CAP(flow_table_properties_nic_transmit_rdma.flow_modify_en), \
  93                 FS_CAP(flow_table_properties_nic_transmit_rdma.modify_root),    \
  94                 FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  95                                .identified_miss_table_mode),                    \
  96                 FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  97                                .flow_table_modify))
  98
  99 #define LEFTOVERS_NUM_LEVELS 1
 100 #define LEFTOVERS_NUM_PRIOS 1
 101
 102 #define BY_PASS_PRIO_NUM_LEVELS 1
 103 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
 104                            LEFTOVERS_NUM_PRIOS)
 105
 106 #define ETHTOOL_PRIO_NUM_LEVELS 1
 107 #define ETHTOOL_NUM_PRIOS 11
 108 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
 109 /* Promiscuous, Vlan, mac, ttc, inner ttc, {UDP/ANY/aRFS/accel/{esp, esp_err}} */
 110 #define KERNEL_NIC_PRIO_NUM_LEVELS 7
 111 #define KERNEL_NIC_NUM_PRIOS 1
 112 /* One more level for tc */
 113 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
 114
 115 #define KERNEL_NIC_TC_NUM_PRIOS  1
 116 #define KERNEL_NIC_TC_NUM_LEVELS 2
 117
 118 #define ANCHOR_NUM_LEVELS 1
 119 #define ANCHOR_NUM_PRIOS 1
 120 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 121
 122 #define OFFLOADS_MAX_FT 2
 123 #define OFFLOADS_NUM_PRIOS 2
 124 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS)
 125
 126 #define LAG_PRIO_NUM_LEVELS 1
 127 #define LAG_NUM_PRIOS 1
 128 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 129
 130 #define KERNEL_TX_IPSEC_NUM_PRIOS  1
 131 #define KERNEL_TX_IPSEC_NUM_LEVELS 1
 132 #define KERNEL_TX_MIN_LEVEL        (KERNEL_TX_IPSEC_NUM_LEVELS)
 133
 134 struct node_caps {
 135         size_t  arr_sz;
 136         long    *caps;
 137 };
 138
 139 static struct init_tree_node {
 140         enum fs_node_type       type;
 141         struct init_tree_node *children;
 142         int ar_size;
 143         struct node_caps caps;
 144         int min_ft_level;
 145         int num_leaf_prios;
 146         int prio;
 147         int num_levels;
 148         enum mlx5_flow_table_miss_action def_miss_action;
 149 } root_fs = {
 150         .type = FS_TYPE_NAMESPACE,
 151         .ar_size = 7,
 152           .children = (struct init_tree_node[]){
 153                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 154                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 155                                   ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 156                                                     BY_PASS_PRIO_NUM_LEVELS))),
 157                   ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 158                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 159                                   ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 160                                                     LAG_PRIO_NUM_LEVELS))),
 161                   ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 162                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 163                                   ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS,
 164                                                     OFFLOADS_MAX_FT))),
 165                   ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 166                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 167                                   ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 168                                                     ETHTOOL_PRIO_NUM_LEVELS))),
 169                   ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 170                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 171                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS,
 172                                                     KERNEL_NIC_TC_NUM_LEVELS),
 173                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 174                                                     KERNEL_NIC_PRIO_NUM_LEVELS))),
 175                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 176                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 177                                   ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS,
 178                                                     LEFTOVERS_NUM_LEVELS))),
 179                   ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 180                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 181                                   ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS,
 182                                                     ANCHOR_NUM_LEVELS))),
 183         }
 184 };
 185
 186 static struct init_tree_node egress_root_fs = {
 187         .type = FS_TYPE_NAMESPACE,
 188 #ifdef CONFIG_MLX5_IPSEC
 189         .ar_size = 2,
 190 #else
 191         .ar_size = 1,
 192 #endif
 193         .children = (struct init_tree_node[]) {
 194                 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 195                          FS_CHAINING_CAPS_EGRESS,
 196                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 197                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 198                                                   BY_PASS_PRIO_NUM_LEVELS))),
 199 #ifdef CONFIG_MLX5_IPSEC
 200                 ADD_PRIO(0, KERNEL_TX_MIN_LEVEL, 0,
 201                          FS_CHAINING_CAPS_EGRESS,
 202                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 203                                 ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS,
 204                                                   KERNEL_TX_IPSEC_NUM_LEVELS))),
 205 #endif
 206         }
 207 };
 208
 209 #define RDMA_RX_BYPASS_PRIO 0
 210 #define RDMA_RX_KERNEL_PRIO 1
 211 static struct init_tree_node rdma_rx_root_fs = {
 212         .type = FS_TYPE_NAMESPACE,
 213         .ar_size = 2,
 214         .children = (struct init_tree_node[]) {
 215                 [RDMA_RX_BYPASS_PRIO] =
 216                 ADD_PRIO(0, MLX5_BY_PASS_NUM_REGULAR_PRIOS, 0,
 217                          FS_CHAINING_CAPS,
 218                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 219                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS,
 220                                                   BY_PASS_PRIO_NUM_LEVELS))),
 221                 [RDMA_RX_KERNEL_PRIO] =
 222                 ADD_PRIO(0, MLX5_BY_PASS_NUM_REGULAR_PRIOS + 1, 0,
 223                          FS_CHAINING_CAPS,
 224                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN,
 225                                 ADD_MULTIPLE_PRIO(1, 1))),
 226         }
 227 };
 228
 229 static struct init_tree_node rdma_tx_root_fs = {
 230         .type = FS_TYPE_NAMESPACE,
 231         .ar_size = 1,
 232         .children = (struct init_tree_node[]) {
 233                 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 234                          FS_CHAINING_CAPS_RDMA_TX,
 235                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 236                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 237                                                   BY_PASS_PRIO_NUM_LEVELS))),
 238         }
 239 };
 240
 241 enum fs_i_lock_class {
 242         FS_LOCK_GRANDPARENT,
 243         FS_LOCK_PARENT,
 244         FS_LOCK_CHILD
 245 };
 246
 247 static const struct rhashtable_params rhash_fte = {
 248         .key_len = sizeof_field(struct fs_fte, val),
 249         .key_offset = offsetof(struct fs_fte, val),
 250         .head_offset = offsetof(struct fs_fte, hash),
 251         .automatic_shrinking = true,
 252         .min_size = 1,
 253 };
 254
 255 static const struct rhashtable_params rhash_fg = {
 256         .key_len = sizeof_field(struct mlx5_flow_group, mask),
 257         .key_offset = offsetof(struct mlx5_flow_group, mask),
 258         .head_offset = offsetof(struct mlx5_flow_group, hash),
 259         .automatic_shrinking = true,
 260         .min_size = 1,
 261
 262 };
 263
 264 static void del_hw_flow_table(struct fs_node *node);
 265 static void del_hw_flow_group(struct fs_node *node);
 266 static void del_hw_fte(struct fs_node *node);
 267 static void del_sw_flow_table(struct fs_node *node);
 268 static void del_sw_flow_group(struct fs_node *node);
 269 static void del_sw_fte(struct fs_node *node);
 270 static void del_sw_prio(struct fs_node *node);
 271 static void del_sw_ns(struct fs_node *node);
 272 /* Delete rule (destination) is special case that
 273  * requires to lock the FTE for all the deletion process.
 274  */
 275 static void del_sw_hw_rule(struct fs_node *node);
 276 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 277                                 struct mlx5_flow_destination *d2);
 278 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 279 static struct mlx5_flow_rule *
 280 find_flow_rule(struct fs_fte *fte,
 281                struct mlx5_flow_destination *dest);
 282
 283 static void tree_init_node(struct fs_node *node,
 284                            void (*del_hw_func)(struct fs_node *),
 285                            void (*del_sw_func)(struct fs_node *))
 286 {
 287         refcount_set(&node->refcount, 1);
 288         INIT_LIST_HEAD(&node->list);
 289         INIT_LIST_HEAD(&node->children);
 290         init_rwsem(&node->lock);
 291         node->del_hw_func = del_hw_func;
 292         node->del_sw_func = del_sw_func;
 293         node->active = false;
 294 }
 295
 296 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 297 {
 298         if (parent)
 299                 refcount_inc(&parent->refcount);
 300         node->parent = parent;
 301
 302         /* Parent is the root */
 303         if (!parent)
 304                 node->root = node;
 305         else
 306                 node->root = parent->root;
 307 }
 308
 309 static int tree_get_node(struct fs_node *node)
 310 {
 311         return refcount_inc_not_zero(&node->refcount);
 312 }
 313
 314 static void nested_down_read_ref_node(struct fs_node *node,
 315                                       enum fs_i_lock_class class)
 316 {
 317         if (node) {
 318                 down_read_nested(&node->lock, class);
 319                 refcount_inc(&node->refcount);
 320         }
 321 }
 322
 323 static void nested_down_write_ref_node(struct fs_node *node,
 324                                        enum fs_i_lock_class class)
 325 {
 326         if (node) {
 327                 down_write_nested(&node->lock, class);
 328                 refcount_inc(&node->refcount);
 329         }
 330 }
 331
 332 static void down_write_ref_node(struct fs_node *node, bool locked)
 333 {
 334         if (node) {
 335                 if (!locked)
 336                         down_write(&node->lock);
 337                 refcount_inc(&node->refcount);
 338         }
 339 }
 340
 341 static void up_read_ref_node(struct fs_node *node)
 342 {
 343         refcount_dec(&node->refcount);
 344         up_read(&node->lock);
 345 }
 346
 347 static void up_write_ref_node(struct fs_node *node, bool locked)
 348 {
 349         refcount_dec(&node->refcount);
 350         if (!locked)
 351                 up_write(&node->lock);
 352 }
 353
 354 static void tree_put_node(struct fs_node *node, bool locked)
 355 {
 356         struct fs_node *parent_node = node->parent;
 357
 358         if (refcount_dec_and_test(&node->refcount)) {
 359                 if (node->del_hw_func)
 360                         node->del_hw_func(node);
 361                 if (parent_node) {
 362                         down_write_ref_node(parent_node, locked);
 363                         list_del_init(&node->list);
 364                 }
 365                 node->del_sw_func(node);
 366                 if (parent_node)
 367                         up_write_ref_node(parent_node, locked);
 368                 node = NULL;
 369         }
 370         if (!node && parent_node)
 371                 tree_put_node(parent_node, locked);
 372 }
 373
 374 static int tree_remove_node(struct fs_node *node, bool locked)
 375 {
 376         if (refcount_read(&node->refcount) > 1) {
 377                 refcount_dec(&node->refcount);
 378                 return -EEXIST;
 379         }
 380         tree_put_node(node, locked);
 381         return 0;
 382 }
 383
 384 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 385                                  unsigned int prio)
 386 {
 387         struct fs_prio *iter_prio;
 388
 389         fs_for_each_prio(iter_prio, ns) {
 390                 if (iter_prio->prio == prio)
 391                         return iter_prio;
 392         }
 393
 394         return NULL;
 395 }
 396
 397 static bool is_fwd_next_action(u32 action)
 398 {
 399         return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
 400                          MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
 401 }
 402
 403 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 404 {
 405         int i;
 406
 407         for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 408                 if (spec->match_value[i] & ~spec->match_criteria[i]) {
 409                         pr_warn("mlx5_core: match_value differs from match_criteria\n");
 410                         return false;
 411                 }
 412
 413         return true;
 414 }
 415
 416 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 417 {
 418         struct fs_node *root;
 419         struct mlx5_flow_namespace *ns;
 420
 421         root = node->root;
 422
 423         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 424                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 425                 return NULL;
 426         }
 427
 428         ns = container_of(root, struct mlx5_flow_namespace, node);
 429         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 430 }
 431
 432 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 433 {
 434         struct mlx5_flow_root_namespace *root = find_root(node);
 435
 436         if (root)
 437                 return root->dev->priv.steering;
 438         return NULL;
 439 }
 440
 441 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 442 {
 443         struct mlx5_flow_root_namespace *root = find_root(node);
 444
 445         if (root)
 446                 return root->dev;
 447         return NULL;
 448 }
 449
 450 static void del_sw_ns(struct fs_node *node)
 451 {
 452         kfree(node);
 453 }
 454
 455 static void del_sw_prio(struct fs_node *node)
 456 {
 457         kfree(node);
 458 }
 459
 460 static void del_hw_flow_table(struct fs_node *node)
 461 {
 462         struct mlx5_flow_root_namespace *root;
 463         struct mlx5_flow_table *ft;
 464         struct mlx5_core_dev *dev;
 465         int err;
 466
 467         fs_get_obj(ft, node);
 468         dev = get_dev(&ft->node);
 469         root = find_root(&ft->node);
 470         trace_mlx5_fs_del_ft(ft);
 471
 472         if (node->active) {
 473                 err = root->cmds->destroy_flow_table(root, ft);
 474                 if (err)
 475                         mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 476         }
 477 }
 478
 479 static void del_sw_flow_table(struct fs_node *node)
 480 {
 481         struct mlx5_flow_table *ft;
 482         struct fs_prio *prio;
 483
 484         fs_get_obj(ft, node);
 485
 486         rhltable_destroy(&ft->fgs_hash);
 487         if (ft->node.parent) {
 488                 fs_get_obj(prio, ft->node.parent);
 489                 prio->num_ft--;
 490         }
 491         kfree(ft);
 492 }
 493
 494 static void modify_fte(struct fs_fte *fte)
 495 {
 496         struct mlx5_flow_root_namespace *root;
 497         struct mlx5_flow_table *ft;
 498         struct mlx5_flow_group *fg;
 499         struct mlx5_core_dev *dev;
 500         int err;
 501
 502         fs_get_obj(fg, fte->node.parent);
 503         fs_get_obj(ft, fg->node.parent);
 504         dev = get_dev(&fte->node);
 505
 506         root = find_root(&ft->node);
 507         err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte);
 508         if (err)
 509                 mlx5_core_warn(dev,
 510                                "%s can't del rule fg id=%d fte_index=%d\n",
 511                                __func__, fg->id, fte->index);
 512         fte->modify_mask = 0;
 513 }
 514
 515 static void del_sw_hw_rule(struct fs_node *node)
 516 {
 517         struct mlx5_flow_rule *rule;
 518         struct fs_fte *fte;
 519
 520         fs_get_obj(rule, node);
 521         fs_get_obj(fte, rule->node.parent);
 522         trace_mlx5_fs_del_rule(rule);
 523         if (is_fwd_next_action(rule->sw_action)) {
 524                 mutex_lock(&rule->dest_attr.ft->lock);
 525                 list_del(&rule->next_ft);
 526                 mutex_unlock(&rule->dest_attr.ft->lock);
 527         }
 528
 529         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 530             --fte->dests_size) {
 531                 fte->modify_mask |=
 532                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 533                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 534                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 535                 goto out;
 536         }
 537
 538         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT &&
 539             --fte->dests_size) {
 540                 fte->modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
 541                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW;
 542                 goto out;
 543         }
 544
 545         if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 546             --fte->dests_size) {
 547                 fte->modify_mask |=
 548                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 549         }
 550 out:
 551         kfree(rule);
 552 }
 553
 554 static void del_hw_fte(struct fs_node *node)
 555 {
 556         struct mlx5_flow_root_namespace *root;
 557         struct mlx5_flow_table *ft;
 558         struct mlx5_flow_group *fg;
 559         struct mlx5_core_dev *dev;
 560         struct fs_fte *fte;
 561         int err;
 562
 563         fs_get_obj(fte, node);
 564         fs_get_obj(fg, fte->node.parent);
 565         fs_get_obj(ft, fg->node.parent);
 566
 567         trace_mlx5_fs_del_fte(fte);
 568         dev = get_dev(&ft->node);
 569         root = find_root(&ft->node);
 570         if (node->active) {
 571                 err = root->cmds->delete_fte(root, ft, fte);
 572                 if (err)
 573                         mlx5_core_warn(dev,
 574                                        "flow steering can't delete fte in index %d of flow group id %d\n",
 575                                        fte->index, fg->id);
 576                 node->active = false;
 577         }
 578 }
 579
 580 static void del_sw_fte(struct fs_node *node)
 581 {
 582         struct mlx5_flow_steering *steering = get_steering(node);
 583         struct mlx5_flow_group *fg;
 584         struct fs_fte *fte;
 585         int err;
 586
 587         fs_get_obj(fte, node);
 588         fs_get_obj(fg, fte->node.parent);
 589
 590         err = rhashtable_remove_fast(&fg->ftes_hash,
 591                                      &fte->hash,
 592                                      rhash_fte);
 593         WARN_ON(err);
 594         ida_free(&fg->fte_allocator, fte->index - fg->start_index);
 595         kmem_cache_free(steering->ftes_cache, fte);
 596 }
 597
 598 static void del_hw_flow_group(struct fs_node *node)
 599 {
 600         struct mlx5_flow_root_namespace *root;
 601         struct mlx5_flow_group *fg;
 602         struct mlx5_flow_table *ft;
 603         struct mlx5_core_dev *dev;
 604
 605         fs_get_obj(fg, node);
 606         fs_get_obj(ft, fg->node.parent);
 607         dev = get_dev(&ft->node);
 608         trace_mlx5_fs_del_fg(fg);
 609
 610         root = find_root(&ft->node);
 611         if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
 612                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 613                                fg->id, ft->id);
 614 }
 615
 616 static void del_sw_flow_group(struct fs_node *node)
 617 {
 618         struct mlx5_flow_steering *steering = get_steering(node);
 619         struct mlx5_flow_group *fg;
 620         struct mlx5_flow_table *ft;
 621         int err;
 622
 623         fs_get_obj(fg, node);
 624         fs_get_obj(ft, fg->node.parent);
 625
 626         rhashtable_destroy(&fg->ftes_hash);
 627         ida_destroy(&fg->fte_allocator);
 628         if (ft->autogroup.active &&
 629             fg->max_ftes == ft->autogroup.group_size &&
 630             fg->start_index < ft->autogroup.max_fte)
 631                 ft->autogroup.num_groups--;
 632         err = rhltable_remove(&ft->fgs_hash,
 633                               &fg->hash,
 634                               rhash_fg);
 635         WARN_ON(err);
 636         kmem_cache_free(steering->fgs_cache, fg);
 637 }
 638
 639 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 640 {
 641         int index;
 642         int ret;
 643
 644         index = ida_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, GFP_KERNEL);
 645         if (index < 0)
 646                 return index;
 647
 648         fte->index = index + fg->start_index;
 649         ret = rhashtable_insert_fast(&fg->ftes_hash,
 650                                      &fte->hash,
 651                                      rhash_fte);
 652         if (ret)
 653                 goto err_ida_remove;
 654
 655         tree_add_node(&fte->node, &fg->node);
 656         list_add_tail(&fte->node.list, &fg->node.children);
 657         return 0;
 658
 659 err_ida_remove:
 660         ida_free(&fg->fte_allocator, index);
 661         return ret;
 662 }
 663
 664 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 665                                 const struct mlx5_flow_spec *spec,
 666                                 struct mlx5_flow_act *flow_act)
 667 {
 668         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 669         struct fs_fte *fte;
 670
 671         fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 672         if (!fte)
 673                 return ERR_PTR(-ENOMEM);
 674
 675         memcpy(fte->val, &spec->match_value, sizeof(fte->val));
 676         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 677         fte->action = *flow_act;
 678         fte->flow_context = spec->flow_context;
 679
 680         tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
 681
 682         return fte;
 683 }
 684
 685 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 686                                struct mlx5_flow_group *fg)
 687 {
 688         rhashtable_destroy(&fg->ftes_hash);
 689         kmem_cache_free(steering->fgs_cache, fg);
 690 }
 691
 692 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 693                                                 u8 match_criteria_enable,
 694                                                 const void *match_criteria,
 695                                                 int start_index,
 696                                                 int end_index)
 697 {
 698         struct mlx5_flow_group *fg;
 699         int ret;
 700
 701         fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 702         if (!fg)
 703                 return ERR_PTR(-ENOMEM);
 704
 705         ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 706         if (ret) {
 707                 kmem_cache_free(steering->fgs_cache, fg);
 708                 return ERR_PTR(ret);
 709         }
 710
 711         ida_init(&fg->fte_allocator);
 712         fg->mask.match_criteria_enable = match_criteria_enable;
 713         memcpy(&fg->mask.match_criteria, match_criteria,
 714                sizeof(fg->mask.match_criteria));
 715         fg->node.type =  FS_TYPE_FLOW_GROUP;
 716         fg->start_index = start_index;
 717         fg->max_ftes = end_index - start_index + 1;
 718
 719         return fg;
 720 }
 721
 722 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 723                                                        u8 match_criteria_enable,
 724                                                        const void *match_criteria,
 725                                                        int start_index,
 726                                                        int end_index,
 727                                                        struct list_head *prev)
 728 {
 729         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 730         struct mlx5_flow_group *fg;
 731         int ret;
 732
 733         fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 734                               start_index, end_index);
 735         if (IS_ERR(fg))
 736                 return fg;
 737
 738         /* initialize refcnt, add to parent list */
 739         ret = rhltable_insert(&ft->fgs_hash,
 740                               &fg->hash,
 741                               rhash_fg);
 742         if (ret) {
 743                 dealloc_flow_group(steering, fg);
 744                 return ERR_PTR(ret);
 745         }
 746
 747         tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 748         tree_add_node(&fg->node, &ft->node);
 749         /* Add node to group list */
 750         list_add(&fg->node.list, prev);
 751         atomic_inc(&ft->node.version);
 752
 753         return fg;
 754 }
 755
 756 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport,
 757                                                 enum fs_flow_table_type table_type,
 758                                                 enum fs_flow_table_op_mod op_mod,
 759                                                 u32 flags)
 760 {
 761         struct mlx5_flow_table *ft;
 762         int ret;
 763
 764         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 765         if (!ft)
 766                 return ERR_PTR(-ENOMEM);
 767
 768         ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 769         if (ret) {
 770                 kfree(ft);
 771                 return ERR_PTR(ret);
 772         }
 773
 774         ft->level = level;
 775         ft->node.type = FS_TYPE_FLOW_TABLE;
 776         ft->op_mod = op_mod;
 777         ft->type = table_type;
 778         ft->vport = vport;
 779         ft->flags = flags;
 780         INIT_LIST_HEAD(&ft->fwd_rules);
 781         mutex_init(&ft->lock);
 782
 783         return ft;
 784 }
 785
 786 /* If reverse is false, then we search for the first flow table in the
 787  * root sub-tree from start(closest from right), else we search for the
 788  * last flow table in the root sub-tree till start(closest from left).
 789  */
 790 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 791                                                          struct list_head *start,
 792                                                          bool reverse)
 793 {
 794 #define list_advance_entry(pos, reverse)                \
 795         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 796
 797 #define list_for_each_advance_continue(pos, head, reverse)      \
 798         for (pos = list_advance_entry(pos, reverse);            \
 799              &pos->list != (head);                              \
 800              pos = list_advance_entry(pos, reverse))
 801
 802         struct fs_node *iter = list_entry(start, struct fs_node, list);
 803         struct mlx5_flow_table *ft = NULL;
 804
 805         if (!root || root->type == FS_TYPE_PRIO_CHAINS)
 806                 return NULL;
 807
 808         list_for_each_advance_continue(iter, &root->children, reverse) {
 809                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 810                         fs_get_obj(ft, iter);
 811                         return ft;
 812                 }
 813                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 814                 if (ft)
 815                         return ft;
 816         }
 817
 818         return ft;
 819 }
 820
 821 /* If reverse is false then return the first flow table in next priority of
 822  * prio in the tree, else return the last flow table in the previous priority
 823  * of prio in the tree.
 824  */
 825 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 826 {
 827         struct mlx5_flow_table *ft = NULL;
 828         struct fs_node *curr_node;
 829         struct fs_node *parent;
 830
 831         parent = prio->node.parent;
 832         curr_node = &prio->node;
 833         while (!ft && parent) {
 834                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 835                 curr_node = parent;
 836                 parent = curr_node->parent;
 837         }
 838         return ft;
 839 }
 840
 841 /* Assuming all the tree is locked by mutex chain lock */
 842 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 843 {
 844         return find_closest_ft(prio, false);
 845 }
 846
 847 /* Assuming all the tree is locked by mutex chain lock */
 848 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 849 {
 850         return find_closest_ft(prio, true);
 851 }
 852
 853 static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft,
 854                                                 struct mlx5_flow_act *flow_act)
 855 {
 856         struct fs_prio *prio;
 857         bool next_ns;
 858
 859         next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
 860         fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent);
 861
 862         return find_next_chained_ft(prio);
 863 }
 864
 865 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 866                                struct fs_prio *prio,
 867                                struct mlx5_flow_table *ft)
 868 {
 869         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 870         struct mlx5_flow_table *iter;
 871         int err;
 872
 873         fs_for_each_ft(iter, prio) {
 874                 err = root->cmds->modify_flow_table(root, iter, ft);
 875                 if (err) {
 876                         mlx5_core_err(dev,
 877                                       "Failed to modify flow table id %d, type %d, err %d\n",
 878                                       iter->id, iter->type, err);
 879                         /* The driver is out of sync with the FW */
 880                         return err;
 881                 }
 882         }
 883         return 0;
 884 }
 885
 886 /* Connect flow tables from previous priority of prio to ft */
 887 static int connect_prev_fts(struct mlx5_core_dev *dev,
 888                             struct mlx5_flow_table *ft,
 889                             struct fs_prio *prio)
 890 {
 891         struct mlx5_flow_table *prev_ft;
 892
 893         prev_ft = find_prev_chained_ft(prio);
 894         if (prev_ft) {
 895                 struct fs_prio *prev_prio;
 896
 897                 fs_get_obj(prev_prio, prev_ft->node.parent);
 898                 return connect_fts_in_prio(dev, prev_prio, ft);
 899         }
 900         return 0;
 901 }
 902
 903 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 904                                  *prio)
 905 {
 906         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 907         struct mlx5_ft_underlay_qp *uqp;
 908         int min_level = INT_MAX;
 909         int err = 0;
 910         u32 qpn;
 911
 912         if (root->root_ft)
 913                 min_level = root->root_ft->level;
 914
 915         if (ft->level >= min_level)
 916                 return 0;
 917
 918         if (list_empty(&root->underlay_qpns)) {
 919                 /* Don't set any QPN (zero) in case QPN list is empty */
 920                 qpn = 0;
 921                 err = root->cmds->update_root_ft(root, ft, qpn, false);
 922         } else {
 923                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
 924                         qpn = uqp->qpn;
 925                         err = root->cmds->update_root_ft(root, ft,
 926                                                          qpn, false);
 927                         if (err)
 928                                 break;
 929                 }
 930         }
 931
 932         if (err)
 933                 mlx5_core_warn(root->dev,
 934                                "Update root flow table of id(%u) qpn(%d) failed\n",
 935                                ft->id, qpn);
 936         else
 937                 root->root_ft = ft;
 938
 939         return err;
 940 }
 941
 942 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 943                                          struct mlx5_flow_destination *dest)
 944 {
 945         struct mlx5_flow_root_namespace *root;
 946         struct mlx5_flow_table *ft;
 947         struct mlx5_flow_group *fg;
 948         struct fs_fte *fte;
 949         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 950         int err = 0;
 951
 952         fs_get_obj(fte, rule->node.parent);
 953         if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 954                 return -EINVAL;
 955         down_write_ref_node(&fte->node, false);
 956         fs_get_obj(fg, fte->node.parent);
 957         fs_get_obj(ft, fg->node.parent);
 958
 959         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 960         root = find_root(&ft->node);
 961         err = root->cmds->update_fte(root, ft, fg,
 962                                      modify_mask, fte);
 963         up_write_ref_node(&fte->node, false);
 964
 965         return err;
 966 }
 967
 968 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
 969                                  struct mlx5_flow_destination *new_dest,
 970                                  struct mlx5_flow_destination *old_dest)
 971 {
 972         int i;
 973
 974         if (!old_dest) {
 975                 if (handle->num_rules != 1)
 976                         return -EINVAL;
 977                 return _mlx5_modify_rule_destination(handle->rule[0],
 978                                                      new_dest);
 979         }
 980
 981         for (i = 0; i < handle->num_rules; i++) {
 982                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
 983                         return _mlx5_modify_rule_destination(handle->rule[i],
 984                                                              new_dest);
 985         }
 986
 987         return -EINVAL;
 988 }
 989
 990 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 991 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 992                              struct mlx5_flow_table *new_next_ft,
 993                              struct mlx5_flow_table *old_next_ft)
 994 {
 995         struct mlx5_flow_destination dest = {};
 996         struct mlx5_flow_rule *iter;
 997         int err = 0;
 998
 999         /* new_next_ft and old_next_ft could be NULL only
1000          * when we create/destroy the anchor flow table.
1001          */
1002         if (!new_next_ft || !old_next_ft)
1003                 return 0;
1004
1005         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1006         dest.ft = new_next_ft;
1007
1008         mutex_lock(&old_next_ft->lock);
1009         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
1010         mutex_unlock(&old_next_ft->lock);
1011         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
1012                 if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) &&
1013                     iter->ft->ns == new_next_ft->ns)
1014                         continue;
1015
1016                 err = _mlx5_modify_rule_destination(iter, &dest);
1017                 if (err)
1018                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
1019                                new_next_ft->id);
1020         }
1021         return 0;
1022 }
1023
1024 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
1025                               struct fs_prio *prio)
1026 {
1027         struct mlx5_flow_table *next_ft;
1028         int err = 0;
1029
1030         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
1031
1032         if (list_empty(&prio->node.children)) {
1033                 err = connect_prev_fts(dev, ft, prio);
1034                 if (err)
1035                         return err;
1036
1037                 next_ft = find_next_chained_ft(prio);
1038                 err = connect_fwd_rules(dev, ft, next_ft);
1039                 if (err)
1040                         return err;
1041         }
1042
1043         if (MLX5_CAP_FLOWTABLE(dev,
1044                                flow_table_properties_nic_receive.modify_root))
1045                 err = update_root_ft_create(ft, prio);
1046         return err;
1047 }
1048
1049 static void list_add_flow_table(struct mlx5_flow_table *ft,
1050                                 struct fs_prio *prio)
1051 {
1052         struct list_head *prev = &prio->node.children;
1053         struct mlx5_flow_table *iter;
1054
1055         fs_for_each_ft(iter, prio) {
1056                 if (iter->level > ft->level)
1057                         break;
1058                 prev = &iter->node.list;
1059         }
1060         list_add(&ft->node.list, prev);
1061 }
1062
1063 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1064                                                         struct mlx5_flow_table_attr *ft_attr,
1065                                                         enum fs_flow_table_op_mod op_mod,
1066                                                         u16 vport)
1067 {
1068         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1069         bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED;
1070         struct mlx5_flow_table *next_ft;
1071         struct fs_prio *fs_prio = NULL;
1072         struct mlx5_flow_table *ft;
1073         int err;
1074
1075         if (!root) {
1076                 pr_err("mlx5: flow steering failed to find root of namespace\n");
1077                 return ERR_PTR(-ENODEV);
1078         }
1079
1080         mutex_lock(&root->chain_lock);
1081         fs_prio = find_prio(ns, ft_attr->prio);
1082         if (!fs_prio) {
1083                 err = -EINVAL;
1084                 goto unlock_root;
1085         }
1086         if (!unmanaged) {
1087                 /* The level is related to the
1088                  * priority level range.
1089                  */
1090                 if (ft_attr->level >= fs_prio->num_levels) {
1091                         err = -ENOSPC;
1092                         goto unlock_root;
1093                 }
1094
1095                 ft_attr->level += fs_prio->start_level;
1096         }
1097
1098         /* The level is related to the
1099          * priority level range.
1100          */
1101         ft = alloc_flow_table(ft_attr->level,
1102                               vport,
1103                               root->table_type,
1104                               op_mod, ft_attr->flags);
1105         if (IS_ERR(ft)) {
1106                 err = PTR_ERR(ft);
1107                 goto unlock_root;
1108         }
1109
1110         tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1111         next_ft = unmanaged ? ft_attr->next_ft :
1112                               find_next_chained_ft(fs_prio);
1113         ft->def_miss_action = ns->def_miss_action;
1114         ft->ns = ns;
1115         err = root->cmds->create_flow_table(root, ft, ft_attr->max_fte, next_ft);
1116         if (err)
1117                 goto free_ft;
1118
1119         if (!unmanaged) {
1120                 err = connect_flow_table(root->dev, ft, fs_prio);
1121                 if (err)
1122                         goto destroy_ft;
1123         }
1124
1125         ft->node.active = true;
1126         down_write_ref_node(&fs_prio->node, false);
1127         if (!unmanaged) {
1128                 tree_add_node(&ft->node, &fs_prio->node);
1129                 list_add_flow_table(ft, fs_prio);
1130         } else {
1131                 ft->node.root = fs_prio->node.root;
1132         }
1133         fs_prio->num_ft++;
1134         up_write_ref_node(&fs_prio->node, false);
1135         mutex_unlock(&root->chain_lock);
1136         trace_mlx5_fs_add_ft(ft);
1137         return ft;
1138 destroy_ft:
1139         root->cmds->destroy_flow_table(root, ft);
1140 free_ft:
1141         rhltable_destroy(&ft->fgs_hash);
1142         kfree(ft);
1143 unlock_root:
1144         mutex_unlock(&root->chain_lock);
1145         return ERR_PTR(err);
1146 }
1147
1148 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1149                                                struct mlx5_flow_table_attr *ft_attr)
1150 {
1151         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1152 }
1153 EXPORT_SYMBOL(mlx5_create_flow_table);
1154
1155 struct mlx5_flow_table *
1156 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1157                              struct mlx5_flow_table_attr *ft_attr, u16 vport)
1158 {
1159         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1160 }
1161
1162 struct mlx5_flow_table*
1163 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1164                                  int prio, u32 level)
1165 {
1166         struct mlx5_flow_table_attr ft_attr = {};
1167
1168         ft_attr.level = level;
1169         ft_attr.prio  = prio;
1170         ft_attr.max_fte = 1;
1171
1172         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1173 }
1174 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1175
1176 #define MAX_FLOW_GROUP_SIZE BIT(24)
1177 struct mlx5_flow_table*
1178 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1179                                     struct mlx5_flow_table_attr *ft_attr)
1180 {
1181         int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1182         int max_num_groups = ft_attr->autogroup.max_num_groups;
1183         struct mlx5_flow_table *ft;
1184         int autogroups_max_fte;
1185
1186         ft = mlx5_create_flow_table(ns, ft_attr);
1187         if (IS_ERR(ft))
1188                 return ft;
1189
1190         autogroups_max_fte = ft->max_fte - num_reserved_entries;
1191         if (max_num_groups > autogroups_max_fte)
1192                 goto err_validate;
1193         if (num_reserved_entries > ft->max_fte)
1194                 goto err_validate;
1195
1196         /* Align the number of groups according to the largest group size */
1197         if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE)
1198                 max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1;
1199
1200         ft->autogroup.active = true;
1201         ft->autogroup.required_groups = max_num_groups;
1202         ft->autogroup.max_fte = autogroups_max_fte;
1203         /* We save place for flow groups in addition to max types */
1204         ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1205
1206         return ft;
1207
1208 err_validate:
1209         mlx5_destroy_flow_table(ft);
1210         return ERR_PTR(-ENOSPC);
1211 }
1212 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1213
1214 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1215                                                u32 *fg_in)
1216 {
1217         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1218         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1219                                             fg_in, match_criteria);
1220         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1221                                             fg_in,
1222                                             match_criteria_enable);
1223         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1224                                    start_flow_index);
1225         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1226                                  end_flow_index);
1227         struct mlx5_flow_group *fg;
1228         int err;
1229
1230         if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1231                 return ERR_PTR(-EPERM);
1232
1233         down_write_ref_node(&ft->node, false);
1234         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1235                                      start_index, end_index,
1236                                      ft->node.children.prev);
1237         up_write_ref_node(&ft->node, false);
1238         if (IS_ERR(fg))
1239                 return fg;
1240
1241         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1242         if (err) {
1243                 tree_put_node(&fg->node, false);
1244                 return ERR_PTR(err);
1245         }
1246         trace_mlx5_fs_add_fg(fg);
1247         fg->node.active = true;
1248
1249         return fg;
1250 }
1251 EXPORT_SYMBOL(mlx5_create_flow_group);
1252
1253 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1254 {
1255         struct mlx5_flow_rule *rule;
1256
1257         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1258         if (!rule)
1259                 return NULL;
1260
1261         INIT_LIST_HEAD(&rule->next_ft);
1262         rule->node.type = FS_TYPE_FLOW_DEST;
1263         if (dest)
1264                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1265
1266         return rule;
1267 }
1268
1269 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1270 {
1271         struct mlx5_flow_handle *handle;
1272
1273         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1274         if (!handle)
1275                 return NULL;
1276
1277         handle->num_rules = num_rules;
1278
1279         return handle;
1280 }
1281
1282 static void destroy_flow_handle(struct fs_fte *fte,
1283                                 struct mlx5_flow_handle *handle,
1284                                 struct mlx5_flow_destination *dest,
1285                                 int i)
1286 {
1287         for (; --i >= 0;) {
1288                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1289                         fte->dests_size--;
1290                         list_del(&handle->rule[i]->node.list);
1291                         kfree(handle->rule[i]);
1292                 }
1293         }
1294         kfree(handle);
1295 }
1296
1297 static struct mlx5_flow_handle *
1298 create_flow_handle(struct fs_fte *fte,
1299                    struct mlx5_flow_destination *dest,
1300                    int dest_num,
1301                    int *modify_mask,
1302                    bool *new_rule)
1303 {
1304         struct mlx5_flow_handle *handle;
1305         struct mlx5_flow_rule *rule = NULL;
1306         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1307         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1308         int type;
1309         int i = 0;
1310
1311         handle = alloc_handle((dest_num) ? dest_num : 1);
1312         if (!handle)
1313                 return ERR_PTR(-ENOMEM);
1314
1315         do {
1316                 if (dest) {
1317                         rule = find_flow_rule(fte, dest + i);
1318                         if (rule) {
1319                                 refcount_inc(&rule->node.refcount);
1320                                 goto rule_found;
1321                         }
1322                 }
1323
1324                 *new_rule = true;
1325                 rule = alloc_rule(dest + i);
1326                 if (!rule)
1327                         goto free_rules;
1328
1329                 /* Add dest to dests list- we need flow tables to be in the
1330                  * end of the list for forward to next prio rules.
1331                  */
1332                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1333                 if (dest &&
1334                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1335                         list_add(&rule->node.list, &fte->node.children);
1336                 else
1337                         list_add_tail(&rule->node.list, &fte->node.children);
1338                 if (dest) {
1339                         fte->dests_size++;
1340
1341                         type = dest[i].type ==
1342                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1343                         *modify_mask |= type ? count : dst;
1344                 }
1345 rule_found:
1346                 handle->rule[i] = rule;
1347         } while (++i < dest_num);
1348
1349         return handle;
1350
1351 free_rules:
1352         destroy_flow_handle(fte, handle, dest, i);
1353         return ERR_PTR(-ENOMEM);
1354 }
1355
1356 /* fte should not be deleted while calling this function */
1357 static struct mlx5_flow_handle *
1358 add_rule_fte(struct fs_fte *fte,
1359              struct mlx5_flow_group *fg,
1360              struct mlx5_flow_destination *dest,
1361              int dest_num,
1362              bool update_action)
1363 {
1364         struct mlx5_flow_root_namespace *root;
1365         struct mlx5_flow_handle *handle;
1366         struct mlx5_flow_table *ft;
1367         int modify_mask = 0;
1368         int err;
1369         bool new_rule = false;
1370
1371         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1372                                     &new_rule);
1373         if (IS_ERR(handle) || !new_rule)
1374                 goto out;
1375
1376         if (update_action)
1377                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1378
1379         fs_get_obj(ft, fg->node.parent);
1380         root = find_root(&fg->node);
1381         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1382                 err = root->cmds->create_fte(root, ft, fg, fte);
1383         else
1384                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1385         if (err)
1386                 goto free_handle;
1387
1388         fte->node.active = true;
1389         fte->status |= FS_FTE_STATUS_EXISTING;
1390         atomic_inc(&fg->node.version);
1391
1392 out:
1393         return handle;
1394
1395 free_handle:
1396         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1397         return ERR_PTR(err);
1398 }
1399
1400 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1401                                                      const struct mlx5_flow_spec *spec)
1402 {
1403         struct list_head *prev = &ft->node.children;
1404         u32 max_fte = ft->autogroup.max_fte;
1405         unsigned int candidate_index = 0;
1406         unsigned int group_size = 0;
1407         struct mlx5_flow_group *fg;
1408
1409         if (!ft->autogroup.active)
1410                 return ERR_PTR(-ENOENT);
1411
1412         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1413                 group_size = ft->autogroup.group_size;
1414
1415         /*  max_fte == ft->autogroup.max_types */
1416         if (group_size == 0)
1417                 group_size = 1;
1418
1419         /* sorted by start_index */
1420         fs_for_each_fg(fg, ft) {
1421                 if (candidate_index + group_size > fg->start_index)
1422                         candidate_index = fg->start_index + fg->max_ftes;
1423                 else
1424                         break;
1425                 prev = &fg->node.list;
1426         }
1427
1428         if (candidate_index + group_size > max_fte)
1429                 return ERR_PTR(-ENOSPC);
1430
1431         fg = alloc_insert_flow_group(ft,
1432                                      spec->match_criteria_enable,
1433                                      spec->match_criteria,
1434                                      candidate_index,
1435                                      candidate_index + group_size - 1,
1436                                      prev);
1437         if (IS_ERR(fg))
1438                 goto out;
1439
1440         if (group_size == ft->autogroup.group_size)
1441                 ft->autogroup.num_groups++;
1442
1443 out:
1444         return fg;
1445 }
1446
1447 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1448                                   struct mlx5_flow_group *fg)
1449 {
1450         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1451         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1452         void *match_criteria_addr;
1453         u8 src_esw_owner_mask_on;
1454         void *misc;
1455         int err;
1456         u32 *in;
1457
1458         in = kvzalloc(inlen, GFP_KERNEL);
1459         if (!in)
1460                 return -ENOMEM;
1461
1462         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1463                  fg->mask.match_criteria_enable);
1464         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1465         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1466                  fg->max_ftes - 1);
1467
1468         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1469                             misc_parameters);
1470         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1471                                          source_eswitch_owner_vhca_id);
1472         MLX5_SET(create_flow_group_in, in,
1473                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1474
1475         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1476                                            in, match_criteria);
1477         memcpy(match_criteria_addr, fg->mask.match_criteria,
1478                sizeof(fg->mask.match_criteria));
1479
1480         err = root->cmds->create_flow_group(root, ft, in, fg);
1481         if (!err) {
1482                 fg->node.active = true;
1483                 trace_mlx5_fs_add_fg(fg);
1484         }
1485
1486         kvfree(in);
1487         return err;
1488 }
1489
1490 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1491                                 struct mlx5_flow_destination *d2)
1492 {
1493         if (d1->type == d2->type) {
1494                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1495                      d1->vport.num == d2->vport.num &&
1496                      d1->vport.flags == d2->vport.flags &&
1497                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1498                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1499                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1500                       (d1->vport.pkt_reformat->id ==
1501                        d2->vport.pkt_reformat->id) : true)) ||
1502                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1503                      d1->ft == d2->ft) ||
1504                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1505                      d1->tir_num == d2->tir_num) ||
1506                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1507                      d1->ft_num == d2->ft_num) ||
1508                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER &&
1509                      d1->sampler_id == d2->sampler_id))
1510                         return true;
1511         }
1512
1513         return false;
1514 }
1515
1516 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1517                                              struct mlx5_flow_destination *dest)
1518 {
1519         struct mlx5_flow_rule *rule;
1520
1521         list_for_each_entry(rule, &fte->node.children, node.list) {
1522                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1523                         return rule;
1524         }
1525         return NULL;
1526 }
1527
1528 static bool check_conflicting_actions(u32 action1, u32 action2)
1529 {
1530         u32 xored_actions = action1 ^ action2;
1531
1532         /* if one rule only wants to count, it's ok */
1533         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1534             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1535                 return false;
1536
1537         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1538                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1539                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1540                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1541                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1542                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1543                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1544                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1545                 return true;
1546
1547         return false;
1548 }
1549
1550 static int check_conflicting_ftes(struct fs_fte *fte,
1551                                   const struct mlx5_flow_context *flow_context,
1552                                   const struct mlx5_flow_act *flow_act)
1553 {
1554         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1555                 mlx5_core_warn(get_dev(&fte->node),
1556                                "Found two FTEs with conflicting actions\n");
1557                 return -EEXIST;
1558         }
1559
1560         if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1561             fte->flow_context.flow_tag != flow_context->flow_tag) {
1562                 mlx5_core_warn(get_dev(&fte->node),
1563                                "FTE flow tag %u already exists with different flow tag %u\n",
1564                                fte->flow_context.flow_tag,
1565                                flow_context->flow_tag);
1566                 return -EEXIST;
1567         }
1568
1569         return 0;
1570 }
1571
1572 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1573                                             const struct mlx5_flow_spec *spec,
1574                                             struct mlx5_flow_act *flow_act,
1575                                             struct mlx5_flow_destination *dest,
1576                                             int dest_num,
1577                                             struct fs_fte *fte)
1578 {
1579         struct mlx5_flow_handle *handle;
1580         int old_action;
1581         int i;
1582         int ret;
1583
1584         ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1585         if (ret)
1586                 return ERR_PTR(ret);
1587
1588         old_action = fte->action.action;
1589         fte->action.action |= flow_act->action;
1590         handle = add_rule_fte(fte, fg, dest, dest_num,
1591                               old_action != flow_act->action);
1592         if (IS_ERR(handle)) {
1593                 fte->action.action = old_action;
1594                 return handle;
1595         }
1596         trace_mlx5_fs_set_fte(fte, false);
1597
1598         for (i = 0; i < handle->num_rules; i++) {
1599                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1600                         tree_add_node(&handle->rule[i]->node, &fte->node);
1601                         trace_mlx5_fs_add_rule(handle->rule[i]);
1602                 }
1603         }
1604         return handle;
1605 }
1606
1607 static bool counter_is_valid(u32 action)
1608 {
1609         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1610                           MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1611                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1612 }
1613
1614 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1615                           struct mlx5_flow_act *flow_act,
1616                           struct mlx5_flow_table *ft)
1617 {
1618         bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1619         u32 action = flow_act->action;
1620
1621         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1622                 return counter_is_valid(action);
1623
1624         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1625                 return true;
1626
1627         if (ignore_level) {
1628                 if (ft->type != FS_FT_FDB &&
1629                     ft->type != FS_FT_NIC_RX)
1630                         return false;
1631
1632                 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1633                     ft->type != dest->ft->type)
1634                         return false;
1635         }
1636
1637         if (!dest || ((dest->type ==
1638             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1639             (dest->ft->level <= ft->level && !ignore_level)))
1640                 return false;
1641         return true;
1642 }
1643
1644 struct match_list {
1645         struct list_head        list;
1646         struct mlx5_flow_group *g;
1647 };
1648
1649 static void free_match_list(struct match_list *head, bool ft_locked)
1650 {
1651         struct match_list *iter, *match_tmp;
1652
1653         list_for_each_entry_safe(iter, match_tmp, &head->list,
1654                                  list) {
1655                 tree_put_node(&iter->g->node, ft_locked);
1656                 list_del(&iter->list);
1657                 kfree(iter);
1658         }
1659 }
1660
1661 static int build_match_list(struct match_list *match_head,
1662                             struct mlx5_flow_table *ft,
1663                             const struct mlx5_flow_spec *spec,
1664                             bool ft_locked)
1665 {
1666         struct rhlist_head *tmp, *list;
1667         struct mlx5_flow_group *g;
1668         int err = 0;
1669
1670         rcu_read_lock();
1671         INIT_LIST_HEAD(&match_head->list);
1672         /* Collect all fgs which has a matching match_criteria */
1673         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1674         /* RCU is atomic, we can't execute FW commands here */
1675         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1676                 struct match_list *curr_match;
1677
1678                 if (unlikely(!tree_get_node(&g->node)))
1679                         continue;
1680
1681                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1682                 if (!curr_match) {
1683                         free_match_list(match_head, ft_locked);
1684                         err = -ENOMEM;
1685                         goto out;
1686                 }
1687                 curr_match->g = g;
1688                 list_add_tail(&curr_match->list, &match_head->list);
1689         }
1690 out:
1691         rcu_read_unlock();
1692         return err;
1693 }
1694
1695 static u64 matched_fgs_get_version(struct list_head *match_head)
1696 {
1697         struct match_list *iter;
1698         u64 version = 0;
1699
1700         list_for_each_entry(iter, match_head, list)
1701                 version += (u64)atomic_read(&iter->g->node.version);
1702         return version;
1703 }
1704
1705 static struct fs_fte *
1706 lookup_fte_locked(struct mlx5_flow_group *g,
1707                   const u32 *match_value,
1708                   bool take_write)
1709 {
1710         struct fs_fte *fte_tmp;
1711
1712         if (take_write)
1713                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1714         else
1715                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1716         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1717                                          rhash_fte);
1718         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1719                 fte_tmp = NULL;
1720                 goto out;
1721         }
1722         if (!fte_tmp->node.active) {
1723                 tree_put_node(&fte_tmp->node, false);
1724                 fte_tmp = NULL;
1725                 goto out;
1726         }
1727
1728         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1729 out:
1730         if (take_write)
1731                 up_write_ref_node(&g->node, false);
1732         else
1733                 up_read_ref_node(&g->node);
1734         return fte_tmp;
1735 }
1736
1737 static struct mlx5_flow_handle *
1738 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1739                        struct list_head *match_head,
1740                        const struct mlx5_flow_spec *spec,
1741                        struct mlx5_flow_act *flow_act,
1742                        struct mlx5_flow_destination *dest,
1743                        int dest_num,
1744                        int ft_version)
1745 {
1746         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1747         struct mlx5_flow_group *g;
1748         struct mlx5_flow_handle *rule;
1749         struct match_list *iter;
1750         bool take_write = false;
1751         struct fs_fte *fte;
1752         u64  version = 0;
1753         int err;
1754
1755         fte = alloc_fte(ft, spec, flow_act);
1756         if (IS_ERR(fte))
1757                 return  ERR_PTR(-ENOMEM);
1758
1759 search_again_locked:
1760         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1761                 goto skip_search;
1762         version = matched_fgs_get_version(match_head);
1763         /* Try to find an fte with identical match value and attempt update its
1764          * action.
1765          */
1766         list_for_each_entry(iter, match_head, list) {
1767                 struct fs_fte *fte_tmp;
1768
1769                 g = iter->g;
1770                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1771                 if (!fte_tmp)
1772                         continue;
1773                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1774                 /* No error check needed here, because insert_fte() is not called */
1775                 up_write_ref_node(&fte_tmp->node, false);
1776                 tree_put_node(&fte_tmp->node, false);
1777                 kmem_cache_free(steering->ftes_cache, fte);
1778                 return rule;
1779         }
1780
1781 skip_search:
1782         /* No group with matching fte found, or we skipped the search.
1783          * Try to add a new fte to any matching fg.
1784          */
1785
1786         /* Check the ft version, for case that new flow group
1787          * was added while the fgs weren't locked
1788          */
1789         if (atomic_read(&ft->node.version) != ft_version) {
1790                 rule = ERR_PTR(-EAGAIN);
1791                 goto out;
1792         }
1793
1794         /* Check the fgs version. If version have changed it could be that an
1795          * FTE with the same match value was added while the fgs weren't
1796          * locked.
1797          */
1798         if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1799             version != matched_fgs_get_version(match_head)) {
1800                 take_write = true;
1801                 goto search_again_locked;
1802         }
1803
1804         list_for_each_entry(iter, match_head, list) {
1805                 g = iter->g;
1806
1807                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1808
1809                 if (!g->node.active) {
1810                         up_write_ref_node(&g->node, false);
1811                         continue;
1812                 }
1813
1814                 err = insert_fte(g, fte);
1815                 if (err) {
1816                         up_write_ref_node(&g->node, false);
1817                         if (err == -ENOSPC)
1818                                 continue;
1819                         kmem_cache_free(steering->ftes_cache, fte);
1820                         return ERR_PTR(err);
1821                 }
1822
1823                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1824                 up_write_ref_node(&g->node, false);
1825                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1826                 up_write_ref_node(&fte->node, false);
1827                 if (IS_ERR(rule))
1828                         tree_put_node(&fte->node, false);
1829                 return rule;
1830         }
1831         rule = ERR_PTR(-ENOENT);
1832 out:
1833         kmem_cache_free(steering->ftes_cache, fte);
1834         return rule;
1835 }
1836
1837 static struct mlx5_flow_handle *
1838 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1839                      const struct mlx5_flow_spec *spec,
1840                      struct mlx5_flow_act *flow_act,
1841                      struct mlx5_flow_destination *dest,
1842                      int dest_num)
1843
1844 {
1845         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1846         struct mlx5_flow_handle *rule;
1847         struct match_list match_head;
1848         struct mlx5_flow_group *g;
1849         bool take_write = false;
1850         struct fs_fte *fte;
1851         int version;
1852         int err;
1853         int i;
1854
1855         if (!check_valid_spec(spec))
1856                 return ERR_PTR(-EINVAL);
1857
1858         for (i = 0; i < dest_num; i++) {
1859                 if (!dest_is_valid(&dest[i], flow_act, ft))
1860                         return ERR_PTR(-EINVAL);
1861         }
1862         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1863 search_again_locked:
1864         version = atomic_read(&ft->node.version);
1865
1866         /* Collect all fgs which has a matching match_criteria */
1867         err = build_match_list(&match_head, ft, spec, take_write);
1868         if (err) {
1869                 if (take_write)
1870                         up_write_ref_node(&ft->node, false);
1871                 else
1872                         up_read_ref_node(&ft->node);
1873                 return ERR_PTR(err);
1874         }
1875
1876         if (!take_write)
1877                 up_read_ref_node(&ft->node);
1878
1879         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1880                                       dest_num, version);
1881         free_match_list(&match_head, take_write);
1882         if (!IS_ERR(rule) ||
1883             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1884                 if (take_write)
1885                         up_write_ref_node(&ft->node, false);
1886                 return rule;
1887         }
1888
1889         if (!take_write) {
1890                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1891                 take_write = true;
1892         }
1893
1894         if (PTR_ERR(rule) == -EAGAIN ||
1895             version != atomic_read(&ft->node.version))
1896                 goto search_again_locked;
1897
1898         g = alloc_auto_flow_group(ft, spec);
1899         if (IS_ERR(g)) {
1900                 rule = ERR_CAST(g);
1901                 up_write_ref_node(&ft->node, false);
1902                 return rule;
1903         }
1904
1905         fte = alloc_fte(ft, spec, flow_act);
1906         if (IS_ERR(fte)) {
1907                 up_write_ref_node(&ft->node, false);
1908                 err = PTR_ERR(fte);
1909                 goto err_alloc_fte;
1910         }
1911
1912         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1913         up_write_ref_node(&ft->node, false);
1914
1915         err = create_auto_flow_group(ft, g);
1916         if (err)
1917                 goto err_release_fg;
1918
1919         err = insert_fte(g, fte);
1920         if (err)
1921                 goto err_release_fg;
1922
1923         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1924         up_write_ref_node(&g->node, false);
1925         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1926         up_write_ref_node(&fte->node, false);
1927         if (IS_ERR(rule))
1928                 tree_put_node(&fte->node, false);
1929         tree_put_node(&g->node, false);
1930         return rule;
1931
1932 err_release_fg:
1933         up_write_ref_node(&g->node, false);
1934         kmem_cache_free(steering->ftes_cache, fte);
1935 err_alloc_fte:
1936         tree_put_node(&g->node, false);
1937         return ERR_PTR(err);
1938 }
1939
1940 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1941 {
1942         return ((ft->type == FS_FT_NIC_RX) &&
1943                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1944 }
1945
1946 struct mlx5_flow_handle *
1947 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1948                     const struct mlx5_flow_spec *spec,
1949                     struct mlx5_flow_act *flow_act,
1950                     struct mlx5_flow_destination *dest,
1951                     int num_dest)
1952 {
1953         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1954         static const struct mlx5_flow_spec zero_spec = {};
1955         struct mlx5_flow_destination *gen_dest = NULL;
1956         struct mlx5_flow_table *next_ft = NULL;
1957         struct mlx5_flow_handle *handle = NULL;
1958         u32 sw_action = flow_act->action;
1959         int i;
1960
1961         if (!spec)
1962                 spec = &zero_spec;
1963
1964         if (!is_fwd_next_action(sw_action))
1965                 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1966
1967         if (!fwd_next_prio_supported(ft))
1968                 return ERR_PTR(-EOPNOTSUPP);
1969
1970         mutex_lock(&root->chain_lock);
1971         next_ft = find_next_fwd_ft(ft, flow_act);
1972         if (!next_ft) {
1973                 handle = ERR_PTR(-EOPNOTSUPP);
1974                 goto unlock;
1975         }
1976
1977         gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
1978                            GFP_KERNEL);
1979         if (!gen_dest) {
1980                 handle = ERR_PTR(-ENOMEM);
1981                 goto unlock;
1982         }
1983         for (i = 0; i < num_dest; i++)
1984                 gen_dest[i] = dest[i];
1985         gen_dest[i].type =
1986                 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1987         gen_dest[i].ft = next_ft;
1988         dest = gen_dest;
1989         num_dest++;
1990         flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
1991                               MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
1992         flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1993         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1994         if (IS_ERR(handle))
1995                 goto unlock;
1996
1997         if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
1998                 mutex_lock(&next_ft->lock);
1999                 list_add(&handle->rule[num_dest - 1]->next_ft,
2000                          &next_ft->fwd_rules);
2001                 mutex_unlock(&next_ft->lock);
2002                 handle->rule[num_dest - 1]->sw_action = sw_action;
2003                 handle->rule[num_dest - 1]->ft = ft;
2004         }
2005 unlock:
2006         mutex_unlock(&root->chain_lock);
2007         kfree(gen_dest);
2008         return handle;
2009 }
2010 EXPORT_SYMBOL(mlx5_add_flow_rules);
2011
2012 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2013 {
2014         struct fs_fte *fte;
2015         int i;
2016
2017         /* In order to consolidate the HW changes we lock the FTE for other
2018          * changes, and increase its refcount, in order not to perform the
2019          * "del" functions of the FTE. Will handle them here.
2020          * The removal of the rules is done under locked FTE.
2021          * After removing all the handle's rules, if there are remaining
2022          * rules, it means we just need to modify the FTE in FW, and
2023          * unlock/decrease the refcount we increased before.
2024          * Otherwise, it means the FTE should be deleted. First delete the
2025          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2026          * the FTE, which will handle the last decrease of the refcount, as
2027          * well as required handling of its parent.
2028          */
2029         fs_get_obj(fte, handle->rule[0]->node.parent);
2030         down_write_ref_node(&fte->node, false);
2031         for (i = handle->num_rules - 1; i >= 0; i--)
2032                 tree_remove_node(&handle->rule[i]->node, true);
2033         if (fte->dests_size) {
2034                 if (fte->modify_mask)
2035                         modify_fte(fte);
2036                 up_write_ref_node(&fte->node, false);
2037         } else if (list_empty(&fte->node.children)) {
2038                 del_hw_fte(&fte->node);
2039                 /* Avoid double call to del_hw_fte */
2040                 fte->node.del_hw_func = NULL;
2041                 up_write_ref_node(&fte->node, false);
2042                 tree_put_node(&fte->node, false);
2043         }
2044         kfree(handle);
2045 }
2046 EXPORT_SYMBOL(mlx5_del_flow_rules);
2047
2048 /* Assuming prio->node.children(flow tables) is sorted by level */
2049 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2050 {
2051         struct fs_prio *prio;
2052
2053         fs_get_obj(prio, ft->node.parent);
2054
2055         if (!list_is_last(&ft->node.list, &prio->node.children))
2056                 return list_next_entry(ft, node.list);
2057         return find_next_chained_ft(prio);
2058 }
2059
2060 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2061 {
2062         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2063         struct mlx5_ft_underlay_qp *uqp;
2064         struct mlx5_flow_table *new_root_ft = NULL;
2065         int err = 0;
2066         u32 qpn;
2067
2068         if (root->root_ft != ft)
2069                 return 0;
2070
2071         new_root_ft = find_next_ft(ft);
2072         if (!new_root_ft) {
2073                 root->root_ft = NULL;
2074                 return 0;
2075         }
2076
2077         if (list_empty(&root->underlay_qpns)) {
2078                 /* Don't set any QPN (zero) in case QPN list is empty */
2079                 qpn = 0;
2080                 err = root->cmds->update_root_ft(root, new_root_ft,
2081                                                  qpn, false);
2082         } else {
2083                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2084                         qpn = uqp->qpn;
2085                         err = root->cmds->update_root_ft(root,
2086                                                          new_root_ft, qpn,
2087                                                          false);
2088                         if (err)
2089                                 break;
2090                 }
2091         }
2092
2093         if (err)
2094                 mlx5_core_warn(root->dev,
2095                                "Update root flow table of id(%u) qpn(%d) failed\n",
2096                                ft->id, qpn);
2097         else
2098                 root->root_ft = new_root_ft;
2099
2100         return 0;
2101 }
2102
2103 /* Connect flow table from previous priority to
2104  * the next flow table.
2105  */
2106 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2107 {
2108         struct mlx5_core_dev *dev = get_dev(&ft->node);
2109         struct mlx5_flow_table *next_ft;
2110         struct fs_prio *prio;
2111         int err = 0;
2112
2113         err = update_root_ft_destroy(ft);
2114         if (err)
2115                 return err;
2116
2117         fs_get_obj(prio, ft->node.parent);
2118         if  (!(list_first_entry(&prio->node.children,
2119                                 struct mlx5_flow_table,
2120                                 node.list) == ft))
2121                 return 0;
2122
2123         next_ft = find_next_chained_ft(prio);
2124         err = connect_fwd_rules(dev, next_ft, ft);
2125         if (err)
2126                 return err;
2127
2128         err = connect_prev_fts(dev, next_ft, prio);
2129         if (err)
2130                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2131                                ft->id);
2132         return err;
2133 }
2134
2135 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2136 {
2137         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2138         int err = 0;
2139
2140         mutex_lock(&root->chain_lock);
2141         if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2142                 err = disconnect_flow_table(ft);
2143         if (err) {
2144                 mutex_unlock(&root->chain_lock);
2145                 return err;
2146         }
2147         if (tree_remove_node(&ft->node, false))
2148                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2149                                ft->id);
2150         mutex_unlock(&root->chain_lock);
2151
2152         return err;
2153 }
2154 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2155
2156 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2157 {
2158         if (tree_remove_node(&fg->node, false))
2159                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2160                                fg->id);
2161 }
2162 EXPORT_SYMBOL(mlx5_destroy_flow_group);
2163
2164 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2165                                                 int n)
2166 {
2167         struct mlx5_flow_steering *steering = dev->priv.steering;
2168
2169         if (!steering || !steering->fdb_sub_ns)
2170                 return NULL;
2171
2172         return steering->fdb_sub_ns[n];
2173 }
2174 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2175
2176 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2177                                                     enum mlx5_flow_namespace_type type)
2178 {
2179         struct mlx5_flow_steering *steering = dev->priv.steering;
2180         struct mlx5_flow_root_namespace *root_ns;
2181         int prio = 0;
2182         struct fs_prio *fs_prio;
2183         struct mlx5_flow_namespace *ns;
2184
2185         if (!steering)
2186                 return NULL;
2187
2188         switch (type) {
2189         case MLX5_FLOW_NAMESPACE_FDB:
2190                 if (steering->fdb_root_ns)
2191                         return &steering->fdb_root_ns->ns;
2192                 return NULL;
2193         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2194                 if (steering->sniffer_rx_root_ns)
2195                         return &steering->sniffer_rx_root_ns->ns;
2196                 return NULL;
2197         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2198                 if (steering->sniffer_tx_root_ns)
2199                         return &steering->sniffer_tx_root_ns->ns;
2200                 return NULL;
2201         default:
2202                 break;
2203         }
2204
2205         if (type == MLX5_FLOW_NAMESPACE_EGRESS ||
2206             type == MLX5_FLOW_NAMESPACE_EGRESS_KERNEL) {
2207                 root_ns = steering->egress_root_ns;
2208                 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2209         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX) {
2210                 root_ns = steering->rdma_rx_root_ns;
2211                 prio = RDMA_RX_BYPASS_PRIO;
2212         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL) {
2213                 root_ns = steering->rdma_rx_root_ns;
2214                 prio = RDMA_RX_KERNEL_PRIO;
2215         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_TX) {
2216                 root_ns = steering->rdma_tx_root_ns;
2217         } else { /* Must be NIC RX */
2218                 root_ns = steering->root_ns;
2219                 prio = type;
2220         }
2221
2222         if (!root_ns)
2223                 return NULL;
2224
2225         fs_prio = find_prio(&root_ns->ns, prio);
2226         if (!fs_prio)
2227                 return NULL;
2228
2229         ns = list_first_entry(&fs_prio->node.children,
2230                               typeof(*ns),
2231                               node.list);
2232
2233         return ns;
2234 }
2235 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2236
2237 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2238                                                               enum mlx5_flow_namespace_type type,
2239                                                               int vport)
2240 {
2241         struct mlx5_flow_steering *steering = dev->priv.steering;
2242
2243         if (!steering)
2244                 return NULL;
2245
2246         switch (type) {
2247         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2248                 if (vport >= steering->esw_egress_acl_vports)
2249                         return NULL;
2250                 if (steering->esw_egress_root_ns &&
2251                     steering->esw_egress_root_ns[vport])
2252                         return &steering->esw_egress_root_ns[vport]->ns;
2253                 else
2254                         return NULL;
2255         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2256                 if (vport >= steering->esw_ingress_acl_vports)
2257                         return NULL;
2258                 if (steering->esw_ingress_root_ns &&
2259                     steering->esw_ingress_root_ns[vport])
2260                         return &steering->esw_ingress_root_ns[vport]->ns;
2261                 else
2262                         return NULL;
2263         default:
2264                 return NULL;
2265         }
2266 }
2267
2268 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2269                                        unsigned int prio,
2270                                        int num_levels,
2271                                        enum fs_node_type type)
2272 {
2273         struct fs_prio *fs_prio;
2274
2275         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2276         if (!fs_prio)
2277                 return ERR_PTR(-ENOMEM);
2278
2279         fs_prio->node.type = type;
2280         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2281         tree_add_node(&fs_prio->node, &ns->node);
2282         fs_prio->num_levels = num_levels;
2283         fs_prio->prio = prio;
2284         list_add_tail(&fs_prio->node.list, &ns->node.children);
2285
2286         return fs_prio;
2287 }
2288
2289 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2290                                               unsigned int prio,
2291                                               int num_levels)
2292 {
2293         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2294 }
2295
2296 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2297                                       unsigned int prio, int num_levels)
2298 {
2299         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2300 }
2301
2302 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2303                                                      *ns)
2304 {
2305         ns->node.type = FS_TYPE_NAMESPACE;
2306
2307         return ns;
2308 }
2309
2310 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2311                                                        int def_miss_act)
2312 {
2313         struct mlx5_flow_namespace      *ns;
2314
2315         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2316         if (!ns)
2317                 return ERR_PTR(-ENOMEM);
2318
2319         fs_init_namespace(ns);
2320         ns->def_miss_action = def_miss_act;
2321         tree_init_node(&ns->node, NULL, del_sw_ns);
2322         tree_add_node(&ns->node, &prio->node);
2323         list_add_tail(&ns->node.list, &prio->node.children);
2324
2325         return ns;
2326 }
2327
2328 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2329                              struct init_tree_node *prio_metadata)
2330 {
2331         struct fs_prio *fs_prio;
2332         int i;
2333
2334         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2335                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2336                 if (IS_ERR(fs_prio))
2337                         return PTR_ERR(fs_prio);
2338         }
2339         return 0;
2340 }
2341
2342 #define FLOW_TABLE_BIT_SZ 1
2343 #define GET_FLOW_TABLE_CAP(dev, offset) \
2344         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2345                         offset / 32)) >>                                        \
2346           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2347 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2348 {
2349         int i;
2350
2351         for (i = 0; i < caps->arr_sz; i++) {
2352                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2353                         return false;
2354         }
2355         return true;
2356 }
2357
2358 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2359                                     struct init_tree_node *init_node,
2360                                     struct fs_node *fs_parent_node,
2361                                     struct init_tree_node *init_parent_node,
2362                                     int prio)
2363 {
2364         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2365                                               flow_table_properties_nic_receive.
2366                                               max_ft_level);
2367         struct mlx5_flow_namespace *fs_ns;
2368         struct fs_prio *fs_prio;
2369         struct fs_node *base;
2370         int i;
2371         int err;
2372
2373         if (init_node->type == FS_TYPE_PRIO) {
2374                 if ((init_node->min_ft_level > max_ft_level) ||
2375                     !has_required_caps(steering->dev, &init_node->caps))
2376                         return 0;
2377
2378                 fs_get_obj(fs_ns, fs_parent_node);
2379                 if (init_node->num_leaf_prios)
2380                         return create_leaf_prios(fs_ns, prio, init_node);
2381                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2382                 if (IS_ERR(fs_prio))
2383                         return PTR_ERR(fs_prio);
2384                 base = &fs_prio->node;
2385         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2386                 fs_get_obj(fs_prio, fs_parent_node);
2387                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2388                 if (IS_ERR(fs_ns))
2389                         return PTR_ERR(fs_ns);
2390                 base = &fs_ns->node;
2391         } else {
2392                 return -EINVAL;
2393         }
2394         prio = 0;
2395         for (i = 0; i < init_node->ar_size; i++) {
2396                 err = init_root_tree_recursive(steering, &init_node->children[i],
2397                                                base, init_node, prio);
2398                 if (err)
2399                         return err;
2400                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2401                     init_node->children[i].num_leaf_prios) {
2402                         prio += init_node->children[i].num_leaf_prios;
2403                 }
2404         }
2405
2406         return 0;
2407 }
2408
2409 static int init_root_tree(struct mlx5_flow_steering *steering,
2410                           struct init_tree_node *init_node,
2411                           struct fs_node *fs_parent_node)
2412 {
2413         int err;
2414         int i;
2415
2416         for (i = 0; i < init_node->ar_size; i++) {
2417                 err = init_root_tree_recursive(steering, &init_node->children[i],
2418                                                fs_parent_node,
2419                                                init_node, i);
2420                 if (err)
2421                         return err;
2422         }
2423         return 0;
2424 }
2425
2426 static void del_sw_root_ns(struct fs_node *node)
2427 {
2428         struct mlx5_flow_root_namespace *root_ns;
2429         struct mlx5_flow_namespace *ns;
2430
2431         fs_get_obj(ns, node);
2432         root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2433         mutex_destroy(&root_ns->chain_lock);
2434         kfree(node);
2435 }
2436
2437 static struct mlx5_flow_root_namespace
2438 *create_root_ns(struct mlx5_flow_steering *steering,
2439                 enum fs_flow_table_type table_type)
2440 {
2441         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2442         struct mlx5_flow_root_namespace *root_ns;
2443         struct mlx5_flow_namespace *ns;
2444
2445         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2446             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2447                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2448
2449         /* Create the root namespace */
2450         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2451         if (!root_ns)
2452                 return NULL;
2453
2454         root_ns->dev = steering->dev;
2455         root_ns->table_type = table_type;
2456         root_ns->cmds = cmds;
2457
2458         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2459
2460         ns = &root_ns->ns;
2461         fs_init_namespace(ns);
2462         mutex_init(&root_ns->chain_lock);
2463         tree_init_node(&ns->node, NULL, del_sw_root_ns);
2464         tree_add_node(&ns->node, NULL);
2465
2466         return root_ns;
2467 }
2468
2469 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2470
2471 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2472 {
2473         struct fs_prio *prio;
2474
2475         fs_for_each_prio(prio, ns) {
2476                  /* This updates prio start_level and num_levels */
2477                 set_prio_attrs_in_prio(prio, acc_level);
2478                 acc_level += prio->num_levels;
2479         }
2480         return acc_level;
2481 }
2482
2483 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2484 {
2485         struct mlx5_flow_namespace *ns;
2486         int acc_level_ns = acc_level;
2487
2488         prio->start_level = acc_level;
2489         fs_for_each_ns(ns, prio) {
2490                 /* This updates start_level and num_levels of ns's priority descendants */
2491                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2492
2493                 /* If this a prio with chains, and we can jump from one chain
2494                  * (namepsace) to another, so we accumulate the levels
2495                  */
2496                 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2497                         acc_level = acc_level_ns;
2498         }
2499
2500         if (!prio->num_levels)
2501                 prio->num_levels = acc_level_ns - prio->start_level;
2502         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2503 }
2504
2505 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2506 {
2507         struct mlx5_flow_namespace *ns = &root_ns->ns;
2508         struct fs_prio *prio;
2509         int start_level = 0;
2510
2511         fs_for_each_prio(prio, ns) {
2512                 set_prio_attrs_in_prio(prio, start_level);
2513                 start_level += prio->num_levels;
2514         }
2515 }
2516
2517 #define ANCHOR_PRIO 0
2518 #define ANCHOR_SIZE 1
2519 #define ANCHOR_LEVEL 0
2520 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2521 {
2522         struct mlx5_flow_namespace *ns = NULL;
2523         struct mlx5_flow_table_attr ft_attr = {};
2524         struct mlx5_flow_table *ft;
2525
2526         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2527         if (WARN_ON(!ns))
2528                 return -EINVAL;
2529
2530         ft_attr.max_fte = ANCHOR_SIZE;
2531         ft_attr.level   = ANCHOR_LEVEL;
2532         ft_attr.prio    = ANCHOR_PRIO;
2533
2534         ft = mlx5_create_flow_table(ns, &ft_attr);
2535         if (IS_ERR(ft)) {
2536                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2537                 return PTR_ERR(ft);
2538         }
2539         return 0;
2540 }
2541
2542 static int init_root_ns(struct mlx5_flow_steering *steering)
2543 {
2544         int err;
2545
2546         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2547         if (!steering->root_ns)
2548                 return -ENOMEM;
2549
2550         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2551         if (err)
2552                 goto out_err;
2553
2554         set_prio_attrs(steering->root_ns);
2555         err = create_anchor_flow_table(steering);
2556         if (err)
2557                 goto out_err;
2558
2559         return 0;
2560
2561 out_err:
2562         cleanup_root_ns(steering->root_ns);
2563         steering->root_ns = NULL;
2564         return err;
2565 }
2566
2567 static void clean_tree(struct fs_node *node)
2568 {
2569         if (node) {
2570                 struct fs_node *iter;
2571                 struct fs_node *temp;
2572
2573                 tree_get_node(node);
2574                 list_for_each_entry_safe(iter, temp, &node->children, list)
2575                         clean_tree(iter);
2576                 tree_put_node(node, false);
2577                 tree_remove_node(node, false);
2578         }
2579 }
2580
2581 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2582 {
2583         if (!root_ns)
2584                 return;
2585
2586         clean_tree(&root_ns->ns.node);
2587 }
2588
2589 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2590 {
2591         struct mlx5_flow_steering *steering = dev->priv.steering;
2592
2593         cleanup_root_ns(steering->root_ns);
2594         cleanup_root_ns(steering->fdb_root_ns);
2595         steering->fdb_root_ns = NULL;
2596         kfree(steering->fdb_sub_ns);
2597         steering->fdb_sub_ns = NULL;
2598         cleanup_root_ns(steering->sniffer_rx_root_ns);
2599         cleanup_root_ns(steering->sniffer_tx_root_ns);
2600         cleanup_root_ns(steering->rdma_rx_root_ns);
2601         cleanup_root_ns(steering->rdma_tx_root_ns);
2602         cleanup_root_ns(steering->egress_root_ns);
2603         mlx5_cleanup_fc_stats(dev);
2604         kmem_cache_destroy(steering->ftes_cache);
2605         kmem_cache_destroy(steering->fgs_cache);
2606         mlx5_ft_pool_destroy(dev);
2607         kfree(steering);
2608 }
2609
2610 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2611 {
2612         struct fs_prio *prio;
2613
2614         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2615         if (!steering->sniffer_tx_root_ns)
2616                 return -ENOMEM;
2617
2618         /* Create single prio */
2619         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2620         return PTR_ERR_OR_ZERO(prio);
2621 }
2622
2623 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2624 {
2625         struct fs_prio *prio;
2626
2627         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2628         if (!steering->sniffer_rx_root_ns)
2629                 return -ENOMEM;
2630
2631         /* Create single prio */
2632         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2633         return PTR_ERR_OR_ZERO(prio);
2634 }
2635
2636 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2637 {
2638         int err;
2639
2640         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2641         if (!steering->rdma_rx_root_ns)
2642                 return -ENOMEM;
2643
2644         err = init_root_tree(steering, &rdma_rx_root_fs,
2645                              &steering->rdma_rx_root_ns->ns.node);
2646         if (err)
2647                 goto out_err;
2648
2649         set_prio_attrs(steering->rdma_rx_root_ns);
2650
2651         return 0;
2652
2653 out_err:
2654         cleanup_root_ns(steering->rdma_rx_root_ns);
2655         steering->rdma_rx_root_ns = NULL;
2656         return err;
2657 }
2658
2659 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2660 {
2661         int err;
2662
2663         steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2664         if (!steering->rdma_tx_root_ns)
2665                 return -ENOMEM;
2666
2667         err = init_root_tree(steering, &rdma_tx_root_fs,
2668                              &steering->rdma_tx_root_ns->ns.node);
2669         if (err)
2670                 goto out_err;
2671
2672         set_prio_attrs(steering->rdma_tx_root_ns);
2673
2674         return 0;
2675
2676 out_err:
2677         cleanup_root_ns(steering->rdma_tx_root_ns);
2678         steering->rdma_tx_root_ns = NULL;
2679         return err;
2680 }
2681
2682 /* FT and tc chains are stored in the same array so we can re-use the
2683  * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2684  * When creating a new ns for each chain store it in the first available slot.
2685  * Assume tc chains are created and stored first and only then the FT chain.
2686  */
2687 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2688                                         struct mlx5_flow_namespace *ns)
2689 {
2690         int chain = 0;
2691
2692         while (steering->fdb_sub_ns[chain])
2693                 ++chain;
2694
2695         steering->fdb_sub_ns[chain] = ns;
2696 }
2697
2698 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2699                                         struct fs_prio *maj_prio)
2700 {
2701         struct mlx5_flow_namespace *ns;
2702         struct fs_prio *min_prio;
2703         int prio;
2704
2705         ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2706         if (IS_ERR(ns))
2707                 return PTR_ERR(ns);
2708
2709         for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2710                 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2711                 if (IS_ERR(min_prio))
2712                         return PTR_ERR(min_prio);
2713         }
2714
2715         store_fdb_sub_ns_prio_chain(steering, ns);
2716
2717         return 0;
2718 }
2719
2720 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2721                              int fs_prio,
2722                              int chains)
2723 {
2724         struct fs_prio *maj_prio;
2725         int levels;
2726         int chain;
2727         int err;
2728
2729         levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2730         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2731                                           fs_prio,
2732                                           levels);
2733         if (IS_ERR(maj_prio))
2734                 return PTR_ERR(maj_prio);
2735
2736         for (chain = 0; chain < chains; chain++) {
2737                 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2738                 if (err)
2739                         return err;
2740         }
2741
2742         return 0;
2743 }
2744
2745 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2746 {
2747         int err;
2748
2749         steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2750                                        sizeof(*steering->fdb_sub_ns),
2751                                        GFP_KERNEL);
2752         if (!steering->fdb_sub_ns)
2753                 return -ENOMEM;
2754
2755         err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
2756         if (err)
2757                 return err;
2758
2759         err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
2760         if (err)
2761                 return err;
2762
2763         return 0;
2764 }
2765
2766 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2767 {
2768         struct fs_prio *maj_prio;
2769         int err;
2770
2771         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2772         if (!steering->fdb_root_ns)
2773                 return -ENOMEM;
2774
2775         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2776                                   1);
2777         if (IS_ERR(maj_prio)) {
2778                 err = PTR_ERR(maj_prio);
2779                 goto out_err;
2780         }
2781         err = create_fdb_fast_path(steering);
2782         if (err)
2783                 goto out_err;
2784
2785         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1);
2786         if (IS_ERR(maj_prio)) {
2787                 err = PTR_ERR(maj_prio);
2788                 goto out_err;
2789         }
2790
2791         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 3);
2792         if (IS_ERR(maj_prio)) {
2793                 err = PTR_ERR(maj_prio);
2794                 goto out_err;
2795         }
2796
2797         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2798         if (IS_ERR(maj_prio)) {
2799                 err = PTR_ERR(maj_prio);
2800                 goto out_err;
2801         }
2802
2803         /* We put this priority last, knowing that nothing will get here
2804          * unless explicitly forwarded to. This is possible because the
2805          * slow path tables have catch all rules and nothing gets passed
2806          * those tables.
2807          */
2808         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
2809         if (IS_ERR(maj_prio)) {
2810                 err = PTR_ERR(maj_prio);
2811                 goto out_err;
2812         }
2813
2814         set_prio_attrs(steering->fdb_root_ns);
2815         return 0;
2816
2817 out_err:
2818         cleanup_root_ns(steering->fdb_root_ns);
2819         kfree(steering->fdb_sub_ns);
2820         steering->fdb_sub_ns = NULL;
2821         steering->fdb_root_ns = NULL;
2822         return err;
2823 }
2824
2825 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2826 {
2827         struct fs_prio *prio;
2828
2829         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2830         if (!steering->esw_egress_root_ns[vport])
2831                 return -ENOMEM;
2832
2833         /* create 1 prio*/
2834         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2835         return PTR_ERR_OR_ZERO(prio);
2836 }
2837
2838 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2839 {
2840         struct fs_prio *prio;
2841
2842         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2843         if (!steering->esw_ingress_root_ns[vport])
2844                 return -ENOMEM;
2845
2846         /* create 1 prio*/
2847         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2848         return PTR_ERR_OR_ZERO(prio);
2849 }
2850
2851 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2852 {
2853         struct mlx5_flow_steering *steering = dev->priv.steering;
2854         int err;
2855         int i;
2856
2857         steering->esw_egress_root_ns =
2858                         kcalloc(total_vports,
2859                                 sizeof(*steering->esw_egress_root_ns),
2860                                 GFP_KERNEL);
2861         if (!steering->esw_egress_root_ns)
2862                 return -ENOMEM;
2863
2864         for (i = 0; i < total_vports; i++) {
2865                 err = init_egress_acl_root_ns(steering, i);
2866                 if (err)
2867                         goto cleanup_root_ns;
2868         }
2869         steering->esw_egress_acl_vports = total_vports;
2870         return 0;
2871
2872 cleanup_root_ns:
2873         for (i--; i >= 0; i--)
2874                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2875         kfree(steering->esw_egress_root_ns);
2876         steering->esw_egress_root_ns = NULL;
2877         return err;
2878 }
2879
2880 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev)
2881 {
2882         struct mlx5_flow_steering *steering = dev->priv.steering;
2883         int i;
2884
2885         if (!steering->esw_egress_root_ns)
2886                 return;
2887
2888         for (i = 0; i < steering->esw_egress_acl_vports; i++)
2889                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2890
2891         kfree(steering->esw_egress_root_ns);
2892         steering->esw_egress_root_ns = NULL;
2893 }
2894
2895 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2896 {
2897         struct mlx5_flow_steering *steering = dev->priv.steering;
2898         int err;
2899         int i;
2900
2901         steering->esw_ingress_root_ns =
2902                         kcalloc(total_vports,
2903                                 sizeof(*steering->esw_ingress_root_ns),
2904                                 GFP_KERNEL);
2905         if (!steering->esw_ingress_root_ns)
2906                 return -ENOMEM;
2907
2908         for (i = 0; i < total_vports; i++) {
2909                 err = init_ingress_acl_root_ns(steering, i);
2910                 if (err)
2911                         goto cleanup_root_ns;
2912         }
2913         steering->esw_ingress_acl_vports = total_vports;
2914         return 0;
2915
2916 cleanup_root_ns:
2917         for (i--; i >= 0; i--)
2918                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2919         kfree(steering->esw_ingress_root_ns);
2920         steering->esw_ingress_root_ns = NULL;
2921         return err;
2922 }
2923
2924 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev)
2925 {
2926         struct mlx5_flow_steering *steering = dev->priv.steering;
2927         int i;
2928
2929         if (!steering->esw_ingress_root_ns)
2930                 return;
2931
2932         for (i = 0; i < steering->esw_ingress_acl_vports; i++)
2933                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2934
2935         kfree(steering->esw_ingress_root_ns);
2936         steering->esw_ingress_root_ns = NULL;
2937 }
2938
2939 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2940 {
2941         int err;
2942
2943         steering->egress_root_ns = create_root_ns(steering,
2944                                                   FS_FT_NIC_TX);
2945         if (!steering->egress_root_ns)
2946                 return -ENOMEM;
2947
2948         err = init_root_tree(steering, &egress_root_fs,
2949                              &steering->egress_root_ns->ns.node);
2950         if (err)
2951                 goto cleanup;
2952         set_prio_attrs(steering->egress_root_ns);
2953         return 0;
2954 cleanup:
2955         cleanup_root_ns(steering->egress_root_ns);
2956         steering->egress_root_ns = NULL;
2957         return err;
2958 }
2959
2960 int mlx5_init_fs(struct mlx5_core_dev *dev)
2961 {
2962         struct mlx5_flow_steering *steering;
2963         int err = 0;
2964
2965         err = mlx5_init_fc_stats(dev);
2966         if (err)
2967                 return err;
2968
2969         err = mlx5_ft_pool_init(dev);
2970         if (err)
2971                 return err;
2972
2973         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2974         if (!steering) {
2975                 err = -ENOMEM;
2976                 goto err;
2977         }
2978
2979         steering->dev = dev;
2980         dev->priv.steering = steering;
2981
2982         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2983                                                 sizeof(struct mlx5_flow_group), 0,
2984                                                 0, NULL);
2985         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2986                                                  0, NULL);
2987         if (!steering->ftes_cache || !steering->fgs_cache) {
2988                 err = -ENOMEM;
2989                 goto err;
2990         }
2991
2992         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2993               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2994              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2995               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2996             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2997                 err = init_root_ns(steering);
2998                 if (err)
2999                         goto err;
3000         }
3001
3002         if (MLX5_ESWITCH_MANAGER(dev)) {
3003                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
3004                         err = init_fdb_root_ns(steering);
3005                         if (err)
3006                                 goto err;
3007                 }
3008         }
3009
3010         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
3011                 err = init_sniffer_rx_root_ns(steering);
3012                 if (err)
3013                         goto err;
3014         }
3015
3016         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
3017                 err = init_sniffer_tx_root_ns(steering);
3018                 if (err)
3019                         goto err;
3020         }
3021
3022         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
3023             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3024                 err = init_rdma_rx_root_ns(steering);
3025                 if (err)
3026                         goto err;
3027         }
3028
3029         if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3030                 err = init_rdma_tx_root_ns(steering);
3031                 if (err)
3032                         goto err;
3033         }
3034
3035         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE ||
3036             MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3037                 err = init_egress_root_ns(steering);
3038                 if (err)
3039                         goto err;
3040         }
3041
3042         return 0;
3043 err:
3044         mlx5_cleanup_fs(dev);
3045         return err;
3046 }
3047
3048 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3049 {
3050         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3051         struct mlx5_ft_underlay_qp *new_uqp;
3052         int err = 0;
3053
3054         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3055         if (!new_uqp)
3056                 return -ENOMEM;
3057
3058         mutex_lock(&root->chain_lock);
3059
3060         if (!root->root_ft) {
3061                 err = -EINVAL;
3062                 goto update_ft_fail;
3063         }
3064
3065         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3066                                          false);
3067         if (err) {
3068                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3069                                underlay_qpn, err);
3070                 goto update_ft_fail;
3071         }
3072
3073         new_uqp->qpn = underlay_qpn;
3074         list_add_tail(&new_uqp->list, &root->underlay_qpns);
3075
3076         mutex_unlock(&root->chain_lock);
3077
3078         return 0;
3079
3080 update_ft_fail:
3081         mutex_unlock(&root->chain_lock);
3082         kfree(new_uqp);
3083         return err;
3084 }
3085 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3086
3087 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3088 {
3089         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3090         struct mlx5_ft_underlay_qp *uqp;
3091         bool found = false;
3092         int err = 0;
3093
3094         mutex_lock(&root->chain_lock);
3095         list_for_each_entry(uqp, &root->underlay_qpns, list) {
3096                 if (uqp->qpn == underlay_qpn) {
3097                         found = true;
3098                         break;
3099                 }
3100         }
3101
3102         if (!found) {
3103                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3104                                underlay_qpn);
3105                 err = -EINVAL;
3106                 goto out;
3107         }
3108
3109         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3110                                          true);
3111         if (err)
3112                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3113                                underlay_qpn, err);
3114
3115         list_del(&uqp->list);
3116         mutex_unlock(&root->chain_lock);
3117         kfree(uqp);
3118
3119         return 0;
3120
3121 out:
3122         mutex_unlock(&root->chain_lock);
3123         return err;
3124 }
3125 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3126
3127 static struct mlx5_flow_root_namespace
3128 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3129 {
3130         struct mlx5_flow_namespace *ns;
3131
3132         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3133             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3134                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3135         else
3136                 ns = mlx5_get_flow_namespace(dev, ns_type);
3137         if (!ns)
3138                 return NULL;
3139
3140         return find_root(&ns->node);
3141 }
3142
3143 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3144                                                  u8 ns_type, u8 num_actions,
3145                                                  void *modify_actions)
3146 {
3147         struct mlx5_flow_root_namespace *root;
3148         struct mlx5_modify_hdr *modify_hdr;
3149         int err;
3150
3151         root = get_root_namespace(dev, ns_type);
3152         if (!root)
3153                 return ERR_PTR(-EOPNOTSUPP);
3154
3155         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3156         if (!modify_hdr)
3157                 return ERR_PTR(-ENOMEM);
3158
3159         modify_hdr->ns_type = ns_type;
3160         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3161                                               modify_actions, modify_hdr);
3162         if (err) {
3163                 kfree(modify_hdr);
3164                 return ERR_PTR(err);
3165         }
3166
3167         return modify_hdr;
3168 }
3169 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3170
3171 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3172                                 struct mlx5_modify_hdr *modify_hdr)
3173 {
3174         struct mlx5_flow_root_namespace *root;
3175
3176         root = get_root_namespace(dev, modify_hdr->ns_type);
3177         if (WARN_ON(!root))
3178                 return;
3179         root->cmds->modify_header_dealloc(root, modify_hdr);
3180         kfree(modify_hdr);
3181 }
3182 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3183
3184 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3185                                                      struct mlx5_pkt_reformat_params *params,
3186                                                      enum mlx5_flow_namespace_type ns_type)
3187 {
3188         struct mlx5_pkt_reformat *pkt_reformat;
3189         struct mlx5_flow_root_namespace *root;
3190         int err;
3191
3192         root = get_root_namespace(dev, ns_type);
3193         if (!root)
3194                 return ERR_PTR(-EOPNOTSUPP);
3195
3196         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3197         if (!pkt_reformat)
3198                 return ERR_PTR(-ENOMEM);
3199
3200         pkt_reformat->ns_type = ns_type;
3201         pkt_reformat->reformat_type = params->type;
3202         err = root->cmds->packet_reformat_alloc(root, params, ns_type,
3203                                                 pkt_reformat);
3204         if (err) {
3205                 kfree(pkt_reformat);
3206                 return ERR_PTR(err);
3207         }
3208
3209         return pkt_reformat;
3210 }
3211 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3212
3213 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3214                                   struct mlx5_pkt_reformat *pkt_reformat)
3215 {
3216         struct mlx5_flow_root_namespace *root;
3217
3218         root = get_root_namespace(dev, pkt_reformat->ns_type);
3219         if (WARN_ON(!root))
3220                 return;
3221         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3222         kfree(pkt_reformat);
3223 }
3224 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3225
3226 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3227                                  struct mlx5_flow_root_namespace *peer_ns)
3228 {
3229         if (peer_ns && ns->mode != peer_ns->mode) {
3230                 mlx5_core_err(ns->dev,
3231                               "Can't peer namespace of different steering mode\n");
3232                 return -EINVAL;
3233         }
3234
3235         return ns->cmds->set_peer(ns, peer_ns);
3236 }
3237
3238 /* This function should be called only at init stage of the namespace.
3239  * It is not safe to call this function while steering operations
3240  * are executed in the namespace.
3241  */
3242 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3243                                  enum mlx5_flow_steering_mode mode)
3244 {
3245         struct mlx5_flow_root_namespace *root;
3246         const struct mlx5_flow_cmds *cmds;
3247         int err;
3248
3249         root = find_root(&ns->node);
3250         if (&root->ns != ns)
3251         /* Can't set cmds to non root namespace */
3252                 return -EINVAL;
3253
3254         if (root->table_type != FS_FT_FDB)
3255                 return -EOPNOTSUPP;
3256
3257         if (root->mode == mode)
3258                 return 0;
3259
3260         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3261                 cmds = mlx5_fs_cmd_get_dr_cmds();
3262         else
3263                 cmds = mlx5_fs_cmd_get_fw_cmds();
3264         if (!cmds)
3265                 return -EOPNOTSUPP;
3266
3267         err = cmds->create_ns(root);
3268         if (err) {
3269                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3270                               err);
3271                 return err;
3272         }
3273
3274         root->cmds->destroy_ns(root);
3275         root->cmds = cmds;
3276         root->mode = mode;
3277
3278         return 0;
3279 }