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 struct mlx5_flow_table*
1177 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1178                                     struct mlx5_flow_table_attr *ft_attr)
1179 {
1180         int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1181         int max_num_groups = ft_attr->autogroup.max_num_groups;
1182         struct mlx5_flow_table *ft;
1183         int autogroups_max_fte;
1184
1185         ft = mlx5_create_flow_table(ns, ft_attr);
1186         if (IS_ERR(ft))
1187                 return ft;
1188
1189         autogroups_max_fte = ft->max_fte - num_reserved_entries;
1190         if (max_num_groups > autogroups_max_fte)
1191                 goto err_validate;
1192         if (num_reserved_entries > ft->max_fte)
1193                 goto err_validate;
1194
1195         ft->autogroup.active = true;
1196         ft->autogroup.required_groups = max_num_groups;
1197         ft->autogroup.max_fte = autogroups_max_fte;
1198         /* We save place for flow groups in addition to max types */
1199         ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1200
1201         return ft;
1202
1203 err_validate:
1204         mlx5_destroy_flow_table(ft);
1205         return ERR_PTR(-ENOSPC);
1206 }
1207 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1208
1209 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1210                                                u32 *fg_in)
1211 {
1212         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1213         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1214                                             fg_in, match_criteria);
1215         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1216                                             fg_in,
1217                                             match_criteria_enable);
1218         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1219                                    start_flow_index);
1220         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1221                                  end_flow_index);
1222         struct mlx5_flow_group *fg;
1223         int err;
1224
1225         if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1226                 return ERR_PTR(-EPERM);
1227
1228         down_write_ref_node(&ft->node, false);
1229         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1230                                      start_index, end_index,
1231                                      ft->node.children.prev);
1232         up_write_ref_node(&ft->node, false);
1233         if (IS_ERR(fg))
1234                 return fg;
1235
1236         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1237         if (err) {
1238                 tree_put_node(&fg->node, false);
1239                 return ERR_PTR(err);
1240         }
1241         trace_mlx5_fs_add_fg(fg);
1242         fg->node.active = true;
1243
1244         return fg;
1245 }
1246 EXPORT_SYMBOL(mlx5_create_flow_group);
1247
1248 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1249 {
1250         struct mlx5_flow_rule *rule;
1251
1252         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1253         if (!rule)
1254                 return NULL;
1255
1256         INIT_LIST_HEAD(&rule->next_ft);
1257         rule->node.type = FS_TYPE_FLOW_DEST;
1258         if (dest)
1259                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1260
1261         return rule;
1262 }
1263
1264 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1265 {
1266         struct mlx5_flow_handle *handle;
1267
1268         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1269         if (!handle)
1270                 return NULL;
1271
1272         handle->num_rules = num_rules;
1273
1274         return handle;
1275 }
1276
1277 static void destroy_flow_handle(struct fs_fte *fte,
1278                                 struct mlx5_flow_handle *handle,
1279                                 struct mlx5_flow_destination *dest,
1280                                 int i)
1281 {
1282         for (; --i >= 0;) {
1283                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1284                         fte->dests_size--;
1285                         list_del(&handle->rule[i]->node.list);
1286                         kfree(handle->rule[i]);
1287                 }
1288         }
1289         kfree(handle);
1290 }
1291
1292 static struct mlx5_flow_handle *
1293 create_flow_handle(struct fs_fte *fte,
1294                    struct mlx5_flow_destination *dest,
1295                    int dest_num,
1296                    int *modify_mask,
1297                    bool *new_rule)
1298 {
1299         struct mlx5_flow_handle *handle;
1300         struct mlx5_flow_rule *rule = NULL;
1301         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1302         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1303         int type;
1304         int i = 0;
1305
1306         handle = alloc_handle((dest_num) ? dest_num : 1);
1307         if (!handle)
1308                 return ERR_PTR(-ENOMEM);
1309
1310         do {
1311                 if (dest) {
1312                         rule = find_flow_rule(fte, dest + i);
1313                         if (rule) {
1314                                 refcount_inc(&rule->node.refcount);
1315                                 goto rule_found;
1316                         }
1317                 }
1318
1319                 *new_rule = true;
1320                 rule = alloc_rule(dest + i);
1321                 if (!rule)
1322                         goto free_rules;
1323
1324                 /* Add dest to dests list- we need flow tables to be in the
1325                  * end of the list for forward to next prio rules.
1326                  */
1327                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1328                 if (dest &&
1329                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1330                         list_add(&rule->node.list, &fte->node.children);
1331                 else
1332                         list_add_tail(&rule->node.list, &fte->node.children);
1333                 if (dest) {
1334                         fte->dests_size++;
1335
1336                         type = dest[i].type ==
1337                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1338                         *modify_mask |= type ? count : dst;
1339                 }
1340 rule_found:
1341                 handle->rule[i] = rule;
1342         } while (++i < dest_num);
1343
1344         return handle;
1345
1346 free_rules:
1347         destroy_flow_handle(fte, handle, dest, i);
1348         return ERR_PTR(-ENOMEM);
1349 }
1350
1351 /* fte should not be deleted while calling this function */
1352 static struct mlx5_flow_handle *
1353 add_rule_fte(struct fs_fte *fte,
1354              struct mlx5_flow_group *fg,
1355              struct mlx5_flow_destination *dest,
1356              int dest_num,
1357              bool update_action)
1358 {
1359         struct mlx5_flow_root_namespace *root;
1360         struct mlx5_flow_handle *handle;
1361         struct mlx5_flow_table *ft;
1362         int modify_mask = 0;
1363         int err;
1364         bool new_rule = false;
1365
1366         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1367                                     &new_rule);
1368         if (IS_ERR(handle) || !new_rule)
1369                 goto out;
1370
1371         if (update_action)
1372                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1373
1374         fs_get_obj(ft, fg->node.parent);
1375         root = find_root(&fg->node);
1376         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1377                 err = root->cmds->create_fte(root, ft, fg, fte);
1378         else
1379                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1380         if (err)
1381                 goto free_handle;
1382
1383         fte->node.active = true;
1384         fte->status |= FS_FTE_STATUS_EXISTING;
1385         atomic_inc(&fg->node.version);
1386
1387 out:
1388         return handle;
1389
1390 free_handle:
1391         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1392         return ERR_PTR(err);
1393 }
1394
1395 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1396                                                      const struct mlx5_flow_spec *spec)
1397 {
1398         struct list_head *prev = &ft->node.children;
1399         u32 max_fte = ft->autogroup.max_fte;
1400         unsigned int candidate_index = 0;
1401         unsigned int group_size = 0;
1402         struct mlx5_flow_group *fg;
1403
1404         if (!ft->autogroup.active)
1405                 return ERR_PTR(-ENOENT);
1406
1407         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1408                 group_size = ft->autogroup.group_size;
1409
1410         /*  max_fte == ft->autogroup.max_types */
1411         if (group_size == 0)
1412                 group_size = 1;
1413
1414         /* sorted by start_index */
1415         fs_for_each_fg(fg, ft) {
1416                 if (candidate_index + group_size > fg->start_index)
1417                         candidate_index = fg->start_index + fg->max_ftes;
1418                 else
1419                         break;
1420                 prev = &fg->node.list;
1421         }
1422
1423         if (candidate_index + group_size > max_fte)
1424                 return ERR_PTR(-ENOSPC);
1425
1426         fg = alloc_insert_flow_group(ft,
1427                                      spec->match_criteria_enable,
1428                                      spec->match_criteria,
1429                                      candidate_index,
1430                                      candidate_index + group_size - 1,
1431                                      prev);
1432         if (IS_ERR(fg))
1433                 goto out;
1434
1435         if (group_size == ft->autogroup.group_size)
1436                 ft->autogroup.num_groups++;
1437
1438 out:
1439         return fg;
1440 }
1441
1442 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1443                                   struct mlx5_flow_group *fg)
1444 {
1445         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1446         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1447         void *match_criteria_addr;
1448         u8 src_esw_owner_mask_on;
1449         void *misc;
1450         int err;
1451         u32 *in;
1452
1453         in = kvzalloc(inlen, GFP_KERNEL);
1454         if (!in)
1455                 return -ENOMEM;
1456
1457         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1458                  fg->mask.match_criteria_enable);
1459         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1460         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1461                  fg->max_ftes - 1);
1462
1463         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1464                             misc_parameters);
1465         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1466                                          source_eswitch_owner_vhca_id);
1467         MLX5_SET(create_flow_group_in, in,
1468                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1469
1470         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1471                                            in, match_criteria);
1472         memcpy(match_criteria_addr, fg->mask.match_criteria,
1473                sizeof(fg->mask.match_criteria));
1474
1475         err = root->cmds->create_flow_group(root, ft, in, fg);
1476         if (!err) {
1477                 fg->node.active = true;
1478                 trace_mlx5_fs_add_fg(fg);
1479         }
1480
1481         kvfree(in);
1482         return err;
1483 }
1484
1485 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1486                                 struct mlx5_flow_destination *d2)
1487 {
1488         if (d1->type == d2->type) {
1489                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1490                      d1->vport.num == d2->vport.num &&
1491                      d1->vport.flags == d2->vport.flags &&
1492                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1493                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1494                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1495                       (d1->vport.pkt_reformat->id ==
1496                        d2->vport.pkt_reformat->id) : true)) ||
1497                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1498                      d1->ft == d2->ft) ||
1499                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1500                      d1->tir_num == d2->tir_num) ||
1501                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1502                      d1->ft_num == d2->ft_num))
1503                         return true;
1504         }
1505
1506         return false;
1507 }
1508
1509 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1510                                              struct mlx5_flow_destination *dest)
1511 {
1512         struct mlx5_flow_rule *rule;
1513
1514         list_for_each_entry(rule, &fte->node.children, node.list) {
1515                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1516                         return rule;
1517         }
1518         return NULL;
1519 }
1520
1521 static bool check_conflicting_actions(u32 action1, u32 action2)
1522 {
1523         u32 xored_actions = action1 ^ action2;
1524
1525         /* if one rule only wants to count, it's ok */
1526         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1527             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1528                 return false;
1529
1530         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1531                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1532                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1533                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1534                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1535                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1536                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1537                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1538                 return true;
1539
1540         return false;
1541 }
1542
1543 static int check_conflicting_ftes(struct fs_fte *fte,
1544                                   const struct mlx5_flow_context *flow_context,
1545                                   const struct mlx5_flow_act *flow_act)
1546 {
1547         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1548                 mlx5_core_warn(get_dev(&fte->node),
1549                                "Found two FTEs with conflicting actions\n");
1550                 return -EEXIST;
1551         }
1552
1553         if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1554             fte->flow_context.flow_tag != flow_context->flow_tag) {
1555                 mlx5_core_warn(get_dev(&fte->node),
1556                                "FTE flow tag %u already exists with different flow tag %u\n",
1557                                fte->flow_context.flow_tag,
1558                                flow_context->flow_tag);
1559                 return -EEXIST;
1560         }
1561
1562         return 0;
1563 }
1564
1565 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1566                                             const struct mlx5_flow_spec *spec,
1567                                             struct mlx5_flow_act *flow_act,
1568                                             struct mlx5_flow_destination *dest,
1569                                             int dest_num,
1570                                             struct fs_fte *fte)
1571 {
1572         struct mlx5_flow_handle *handle;
1573         int old_action;
1574         int i;
1575         int ret;
1576
1577         ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1578         if (ret)
1579                 return ERR_PTR(ret);
1580
1581         old_action = fte->action.action;
1582         fte->action.action |= flow_act->action;
1583         handle = add_rule_fte(fte, fg, dest, dest_num,
1584                               old_action != flow_act->action);
1585         if (IS_ERR(handle)) {
1586                 fte->action.action = old_action;
1587                 return handle;
1588         }
1589         trace_mlx5_fs_set_fte(fte, false);
1590
1591         for (i = 0; i < handle->num_rules; i++) {
1592                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1593                         tree_add_node(&handle->rule[i]->node, &fte->node);
1594                         trace_mlx5_fs_add_rule(handle->rule[i]);
1595                 }
1596         }
1597         return handle;
1598 }
1599
1600 static bool counter_is_valid(u32 action)
1601 {
1602         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1603                           MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1604                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1605 }
1606
1607 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1608                           struct mlx5_flow_act *flow_act,
1609                           struct mlx5_flow_table *ft)
1610 {
1611         bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1612         u32 action = flow_act->action;
1613
1614         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1615                 return counter_is_valid(action);
1616
1617         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1618                 return true;
1619
1620         if (ignore_level) {
1621                 if (ft->type != FS_FT_FDB &&
1622                     ft->type != FS_FT_NIC_RX)
1623                         return false;
1624
1625                 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1626                     ft->type != dest->ft->type)
1627                         return false;
1628         }
1629
1630         if (!dest || ((dest->type ==
1631             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1632             (dest->ft->level <= ft->level && !ignore_level)))
1633                 return false;
1634         return true;
1635 }
1636
1637 struct match_list {
1638         struct list_head        list;
1639         struct mlx5_flow_group *g;
1640 };
1641
1642 static void free_match_list(struct match_list *head, bool ft_locked)
1643 {
1644         struct match_list *iter, *match_tmp;
1645
1646         list_for_each_entry_safe(iter, match_tmp, &head->list,
1647                                  list) {
1648                 tree_put_node(&iter->g->node, ft_locked);
1649                 list_del(&iter->list);
1650                 kfree(iter);
1651         }
1652 }
1653
1654 static int build_match_list(struct match_list *match_head,
1655                             struct mlx5_flow_table *ft,
1656                             const struct mlx5_flow_spec *spec,
1657                             bool ft_locked)
1658 {
1659         struct rhlist_head *tmp, *list;
1660         struct mlx5_flow_group *g;
1661         int err = 0;
1662
1663         rcu_read_lock();
1664         INIT_LIST_HEAD(&match_head->list);
1665         /* Collect all fgs which has a matching match_criteria */
1666         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1667         /* RCU is atomic, we can't execute FW commands here */
1668         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1669                 struct match_list *curr_match;
1670
1671                 if (unlikely(!tree_get_node(&g->node)))
1672                         continue;
1673
1674                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1675                 if (!curr_match) {
1676                         free_match_list(match_head, ft_locked);
1677                         err = -ENOMEM;
1678                         goto out;
1679                 }
1680                 curr_match->g = g;
1681                 list_add_tail(&curr_match->list, &match_head->list);
1682         }
1683 out:
1684         rcu_read_unlock();
1685         return err;
1686 }
1687
1688 static u64 matched_fgs_get_version(struct list_head *match_head)
1689 {
1690         struct match_list *iter;
1691         u64 version = 0;
1692
1693         list_for_each_entry(iter, match_head, list)
1694                 version += (u64)atomic_read(&iter->g->node.version);
1695         return version;
1696 }
1697
1698 static struct fs_fte *
1699 lookup_fte_locked(struct mlx5_flow_group *g,
1700                   const u32 *match_value,
1701                   bool take_write)
1702 {
1703         struct fs_fte *fte_tmp;
1704
1705         if (take_write)
1706                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1707         else
1708                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1709         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1710                                          rhash_fte);
1711         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1712                 fte_tmp = NULL;
1713                 goto out;
1714         }
1715         if (!fte_tmp->node.active) {
1716                 tree_put_node(&fte_tmp->node, false);
1717                 fte_tmp = NULL;
1718                 goto out;
1719         }
1720
1721         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1722 out:
1723         if (take_write)
1724                 up_write_ref_node(&g->node, false);
1725         else
1726                 up_read_ref_node(&g->node);
1727         return fte_tmp;
1728 }
1729
1730 static struct mlx5_flow_handle *
1731 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1732                        struct list_head *match_head,
1733                        const struct mlx5_flow_spec *spec,
1734                        struct mlx5_flow_act *flow_act,
1735                        struct mlx5_flow_destination *dest,
1736                        int dest_num,
1737                        int ft_version)
1738 {
1739         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1740         struct mlx5_flow_group *g;
1741         struct mlx5_flow_handle *rule;
1742         struct match_list *iter;
1743         bool take_write = false;
1744         struct fs_fte *fte;
1745         u64  version = 0;
1746         int err;
1747
1748         fte = alloc_fte(ft, spec, flow_act);
1749         if (IS_ERR(fte))
1750                 return  ERR_PTR(-ENOMEM);
1751
1752 search_again_locked:
1753         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1754                 goto skip_search;
1755         version = matched_fgs_get_version(match_head);
1756         /* Try to find an fte with identical match value and attempt update its
1757          * action.
1758          */
1759         list_for_each_entry(iter, match_head, list) {
1760                 struct fs_fte *fte_tmp;
1761
1762                 g = iter->g;
1763                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1764                 if (!fte_tmp)
1765                         continue;
1766                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1767                 /* No error check needed here, because insert_fte() is not called */
1768                 up_write_ref_node(&fte_tmp->node, false);
1769                 tree_put_node(&fte_tmp->node, false);
1770                 kmem_cache_free(steering->ftes_cache, fte);
1771                 return rule;
1772         }
1773
1774 skip_search:
1775         /* No group with matching fte found, or we skipped the search.
1776          * Try to add a new fte to any matching fg.
1777          */
1778
1779         /* Check the ft version, for case that new flow group
1780          * was added while the fgs weren't locked
1781          */
1782         if (atomic_read(&ft->node.version) != ft_version) {
1783                 rule = ERR_PTR(-EAGAIN);
1784                 goto out;
1785         }
1786
1787         /* Check the fgs version. If version have changed it could be that an
1788          * FTE with the same match value was added while the fgs weren't
1789          * locked.
1790          */
1791         if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1792             version != matched_fgs_get_version(match_head)) {
1793                 take_write = true;
1794                 goto search_again_locked;
1795         }
1796
1797         list_for_each_entry(iter, match_head, list) {
1798                 g = iter->g;
1799
1800                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1801
1802                 if (!g->node.active) {
1803                         up_write_ref_node(&g->node, false);
1804                         continue;
1805                 }
1806
1807                 err = insert_fte(g, fte);
1808                 if (err) {
1809                         up_write_ref_node(&g->node, false);
1810                         if (err == -ENOSPC)
1811                                 continue;
1812                         kmem_cache_free(steering->ftes_cache, fte);
1813                         return ERR_PTR(err);
1814                 }
1815
1816                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1817                 up_write_ref_node(&g->node, false);
1818                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1819                 up_write_ref_node(&fte->node, false);
1820                 if (IS_ERR(rule))
1821                         tree_put_node(&fte->node, false);
1822                 return rule;
1823         }
1824         rule = ERR_PTR(-ENOENT);
1825 out:
1826         kmem_cache_free(steering->ftes_cache, fte);
1827         return rule;
1828 }
1829
1830 static struct mlx5_flow_handle *
1831 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1832                      const struct mlx5_flow_spec *spec,
1833                      struct mlx5_flow_act *flow_act,
1834                      struct mlx5_flow_destination *dest,
1835                      int dest_num)
1836
1837 {
1838         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1839         struct mlx5_flow_handle *rule;
1840         struct match_list match_head;
1841         struct mlx5_flow_group *g;
1842         bool take_write = false;
1843         struct fs_fte *fte;
1844         int version;
1845         int err;
1846         int i;
1847
1848         if (!check_valid_spec(spec))
1849                 return ERR_PTR(-EINVAL);
1850
1851         for (i = 0; i < dest_num; i++) {
1852                 if (!dest_is_valid(&dest[i], flow_act, ft))
1853                         return ERR_PTR(-EINVAL);
1854         }
1855         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1856 search_again_locked:
1857         version = atomic_read(&ft->node.version);
1858
1859         /* Collect all fgs which has a matching match_criteria */
1860         err = build_match_list(&match_head, ft, spec, take_write);
1861         if (err) {
1862                 if (take_write)
1863                         up_write_ref_node(&ft->node, false);
1864                 else
1865                         up_read_ref_node(&ft->node);
1866                 return ERR_PTR(err);
1867         }
1868
1869         if (!take_write)
1870                 up_read_ref_node(&ft->node);
1871
1872         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1873                                       dest_num, version);
1874         free_match_list(&match_head, take_write);
1875         if (!IS_ERR(rule) ||
1876             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1877                 if (take_write)
1878                         up_write_ref_node(&ft->node, false);
1879                 return rule;
1880         }
1881
1882         if (!take_write) {
1883                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1884                 take_write = true;
1885         }
1886
1887         if (PTR_ERR(rule) == -EAGAIN ||
1888             version != atomic_read(&ft->node.version))
1889                 goto search_again_locked;
1890
1891         g = alloc_auto_flow_group(ft, spec);
1892         if (IS_ERR(g)) {
1893                 rule = ERR_CAST(g);
1894                 up_write_ref_node(&ft->node, false);
1895                 return rule;
1896         }
1897
1898         fte = alloc_fte(ft, spec, flow_act);
1899         if (IS_ERR(fte)) {
1900                 up_write_ref_node(&ft->node, false);
1901                 err = PTR_ERR(fte);
1902                 goto err_alloc_fte;
1903         }
1904
1905         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1906         up_write_ref_node(&ft->node, false);
1907
1908         err = create_auto_flow_group(ft, g);
1909         if (err)
1910                 goto err_release_fg;
1911
1912         err = insert_fte(g, fte);
1913         if (err)
1914                 goto err_release_fg;
1915
1916         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1917         up_write_ref_node(&g->node, false);
1918         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1919         up_write_ref_node(&fte->node, false);
1920         if (IS_ERR(rule))
1921                 tree_put_node(&fte->node, false);
1922         tree_put_node(&g->node, false);
1923         return rule;
1924
1925 err_release_fg:
1926         up_write_ref_node(&g->node, false);
1927         kmem_cache_free(steering->ftes_cache, fte);
1928 err_alloc_fte:
1929         tree_put_node(&g->node, false);
1930         return ERR_PTR(err);
1931 }
1932
1933 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1934 {
1935         return ((ft->type == FS_FT_NIC_RX) &&
1936                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1937 }
1938
1939 struct mlx5_flow_handle *
1940 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1941                     const struct mlx5_flow_spec *spec,
1942                     struct mlx5_flow_act *flow_act,
1943                     struct mlx5_flow_destination *dest,
1944                     int num_dest)
1945 {
1946         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1947         static const struct mlx5_flow_spec zero_spec = {};
1948         struct mlx5_flow_destination *gen_dest = NULL;
1949         struct mlx5_flow_table *next_ft = NULL;
1950         struct mlx5_flow_handle *handle = NULL;
1951         u32 sw_action = flow_act->action;
1952         int i;
1953
1954         if (!spec)
1955                 spec = &zero_spec;
1956
1957         if (!is_fwd_next_action(sw_action))
1958                 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1959
1960         if (!fwd_next_prio_supported(ft))
1961                 return ERR_PTR(-EOPNOTSUPP);
1962
1963         mutex_lock(&root->chain_lock);
1964         next_ft = find_next_fwd_ft(ft, flow_act);
1965         if (!next_ft) {
1966                 handle = ERR_PTR(-EOPNOTSUPP);
1967                 goto unlock;
1968         }
1969
1970         gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
1971                            GFP_KERNEL);
1972         if (!gen_dest) {
1973                 handle = ERR_PTR(-ENOMEM);
1974                 goto unlock;
1975         }
1976         for (i = 0; i < num_dest; i++)
1977                 gen_dest[i] = dest[i];
1978         gen_dest[i].type =
1979                 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1980         gen_dest[i].ft = next_ft;
1981         dest = gen_dest;
1982         num_dest++;
1983         flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
1984                               MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
1985         flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1986         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1987         if (IS_ERR(handle))
1988                 goto unlock;
1989
1990         if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
1991                 mutex_lock(&next_ft->lock);
1992                 list_add(&handle->rule[num_dest - 1]->next_ft,
1993                          &next_ft->fwd_rules);
1994                 mutex_unlock(&next_ft->lock);
1995                 handle->rule[num_dest - 1]->sw_action = sw_action;
1996                 handle->rule[num_dest - 1]->ft = ft;
1997         }
1998 unlock:
1999         mutex_unlock(&root->chain_lock);
2000         kfree(gen_dest);
2001         return handle;
2002 }
2003 EXPORT_SYMBOL(mlx5_add_flow_rules);
2004
2005 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2006 {
2007         struct fs_fte *fte;
2008         int i;
2009
2010         /* In order to consolidate the HW changes we lock the FTE for other
2011          * changes, and increase its refcount, in order not to perform the
2012          * "del" functions of the FTE. Will handle them here.
2013          * The removal of the rules is done under locked FTE.
2014          * After removing all the handle's rules, if there are remaining
2015          * rules, it means we just need to modify the FTE in FW, and
2016          * unlock/decrease the refcount we increased before.
2017          * Otherwise, it means the FTE should be deleted. First delete the
2018          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2019          * the FTE, which will handle the last decrease of the refcount, as
2020          * well as required handling of its parent.
2021          */
2022         fs_get_obj(fte, handle->rule[0]->node.parent);
2023         down_write_ref_node(&fte->node, false);
2024         for (i = handle->num_rules - 1; i >= 0; i--)
2025                 tree_remove_node(&handle->rule[i]->node, true);
2026         if (fte->dests_size) {
2027                 if (fte->modify_mask)
2028                         modify_fte(fte);
2029                 up_write_ref_node(&fte->node, false);
2030         } else if (list_empty(&fte->node.children)) {
2031                 del_hw_fte(&fte->node);
2032                 /* Avoid double call to del_hw_fte */
2033                 fte->node.del_hw_func = NULL;
2034                 up_write_ref_node(&fte->node, false);
2035                 tree_put_node(&fte->node, false);
2036         }
2037         kfree(handle);
2038 }
2039 EXPORT_SYMBOL(mlx5_del_flow_rules);
2040
2041 /* Assuming prio->node.children(flow tables) is sorted by level */
2042 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2043 {
2044         struct fs_prio *prio;
2045
2046         fs_get_obj(prio, ft->node.parent);
2047
2048         if (!list_is_last(&ft->node.list, &prio->node.children))
2049                 return list_next_entry(ft, node.list);
2050         return find_next_chained_ft(prio);
2051 }
2052
2053 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2054 {
2055         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2056         struct mlx5_ft_underlay_qp *uqp;
2057         struct mlx5_flow_table *new_root_ft = NULL;
2058         int err = 0;
2059         u32 qpn;
2060
2061         if (root->root_ft != ft)
2062                 return 0;
2063
2064         new_root_ft = find_next_ft(ft);
2065         if (!new_root_ft) {
2066                 root->root_ft = NULL;
2067                 return 0;
2068         }
2069
2070         if (list_empty(&root->underlay_qpns)) {
2071                 /* Don't set any QPN (zero) in case QPN list is empty */
2072                 qpn = 0;
2073                 err = root->cmds->update_root_ft(root, new_root_ft,
2074                                                  qpn, false);
2075         } else {
2076                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2077                         qpn = uqp->qpn;
2078                         err = root->cmds->update_root_ft(root,
2079                                                          new_root_ft, qpn,
2080                                                          false);
2081                         if (err)
2082                                 break;
2083                 }
2084         }
2085
2086         if (err)
2087                 mlx5_core_warn(root->dev,
2088                                "Update root flow table of id(%u) qpn(%d) failed\n",
2089                                ft->id, qpn);
2090         else
2091                 root->root_ft = new_root_ft;
2092
2093         return 0;
2094 }
2095
2096 /* Connect flow table from previous priority to
2097  * the next flow table.
2098  */
2099 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2100 {
2101         struct mlx5_core_dev *dev = get_dev(&ft->node);
2102         struct mlx5_flow_table *next_ft;
2103         struct fs_prio *prio;
2104         int err = 0;
2105
2106         err = update_root_ft_destroy(ft);
2107         if (err)
2108                 return err;
2109
2110         fs_get_obj(prio, ft->node.parent);
2111         if  (!(list_first_entry(&prio->node.children,
2112                                 struct mlx5_flow_table,
2113                                 node.list) == ft))
2114                 return 0;
2115
2116         next_ft = find_next_chained_ft(prio);
2117         err = connect_fwd_rules(dev, next_ft, ft);
2118         if (err)
2119                 return err;
2120
2121         err = connect_prev_fts(dev, next_ft, prio);
2122         if (err)
2123                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2124                                ft->id);
2125         return err;
2126 }
2127
2128 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2129 {
2130         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2131         int err = 0;
2132
2133         mutex_lock(&root->chain_lock);
2134         if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2135                 err = disconnect_flow_table(ft);
2136         if (err) {
2137                 mutex_unlock(&root->chain_lock);
2138                 return err;
2139         }
2140         if (tree_remove_node(&ft->node, false))
2141                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2142                                ft->id);
2143         mutex_unlock(&root->chain_lock);
2144
2145         return err;
2146 }
2147 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2148
2149 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2150 {
2151         if (tree_remove_node(&fg->node, false))
2152                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2153                                fg->id);
2154 }
2155 EXPORT_SYMBOL(mlx5_destroy_flow_group);
2156
2157 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2158                                                 int n)
2159 {
2160         struct mlx5_flow_steering *steering = dev->priv.steering;
2161
2162         if (!steering || !steering->fdb_sub_ns)
2163                 return NULL;
2164
2165         return steering->fdb_sub_ns[n];
2166 }
2167 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2168
2169 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2170                                                     enum mlx5_flow_namespace_type type)
2171 {
2172         struct mlx5_flow_steering *steering = dev->priv.steering;
2173         struct mlx5_flow_root_namespace *root_ns;
2174         int prio = 0;
2175         struct fs_prio *fs_prio;
2176         struct mlx5_flow_namespace *ns;
2177
2178         if (!steering)
2179                 return NULL;
2180
2181         switch (type) {
2182         case MLX5_FLOW_NAMESPACE_FDB:
2183                 if (steering->fdb_root_ns)
2184                         return &steering->fdb_root_ns->ns;
2185                 return NULL;
2186         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2187                 if (steering->sniffer_rx_root_ns)
2188                         return &steering->sniffer_rx_root_ns->ns;
2189                 return NULL;
2190         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2191                 if (steering->sniffer_tx_root_ns)
2192                         return &steering->sniffer_tx_root_ns->ns;
2193                 return NULL;
2194         default:
2195                 break;
2196         }
2197
2198         if (type == MLX5_FLOW_NAMESPACE_EGRESS ||
2199             type == MLX5_FLOW_NAMESPACE_EGRESS_KERNEL) {
2200                 root_ns = steering->egress_root_ns;
2201                 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2202         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX) {
2203                 root_ns = steering->rdma_rx_root_ns;
2204                 prio = RDMA_RX_BYPASS_PRIO;
2205         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL) {
2206                 root_ns = steering->rdma_rx_root_ns;
2207                 prio = RDMA_RX_KERNEL_PRIO;
2208         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_TX) {
2209                 root_ns = steering->rdma_tx_root_ns;
2210         } else { /* Must be NIC RX */
2211                 root_ns = steering->root_ns;
2212                 prio = type;
2213         }
2214
2215         if (!root_ns)
2216                 return NULL;
2217
2218         fs_prio = find_prio(&root_ns->ns, prio);
2219         if (!fs_prio)
2220                 return NULL;
2221
2222         ns = list_first_entry(&fs_prio->node.children,
2223                               typeof(*ns),
2224                               node.list);
2225
2226         return ns;
2227 }
2228 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2229
2230 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2231                                                               enum mlx5_flow_namespace_type type,
2232                                                               int vport)
2233 {
2234         struct mlx5_flow_steering *steering = dev->priv.steering;
2235
2236         if (!steering)
2237                 return NULL;
2238
2239         switch (type) {
2240         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2241                 if (vport >= steering->esw_egress_acl_vports)
2242                         return NULL;
2243                 if (steering->esw_egress_root_ns &&
2244                     steering->esw_egress_root_ns[vport])
2245                         return &steering->esw_egress_root_ns[vport]->ns;
2246                 else
2247                         return NULL;
2248         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2249                 if (vport >= steering->esw_ingress_acl_vports)
2250                         return NULL;
2251                 if (steering->esw_ingress_root_ns &&
2252                     steering->esw_ingress_root_ns[vport])
2253                         return &steering->esw_ingress_root_ns[vport]->ns;
2254                 else
2255                         return NULL;
2256         default:
2257                 return NULL;
2258         }
2259 }
2260
2261 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2262                                        unsigned int prio,
2263                                        int num_levels,
2264                                        enum fs_node_type type)
2265 {
2266         struct fs_prio *fs_prio;
2267
2268         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2269         if (!fs_prio)
2270                 return ERR_PTR(-ENOMEM);
2271
2272         fs_prio->node.type = type;
2273         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2274         tree_add_node(&fs_prio->node, &ns->node);
2275         fs_prio->num_levels = num_levels;
2276         fs_prio->prio = prio;
2277         list_add_tail(&fs_prio->node.list, &ns->node.children);
2278
2279         return fs_prio;
2280 }
2281
2282 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2283                                               unsigned int prio,
2284                                               int num_levels)
2285 {
2286         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2287 }
2288
2289 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2290                                       unsigned int prio, int num_levels)
2291 {
2292         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2293 }
2294
2295 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2296                                                      *ns)
2297 {
2298         ns->node.type = FS_TYPE_NAMESPACE;
2299
2300         return ns;
2301 }
2302
2303 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2304                                                        int def_miss_act)
2305 {
2306         struct mlx5_flow_namespace      *ns;
2307
2308         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2309         if (!ns)
2310                 return ERR_PTR(-ENOMEM);
2311
2312         fs_init_namespace(ns);
2313         ns->def_miss_action = def_miss_act;
2314         tree_init_node(&ns->node, NULL, del_sw_ns);
2315         tree_add_node(&ns->node, &prio->node);
2316         list_add_tail(&ns->node.list, &prio->node.children);
2317
2318         return ns;
2319 }
2320
2321 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2322                              struct init_tree_node *prio_metadata)
2323 {
2324         struct fs_prio *fs_prio;
2325         int i;
2326
2327         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2328                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2329                 if (IS_ERR(fs_prio))
2330                         return PTR_ERR(fs_prio);
2331         }
2332         return 0;
2333 }
2334
2335 #define FLOW_TABLE_BIT_SZ 1
2336 #define GET_FLOW_TABLE_CAP(dev, offset) \
2337         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2338                         offset / 32)) >>                                        \
2339           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2340 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2341 {
2342         int i;
2343
2344         for (i = 0; i < caps->arr_sz; i++) {
2345                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2346                         return false;
2347         }
2348         return true;
2349 }
2350
2351 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2352                                     struct init_tree_node *init_node,
2353                                     struct fs_node *fs_parent_node,
2354                                     struct init_tree_node *init_parent_node,
2355                                     int prio)
2356 {
2357         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2358                                               flow_table_properties_nic_receive.
2359                                               max_ft_level);
2360         struct mlx5_flow_namespace *fs_ns;
2361         struct fs_prio *fs_prio;
2362         struct fs_node *base;
2363         int i;
2364         int err;
2365
2366         if (init_node->type == FS_TYPE_PRIO) {
2367                 if ((init_node->min_ft_level > max_ft_level) ||
2368                     !has_required_caps(steering->dev, &init_node->caps))
2369                         return 0;
2370
2371                 fs_get_obj(fs_ns, fs_parent_node);
2372                 if (init_node->num_leaf_prios)
2373                         return create_leaf_prios(fs_ns, prio, init_node);
2374                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2375                 if (IS_ERR(fs_prio))
2376                         return PTR_ERR(fs_prio);
2377                 base = &fs_prio->node;
2378         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2379                 fs_get_obj(fs_prio, fs_parent_node);
2380                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2381                 if (IS_ERR(fs_ns))
2382                         return PTR_ERR(fs_ns);
2383                 base = &fs_ns->node;
2384         } else {
2385                 return -EINVAL;
2386         }
2387         prio = 0;
2388         for (i = 0; i < init_node->ar_size; i++) {
2389                 err = init_root_tree_recursive(steering, &init_node->children[i],
2390                                                base, init_node, prio);
2391                 if (err)
2392                         return err;
2393                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2394                     init_node->children[i].num_leaf_prios) {
2395                         prio += init_node->children[i].num_leaf_prios;
2396                 }
2397         }
2398
2399         return 0;
2400 }
2401
2402 static int init_root_tree(struct mlx5_flow_steering *steering,
2403                           struct init_tree_node *init_node,
2404                           struct fs_node *fs_parent_node)
2405 {
2406         int err;
2407         int i;
2408
2409         for (i = 0; i < init_node->ar_size; i++) {
2410                 err = init_root_tree_recursive(steering, &init_node->children[i],
2411                                                fs_parent_node,
2412                                                init_node, i);
2413                 if (err)
2414                         return err;
2415         }
2416         return 0;
2417 }
2418
2419 static void del_sw_root_ns(struct fs_node *node)
2420 {
2421         struct mlx5_flow_root_namespace *root_ns;
2422         struct mlx5_flow_namespace *ns;
2423
2424         fs_get_obj(ns, node);
2425         root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2426         mutex_destroy(&root_ns->chain_lock);
2427         kfree(node);
2428 }
2429
2430 static struct mlx5_flow_root_namespace
2431 *create_root_ns(struct mlx5_flow_steering *steering,
2432                 enum fs_flow_table_type table_type)
2433 {
2434         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2435         struct mlx5_flow_root_namespace *root_ns;
2436         struct mlx5_flow_namespace *ns;
2437
2438         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2439             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2440                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2441
2442         /* Create the root namespace */
2443         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2444         if (!root_ns)
2445                 return NULL;
2446
2447         root_ns->dev = steering->dev;
2448         root_ns->table_type = table_type;
2449         root_ns->cmds = cmds;
2450
2451         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2452
2453         ns = &root_ns->ns;
2454         fs_init_namespace(ns);
2455         mutex_init(&root_ns->chain_lock);
2456         tree_init_node(&ns->node, NULL, del_sw_root_ns);
2457         tree_add_node(&ns->node, NULL);
2458
2459         return root_ns;
2460 }
2461
2462 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2463
2464 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2465 {
2466         struct fs_prio *prio;
2467
2468         fs_for_each_prio(prio, ns) {
2469                  /* This updates prio start_level and num_levels */
2470                 set_prio_attrs_in_prio(prio, acc_level);
2471                 acc_level += prio->num_levels;
2472         }
2473         return acc_level;
2474 }
2475
2476 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2477 {
2478         struct mlx5_flow_namespace *ns;
2479         int acc_level_ns = acc_level;
2480
2481         prio->start_level = acc_level;
2482         fs_for_each_ns(ns, prio) {
2483                 /* This updates start_level and num_levels of ns's priority descendants */
2484                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2485
2486                 /* If this a prio with chains, and we can jump from one chain
2487                  * (namepsace) to another, so we accumulate the levels
2488                  */
2489                 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2490                         acc_level = acc_level_ns;
2491         }
2492
2493         if (!prio->num_levels)
2494                 prio->num_levels = acc_level_ns - prio->start_level;
2495         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2496 }
2497
2498 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2499 {
2500         struct mlx5_flow_namespace *ns = &root_ns->ns;
2501         struct fs_prio *prio;
2502         int start_level = 0;
2503
2504         fs_for_each_prio(prio, ns) {
2505                 set_prio_attrs_in_prio(prio, start_level);
2506                 start_level += prio->num_levels;
2507         }
2508 }
2509
2510 #define ANCHOR_PRIO 0
2511 #define ANCHOR_SIZE 1
2512 #define ANCHOR_LEVEL 0
2513 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2514 {
2515         struct mlx5_flow_namespace *ns = NULL;
2516         struct mlx5_flow_table_attr ft_attr = {};
2517         struct mlx5_flow_table *ft;
2518
2519         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2520         if (WARN_ON(!ns))
2521                 return -EINVAL;
2522
2523         ft_attr.max_fte = ANCHOR_SIZE;
2524         ft_attr.level   = ANCHOR_LEVEL;
2525         ft_attr.prio    = ANCHOR_PRIO;
2526
2527         ft = mlx5_create_flow_table(ns, &ft_attr);
2528         if (IS_ERR(ft)) {
2529                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2530                 return PTR_ERR(ft);
2531         }
2532         return 0;
2533 }
2534
2535 static int init_root_ns(struct mlx5_flow_steering *steering)
2536 {
2537         int err;
2538
2539         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2540         if (!steering->root_ns)
2541                 return -ENOMEM;
2542
2543         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2544         if (err)
2545                 goto out_err;
2546
2547         set_prio_attrs(steering->root_ns);
2548         err = create_anchor_flow_table(steering);
2549         if (err)
2550                 goto out_err;
2551
2552         return 0;
2553
2554 out_err:
2555         cleanup_root_ns(steering->root_ns);
2556         steering->root_ns = NULL;
2557         return err;
2558 }
2559
2560 static void clean_tree(struct fs_node *node)
2561 {
2562         if (node) {
2563                 struct fs_node *iter;
2564                 struct fs_node *temp;
2565
2566                 tree_get_node(node);
2567                 list_for_each_entry_safe(iter, temp, &node->children, list)
2568                         clean_tree(iter);
2569                 tree_put_node(node, false);
2570                 tree_remove_node(node, false);
2571         }
2572 }
2573
2574 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2575 {
2576         if (!root_ns)
2577                 return;
2578
2579         clean_tree(&root_ns->ns.node);
2580 }
2581
2582 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2583 {
2584         struct mlx5_flow_steering *steering = dev->priv.steering;
2585
2586         cleanup_root_ns(steering->root_ns);
2587         cleanup_root_ns(steering->fdb_root_ns);
2588         steering->fdb_root_ns = NULL;
2589         kfree(steering->fdb_sub_ns);
2590         steering->fdb_sub_ns = NULL;
2591         cleanup_root_ns(steering->sniffer_rx_root_ns);
2592         cleanup_root_ns(steering->sniffer_tx_root_ns);
2593         cleanup_root_ns(steering->rdma_rx_root_ns);
2594         cleanup_root_ns(steering->rdma_tx_root_ns);
2595         cleanup_root_ns(steering->egress_root_ns);
2596         mlx5_cleanup_fc_stats(dev);
2597         kmem_cache_destroy(steering->ftes_cache);
2598         kmem_cache_destroy(steering->fgs_cache);
2599         mlx5_ft_pool_destroy(dev);
2600         kfree(steering);
2601 }
2602
2603 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2604 {
2605         struct fs_prio *prio;
2606
2607         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2608         if (!steering->sniffer_tx_root_ns)
2609                 return -ENOMEM;
2610
2611         /* Create single prio */
2612         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2613         return PTR_ERR_OR_ZERO(prio);
2614 }
2615
2616 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2617 {
2618         struct fs_prio *prio;
2619
2620         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2621         if (!steering->sniffer_rx_root_ns)
2622                 return -ENOMEM;
2623
2624         /* Create single prio */
2625         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2626         return PTR_ERR_OR_ZERO(prio);
2627 }
2628
2629 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2630 {
2631         int err;
2632
2633         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2634         if (!steering->rdma_rx_root_ns)
2635                 return -ENOMEM;
2636
2637         err = init_root_tree(steering, &rdma_rx_root_fs,
2638                              &steering->rdma_rx_root_ns->ns.node);
2639         if (err)
2640                 goto out_err;
2641
2642         set_prio_attrs(steering->rdma_rx_root_ns);
2643
2644         return 0;
2645
2646 out_err:
2647         cleanup_root_ns(steering->rdma_rx_root_ns);
2648         steering->rdma_rx_root_ns = NULL;
2649         return err;
2650 }
2651
2652 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2653 {
2654         int err;
2655
2656         steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2657         if (!steering->rdma_tx_root_ns)
2658                 return -ENOMEM;
2659
2660         err = init_root_tree(steering, &rdma_tx_root_fs,
2661                              &steering->rdma_tx_root_ns->ns.node);
2662         if (err)
2663                 goto out_err;
2664
2665         set_prio_attrs(steering->rdma_tx_root_ns);
2666
2667         return 0;
2668
2669 out_err:
2670         cleanup_root_ns(steering->rdma_tx_root_ns);
2671         steering->rdma_tx_root_ns = NULL;
2672         return err;
2673 }
2674
2675 /* FT and tc chains are stored in the same array so we can re-use the
2676  * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2677  * When creating a new ns for each chain store it in the first available slot.
2678  * Assume tc chains are created and stored first and only then the FT chain.
2679  */
2680 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2681                                         struct mlx5_flow_namespace *ns)
2682 {
2683         int chain = 0;
2684
2685         while (steering->fdb_sub_ns[chain])
2686                 ++chain;
2687
2688         steering->fdb_sub_ns[chain] = ns;
2689 }
2690
2691 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2692                                         struct fs_prio *maj_prio)
2693 {
2694         struct mlx5_flow_namespace *ns;
2695         struct fs_prio *min_prio;
2696         int prio;
2697
2698         ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2699         if (IS_ERR(ns))
2700                 return PTR_ERR(ns);
2701
2702         for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2703                 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2704                 if (IS_ERR(min_prio))
2705                         return PTR_ERR(min_prio);
2706         }
2707
2708         store_fdb_sub_ns_prio_chain(steering, ns);
2709
2710         return 0;
2711 }
2712
2713 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2714                              int fs_prio,
2715                              int chains)
2716 {
2717         struct fs_prio *maj_prio;
2718         int levels;
2719         int chain;
2720         int err;
2721
2722         levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2723         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2724                                           fs_prio,
2725                                           levels);
2726         if (IS_ERR(maj_prio))
2727                 return PTR_ERR(maj_prio);
2728
2729         for (chain = 0; chain < chains; chain++) {
2730                 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2731                 if (err)
2732                         return err;
2733         }
2734
2735         return 0;
2736 }
2737
2738 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2739 {
2740         int err;
2741
2742         steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2743                                        sizeof(*steering->fdb_sub_ns),
2744                                        GFP_KERNEL);
2745         if (!steering->fdb_sub_ns)
2746                 return -ENOMEM;
2747
2748         err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
2749         if (err)
2750                 return err;
2751
2752         err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
2753         if (err)
2754                 return err;
2755
2756         return 0;
2757 }
2758
2759 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2760 {
2761         struct fs_prio *maj_prio;
2762         int err;
2763
2764         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2765         if (!steering->fdb_root_ns)
2766                 return -ENOMEM;
2767
2768         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2769                                   1);
2770         if (IS_ERR(maj_prio)) {
2771                 err = PTR_ERR(maj_prio);
2772                 goto out_err;
2773         }
2774         err = create_fdb_fast_path(steering);
2775         if (err)
2776                 goto out_err;
2777
2778         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2779         if (IS_ERR(maj_prio)) {
2780                 err = PTR_ERR(maj_prio);
2781                 goto out_err;
2782         }
2783
2784         /* We put this priority last, knowing that nothing will get here
2785          * unless explicitly forwarded to. This is possible because the
2786          * slow path tables have catch all rules and nothing gets passed
2787          * those tables.
2788          */
2789         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
2790         if (IS_ERR(maj_prio)) {
2791                 err = PTR_ERR(maj_prio);
2792                 goto out_err;
2793         }
2794
2795         set_prio_attrs(steering->fdb_root_ns);
2796         return 0;
2797
2798 out_err:
2799         cleanup_root_ns(steering->fdb_root_ns);
2800         kfree(steering->fdb_sub_ns);
2801         steering->fdb_sub_ns = NULL;
2802         steering->fdb_root_ns = NULL;
2803         return err;
2804 }
2805
2806 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2807 {
2808         struct fs_prio *prio;
2809
2810         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2811         if (!steering->esw_egress_root_ns[vport])
2812                 return -ENOMEM;
2813
2814         /* create 1 prio*/
2815         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2816         return PTR_ERR_OR_ZERO(prio);
2817 }
2818
2819 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2820 {
2821         struct fs_prio *prio;
2822
2823         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2824         if (!steering->esw_ingress_root_ns[vport])
2825                 return -ENOMEM;
2826
2827         /* create 1 prio*/
2828         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2829         return PTR_ERR_OR_ZERO(prio);
2830 }
2831
2832 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2833 {
2834         struct mlx5_flow_steering *steering = dev->priv.steering;
2835         int err;
2836         int i;
2837
2838         steering->esw_egress_root_ns =
2839                         kcalloc(total_vports,
2840                                 sizeof(*steering->esw_egress_root_ns),
2841                                 GFP_KERNEL);
2842         if (!steering->esw_egress_root_ns)
2843                 return -ENOMEM;
2844
2845         for (i = 0; i < total_vports; i++) {
2846                 err = init_egress_acl_root_ns(steering, i);
2847                 if (err)
2848                         goto cleanup_root_ns;
2849         }
2850         steering->esw_egress_acl_vports = total_vports;
2851         return 0;
2852
2853 cleanup_root_ns:
2854         for (i--; i >= 0; i--)
2855                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2856         kfree(steering->esw_egress_root_ns);
2857         steering->esw_egress_root_ns = NULL;
2858         return err;
2859 }
2860
2861 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev)
2862 {
2863         struct mlx5_flow_steering *steering = dev->priv.steering;
2864         int i;
2865
2866         if (!steering->esw_egress_root_ns)
2867                 return;
2868
2869         for (i = 0; i < steering->esw_egress_acl_vports; i++)
2870                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2871
2872         kfree(steering->esw_egress_root_ns);
2873         steering->esw_egress_root_ns = NULL;
2874 }
2875
2876 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2877 {
2878         struct mlx5_flow_steering *steering = dev->priv.steering;
2879         int err;
2880         int i;
2881
2882         steering->esw_ingress_root_ns =
2883                         kcalloc(total_vports,
2884                                 sizeof(*steering->esw_ingress_root_ns),
2885                                 GFP_KERNEL);
2886         if (!steering->esw_ingress_root_ns)
2887                 return -ENOMEM;
2888
2889         for (i = 0; i < total_vports; i++) {
2890                 err = init_ingress_acl_root_ns(steering, i);
2891                 if (err)
2892                         goto cleanup_root_ns;
2893         }
2894         steering->esw_ingress_acl_vports = total_vports;
2895         return 0;
2896
2897 cleanup_root_ns:
2898         for (i--; i >= 0; i--)
2899                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2900         kfree(steering->esw_ingress_root_ns);
2901         steering->esw_ingress_root_ns = NULL;
2902         return err;
2903 }
2904
2905 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev)
2906 {
2907         struct mlx5_flow_steering *steering = dev->priv.steering;
2908         int i;
2909
2910         if (!steering->esw_ingress_root_ns)
2911                 return;
2912
2913         for (i = 0; i < steering->esw_ingress_acl_vports; i++)
2914                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2915
2916         kfree(steering->esw_ingress_root_ns);
2917         steering->esw_ingress_root_ns = NULL;
2918 }
2919
2920 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2921 {
2922         int err;
2923
2924         steering->egress_root_ns = create_root_ns(steering,
2925                                                   FS_FT_NIC_TX);
2926         if (!steering->egress_root_ns)
2927                 return -ENOMEM;
2928
2929         err = init_root_tree(steering, &egress_root_fs,
2930                              &steering->egress_root_ns->ns.node);
2931         if (err)
2932                 goto cleanup;
2933         set_prio_attrs(steering->egress_root_ns);
2934         return 0;
2935 cleanup:
2936         cleanup_root_ns(steering->egress_root_ns);
2937         steering->egress_root_ns = NULL;
2938         return err;
2939 }
2940
2941 int mlx5_init_fs(struct mlx5_core_dev *dev)
2942 {
2943         struct mlx5_flow_steering *steering;
2944         int err = 0;
2945
2946         err = mlx5_init_fc_stats(dev);
2947         if (err)
2948                 return err;
2949
2950         err = mlx5_ft_pool_init(dev);
2951         if (err)
2952                 return err;
2953
2954         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2955         if (!steering)
2956                 goto err;
2957         steering->dev = dev;
2958         dev->priv.steering = steering;
2959
2960         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2961                                                 sizeof(struct mlx5_flow_group), 0,
2962                                                 0, NULL);
2963         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2964                                                  0, NULL);
2965         if (!steering->ftes_cache || !steering->fgs_cache) {
2966                 err = -ENOMEM;
2967                 goto err;
2968         }
2969
2970         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2971               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2972              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2973               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2974             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2975                 err = init_root_ns(steering);
2976                 if (err)
2977                         goto err;
2978         }
2979
2980         if (MLX5_ESWITCH_MANAGER(dev)) {
2981                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2982                         err = init_fdb_root_ns(steering);
2983                         if (err)
2984                                 goto err;
2985                 }
2986         }
2987
2988         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2989                 err = init_sniffer_rx_root_ns(steering);
2990                 if (err)
2991                         goto err;
2992         }
2993
2994         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2995                 err = init_sniffer_tx_root_ns(steering);
2996                 if (err)
2997                         goto err;
2998         }
2999
3000         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
3001             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3002                 err = init_rdma_rx_root_ns(steering);
3003                 if (err)
3004                         goto err;
3005         }
3006
3007         if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3008                 err = init_rdma_tx_root_ns(steering);
3009                 if (err)
3010                         goto err;
3011         }
3012
3013         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE ||
3014             MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3015                 err = init_egress_root_ns(steering);
3016                 if (err)
3017                         goto err;
3018         }
3019
3020         return 0;
3021 err:
3022         mlx5_cleanup_fs(dev);
3023         return err;
3024 }
3025
3026 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3027 {
3028         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3029         struct mlx5_ft_underlay_qp *new_uqp;
3030         int err = 0;
3031
3032         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3033         if (!new_uqp)
3034                 return -ENOMEM;
3035
3036         mutex_lock(&root->chain_lock);
3037
3038         if (!root->root_ft) {
3039                 err = -EINVAL;
3040                 goto update_ft_fail;
3041         }
3042
3043         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3044                                          false);
3045         if (err) {
3046                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3047                                underlay_qpn, err);
3048                 goto update_ft_fail;
3049         }
3050
3051         new_uqp->qpn = underlay_qpn;
3052         list_add_tail(&new_uqp->list, &root->underlay_qpns);
3053
3054         mutex_unlock(&root->chain_lock);
3055
3056         return 0;
3057
3058 update_ft_fail:
3059         mutex_unlock(&root->chain_lock);
3060         kfree(new_uqp);
3061         return err;
3062 }
3063 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3064
3065 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3066 {
3067         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3068         struct mlx5_ft_underlay_qp *uqp;
3069         bool found = false;
3070         int err = 0;
3071
3072         mutex_lock(&root->chain_lock);
3073         list_for_each_entry(uqp, &root->underlay_qpns, list) {
3074                 if (uqp->qpn == underlay_qpn) {
3075                         found = true;
3076                         break;
3077                 }
3078         }
3079
3080         if (!found) {
3081                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3082                                underlay_qpn);
3083                 err = -EINVAL;
3084                 goto out;
3085         }
3086
3087         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3088                                          true);
3089         if (err)
3090                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3091                                underlay_qpn, err);
3092
3093         list_del(&uqp->list);
3094         mutex_unlock(&root->chain_lock);
3095         kfree(uqp);
3096
3097         return 0;
3098
3099 out:
3100         mutex_unlock(&root->chain_lock);
3101         return err;
3102 }
3103 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3104
3105 static struct mlx5_flow_root_namespace
3106 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3107 {
3108         struct mlx5_flow_namespace *ns;
3109
3110         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3111             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3112                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3113         else
3114                 ns = mlx5_get_flow_namespace(dev, ns_type);
3115         if (!ns)
3116                 return NULL;
3117
3118         return find_root(&ns->node);
3119 }
3120
3121 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3122                                                  u8 ns_type, u8 num_actions,
3123                                                  void *modify_actions)
3124 {
3125         struct mlx5_flow_root_namespace *root;
3126         struct mlx5_modify_hdr *modify_hdr;
3127         int err;
3128
3129         root = get_root_namespace(dev, ns_type);
3130         if (!root)
3131                 return ERR_PTR(-EOPNOTSUPP);
3132
3133         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3134         if (!modify_hdr)
3135                 return ERR_PTR(-ENOMEM);
3136
3137         modify_hdr->ns_type = ns_type;
3138         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3139                                               modify_actions, modify_hdr);
3140         if (err) {
3141                 kfree(modify_hdr);
3142                 return ERR_PTR(err);
3143         }
3144
3145         return modify_hdr;
3146 }
3147 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3148
3149 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3150                                 struct mlx5_modify_hdr *modify_hdr)
3151 {
3152         struct mlx5_flow_root_namespace *root;
3153
3154         root = get_root_namespace(dev, modify_hdr->ns_type);
3155         if (WARN_ON(!root))
3156                 return;
3157         root->cmds->modify_header_dealloc(root, modify_hdr);
3158         kfree(modify_hdr);
3159 }
3160 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3161
3162 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3163                                                      int reformat_type,
3164                                                      size_t size,
3165                                                      void *reformat_data,
3166                                                      enum mlx5_flow_namespace_type ns_type)
3167 {
3168         struct mlx5_pkt_reformat *pkt_reformat;
3169         struct mlx5_flow_root_namespace *root;
3170         int err;
3171
3172         root = get_root_namespace(dev, ns_type);
3173         if (!root)
3174                 return ERR_PTR(-EOPNOTSUPP);
3175
3176         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3177         if (!pkt_reformat)
3178                 return ERR_PTR(-ENOMEM);
3179
3180         pkt_reformat->ns_type = ns_type;
3181         pkt_reformat->reformat_type = reformat_type;
3182         err = root->cmds->packet_reformat_alloc(root, reformat_type, size,
3183                                                 reformat_data, ns_type,
3184                                                 pkt_reformat);
3185         if (err) {
3186                 kfree(pkt_reformat);
3187                 return ERR_PTR(err);
3188         }
3189
3190         return pkt_reformat;
3191 }
3192 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3193
3194 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3195                                   struct mlx5_pkt_reformat *pkt_reformat)
3196 {
3197         struct mlx5_flow_root_namespace *root;
3198
3199         root = get_root_namespace(dev, pkt_reformat->ns_type);
3200         if (WARN_ON(!root))
3201                 return;
3202         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3203         kfree(pkt_reformat);
3204 }
3205 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3206
3207 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3208                                  struct mlx5_flow_root_namespace *peer_ns)
3209 {
3210         if (peer_ns && ns->mode != peer_ns->mode) {
3211                 mlx5_core_err(ns->dev,
3212                               "Can't peer namespace of different steering mode\n");
3213                 return -EINVAL;
3214         }
3215
3216         return ns->cmds->set_peer(ns, peer_ns);
3217 }
3218
3219 /* This function should be called only at init stage of the namespace.
3220  * It is not safe to call this function while steering operations
3221  * are executed in the namespace.
3222  */
3223 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3224                                  enum mlx5_flow_steering_mode mode)
3225 {
3226         struct mlx5_flow_root_namespace *root;
3227         const struct mlx5_flow_cmds *cmds;
3228         int err;
3229
3230         root = find_root(&ns->node);
3231         if (&root->ns != ns)
3232         /* Can't set cmds to non root namespace */
3233                 return -EINVAL;
3234
3235         if (root->table_type != FS_FT_FDB)
3236                 return -EOPNOTSUPP;
3237
3238         if (root->mode == mode)
3239                 return 0;
3240
3241         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3242                 cmds = mlx5_fs_cmd_get_dr_cmds();
3243         else
3244                 cmds = mlx5_fs_cmd_get_fw_cmds();
3245         if (!cmds)
3246                 return -EOPNOTSUPP;
3247
3248         err = cmds->create_ns(root);
3249         if (err) {
3250                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3251                               err);
3252                 return err;
3253         }
3254
3255         root->cmds->destroy_ns(root);
3256         root->cmds = cmds;
3257         root->mode = mode;
3258
3259         return 0;
3260 }