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 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, *first_ft;
1028         int err = 0;
1029
1030         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
1031
1032         first_ft = list_first_entry_or_null(&prio->node.children,
1033                                             struct mlx5_flow_table, node.list);
1034         if (!first_ft || first_ft->level > ft->level) {
1035                 err = connect_prev_fts(dev, ft, prio);
1036                 if (err)
1037                         return err;
1038
1039                 next_ft = first_ft ? first_ft : find_next_chained_ft(prio);
1040                 err = connect_fwd_rules(dev, ft, next_ft);
1041                 if (err)
1042                         return err;
1043         }
1044
1045         if (MLX5_CAP_FLOWTABLE(dev,
1046                                flow_table_properties_nic_receive.modify_root))
1047                 err = update_root_ft_create(ft, prio);
1048         return err;
1049 }
1050
1051 static void list_add_flow_table(struct mlx5_flow_table *ft,
1052                                 struct fs_prio *prio)
1053 {
1054         struct list_head *prev = &prio->node.children;
1055         struct mlx5_flow_table *iter;
1056
1057         fs_for_each_ft(iter, prio) {
1058                 if (iter->level > ft->level)
1059                         break;
1060                 prev = &iter->node.list;
1061         }
1062         list_add(&ft->node.list, prev);
1063 }
1064
1065 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1066                                                         struct mlx5_flow_table_attr *ft_attr,
1067                                                         enum fs_flow_table_op_mod op_mod,
1068                                                         u16 vport)
1069 {
1070         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1071         bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED;
1072         struct mlx5_flow_table *next_ft;
1073         struct fs_prio *fs_prio = NULL;
1074         struct mlx5_flow_table *ft;
1075         int err;
1076
1077         if (!root) {
1078                 pr_err("mlx5: flow steering failed to find root of namespace\n");
1079                 return ERR_PTR(-ENODEV);
1080         }
1081
1082         mutex_lock(&root->chain_lock);
1083         fs_prio = find_prio(ns, ft_attr->prio);
1084         if (!fs_prio) {
1085                 err = -EINVAL;
1086                 goto unlock_root;
1087         }
1088         if (!unmanaged) {
1089                 /* The level is related to the
1090                  * priority level range.
1091                  */
1092                 if (ft_attr->level >= fs_prio->num_levels) {
1093                         err = -ENOSPC;
1094                         goto unlock_root;
1095                 }
1096
1097                 ft_attr->level += fs_prio->start_level;
1098         }
1099
1100         /* The level is related to the
1101          * priority level range.
1102          */
1103         ft = alloc_flow_table(ft_attr->level,
1104                               vport,
1105                               root->table_type,
1106                               op_mod, ft_attr->flags);
1107         if (IS_ERR(ft)) {
1108                 err = PTR_ERR(ft);
1109                 goto unlock_root;
1110         }
1111
1112         tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1113         next_ft = unmanaged ? ft_attr->next_ft :
1114                               find_next_chained_ft(fs_prio);
1115         ft->def_miss_action = ns->def_miss_action;
1116         ft->ns = ns;
1117         err = root->cmds->create_flow_table(root, ft, ft_attr->max_fte, next_ft);
1118         if (err)
1119                 goto free_ft;
1120
1121         if (!unmanaged) {
1122                 err = connect_flow_table(root->dev, ft, fs_prio);
1123                 if (err)
1124                         goto destroy_ft;
1125         }
1126
1127         ft->node.active = true;
1128         down_write_ref_node(&fs_prio->node, false);
1129         if (!unmanaged) {
1130                 tree_add_node(&ft->node, &fs_prio->node);
1131                 list_add_flow_table(ft, fs_prio);
1132         } else {
1133                 ft->node.root = fs_prio->node.root;
1134         }
1135         fs_prio->num_ft++;
1136         up_write_ref_node(&fs_prio->node, false);
1137         mutex_unlock(&root->chain_lock);
1138         trace_mlx5_fs_add_ft(ft);
1139         return ft;
1140 destroy_ft:
1141         root->cmds->destroy_flow_table(root, ft);
1142 free_ft:
1143         rhltable_destroy(&ft->fgs_hash);
1144         kfree(ft);
1145 unlock_root:
1146         mutex_unlock(&root->chain_lock);
1147         return ERR_PTR(err);
1148 }
1149
1150 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1151                                                struct mlx5_flow_table_attr *ft_attr)
1152 {
1153         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1154 }
1155 EXPORT_SYMBOL(mlx5_create_flow_table);
1156
1157 struct mlx5_flow_table *
1158 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1159                              struct mlx5_flow_table_attr *ft_attr, u16 vport)
1160 {
1161         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1162 }
1163
1164 struct mlx5_flow_table*
1165 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1166                                  int prio, u32 level)
1167 {
1168         struct mlx5_flow_table_attr ft_attr = {};
1169
1170         ft_attr.level = level;
1171         ft_attr.prio  = prio;
1172         ft_attr.max_fte = 1;
1173
1174         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1175 }
1176 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1177
1178 #define MAX_FLOW_GROUP_SIZE BIT(24)
1179 struct mlx5_flow_table*
1180 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1181                                     struct mlx5_flow_table_attr *ft_attr)
1182 {
1183         int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1184         int max_num_groups = ft_attr->autogroup.max_num_groups;
1185         struct mlx5_flow_table *ft;
1186         int autogroups_max_fte;
1187
1188         ft = mlx5_create_flow_table(ns, ft_attr);
1189         if (IS_ERR(ft))
1190                 return ft;
1191
1192         autogroups_max_fte = ft->max_fte - num_reserved_entries;
1193         if (max_num_groups > autogroups_max_fte)
1194                 goto err_validate;
1195         if (num_reserved_entries > ft->max_fte)
1196                 goto err_validate;
1197
1198         /* Align the number of groups according to the largest group size */
1199         if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE)
1200                 max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1;
1201
1202         ft->autogroup.active = true;
1203         ft->autogroup.required_groups = max_num_groups;
1204         ft->autogroup.max_fte = autogroups_max_fte;
1205         /* We save place for flow groups in addition to max types */
1206         ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1207
1208         return ft;
1209
1210 err_validate:
1211         mlx5_destroy_flow_table(ft);
1212         return ERR_PTR(-ENOSPC);
1213 }
1214 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1215
1216 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1217                                                u32 *fg_in)
1218 {
1219         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1220         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1221                                             fg_in, match_criteria);
1222         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1223                                             fg_in,
1224                                             match_criteria_enable);
1225         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1226                                    start_flow_index);
1227         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1228                                  end_flow_index);
1229         struct mlx5_flow_group *fg;
1230         int err;
1231
1232         if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1233                 return ERR_PTR(-EPERM);
1234
1235         down_write_ref_node(&ft->node, false);
1236         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1237                                      start_index, end_index,
1238                                      ft->node.children.prev);
1239         up_write_ref_node(&ft->node, false);
1240         if (IS_ERR(fg))
1241                 return fg;
1242
1243         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1244         if (err) {
1245                 tree_put_node(&fg->node, false);
1246                 return ERR_PTR(err);
1247         }
1248         trace_mlx5_fs_add_fg(fg);
1249         fg->node.active = true;
1250
1251         return fg;
1252 }
1253 EXPORT_SYMBOL(mlx5_create_flow_group);
1254
1255 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1256 {
1257         struct mlx5_flow_rule *rule;
1258
1259         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1260         if (!rule)
1261                 return NULL;
1262
1263         INIT_LIST_HEAD(&rule->next_ft);
1264         rule->node.type = FS_TYPE_FLOW_DEST;
1265         if (dest)
1266                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1267
1268         return rule;
1269 }
1270
1271 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1272 {
1273         struct mlx5_flow_handle *handle;
1274
1275         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1276         if (!handle)
1277                 return NULL;
1278
1279         handle->num_rules = num_rules;
1280
1281         return handle;
1282 }
1283
1284 static void destroy_flow_handle(struct fs_fte *fte,
1285                                 struct mlx5_flow_handle *handle,
1286                                 struct mlx5_flow_destination *dest,
1287                                 int i)
1288 {
1289         for (; --i >= 0;) {
1290                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1291                         fte->dests_size--;
1292                         list_del(&handle->rule[i]->node.list);
1293                         kfree(handle->rule[i]);
1294                 }
1295         }
1296         kfree(handle);
1297 }
1298
1299 static struct mlx5_flow_handle *
1300 create_flow_handle(struct fs_fte *fte,
1301                    struct mlx5_flow_destination *dest,
1302                    int dest_num,
1303                    int *modify_mask,
1304                    bool *new_rule)
1305 {
1306         struct mlx5_flow_handle *handle;
1307         struct mlx5_flow_rule *rule = NULL;
1308         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1309         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1310         int type;
1311         int i = 0;
1312
1313         handle = alloc_handle((dest_num) ? dest_num : 1);
1314         if (!handle)
1315                 return ERR_PTR(-ENOMEM);
1316
1317         do {
1318                 if (dest) {
1319                         rule = find_flow_rule(fte, dest + i);
1320                         if (rule) {
1321                                 refcount_inc(&rule->node.refcount);
1322                                 goto rule_found;
1323                         }
1324                 }
1325
1326                 *new_rule = true;
1327                 rule = alloc_rule(dest + i);
1328                 if (!rule)
1329                         goto free_rules;
1330
1331                 /* Add dest to dests list- we need flow tables to be in the
1332                  * end of the list for forward to next prio rules.
1333                  */
1334                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1335                 if (dest &&
1336                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1337                         list_add(&rule->node.list, &fte->node.children);
1338                 else
1339                         list_add_tail(&rule->node.list, &fte->node.children);
1340                 if (dest) {
1341                         fte->dests_size++;
1342
1343                         type = dest[i].type ==
1344                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1345                         *modify_mask |= type ? count : dst;
1346                 }
1347 rule_found:
1348                 handle->rule[i] = rule;
1349         } while (++i < dest_num);
1350
1351         return handle;
1352
1353 free_rules:
1354         destroy_flow_handle(fte, handle, dest, i);
1355         return ERR_PTR(-ENOMEM);
1356 }
1357
1358 /* fte should not be deleted while calling this function */
1359 static struct mlx5_flow_handle *
1360 add_rule_fte(struct fs_fte *fte,
1361              struct mlx5_flow_group *fg,
1362              struct mlx5_flow_destination *dest,
1363              int dest_num,
1364              bool update_action)
1365 {
1366         struct mlx5_flow_root_namespace *root;
1367         struct mlx5_flow_handle *handle;
1368         struct mlx5_flow_table *ft;
1369         int modify_mask = 0;
1370         int err;
1371         bool new_rule = false;
1372
1373         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1374                                     &new_rule);
1375         if (IS_ERR(handle) || !new_rule)
1376                 goto out;
1377
1378         if (update_action)
1379                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1380
1381         fs_get_obj(ft, fg->node.parent);
1382         root = find_root(&fg->node);
1383         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1384                 err = root->cmds->create_fte(root, ft, fg, fte);
1385         else
1386                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1387         if (err)
1388                 goto free_handle;
1389
1390         fte->node.active = true;
1391         fte->status |= FS_FTE_STATUS_EXISTING;
1392         atomic_inc(&fg->node.version);
1393
1394 out:
1395         return handle;
1396
1397 free_handle:
1398         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1399         return ERR_PTR(err);
1400 }
1401
1402 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1403                                                      const struct mlx5_flow_spec *spec)
1404 {
1405         struct list_head *prev = &ft->node.children;
1406         u32 max_fte = ft->autogroup.max_fte;
1407         unsigned int candidate_index = 0;
1408         unsigned int group_size = 0;
1409         struct mlx5_flow_group *fg;
1410
1411         if (!ft->autogroup.active)
1412                 return ERR_PTR(-ENOENT);
1413
1414         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1415                 group_size = ft->autogroup.group_size;
1416
1417         /*  max_fte == ft->autogroup.max_types */
1418         if (group_size == 0)
1419                 group_size = 1;
1420
1421         /* sorted by start_index */
1422         fs_for_each_fg(fg, ft) {
1423                 if (candidate_index + group_size > fg->start_index)
1424                         candidate_index = fg->start_index + fg->max_ftes;
1425                 else
1426                         break;
1427                 prev = &fg->node.list;
1428         }
1429
1430         if (candidate_index + group_size > max_fte)
1431                 return ERR_PTR(-ENOSPC);
1432
1433         fg = alloc_insert_flow_group(ft,
1434                                      spec->match_criteria_enable,
1435                                      spec->match_criteria,
1436                                      candidate_index,
1437                                      candidate_index + group_size - 1,
1438                                      prev);
1439         if (IS_ERR(fg))
1440                 goto out;
1441
1442         if (group_size == ft->autogroup.group_size)
1443                 ft->autogroup.num_groups++;
1444
1445 out:
1446         return fg;
1447 }
1448
1449 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1450                                   struct mlx5_flow_group *fg)
1451 {
1452         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1453         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1454         void *match_criteria_addr;
1455         u8 src_esw_owner_mask_on;
1456         void *misc;
1457         int err;
1458         u32 *in;
1459
1460         in = kvzalloc(inlen, GFP_KERNEL);
1461         if (!in)
1462                 return -ENOMEM;
1463
1464         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1465                  fg->mask.match_criteria_enable);
1466         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1467         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1468                  fg->max_ftes - 1);
1469
1470         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1471                             misc_parameters);
1472         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1473                                          source_eswitch_owner_vhca_id);
1474         MLX5_SET(create_flow_group_in, in,
1475                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1476
1477         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1478                                            in, match_criteria);
1479         memcpy(match_criteria_addr, fg->mask.match_criteria,
1480                sizeof(fg->mask.match_criteria));
1481
1482         err = root->cmds->create_flow_group(root, ft, in, fg);
1483         if (!err) {
1484                 fg->node.active = true;
1485                 trace_mlx5_fs_add_fg(fg);
1486         }
1487
1488         kvfree(in);
1489         return err;
1490 }
1491
1492 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1493                                 struct mlx5_flow_destination *d2)
1494 {
1495         if (d1->type == d2->type) {
1496                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1497                      d1->vport.num == d2->vport.num &&
1498                      d1->vport.flags == d2->vport.flags &&
1499                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1500                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1501                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1502                       (d1->vport.pkt_reformat->id ==
1503                        d2->vport.pkt_reformat->id) : true)) ||
1504                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1505                      d1->ft == d2->ft) ||
1506                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1507                      d1->tir_num == d2->tir_num) ||
1508                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1509                      d1->ft_num == d2->ft_num) ||
1510                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER &&
1511                      d1->sampler_id == d2->sampler_id))
1512                         return true;
1513         }
1514
1515         return false;
1516 }
1517
1518 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1519                                              struct mlx5_flow_destination *dest)
1520 {
1521         struct mlx5_flow_rule *rule;
1522
1523         list_for_each_entry(rule, &fte->node.children, node.list) {
1524                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1525                         return rule;
1526         }
1527         return NULL;
1528 }
1529
1530 static bool check_conflicting_actions(u32 action1, u32 action2)
1531 {
1532         u32 xored_actions = action1 ^ action2;
1533
1534         /* if one rule only wants to count, it's ok */
1535         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1536             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1537                 return false;
1538
1539         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1540                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1541                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1542                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1543                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1544                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1545                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1546                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1547                 return true;
1548
1549         return false;
1550 }
1551
1552 static int check_conflicting_ftes(struct fs_fte *fte,
1553                                   const struct mlx5_flow_context *flow_context,
1554                                   const struct mlx5_flow_act *flow_act)
1555 {
1556         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1557                 mlx5_core_warn(get_dev(&fte->node),
1558                                "Found two FTEs with conflicting actions\n");
1559                 return -EEXIST;
1560         }
1561
1562         if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1563             fte->flow_context.flow_tag != flow_context->flow_tag) {
1564                 mlx5_core_warn(get_dev(&fte->node),
1565                                "FTE flow tag %u already exists with different flow tag %u\n",
1566                                fte->flow_context.flow_tag,
1567                                flow_context->flow_tag);
1568                 return -EEXIST;
1569         }
1570
1571         return 0;
1572 }
1573
1574 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1575                                             const struct mlx5_flow_spec *spec,
1576                                             struct mlx5_flow_act *flow_act,
1577                                             struct mlx5_flow_destination *dest,
1578                                             int dest_num,
1579                                             struct fs_fte *fte)
1580 {
1581         struct mlx5_flow_handle *handle;
1582         int old_action;
1583         int i;
1584         int ret;
1585
1586         ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1587         if (ret)
1588                 return ERR_PTR(ret);
1589
1590         old_action = fte->action.action;
1591         fte->action.action |= flow_act->action;
1592         handle = add_rule_fte(fte, fg, dest, dest_num,
1593                               old_action != flow_act->action);
1594         if (IS_ERR(handle)) {
1595                 fte->action.action = old_action;
1596                 return handle;
1597         }
1598         trace_mlx5_fs_set_fte(fte, false);
1599
1600         for (i = 0; i < handle->num_rules; i++) {
1601                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1602                         tree_add_node(&handle->rule[i]->node, &fte->node);
1603                         trace_mlx5_fs_add_rule(handle->rule[i]);
1604                 }
1605         }
1606         return handle;
1607 }
1608
1609 static bool counter_is_valid(u32 action)
1610 {
1611         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1612                           MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1613                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1614 }
1615
1616 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1617                           struct mlx5_flow_act *flow_act,
1618                           struct mlx5_flow_table *ft)
1619 {
1620         bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1621         u32 action = flow_act->action;
1622
1623         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1624                 return counter_is_valid(action);
1625
1626         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1627                 return true;
1628
1629         if (ignore_level) {
1630                 if (ft->type != FS_FT_FDB &&
1631                     ft->type != FS_FT_NIC_RX)
1632                         return false;
1633
1634                 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1635                     ft->type != dest->ft->type)
1636                         return false;
1637         }
1638
1639         if (!dest || ((dest->type ==
1640             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1641             (dest->ft->level <= ft->level && !ignore_level)))
1642                 return false;
1643         return true;
1644 }
1645
1646 struct match_list {
1647         struct list_head        list;
1648         struct mlx5_flow_group *g;
1649 };
1650
1651 static void free_match_list(struct match_list *head, bool ft_locked)
1652 {
1653         struct match_list *iter, *match_tmp;
1654
1655         list_for_each_entry_safe(iter, match_tmp, &head->list,
1656                                  list) {
1657                 tree_put_node(&iter->g->node, ft_locked);
1658                 list_del(&iter->list);
1659                 kfree(iter);
1660         }
1661 }
1662
1663 static int build_match_list(struct match_list *match_head,
1664                             struct mlx5_flow_table *ft,
1665                             const struct mlx5_flow_spec *spec,
1666                             bool ft_locked)
1667 {
1668         struct rhlist_head *tmp, *list;
1669         struct mlx5_flow_group *g;
1670         int err = 0;
1671
1672         rcu_read_lock();
1673         INIT_LIST_HEAD(&match_head->list);
1674         /* Collect all fgs which has a matching match_criteria */
1675         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1676         /* RCU is atomic, we can't execute FW commands here */
1677         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1678                 struct match_list *curr_match;
1679
1680                 if (unlikely(!tree_get_node(&g->node)))
1681                         continue;
1682
1683                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1684                 if (!curr_match) {
1685                         rcu_read_unlock();
1686                         free_match_list(match_head, ft_locked);
1687                         return -ENOMEM;
1688                 }
1689                 curr_match->g = g;
1690                 list_add_tail(&curr_match->list, &match_head->list);
1691         }
1692         rcu_read_unlock();
1693         return err;
1694 }
1695
1696 static u64 matched_fgs_get_version(struct list_head *match_head)
1697 {
1698         struct match_list *iter;
1699         u64 version = 0;
1700
1701         list_for_each_entry(iter, match_head, list)
1702                 version += (u64)atomic_read(&iter->g->node.version);
1703         return version;
1704 }
1705
1706 static struct fs_fte *
1707 lookup_fte_locked(struct mlx5_flow_group *g,
1708                   const u32 *match_value,
1709                   bool take_write)
1710 {
1711         struct fs_fte *fte_tmp;
1712
1713         if (take_write)
1714                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1715         else
1716                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1717         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1718                                          rhash_fte);
1719         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1720                 fte_tmp = NULL;
1721                 goto out;
1722         }
1723         if (!fte_tmp->node.active) {
1724                 tree_put_node(&fte_tmp->node, false);
1725                 fte_tmp = NULL;
1726                 goto out;
1727         }
1728
1729         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1730 out:
1731         if (take_write)
1732                 up_write_ref_node(&g->node, false);
1733         else
1734                 up_read_ref_node(&g->node);
1735         return fte_tmp;
1736 }
1737
1738 static struct mlx5_flow_handle *
1739 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1740                        struct list_head *match_head,
1741                        const struct mlx5_flow_spec *spec,
1742                        struct mlx5_flow_act *flow_act,
1743                        struct mlx5_flow_destination *dest,
1744                        int dest_num,
1745                        int ft_version)
1746 {
1747         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1748         struct mlx5_flow_group *g;
1749         struct mlx5_flow_handle *rule;
1750         struct match_list *iter;
1751         bool take_write = false;
1752         struct fs_fte *fte;
1753         u64  version = 0;
1754         int err;
1755
1756         fte = alloc_fte(ft, spec, flow_act);
1757         if (IS_ERR(fte))
1758                 return  ERR_PTR(-ENOMEM);
1759
1760 search_again_locked:
1761         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1762                 goto skip_search;
1763         version = matched_fgs_get_version(match_head);
1764         /* Try to find an fte with identical match value and attempt update its
1765          * action.
1766          */
1767         list_for_each_entry(iter, match_head, list) {
1768                 struct fs_fte *fte_tmp;
1769
1770                 g = iter->g;
1771                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1772                 if (!fte_tmp)
1773                         continue;
1774                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1775                 /* No error check needed here, because insert_fte() is not called */
1776                 up_write_ref_node(&fte_tmp->node, false);
1777                 tree_put_node(&fte_tmp->node, false);
1778                 kmem_cache_free(steering->ftes_cache, fte);
1779                 return rule;
1780         }
1781
1782 skip_search:
1783         /* No group with matching fte found, or we skipped the search.
1784          * Try to add a new fte to any matching fg.
1785          */
1786
1787         /* Check the ft version, for case that new flow group
1788          * was added while the fgs weren't locked
1789          */
1790         if (atomic_read(&ft->node.version) != ft_version) {
1791                 rule = ERR_PTR(-EAGAIN);
1792                 goto out;
1793         }
1794
1795         /* Check the fgs version. If version have changed it could be that an
1796          * FTE with the same match value was added while the fgs weren't
1797          * locked.
1798          */
1799         if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1800             version != matched_fgs_get_version(match_head)) {
1801                 take_write = true;
1802                 goto search_again_locked;
1803         }
1804
1805         list_for_each_entry(iter, match_head, list) {
1806                 g = iter->g;
1807
1808                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1809
1810                 if (!g->node.active) {
1811                         up_write_ref_node(&g->node, false);
1812                         continue;
1813                 }
1814
1815                 err = insert_fte(g, fte);
1816                 if (err) {
1817                         up_write_ref_node(&g->node, false);
1818                         if (err == -ENOSPC)
1819                                 continue;
1820                         kmem_cache_free(steering->ftes_cache, fte);
1821                         return ERR_PTR(err);
1822                 }
1823
1824                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1825                 up_write_ref_node(&g->node, false);
1826                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1827                 up_write_ref_node(&fte->node, false);
1828                 if (IS_ERR(rule))
1829                         tree_put_node(&fte->node, false);
1830                 return rule;
1831         }
1832         rule = ERR_PTR(-ENOENT);
1833 out:
1834         kmem_cache_free(steering->ftes_cache, fte);
1835         return rule;
1836 }
1837
1838 static struct mlx5_flow_handle *
1839 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1840                      const struct mlx5_flow_spec *spec,
1841                      struct mlx5_flow_act *flow_act,
1842                      struct mlx5_flow_destination *dest,
1843                      int dest_num)
1844
1845 {
1846         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1847         struct mlx5_flow_handle *rule;
1848         struct match_list match_head;
1849         struct mlx5_flow_group *g;
1850         bool take_write = false;
1851         struct fs_fte *fte;
1852         int version;
1853         int err;
1854         int i;
1855
1856         if (!check_valid_spec(spec))
1857                 return ERR_PTR(-EINVAL);
1858
1859         for (i = 0; i < dest_num; i++) {
1860                 if (!dest_is_valid(&dest[i], flow_act, ft))
1861                         return ERR_PTR(-EINVAL);
1862         }
1863         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1864 search_again_locked:
1865         version = atomic_read(&ft->node.version);
1866
1867         /* Collect all fgs which has a matching match_criteria */
1868         err = build_match_list(&match_head, ft, spec, take_write);
1869         if (err) {
1870                 if (take_write)
1871                         up_write_ref_node(&ft->node, false);
1872                 else
1873                         up_read_ref_node(&ft->node);
1874                 return ERR_PTR(err);
1875         }
1876
1877         if (!take_write)
1878                 up_read_ref_node(&ft->node);
1879
1880         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1881                                       dest_num, version);
1882         free_match_list(&match_head, take_write);
1883         if (!IS_ERR(rule) ||
1884             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1885                 if (take_write)
1886                         up_write_ref_node(&ft->node, false);
1887                 return rule;
1888         }
1889
1890         if (!take_write) {
1891                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1892                 take_write = true;
1893         }
1894
1895         if (PTR_ERR(rule) == -EAGAIN ||
1896             version != atomic_read(&ft->node.version))
1897                 goto search_again_locked;
1898
1899         g = alloc_auto_flow_group(ft, spec);
1900         if (IS_ERR(g)) {
1901                 rule = ERR_CAST(g);
1902                 up_write_ref_node(&ft->node, false);
1903                 return rule;
1904         }
1905
1906         fte = alloc_fte(ft, spec, flow_act);
1907         if (IS_ERR(fte)) {
1908                 up_write_ref_node(&ft->node, false);
1909                 err = PTR_ERR(fte);
1910                 goto err_alloc_fte;
1911         }
1912
1913         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1914         up_write_ref_node(&ft->node, false);
1915
1916         err = create_auto_flow_group(ft, g);
1917         if (err)
1918                 goto err_release_fg;
1919
1920         err = insert_fte(g, fte);
1921         if (err)
1922                 goto err_release_fg;
1923
1924         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1925         up_write_ref_node(&g->node, false);
1926         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1927         up_write_ref_node(&fte->node, false);
1928         if (IS_ERR(rule))
1929                 tree_put_node(&fte->node, false);
1930         tree_put_node(&g->node, false);
1931         return rule;
1932
1933 err_release_fg:
1934         up_write_ref_node(&g->node, false);
1935         kmem_cache_free(steering->ftes_cache, fte);
1936 err_alloc_fte:
1937         tree_put_node(&g->node, false);
1938         return ERR_PTR(err);
1939 }
1940
1941 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1942 {
1943         return ((ft->type == FS_FT_NIC_RX) &&
1944                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1945 }
1946
1947 struct mlx5_flow_handle *
1948 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1949                     const struct mlx5_flow_spec *spec,
1950                     struct mlx5_flow_act *flow_act,
1951                     struct mlx5_flow_destination *dest,
1952                     int num_dest)
1953 {
1954         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1955         static const struct mlx5_flow_spec zero_spec = {};
1956         struct mlx5_flow_destination *gen_dest = NULL;
1957         struct mlx5_flow_table *next_ft = NULL;
1958         struct mlx5_flow_handle *handle = NULL;
1959         u32 sw_action = flow_act->action;
1960         int i;
1961
1962         if (!spec)
1963                 spec = &zero_spec;
1964
1965         if (!is_fwd_next_action(sw_action))
1966                 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1967
1968         if (!fwd_next_prio_supported(ft))
1969                 return ERR_PTR(-EOPNOTSUPP);
1970
1971         mutex_lock(&root->chain_lock);
1972         next_ft = find_next_fwd_ft(ft, flow_act);
1973         if (!next_ft) {
1974                 handle = ERR_PTR(-EOPNOTSUPP);
1975                 goto unlock;
1976         }
1977
1978         gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
1979                            GFP_KERNEL);
1980         if (!gen_dest) {
1981                 handle = ERR_PTR(-ENOMEM);
1982                 goto unlock;
1983         }
1984         for (i = 0; i < num_dest; i++)
1985                 gen_dest[i] = dest[i];
1986         gen_dest[i].type =
1987                 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1988         gen_dest[i].ft = next_ft;
1989         dest = gen_dest;
1990         num_dest++;
1991         flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
1992                               MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
1993         flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1994         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1995         if (IS_ERR(handle))
1996                 goto unlock;
1997
1998         if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
1999                 mutex_lock(&next_ft->lock);
2000                 list_add(&handle->rule[num_dest - 1]->next_ft,
2001                          &next_ft->fwd_rules);
2002                 mutex_unlock(&next_ft->lock);
2003                 handle->rule[num_dest - 1]->sw_action = sw_action;
2004                 handle->rule[num_dest - 1]->ft = ft;
2005         }
2006 unlock:
2007         mutex_unlock(&root->chain_lock);
2008         kfree(gen_dest);
2009         return handle;
2010 }
2011 EXPORT_SYMBOL(mlx5_add_flow_rules);
2012
2013 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2014 {
2015         struct fs_fte *fte;
2016         int i;
2017
2018         /* In order to consolidate the HW changes we lock the FTE for other
2019          * changes, and increase its refcount, in order not to perform the
2020          * "del" functions of the FTE. Will handle them here.
2021          * The removal of the rules is done under locked FTE.
2022          * After removing all the handle's rules, if there are remaining
2023          * rules, it means we just need to modify the FTE in FW, and
2024          * unlock/decrease the refcount we increased before.
2025          * Otherwise, it means the FTE should be deleted. First delete the
2026          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2027          * the FTE, which will handle the last decrease of the refcount, as
2028          * well as required handling of its parent.
2029          */
2030         fs_get_obj(fte, handle->rule[0]->node.parent);
2031         down_write_ref_node(&fte->node, false);
2032         for (i = handle->num_rules - 1; i >= 0; i--)
2033                 tree_remove_node(&handle->rule[i]->node, true);
2034         if (fte->dests_size) {
2035                 if (fte->modify_mask)
2036                         modify_fte(fte);
2037                 up_write_ref_node(&fte->node, false);
2038         } else if (list_empty(&fte->node.children)) {
2039                 del_hw_fte(&fte->node);
2040                 /* Avoid double call to del_hw_fte */
2041                 fte->node.del_hw_func = NULL;
2042                 up_write_ref_node(&fte->node, false);
2043                 tree_put_node(&fte->node, false);
2044         }
2045         kfree(handle);
2046 }
2047 EXPORT_SYMBOL(mlx5_del_flow_rules);
2048
2049 /* Assuming prio->node.children(flow tables) is sorted by level */
2050 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2051 {
2052         struct fs_prio *prio;
2053
2054         fs_get_obj(prio, ft->node.parent);
2055
2056         if (!list_is_last(&ft->node.list, &prio->node.children))
2057                 return list_next_entry(ft, node.list);
2058         return find_next_chained_ft(prio);
2059 }
2060
2061 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2062 {
2063         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2064         struct mlx5_ft_underlay_qp *uqp;
2065         struct mlx5_flow_table *new_root_ft = NULL;
2066         int err = 0;
2067         u32 qpn;
2068
2069         if (root->root_ft != ft)
2070                 return 0;
2071
2072         new_root_ft = find_next_ft(ft);
2073         if (!new_root_ft) {
2074                 root->root_ft = NULL;
2075                 return 0;
2076         }
2077
2078         if (list_empty(&root->underlay_qpns)) {
2079                 /* Don't set any QPN (zero) in case QPN list is empty */
2080                 qpn = 0;
2081                 err = root->cmds->update_root_ft(root, new_root_ft,
2082                                                  qpn, false);
2083         } else {
2084                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2085                         qpn = uqp->qpn;
2086                         err = root->cmds->update_root_ft(root,
2087                                                          new_root_ft, qpn,
2088                                                          false);
2089                         if (err)
2090                                 break;
2091                 }
2092         }
2093
2094         if (err)
2095                 mlx5_core_warn(root->dev,
2096                                "Update root flow table of id(%u) qpn(%d) failed\n",
2097                                ft->id, qpn);
2098         else
2099                 root->root_ft = new_root_ft;
2100
2101         return 0;
2102 }
2103
2104 /* Connect flow table from previous priority to
2105  * the next flow table.
2106  */
2107 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2108 {
2109         struct mlx5_core_dev *dev = get_dev(&ft->node);
2110         struct mlx5_flow_table *next_ft;
2111         struct fs_prio *prio;
2112         int err = 0;
2113
2114         err = update_root_ft_destroy(ft);
2115         if (err)
2116                 return err;
2117
2118         fs_get_obj(prio, ft->node.parent);
2119         if  (!(list_first_entry(&prio->node.children,
2120                                 struct mlx5_flow_table,
2121                                 node.list) == ft))
2122                 return 0;
2123
2124         next_ft = find_next_ft(ft);
2125         err = connect_fwd_rules(dev, next_ft, ft);
2126         if (err)
2127                 return err;
2128
2129         err = connect_prev_fts(dev, next_ft, prio);
2130         if (err)
2131                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2132                                ft->id);
2133         return err;
2134 }
2135
2136 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2137 {
2138         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2139         int err = 0;
2140
2141         mutex_lock(&root->chain_lock);
2142         if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2143                 err = disconnect_flow_table(ft);
2144         if (err) {
2145                 mutex_unlock(&root->chain_lock);
2146                 return err;
2147         }
2148         if (tree_remove_node(&ft->node, false))
2149                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2150                                ft->id);
2151         mutex_unlock(&root->chain_lock);
2152
2153         return err;
2154 }
2155 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2156
2157 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2158 {
2159         if (tree_remove_node(&fg->node, false))
2160                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2161                                fg->id);
2162 }
2163 EXPORT_SYMBOL(mlx5_destroy_flow_group);
2164
2165 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2166                                                 int n)
2167 {
2168         struct mlx5_flow_steering *steering = dev->priv.steering;
2169
2170         if (!steering || !steering->fdb_sub_ns)
2171                 return NULL;
2172
2173         return steering->fdb_sub_ns[n];
2174 }
2175 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2176
2177 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2178                                                     enum mlx5_flow_namespace_type type)
2179 {
2180         struct mlx5_flow_steering *steering = dev->priv.steering;
2181         struct mlx5_flow_root_namespace *root_ns;
2182         int prio = 0;
2183         struct fs_prio *fs_prio;
2184         struct mlx5_flow_namespace *ns;
2185
2186         if (!steering)
2187                 return NULL;
2188
2189         switch (type) {
2190         case MLX5_FLOW_NAMESPACE_FDB:
2191                 if (steering->fdb_root_ns)
2192                         return &steering->fdb_root_ns->ns;
2193                 return NULL;
2194         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2195                 if (steering->sniffer_rx_root_ns)
2196                         return &steering->sniffer_rx_root_ns->ns;
2197                 return NULL;
2198         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2199                 if (steering->sniffer_tx_root_ns)
2200                         return &steering->sniffer_tx_root_ns->ns;
2201                 return NULL;
2202         default:
2203                 break;
2204         }
2205
2206         if (type == MLX5_FLOW_NAMESPACE_EGRESS ||
2207             type == MLX5_FLOW_NAMESPACE_EGRESS_KERNEL) {
2208                 root_ns = steering->egress_root_ns;
2209                 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2210         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX) {
2211                 root_ns = steering->rdma_rx_root_ns;
2212                 prio = RDMA_RX_BYPASS_PRIO;
2213         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL) {
2214                 root_ns = steering->rdma_rx_root_ns;
2215                 prio = RDMA_RX_KERNEL_PRIO;
2216         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_TX) {
2217                 root_ns = steering->rdma_tx_root_ns;
2218         } else { /* Must be NIC RX */
2219                 root_ns = steering->root_ns;
2220                 prio = type;
2221         }
2222
2223         if (!root_ns)
2224                 return NULL;
2225
2226         fs_prio = find_prio(&root_ns->ns, prio);
2227         if (!fs_prio)
2228                 return NULL;
2229
2230         ns = list_first_entry(&fs_prio->node.children,
2231                               typeof(*ns),
2232                               node.list);
2233
2234         return ns;
2235 }
2236 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2237
2238 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2239                                                               enum mlx5_flow_namespace_type type,
2240                                                               int vport)
2241 {
2242         struct mlx5_flow_steering *steering = dev->priv.steering;
2243
2244         if (!steering)
2245                 return NULL;
2246
2247         switch (type) {
2248         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2249                 if (vport >= steering->esw_egress_acl_vports)
2250                         return NULL;
2251                 if (steering->esw_egress_root_ns &&
2252                     steering->esw_egress_root_ns[vport])
2253                         return &steering->esw_egress_root_ns[vport]->ns;
2254                 else
2255                         return NULL;
2256         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2257                 if (vport >= steering->esw_ingress_acl_vports)
2258                         return NULL;
2259                 if (steering->esw_ingress_root_ns &&
2260                     steering->esw_ingress_root_ns[vport])
2261                         return &steering->esw_ingress_root_ns[vport]->ns;
2262                 else
2263                         return NULL;
2264         default:
2265                 return NULL;
2266         }
2267 }
2268
2269 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2270                                        unsigned int prio,
2271                                        int num_levels,
2272                                        enum fs_node_type type)
2273 {
2274         struct fs_prio *fs_prio;
2275
2276         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2277         if (!fs_prio)
2278                 return ERR_PTR(-ENOMEM);
2279
2280         fs_prio->node.type = type;
2281         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2282         tree_add_node(&fs_prio->node, &ns->node);
2283         fs_prio->num_levels = num_levels;
2284         fs_prio->prio = prio;
2285         list_add_tail(&fs_prio->node.list, &ns->node.children);
2286
2287         return fs_prio;
2288 }
2289
2290 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2291                                               unsigned int prio,
2292                                               int num_levels)
2293 {
2294         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2295 }
2296
2297 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2298                                       unsigned int prio, int num_levels)
2299 {
2300         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2301 }
2302
2303 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2304                                                      *ns)
2305 {
2306         ns->node.type = FS_TYPE_NAMESPACE;
2307
2308         return ns;
2309 }
2310
2311 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2312                                                        int def_miss_act)
2313 {
2314         struct mlx5_flow_namespace      *ns;
2315
2316         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2317         if (!ns)
2318                 return ERR_PTR(-ENOMEM);
2319
2320         fs_init_namespace(ns);
2321         ns->def_miss_action = def_miss_act;
2322         tree_init_node(&ns->node, NULL, del_sw_ns);
2323         tree_add_node(&ns->node, &prio->node);
2324         list_add_tail(&ns->node.list, &prio->node.children);
2325
2326         return ns;
2327 }
2328
2329 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2330                              struct init_tree_node *prio_metadata)
2331 {
2332         struct fs_prio *fs_prio;
2333         int i;
2334
2335         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2336                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2337                 if (IS_ERR(fs_prio))
2338                         return PTR_ERR(fs_prio);
2339         }
2340         return 0;
2341 }
2342
2343 #define FLOW_TABLE_BIT_SZ 1
2344 #define GET_FLOW_TABLE_CAP(dev, offset) \
2345         ((be32_to_cpu(*((__be32 *)(dev->caps.hca[MLX5_CAP_FLOW_TABLE]->cur) +   \
2346                         offset / 32)) >>                                        \
2347           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2348 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2349 {
2350         int i;
2351
2352         for (i = 0; i < caps->arr_sz; i++) {
2353                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2354                         return false;
2355         }
2356         return true;
2357 }
2358
2359 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2360                                     struct init_tree_node *init_node,
2361                                     struct fs_node *fs_parent_node,
2362                                     struct init_tree_node *init_parent_node,
2363                                     int prio)
2364 {
2365         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2366                                               flow_table_properties_nic_receive.
2367                                               max_ft_level);
2368         struct mlx5_flow_namespace *fs_ns;
2369         struct fs_prio *fs_prio;
2370         struct fs_node *base;
2371         int i;
2372         int err;
2373
2374         if (init_node->type == FS_TYPE_PRIO) {
2375                 if ((init_node->min_ft_level > max_ft_level) ||
2376                     !has_required_caps(steering->dev, &init_node->caps))
2377                         return 0;
2378
2379                 fs_get_obj(fs_ns, fs_parent_node);
2380                 if (init_node->num_leaf_prios)
2381                         return create_leaf_prios(fs_ns, prio, init_node);
2382                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2383                 if (IS_ERR(fs_prio))
2384                         return PTR_ERR(fs_prio);
2385                 base = &fs_prio->node;
2386         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2387                 fs_get_obj(fs_prio, fs_parent_node);
2388                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2389                 if (IS_ERR(fs_ns))
2390                         return PTR_ERR(fs_ns);
2391                 base = &fs_ns->node;
2392         } else {
2393                 return -EINVAL;
2394         }
2395         prio = 0;
2396         for (i = 0; i < init_node->ar_size; i++) {
2397                 err = init_root_tree_recursive(steering, &init_node->children[i],
2398                                                base, init_node, prio);
2399                 if (err)
2400                         return err;
2401                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2402                     init_node->children[i].num_leaf_prios) {
2403                         prio += init_node->children[i].num_leaf_prios;
2404                 }
2405         }
2406
2407         return 0;
2408 }
2409
2410 static int init_root_tree(struct mlx5_flow_steering *steering,
2411                           struct init_tree_node *init_node,
2412                           struct fs_node *fs_parent_node)
2413 {
2414         int err;
2415         int i;
2416
2417         for (i = 0; i < init_node->ar_size; i++) {
2418                 err = init_root_tree_recursive(steering, &init_node->children[i],
2419                                                fs_parent_node,
2420                                                init_node, i);
2421                 if (err)
2422                         return err;
2423         }
2424         return 0;
2425 }
2426
2427 static void del_sw_root_ns(struct fs_node *node)
2428 {
2429         struct mlx5_flow_root_namespace *root_ns;
2430         struct mlx5_flow_namespace *ns;
2431
2432         fs_get_obj(ns, node);
2433         root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2434         mutex_destroy(&root_ns->chain_lock);
2435         kfree(node);
2436 }
2437
2438 static struct mlx5_flow_root_namespace
2439 *create_root_ns(struct mlx5_flow_steering *steering,
2440                 enum fs_flow_table_type table_type)
2441 {
2442         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2443         struct mlx5_flow_root_namespace *root_ns;
2444         struct mlx5_flow_namespace *ns;
2445
2446         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2447             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2448                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2449
2450         /* Create the root namespace */
2451         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2452         if (!root_ns)
2453                 return NULL;
2454
2455         root_ns->dev = steering->dev;
2456         root_ns->table_type = table_type;
2457         root_ns->cmds = cmds;
2458
2459         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2460
2461         ns = &root_ns->ns;
2462         fs_init_namespace(ns);
2463         mutex_init(&root_ns->chain_lock);
2464         tree_init_node(&ns->node, NULL, del_sw_root_ns);
2465         tree_add_node(&ns->node, NULL);
2466
2467         return root_ns;
2468 }
2469
2470 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2471
2472 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2473 {
2474         struct fs_prio *prio;
2475
2476         fs_for_each_prio(prio, ns) {
2477                  /* This updates prio start_level and num_levels */
2478                 set_prio_attrs_in_prio(prio, acc_level);
2479                 acc_level += prio->num_levels;
2480         }
2481         return acc_level;
2482 }
2483
2484 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2485 {
2486         struct mlx5_flow_namespace *ns;
2487         int acc_level_ns = acc_level;
2488
2489         prio->start_level = acc_level;
2490         fs_for_each_ns(ns, prio) {
2491                 /* This updates start_level and num_levels of ns's priority descendants */
2492                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2493
2494                 /* If this a prio with chains, and we can jump from one chain
2495                  * (namespace) to another, so we accumulate the levels
2496                  */
2497                 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2498                         acc_level = acc_level_ns;
2499         }
2500
2501         if (!prio->num_levels)
2502                 prio->num_levels = acc_level_ns - prio->start_level;
2503         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2504 }
2505
2506 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2507 {
2508         struct mlx5_flow_namespace *ns = &root_ns->ns;
2509         struct fs_prio *prio;
2510         int start_level = 0;
2511
2512         fs_for_each_prio(prio, ns) {
2513                 set_prio_attrs_in_prio(prio, start_level);
2514                 start_level += prio->num_levels;
2515         }
2516 }
2517
2518 #define ANCHOR_PRIO 0
2519 #define ANCHOR_SIZE 1
2520 #define ANCHOR_LEVEL 0
2521 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2522 {
2523         struct mlx5_flow_namespace *ns = NULL;
2524         struct mlx5_flow_table_attr ft_attr = {};
2525         struct mlx5_flow_table *ft;
2526
2527         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2528         if (WARN_ON(!ns))
2529                 return -EINVAL;
2530
2531         ft_attr.max_fte = ANCHOR_SIZE;
2532         ft_attr.level   = ANCHOR_LEVEL;
2533         ft_attr.prio    = ANCHOR_PRIO;
2534
2535         ft = mlx5_create_flow_table(ns, &ft_attr);
2536         if (IS_ERR(ft)) {
2537                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2538                 return PTR_ERR(ft);
2539         }
2540         return 0;
2541 }
2542
2543 static int init_root_ns(struct mlx5_flow_steering *steering)
2544 {
2545         int err;
2546
2547         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2548         if (!steering->root_ns)
2549                 return -ENOMEM;
2550
2551         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2552         if (err)
2553                 goto out_err;
2554
2555         set_prio_attrs(steering->root_ns);
2556         err = create_anchor_flow_table(steering);
2557         if (err)
2558                 goto out_err;
2559
2560         return 0;
2561
2562 out_err:
2563         cleanup_root_ns(steering->root_ns);
2564         steering->root_ns = NULL;
2565         return err;
2566 }
2567
2568 static void clean_tree(struct fs_node *node)
2569 {
2570         if (node) {
2571                 struct fs_node *iter;
2572                 struct fs_node *temp;
2573
2574                 tree_get_node(node);
2575                 list_for_each_entry_safe(iter, temp, &node->children, list)
2576                         clean_tree(iter);
2577                 tree_put_node(node, false);
2578                 tree_remove_node(node, false);
2579         }
2580 }
2581
2582 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2583 {
2584         if (!root_ns)
2585                 return;
2586
2587         clean_tree(&root_ns->ns.node);
2588 }
2589
2590 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2591 {
2592         struct mlx5_flow_steering *steering = dev->priv.steering;
2593
2594         cleanup_root_ns(steering->root_ns);
2595         cleanup_root_ns(steering->fdb_root_ns);
2596         steering->fdb_root_ns = NULL;
2597         kfree(steering->fdb_sub_ns);
2598         steering->fdb_sub_ns = NULL;
2599         cleanup_root_ns(steering->sniffer_rx_root_ns);
2600         cleanup_root_ns(steering->sniffer_tx_root_ns);
2601         cleanup_root_ns(steering->rdma_rx_root_ns);
2602         cleanup_root_ns(steering->rdma_tx_root_ns);
2603         cleanup_root_ns(steering->egress_root_ns);
2604         mlx5_cleanup_fc_stats(dev);
2605         kmem_cache_destroy(steering->ftes_cache);
2606         kmem_cache_destroy(steering->fgs_cache);
2607         mlx5_ft_pool_destroy(dev);
2608         kfree(steering);
2609 }
2610
2611 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2612 {
2613         struct fs_prio *prio;
2614
2615         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2616         if (!steering->sniffer_tx_root_ns)
2617                 return -ENOMEM;
2618
2619         /* Create single prio */
2620         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2621         return PTR_ERR_OR_ZERO(prio);
2622 }
2623
2624 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2625 {
2626         struct fs_prio *prio;
2627
2628         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2629         if (!steering->sniffer_rx_root_ns)
2630                 return -ENOMEM;
2631
2632         /* Create single prio */
2633         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2634         return PTR_ERR_OR_ZERO(prio);
2635 }
2636
2637 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2638 {
2639         int err;
2640
2641         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2642         if (!steering->rdma_rx_root_ns)
2643                 return -ENOMEM;
2644
2645         err = init_root_tree(steering, &rdma_rx_root_fs,
2646                              &steering->rdma_rx_root_ns->ns.node);
2647         if (err)
2648                 goto out_err;
2649
2650         set_prio_attrs(steering->rdma_rx_root_ns);
2651
2652         return 0;
2653
2654 out_err:
2655         cleanup_root_ns(steering->rdma_rx_root_ns);
2656         steering->rdma_rx_root_ns = NULL;
2657         return err;
2658 }
2659
2660 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2661 {
2662         int err;
2663
2664         steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2665         if (!steering->rdma_tx_root_ns)
2666                 return -ENOMEM;
2667
2668         err = init_root_tree(steering, &rdma_tx_root_fs,
2669                              &steering->rdma_tx_root_ns->ns.node);
2670         if (err)
2671                 goto out_err;
2672
2673         set_prio_attrs(steering->rdma_tx_root_ns);
2674
2675         return 0;
2676
2677 out_err:
2678         cleanup_root_ns(steering->rdma_tx_root_ns);
2679         steering->rdma_tx_root_ns = NULL;
2680         return err;
2681 }
2682
2683 /* FT and tc chains are stored in the same array so we can re-use the
2684  * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2685  * When creating a new ns for each chain store it in the first available slot.
2686  * Assume tc chains are created and stored first and only then the FT chain.
2687  */
2688 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2689                                         struct mlx5_flow_namespace *ns)
2690 {
2691         int chain = 0;
2692
2693         while (steering->fdb_sub_ns[chain])
2694                 ++chain;
2695
2696         steering->fdb_sub_ns[chain] = ns;
2697 }
2698
2699 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2700                                         struct fs_prio *maj_prio)
2701 {
2702         struct mlx5_flow_namespace *ns;
2703         struct fs_prio *min_prio;
2704         int prio;
2705
2706         ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2707         if (IS_ERR(ns))
2708                 return PTR_ERR(ns);
2709
2710         for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2711                 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2712                 if (IS_ERR(min_prio))
2713                         return PTR_ERR(min_prio);
2714         }
2715
2716         store_fdb_sub_ns_prio_chain(steering, ns);
2717
2718         return 0;
2719 }
2720
2721 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2722                              int fs_prio,
2723                              int chains)
2724 {
2725         struct fs_prio *maj_prio;
2726         int levels;
2727         int chain;
2728         int err;
2729
2730         levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2731         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2732                                           fs_prio,
2733                                           levels);
2734         if (IS_ERR(maj_prio))
2735                 return PTR_ERR(maj_prio);
2736
2737         for (chain = 0; chain < chains; chain++) {
2738                 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2739                 if (err)
2740                         return err;
2741         }
2742
2743         return 0;
2744 }
2745
2746 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2747 {
2748         int err;
2749
2750         steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2751                                        sizeof(*steering->fdb_sub_ns),
2752                                        GFP_KERNEL);
2753         if (!steering->fdb_sub_ns)
2754                 return -ENOMEM;
2755
2756         err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
2757         if (err)
2758                 return err;
2759
2760         err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
2761         if (err)
2762                 return err;
2763
2764         return 0;
2765 }
2766
2767 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2768 {
2769         struct fs_prio *maj_prio;
2770         int err;
2771
2772         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2773         if (!steering->fdb_root_ns)
2774                 return -ENOMEM;
2775
2776         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2777                                   1);
2778         if (IS_ERR(maj_prio)) {
2779                 err = PTR_ERR(maj_prio);
2780                 goto out_err;
2781         }
2782         err = create_fdb_fast_path(steering);
2783         if (err)
2784                 goto out_err;
2785
2786         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1);
2787         if (IS_ERR(maj_prio)) {
2788                 err = PTR_ERR(maj_prio);
2789                 goto out_err;
2790         }
2791
2792         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 3);
2793         if (IS_ERR(maj_prio)) {
2794                 err = PTR_ERR(maj_prio);
2795                 goto out_err;
2796         }
2797
2798         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2799         if (IS_ERR(maj_prio)) {
2800                 err = PTR_ERR(maj_prio);
2801                 goto out_err;
2802         }
2803
2804         /* We put this priority last, knowing that nothing will get here
2805          * unless explicitly forwarded to. This is possible because the
2806          * slow path tables have catch all rules and nothing gets passed
2807          * those tables.
2808          */
2809         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
2810         if (IS_ERR(maj_prio)) {
2811                 err = PTR_ERR(maj_prio);
2812                 goto out_err;
2813         }
2814
2815         set_prio_attrs(steering->fdb_root_ns);
2816         return 0;
2817
2818 out_err:
2819         cleanup_root_ns(steering->fdb_root_ns);
2820         kfree(steering->fdb_sub_ns);
2821         steering->fdb_sub_ns = NULL;
2822         steering->fdb_root_ns = NULL;
2823         return err;
2824 }
2825
2826 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2827 {
2828         struct fs_prio *prio;
2829
2830         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2831         if (!steering->esw_egress_root_ns[vport])
2832                 return -ENOMEM;
2833
2834         /* create 1 prio*/
2835         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2836         return PTR_ERR_OR_ZERO(prio);
2837 }
2838
2839 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2840 {
2841         struct fs_prio *prio;
2842
2843         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2844         if (!steering->esw_ingress_root_ns[vport])
2845                 return -ENOMEM;
2846
2847         /* create 1 prio*/
2848         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2849         return PTR_ERR_OR_ZERO(prio);
2850 }
2851
2852 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2853 {
2854         struct mlx5_flow_steering *steering = dev->priv.steering;
2855         int err;
2856         int i;
2857
2858         steering->esw_egress_root_ns =
2859                         kcalloc(total_vports,
2860                                 sizeof(*steering->esw_egress_root_ns),
2861                                 GFP_KERNEL);
2862         if (!steering->esw_egress_root_ns)
2863                 return -ENOMEM;
2864
2865         for (i = 0; i < total_vports; i++) {
2866                 err = init_egress_acl_root_ns(steering, i);
2867                 if (err)
2868                         goto cleanup_root_ns;
2869         }
2870         steering->esw_egress_acl_vports = total_vports;
2871         return 0;
2872
2873 cleanup_root_ns:
2874         for (i--; i >= 0; i--)
2875                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2876         kfree(steering->esw_egress_root_ns);
2877         steering->esw_egress_root_ns = NULL;
2878         return err;
2879 }
2880
2881 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev)
2882 {
2883         struct mlx5_flow_steering *steering = dev->priv.steering;
2884         int i;
2885
2886         if (!steering->esw_egress_root_ns)
2887                 return;
2888
2889         for (i = 0; i < steering->esw_egress_acl_vports; i++)
2890                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2891
2892         kfree(steering->esw_egress_root_ns);
2893         steering->esw_egress_root_ns = NULL;
2894 }
2895
2896 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2897 {
2898         struct mlx5_flow_steering *steering = dev->priv.steering;
2899         int err;
2900         int i;
2901
2902         steering->esw_ingress_root_ns =
2903                         kcalloc(total_vports,
2904                                 sizeof(*steering->esw_ingress_root_ns),
2905                                 GFP_KERNEL);
2906         if (!steering->esw_ingress_root_ns)
2907                 return -ENOMEM;
2908
2909         for (i = 0; i < total_vports; i++) {
2910                 err = init_ingress_acl_root_ns(steering, i);
2911                 if (err)
2912                         goto cleanup_root_ns;
2913         }
2914         steering->esw_ingress_acl_vports = total_vports;
2915         return 0;
2916
2917 cleanup_root_ns:
2918         for (i--; i >= 0; i--)
2919                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2920         kfree(steering->esw_ingress_root_ns);
2921         steering->esw_ingress_root_ns = NULL;
2922         return err;
2923 }
2924
2925 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev)
2926 {
2927         struct mlx5_flow_steering *steering = dev->priv.steering;
2928         int i;
2929
2930         if (!steering->esw_ingress_root_ns)
2931                 return;
2932
2933         for (i = 0; i < steering->esw_ingress_acl_vports; i++)
2934                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2935
2936         kfree(steering->esw_ingress_root_ns);
2937         steering->esw_ingress_root_ns = NULL;
2938 }
2939
2940 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2941 {
2942         int err;
2943
2944         steering->egress_root_ns = create_root_ns(steering,
2945                                                   FS_FT_NIC_TX);
2946         if (!steering->egress_root_ns)
2947                 return -ENOMEM;
2948
2949         err = init_root_tree(steering, &egress_root_fs,
2950                              &steering->egress_root_ns->ns.node);
2951         if (err)
2952                 goto cleanup;
2953         set_prio_attrs(steering->egress_root_ns);
2954         return 0;
2955 cleanup:
2956         cleanup_root_ns(steering->egress_root_ns);
2957         steering->egress_root_ns = NULL;
2958         return err;
2959 }
2960
2961 int mlx5_init_fs(struct mlx5_core_dev *dev)
2962 {
2963         struct mlx5_flow_steering *steering;
2964         int err = 0;
2965
2966         err = mlx5_init_fc_stats(dev);
2967         if (err)
2968                 return err;
2969
2970         err = mlx5_ft_pool_init(dev);
2971         if (err)
2972                 return err;
2973
2974         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2975         if (!steering) {
2976                 err = -ENOMEM;
2977                 goto err;
2978         }
2979
2980         steering->dev = dev;
2981         dev->priv.steering = steering;
2982
2983         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2984                                                 sizeof(struct mlx5_flow_group), 0,
2985                                                 0, NULL);
2986         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2987                                                  0, NULL);
2988         if (!steering->ftes_cache || !steering->fgs_cache) {
2989                 err = -ENOMEM;
2990                 goto err;
2991         }
2992
2993         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2994               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2995              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2996               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2997             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2998                 err = init_root_ns(steering);
2999                 if (err)
3000                         goto err;
3001         }
3002
3003         if (MLX5_ESWITCH_MANAGER(dev)) {
3004                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
3005                         err = init_fdb_root_ns(steering);
3006                         if (err)
3007                                 goto err;
3008                 }
3009         }
3010
3011         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
3012                 err = init_sniffer_rx_root_ns(steering);
3013                 if (err)
3014                         goto err;
3015         }
3016
3017         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
3018                 err = init_sniffer_tx_root_ns(steering);
3019                 if (err)
3020                         goto err;
3021         }
3022
3023         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
3024             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3025                 err = init_rdma_rx_root_ns(steering);
3026                 if (err)
3027                         goto err;
3028         }
3029
3030         if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3031                 err = init_rdma_tx_root_ns(steering);
3032                 if (err)
3033                         goto err;
3034         }
3035
3036         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE ||
3037             MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3038                 err = init_egress_root_ns(steering);
3039                 if (err)
3040                         goto err;
3041         }
3042
3043         return 0;
3044 err:
3045         mlx5_cleanup_fs(dev);
3046         return err;
3047 }
3048
3049 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3050 {
3051         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3052         struct mlx5_ft_underlay_qp *new_uqp;
3053         int err = 0;
3054
3055         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3056         if (!new_uqp)
3057                 return -ENOMEM;
3058
3059         mutex_lock(&root->chain_lock);
3060
3061         if (!root->root_ft) {
3062                 err = -EINVAL;
3063                 goto update_ft_fail;
3064         }
3065
3066         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3067                                          false);
3068         if (err) {
3069                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3070                                underlay_qpn, err);
3071                 goto update_ft_fail;
3072         }
3073
3074         new_uqp->qpn = underlay_qpn;
3075         list_add_tail(&new_uqp->list, &root->underlay_qpns);
3076
3077         mutex_unlock(&root->chain_lock);
3078
3079         return 0;
3080
3081 update_ft_fail:
3082         mutex_unlock(&root->chain_lock);
3083         kfree(new_uqp);
3084         return err;
3085 }
3086 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3087
3088 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3089 {
3090         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3091         struct mlx5_ft_underlay_qp *uqp;
3092         bool found = false;
3093         int err = 0;
3094
3095         mutex_lock(&root->chain_lock);
3096         list_for_each_entry(uqp, &root->underlay_qpns, list) {
3097                 if (uqp->qpn == underlay_qpn) {
3098                         found = true;
3099                         break;
3100                 }
3101         }
3102
3103         if (!found) {
3104                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3105                                underlay_qpn);
3106                 err = -EINVAL;
3107                 goto out;
3108         }
3109
3110         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3111                                          true);
3112         if (err)
3113                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3114                                underlay_qpn, err);
3115
3116         list_del(&uqp->list);
3117         mutex_unlock(&root->chain_lock);
3118         kfree(uqp);
3119
3120         return 0;
3121
3122 out:
3123         mutex_unlock(&root->chain_lock);
3124         return err;
3125 }
3126 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3127
3128 static struct mlx5_flow_root_namespace
3129 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3130 {
3131         struct mlx5_flow_namespace *ns;
3132
3133         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3134             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3135                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3136         else
3137                 ns = mlx5_get_flow_namespace(dev, ns_type);
3138         if (!ns)
3139                 return NULL;
3140
3141         return find_root(&ns->node);
3142 }
3143
3144 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3145                                                  u8 ns_type, u8 num_actions,
3146                                                  void *modify_actions)
3147 {
3148         struct mlx5_flow_root_namespace *root;
3149         struct mlx5_modify_hdr *modify_hdr;
3150         int err;
3151
3152         root = get_root_namespace(dev, ns_type);
3153         if (!root)
3154                 return ERR_PTR(-EOPNOTSUPP);
3155
3156         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3157         if (!modify_hdr)
3158                 return ERR_PTR(-ENOMEM);
3159
3160         modify_hdr->ns_type = ns_type;
3161         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3162                                               modify_actions, modify_hdr);
3163         if (err) {
3164                 kfree(modify_hdr);
3165                 return ERR_PTR(err);
3166         }
3167
3168         return modify_hdr;
3169 }
3170 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3171
3172 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3173                                 struct mlx5_modify_hdr *modify_hdr)
3174 {
3175         struct mlx5_flow_root_namespace *root;
3176
3177         root = get_root_namespace(dev, modify_hdr->ns_type);
3178         if (WARN_ON(!root))
3179                 return;
3180         root->cmds->modify_header_dealloc(root, modify_hdr);
3181         kfree(modify_hdr);
3182 }
3183 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3184
3185 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3186                                                      struct mlx5_pkt_reformat_params *params,
3187                                                      enum mlx5_flow_namespace_type ns_type)
3188 {
3189         struct mlx5_pkt_reformat *pkt_reformat;
3190         struct mlx5_flow_root_namespace *root;
3191         int err;
3192
3193         root = get_root_namespace(dev, ns_type);
3194         if (!root)
3195                 return ERR_PTR(-EOPNOTSUPP);
3196
3197         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3198         if (!pkt_reformat)
3199                 return ERR_PTR(-ENOMEM);
3200
3201         pkt_reformat->ns_type = ns_type;
3202         pkt_reformat->reformat_type = params->type;
3203         err = root->cmds->packet_reformat_alloc(root, params, ns_type,
3204                                                 pkt_reformat);
3205         if (err) {
3206                 kfree(pkt_reformat);
3207                 return ERR_PTR(err);
3208         }
3209
3210         return pkt_reformat;
3211 }
3212 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3213
3214 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3215                                   struct mlx5_pkt_reformat *pkt_reformat)
3216 {
3217         struct mlx5_flow_root_namespace *root;
3218
3219         root = get_root_namespace(dev, pkt_reformat->ns_type);
3220         if (WARN_ON(!root))
3221                 return;
3222         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3223         kfree(pkt_reformat);
3224 }
3225 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3226
3227 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3228                                  struct mlx5_flow_root_namespace *peer_ns)
3229 {
3230         if (peer_ns && ns->mode != peer_ns->mode) {
3231                 mlx5_core_err(ns->dev,
3232                               "Can't peer namespace of different steering mode\n");
3233                 return -EINVAL;
3234         }
3235
3236         return ns->cmds->set_peer(ns, peer_ns);
3237 }
3238
3239 /* This function should be called only at init stage of the namespace.
3240  * It is not safe to call this function while steering operations
3241  * are executed in the namespace.
3242  */
3243 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3244                                  enum mlx5_flow_steering_mode mode)
3245 {
3246         struct mlx5_flow_root_namespace *root;
3247         const struct mlx5_flow_cmds *cmds;
3248         int err;
3249
3250         root = find_root(&ns->node);
3251         if (&root->ns != ns)
3252         /* Can't set cmds to non root namespace */
3253                 return -EINVAL;
3254
3255         if (root->table_type != FS_FT_FDB)
3256                 return -EOPNOTSUPP;
3257
3258         if (root->mode == mode)
3259                 return 0;
3260
3261         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3262                 cmds = mlx5_fs_cmd_get_dr_cmds();
3263         else
3264                 cmds = mlx5_fs_cmd_get_fw_cmds();
3265         if (!cmds)
3266                 return -EOPNOTSUPP;
3267
3268         err = cmds->create_ns(root);
3269         if (err) {
3270                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3271                               err);
3272                 return err;
3273         }
3274
3275         root->cmds->destroy_ns(root);
3276         root->cmds = cmds;
3277         root->mode = mode;
3278
3279         return 0;
3280 }