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
  44 #define INIT_TREE_NODE_ARRAY_SIZE(...)  (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
  45                                          sizeof(struct init_tree_node))
  46
  47 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
  48                  ...) {.type = FS_TYPE_PRIO,\
  49         .min_ft_level = min_level_val,\
  50         .num_levels = num_levels_val,\
  51         .num_leaf_prios = num_prios_val,\
  52         .caps = caps_val,\
  53         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  54         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  55 }
  56
  57 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
  58         ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
  59                  __VA_ARGS__)\
  60
  61 #define ADD_NS(def_miss_act, ...) {.type = FS_TYPE_NAMESPACE,   \
  62         .def_miss_action = def_miss_act,\
  63         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  64         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  65 }
  66
  67 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  68                                    sizeof(long))
  69
  70 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  71
  72 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  73                                .caps = (long[]) {__VA_ARGS__} }
  74
  75 #define FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  76                                            FS_CAP(flow_table_properties_nic_receive.modify_root), \
  77                                            FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  78                                            FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  79
  80 #define FS_CHAINING_CAPS_EGRESS                                                \
  81         FS_REQUIRED_CAPS(                                                      \
  82                 FS_CAP(flow_table_properties_nic_transmit.flow_modify_en),     \
  83                 FS_CAP(flow_table_properties_nic_transmit.modify_root),        \
  84                 FS_CAP(flow_table_properties_nic_transmit                      \
  85                                .identified_miss_table_mode),                   \
  86                 FS_CAP(flow_table_properties_nic_transmit.flow_table_modify))
  87
  88 #define FS_CHAINING_CAPS_RDMA_TX                                                \
  89         FS_REQUIRED_CAPS(                                                       \
  90                 FS_CAP(flow_table_properties_nic_transmit_rdma.flow_modify_en), \
  91                 FS_CAP(flow_table_properties_nic_transmit_rdma.modify_root),    \
  92                 FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  93                                .identified_miss_table_mode),                    \
  94                 FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  95                                .flow_table_modify))
  96
  97 #define LEFTOVERS_NUM_LEVELS 1
  98 #define LEFTOVERS_NUM_PRIOS 1
  99
 100 #define RDMA_RX_COUNTERS_PRIO_NUM_LEVELS 1
 101 #define RDMA_TX_COUNTERS_PRIO_NUM_LEVELS 1
 102
 103 #define BY_PASS_PRIO_NUM_LEVELS 1
 104 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
 105                            LEFTOVERS_NUM_PRIOS)
 106
 107 #define ETHTOOL_PRIO_NUM_LEVELS 1
 108 #define ETHTOOL_NUM_PRIOS 11
 109 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
 110 /* Promiscuous, Vlan, mac, ttc, inner ttc, {UDP/ANY/aRFS/accel/{esp, esp_err}} */
 111 #define KERNEL_NIC_PRIO_NUM_LEVELS 7
 112 #define KERNEL_NIC_NUM_PRIOS 1
 113 /* One more level for tc */
 114 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
 115
 116 #define KERNEL_NIC_TC_NUM_PRIOS  1
 117 #define KERNEL_NIC_TC_NUM_LEVELS 3
 118
 119 #define ANCHOR_NUM_LEVELS 1
 120 #define ANCHOR_NUM_PRIOS 1
 121 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 122
 123 #define OFFLOADS_MAX_FT 2
 124 #define OFFLOADS_NUM_PRIOS 2
 125 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS)
 126
 127 #define LAG_PRIO_NUM_LEVELS 1
 128 #define LAG_NUM_PRIOS 1
 129 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 130
 131 #define KERNEL_TX_IPSEC_NUM_PRIOS  1
 132 #define KERNEL_TX_IPSEC_NUM_LEVELS 1
 133 #define KERNEL_TX_MIN_LEVEL        (KERNEL_TX_IPSEC_NUM_LEVELS)
 134
 135 struct node_caps {
 136         size_t  arr_sz;
 137         long    *caps;
 138 };
 139
 140 static struct init_tree_node {
 141         enum fs_node_type       type;
 142         struct init_tree_node *children;
 143         int ar_size;
 144         struct node_caps caps;
 145         int min_ft_level;
 146         int num_leaf_prios;
 147         int prio;
 148         int num_levels;
 149         enum mlx5_flow_table_miss_action def_miss_action;
 150 } root_fs = {
 151         .type = FS_TYPE_NAMESPACE,
 152         .ar_size = 7,
 153           .children = (struct init_tree_node[]){
 154                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 155                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 156                                   ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 157                                                     BY_PASS_PRIO_NUM_LEVELS))),
 158                   ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 159                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 160                                   ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 161                                                     LAG_PRIO_NUM_LEVELS))),
 162                   ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 163                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 164                                   ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS,
 165                                                     OFFLOADS_MAX_FT))),
 166                   ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 167                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 168                                   ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 169                                                     ETHTOOL_PRIO_NUM_LEVELS))),
 170                   ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 171                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 172                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS,
 173                                                     KERNEL_NIC_TC_NUM_LEVELS),
 174                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 175                                                     KERNEL_NIC_PRIO_NUM_LEVELS))),
 176                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 177                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 178                                   ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS,
 179                                                     LEFTOVERS_NUM_LEVELS))),
 180                   ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 181                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 182                                   ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS,
 183                                                     ANCHOR_NUM_LEVELS))),
 184         }
 185 };
 186
 187 static struct init_tree_node egress_root_fs = {
 188         .type = FS_TYPE_NAMESPACE,
 189         .ar_size = 2,
 190         .children = (struct init_tree_node[]) {
 191                 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 192                          FS_CHAINING_CAPS_EGRESS,
 193                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 194                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 195                                                   BY_PASS_PRIO_NUM_LEVELS))),
 196                 ADD_PRIO(0, KERNEL_TX_MIN_LEVEL, 0,
 197                          FS_CHAINING_CAPS_EGRESS,
 198                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 199                                 ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS,
 200                                                   KERNEL_TX_IPSEC_NUM_LEVELS))),
 201         }
 202 };
 203
 204 enum {
 205         RDMA_RX_COUNTERS_PRIO,
 206         RDMA_RX_BYPASS_PRIO,
 207         RDMA_RX_KERNEL_PRIO,
 208 };
 209
 210 #define RDMA_RX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_REGULAR_PRIOS
 211 #define RDMA_RX_KERNEL_MIN_LEVEL (RDMA_RX_BYPASS_MIN_LEVEL + 1)
 212 #define RDMA_RX_COUNTERS_MIN_LEVEL (RDMA_RX_KERNEL_MIN_LEVEL + 2)
 213
 214 static struct init_tree_node rdma_rx_root_fs = {
 215         .type = FS_TYPE_NAMESPACE,
 216         .ar_size = 3,
 217         .children = (struct init_tree_node[]) {
 218                 [RDMA_RX_COUNTERS_PRIO] =
 219                 ADD_PRIO(0, RDMA_RX_COUNTERS_MIN_LEVEL, 0,
 220                          FS_CHAINING_CAPS,
 221                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 222                                 ADD_MULTIPLE_PRIO(MLX5_RDMA_RX_NUM_COUNTERS_PRIOS,
 223                                                   RDMA_RX_COUNTERS_PRIO_NUM_LEVELS))),
 224                 [RDMA_RX_BYPASS_PRIO] =
 225                 ADD_PRIO(0, RDMA_RX_BYPASS_MIN_LEVEL, 0,
 226                          FS_CHAINING_CAPS,
 227                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 228                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS,
 229                                                   BY_PASS_PRIO_NUM_LEVELS))),
 230                 [RDMA_RX_KERNEL_PRIO] =
 231                 ADD_PRIO(0, RDMA_RX_KERNEL_MIN_LEVEL, 0,
 232                          FS_CHAINING_CAPS,
 233                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN,
 234                                 ADD_MULTIPLE_PRIO(1, 1))),
 235         }
 236 };
 237
 238 enum {
 239         RDMA_TX_COUNTERS_PRIO,
 240         RDMA_TX_BYPASS_PRIO,
 241 };
 242
 243 #define RDMA_TX_BYPASS_MIN_LEVEL MLX5_BY_PASS_NUM_PRIOS
 244 #define RDMA_TX_COUNTERS_MIN_LEVEL (RDMA_TX_BYPASS_MIN_LEVEL + 1)
 245
 246 static struct init_tree_node rdma_tx_root_fs = {
 247         .type = FS_TYPE_NAMESPACE,
 248         .ar_size = 2,
 249         .children = (struct init_tree_node[]) {
 250                 [RDMA_TX_COUNTERS_PRIO] =
 251                 ADD_PRIO(0, RDMA_TX_COUNTERS_MIN_LEVEL, 0,
 252                          FS_CHAINING_CAPS,
 253                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 254                                 ADD_MULTIPLE_PRIO(MLX5_RDMA_TX_NUM_COUNTERS_PRIOS,
 255                                                   RDMA_TX_COUNTERS_PRIO_NUM_LEVELS))),
 256                 [RDMA_TX_BYPASS_PRIO] =
 257                 ADD_PRIO(0, RDMA_TX_BYPASS_MIN_LEVEL, 0,
 258                          FS_CHAINING_CAPS_RDMA_TX,
 259                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 260                                 ADD_MULTIPLE_PRIO(RDMA_TX_BYPASS_MIN_LEVEL,
 261                                                   BY_PASS_PRIO_NUM_LEVELS))),
 262         }
 263 };
 264
 265 enum fs_i_lock_class {
 266         FS_LOCK_GRANDPARENT,
 267         FS_LOCK_PARENT,
 268         FS_LOCK_CHILD
 269 };
 270
 271 static const struct rhashtable_params rhash_fte = {
 272         .key_len = sizeof_field(struct fs_fte, val),
 273         .key_offset = offsetof(struct fs_fte, val),
 274         .head_offset = offsetof(struct fs_fte, hash),
 275         .automatic_shrinking = true,
 276         .min_size = 1,
 277 };
 278
 279 static const struct rhashtable_params rhash_fg = {
 280         .key_len = sizeof_field(struct mlx5_flow_group, mask),
 281         .key_offset = offsetof(struct mlx5_flow_group, mask),
 282         .head_offset = offsetof(struct mlx5_flow_group, hash),
 283         .automatic_shrinking = true,
 284         .min_size = 1,
 285
 286 };
 287
 288 static void del_hw_flow_table(struct fs_node *node);
 289 static void del_hw_flow_group(struct fs_node *node);
 290 static void del_hw_fte(struct fs_node *node);
 291 static void del_sw_flow_table(struct fs_node *node);
 292 static void del_sw_flow_group(struct fs_node *node);
 293 static void del_sw_fte(struct fs_node *node);
 294 static void del_sw_prio(struct fs_node *node);
 295 static void del_sw_ns(struct fs_node *node);
 296 /* Delete rule (destination) is special case that
 297  * requires to lock the FTE for all the deletion process.
 298  */
 299 static void del_sw_hw_rule(struct fs_node *node);
 300 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 301                                 struct mlx5_flow_destination *d2);
 302 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 303 static struct mlx5_flow_rule *
 304 find_flow_rule(struct fs_fte *fte,
 305                struct mlx5_flow_destination *dest);
 306
 307 static void tree_init_node(struct fs_node *node,
 308                            void (*del_hw_func)(struct fs_node *),
 309                            void (*del_sw_func)(struct fs_node *))
 310 {
 311         refcount_set(&node->refcount, 1);
 312         INIT_LIST_HEAD(&node->list);
 313         INIT_LIST_HEAD(&node->children);
 314         init_rwsem(&node->lock);
 315         node->del_hw_func = del_hw_func;
 316         node->del_sw_func = del_sw_func;
 317         node->active = false;
 318 }
 319
 320 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 321 {
 322         if (parent)
 323                 refcount_inc(&parent->refcount);
 324         node->parent = parent;
 325
 326         /* Parent is the root */
 327         if (!parent)
 328                 node->root = node;
 329         else
 330                 node->root = parent->root;
 331 }
 332
 333 static int tree_get_node(struct fs_node *node)
 334 {
 335         return refcount_inc_not_zero(&node->refcount);
 336 }
 337
 338 static void nested_down_read_ref_node(struct fs_node *node,
 339                                       enum fs_i_lock_class class)
 340 {
 341         if (node) {
 342                 down_read_nested(&node->lock, class);
 343                 refcount_inc(&node->refcount);
 344         }
 345 }
 346
 347 static void nested_down_write_ref_node(struct fs_node *node,
 348                                        enum fs_i_lock_class class)
 349 {
 350         if (node) {
 351                 down_write_nested(&node->lock, class);
 352                 refcount_inc(&node->refcount);
 353         }
 354 }
 355
 356 static void down_write_ref_node(struct fs_node *node, bool locked)
 357 {
 358         if (node) {
 359                 if (!locked)
 360                         down_write(&node->lock);
 361                 refcount_inc(&node->refcount);
 362         }
 363 }
 364
 365 static void up_read_ref_node(struct fs_node *node)
 366 {
 367         refcount_dec(&node->refcount);
 368         up_read(&node->lock);
 369 }
 370
 371 static void up_write_ref_node(struct fs_node *node, bool locked)
 372 {
 373         refcount_dec(&node->refcount);
 374         if (!locked)
 375                 up_write(&node->lock);
 376 }
 377
 378 static void tree_put_node(struct fs_node *node, bool locked)
 379 {
 380         struct fs_node *parent_node = node->parent;
 381
 382         if (refcount_dec_and_test(&node->refcount)) {
 383                 if (node->del_hw_func)
 384                         node->del_hw_func(node);
 385                 if (parent_node) {
 386                         down_write_ref_node(parent_node, locked);
 387                         list_del_init(&node->list);
 388                 }
 389                 node->del_sw_func(node);
 390                 if (parent_node)
 391                         up_write_ref_node(parent_node, locked);
 392                 node = NULL;
 393         }
 394         if (!node && parent_node)
 395                 tree_put_node(parent_node, locked);
 396 }
 397
 398 static int tree_remove_node(struct fs_node *node, bool locked)
 399 {
 400         if (refcount_read(&node->refcount) > 1) {
 401                 refcount_dec(&node->refcount);
 402                 return -EEXIST;
 403         }
 404         tree_put_node(node, locked);
 405         return 0;
 406 }
 407
 408 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 409                                  unsigned int prio)
 410 {
 411         struct fs_prio *iter_prio;
 412
 413         fs_for_each_prio(iter_prio, ns) {
 414                 if (iter_prio->prio == prio)
 415                         return iter_prio;
 416         }
 417
 418         return NULL;
 419 }
 420
 421 static bool is_fwd_next_action(u32 action)
 422 {
 423         return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
 424                          MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
 425 }
 426
 427 static bool is_fwd_dest_type(enum mlx5_flow_destination_type type)
 428 {
 429         return type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM ||
 430                 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE ||
 431                 type == MLX5_FLOW_DESTINATION_TYPE_UPLINK ||
 432                 type == MLX5_FLOW_DESTINATION_TYPE_VPORT ||
 433                 type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER ||
 434                 type == MLX5_FLOW_DESTINATION_TYPE_TIR;
 435 }
 436
 437 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 438 {
 439         int i;
 440
 441         for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 442                 if (spec->match_value[i] & ~spec->match_criteria[i]) {
 443                         pr_warn("mlx5_core: match_value differs from match_criteria\n");
 444                         return false;
 445                 }
 446
 447         return true;
 448 }
 449
 450 struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 451 {
 452         struct fs_node *root;
 453         struct mlx5_flow_namespace *ns;
 454
 455         root = node->root;
 456
 457         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 458                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 459                 return NULL;
 460         }
 461
 462         ns = container_of(root, struct mlx5_flow_namespace, node);
 463         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 464 }
 465
 466 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 467 {
 468         struct mlx5_flow_root_namespace *root = find_root(node);
 469
 470         if (root)
 471                 return root->dev->priv.steering;
 472         return NULL;
 473 }
 474
 475 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 476 {
 477         struct mlx5_flow_root_namespace *root = find_root(node);
 478
 479         if (root)
 480                 return root->dev;
 481         return NULL;
 482 }
 483
 484 static void del_sw_ns(struct fs_node *node)
 485 {
 486         kfree(node);
 487 }
 488
 489 static void del_sw_prio(struct fs_node *node)
 490 {
 491         kfree(node);
 492 }
 493
 494 static void del_hw_flow_table(struct fs_node *node)
 495 {
 496         struct mlx5_flow_root_namespace *root;
 497         struct mlx5_flow_table *ft;
 498         struct mlx5_core_dev *dev;
 499         int err;
 500
 501         fs_get_obj(ft, node);
 502         dev = get_dev(&ft->node);
 503         root = find_root(&ft->node);
 504         trace_mlx5_fs_del_ft(ft);
 505
 506         if (node->active) {
 507                 err = root->cmds->destroy_flow_table(root, ft);
 508                 if (err)
 509                         mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 510         }
 511 }
 512
 513 static void del_sw_flow_table(struct fs_node *node)
 514 {
 515         struct mlx5_flow_table *ft;
 516         struct fs_prio *prio;
 517
 518         fs_get_obj(ft, node);
 519
 520         rhltable_destroy(&ft->fgs_hash);
 521         if (ft->node.parent) {
 522                 fs_get_obj(prio, ft->node.parent);
 523                 prio->num_ft--;
 524         }
 525         kfree(ft);
 526 }
 527
 528 static void modify_fte(struct fs_fte *fte)
 529 {
 530         struct mlx5_flow_root_namespace *root;
 531         struct mlx5_flow_table *ft;
 532         struct mlx5_flow_group *fg;
 533         struct mlx5_core_dev *dev;
 534         int err;
 535
 536         fs_get_obj(fg, fte->node.parent);
 537         fs_get_obj(ft, fg->node.parent);
 538         dev = get_dev(&fte->node);
 539
 540         root = find_root(&ft->node);
 541         err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte);
 542         if (err)
 543                 mlx5_core_warn(dev,
 544                                "%s can't del rule fg id=%d fte_index=%d\n",
 545                                __func__, fg->id, fte->index);
 546         fte->modify_mask = 0;
 547 }
 548
 549 static void del_sw_hw_rule(struct fs_node *node)
 550 {
 551         struct mlx5_flow_rule *rule;
 552         struct fs_fte *fte;
 553
 554         fs_get_obj(rule, node);
 555         fs_get_obj(fte, rule->node.parent);
 556         trace_mlx5_fs_del_rule(rule);
 557         if (is_fwd_next_action(rule->sw_action)) {
 558                 mutex_lock(&rule->dest_attr.ft->lock);
 559                 list_del(&rule->next_ft);
 560                 mutex_unlock(&rule->dest_attr.ft->lock);
 561         }
 562
 563         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER) {
 564                 --fte->dests_size;
 565                 fte->modify_mask |=
 566                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 567                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 568                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 569                 goto out;
 570         }
 571
 572         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT) {
 573                 --fte->dests_size;
 574                 fte->modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
 575                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW;
 576                 goto out;
 577         }
 578
 579         if (is_fwd_dest_type(rule->dest_attr.type)) {
 580                 --fte->dests_size;
 581                 --fte->fwd_dests;
 582
 583                 if (!fte->fwd_dests)
 584                         fte->action.action &=
 585                                 ~MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
 586                 fte->modify_mask |=
 587                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 588                 goto out;
 589         }
 590 out:
 591         kfree(rule);
 592 }
 593
 594 static void del_hw_fte(struct fs_node *node)
 595 {
 596         struct mlx5_flow_root_namespace *root;
 597         struct mlx5_flow_table *ft;
 598         struct mlx5_flow_group *fg;
 599         struct mlx5_core_dev *dev;
 600         struct fs_fte *fte;
 601         int err;
 602
 603         fs_get_obj(fte, node);
 604         fs_get_obj(fg, fte->node.parent);
 605         fs_get_obj(ft, fg->node.parent);
 606
 607         trace_mlx5_fs_del_fte(fte);
 608         WARN_ON(fte->dests_size);
 609         dev = get_dev(&ft->node);
 610         root = find_root(&ft->node);
 611         if (node->active) {
 612                 err = root->cmds->delete_fte(root, ft, fte);
 613                 if (err)
 614                         mlx5_core_warn(dev,
 615                                        "flow steering can't delete fte in index %d of flow group id %d\n",
 616                                        fte->index, fg->id);
 617                 node->active = false;
 618         }
 619 }
 620
 621 static void del_sw_fte(struct fs_node *node)
 622 {
 623         struct mlx5_flow_steering *steering = get_steering(node);
 624         struct mlx5_flow_group *fg;
 625         struct fs_fte *fte;
 626         int err;
 627
 628         fs_get_obj(fte, node);
 629         fs_get_obj(fg, fte->node.parent);
 630
 631         err = rhashtable_remove_fast(&fg->ftes_hash,
 632                                      &fte->hash,
 633                                      rhash_fte);
 634         WARN_ON(err);
 635         ida_free(&fg->fte_allocator, fte->index - fg->start_index);
 636         kmem_cache_free(steering->ftes_cache, fte);
 637 }
 638
 639 static void del_hw_flow_group(struct fs_node *node)
 640 {
 641         struct mlx5_flow_root_namespace *root;
 642         struct mlx5_flow_group *fg;
 643         struct mlx5_flow_table *ft;
 644         struct mlx5_core_dev *dev;
 645
 646         fs_get_obj(fg, node);
 647         fs_get_obj(ft, fg->node.parent);
 648         dev = get_dev(&ft->node);
 649         trace_mlx5_fs_del_fg(fg);
 650
 651         root = find_root(&ft->node);
 652         if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
 653                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 654                                fg->id, ft->id);
 655 }
 656
 657 static void del_sw_flow_group(struct fs_node *node)
 658 {
 659         struct mlx5_flow_steering *steering = get_steering(node);
 660         struct mlx5_flow_group *fg;
 661         struct mlx5_flow_table *ft;
 662         int err;
 663
 664         fs_get_obj(fg, node);
 665         fs_get_obj(ft, fg->node.parent);
 666
 667         rhashtable_destroy(&fg->ftes_hash);
 668         ida_destroy(&fg->fte_allocator);
 669         if (ft->autogroup.active &&
 670             fg->max_ftes == ft->autogroup.group_size &&
 671             fg->start_index < ft->autogroup.max_fte)
 672                 ft->autogroup.num_groups--;
 673         err = rhltable_remove(&ft->fgs_hash,
 674                               &fg->hash,
 675                               rhash_fg);
 676         WARN_ON(err);
 677         kmem_cache_free(steering->fgs_cache, fg);
 678 }
 679
 680 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 681 {
 682         int index;
 683         int ret;
 684
 685         index = ida_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, GFP_KERNEL);
 686         if (index < 0)
 687                 return index;
 688
 689         fte->index = index + fg->start_index;
 690         ret = rhashtable_insert_fast(&fg->ftes_hash,
 691                                      &fte->hash,
 692                                      rhash_fte);
 693         if (ret)
 694                 goto err_ida_remove;
 695
 696         tree_add_node(&fte->node, &fg->node);
 697         list_add_tail(&fte->node.list, &fg->node.children);
 698         return 0;
 699
 700 err_ida_remove:
 701         ida_free(&fg->fte_allocator, index);
 702         return ret;
 703 }
 704
 705 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 706                                 const struct mlx5_flow_spec *spec,
 707                                 struct mlx5_flow_act *flow_act)
 708 {
 709         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 710         struct fs_fte *fte;
 711
 712         fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 713         if (!fte)
 714                 return ERR_PTR(-ENOMEM);
 715
 716         memcpy(fte->val, &spec->match_value, sizeof(fte->val));
 717         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 718         fte->action = *flow_act;
 719         fte->flow_context = spec->flow_context;
 720
 721         tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
 722
 723         return fte;
 724 }
 725
 726 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 727                                struct mlx5_flow_group *fg)
 728 {
 729         rhashtable_destroy(&fg->ftes_hash);
 730         kmem_cache_free(steering->fgs_cache, fg);
 731 }
 732
 733 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 734                                                 u8 match_criteria_enable,
 735                                                 const void *match_criteria,
 736                                                 int start_index,
 737                                                 int end_index)
 738 {
 739         struct mlx5_flow_group *fg;
 740         int ret;
 741
 742         fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 743         if (!fg)
 744                 return ERR_PTR(-ENOMEM);
 745
 746         ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 747         if (ret) {
 748                 kmem_cache_free(steering->fgs_cache, fg);
 749                 return ERR_PTR(ret);
 750         }
 751
 752         ida_init(&fg->fte_allocator);
 753         fg->mask.match_criteria_enable = match_criteria_enable;
 754         memcpy(&fg->mask.match_criteria, match_criteria,
 755                sizeof(fg->mask.match_criteria));
 756         fg->node.type =  FS_TYPE_FLOW_GROUP;
 757         fg->start_index = start_index;
 758         fg->max_ftes = end_index - start_index + 1;
 759
 760         return fg;
 761 }
 762
 763 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 764                                                        u8 match_criteria_enable,
 765                                                        const void *match_criteria,
 766                                                        int start_index,
 767                                                        int end_index,
 768                                                        struct list_head *prev)
 769 {
 770         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 771         struct mlx5_flow_group *fg;
 772         int ret;
 773
 774         fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 775                               start_index, end_index);
 776         if (IS_ERR(fg))
 777                 return fg;
 778
 779         /* initialize refcnt, add to parent list */
 780         ret = rhltable_insert(&ft->fgs_hash,
 781                               &fg->hash,
 782                               rhash_fg);
 783         if (ret) {
 784                 dealloc_flow_group(steering, fg);
 785                 return ERR_PTR(ret);
 786         }
 787
 788         tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 789         tree_add_node(&fg->node, &ft->node);
 790         /* Add node to group list */
 791         list_add(&fg->node.list, prev);
 792         atomic_inc(&ft->node.version);
 793
 794         return fg;
 795 }
 796
 797 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport,
 798                                                 enum fs_flow_table_type table_type,
 799                                                 enum fs_flow_table_op_mod op_mod,
 800                                                 u32 flags)
 801 {
 802         struct mlx5_flow_table *ft;
 803         int ret;
 804
 805         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 806         if (!ft)
 807                 return ERR_PTR(-ENOMEM);
 808
 809         ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 810         if (ret) {
 811                 kfree(ft);
 812                 return ERR_PTR(ret);
 813         }
 814
 815         ft->level = level;
 816         ft->node.type = FS_TYPE_FLOW_TABLE;
 817         ft->op_mod = op_mod;
 818         ft->type = table_type;
 819         ft->vport = vport;
 820         ft->flags = flags;
 821         INIT_LIST_HEAD(&ft->fwd_rules);
 822         mutex_init(&ft->lock);
 823
 824         return ft;
 825 }
 826
 827 /* If reverse is false, then we search for the first flow table in the
 828  * root sub-tree from start(closest from right), else we search for the
 829  * last flow table in the root sub-tree till start(closest from left).
 830  */
 831 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 832                                                          struct list_head *start,
 833                                                          bool reverse)
 834 {
 835 #define list_advance_entry(pos, reverse)                \
 836         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 837
 838 #define list_for_each_advance_continue(pos, head, reverse)      \
 839         for (pos = list_advance_entry(pos, reverse);            \
 840              &pos->list != (head);                              \
 841              pos = list_advance_entry(pos, reverse))
 842
 843         struct fs_node *iter = list_entry(start, struct fs_node, list);
 844         struct mlx5_flow_table *ft = NULL;
 845
 846         if (!root || root->type == FS_TYPE_PRIO_CHAINS)
 847                 return NULL;
 848
 849         list_for_each_advance_continue(iter, &root->children, reverse) {
 850                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 851                         fs_get_obj(ft, iter);
 852                         return ft;
 853                 }
 854                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 855                 if (ft)
 856                         return ft;
 857         }
 858
 859         return ft;
 860 }
 861
 862 /* If reverse is false then return the first flow table in next priority of
 863  * prio in the tree, else return the last flow table in the previous priority
 864  * of prio in the tree.
 865  */
 866 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 867 {
 868         struct mlx5_flow_table *ft = NULL;
 869         struct fs_node *curr_node;
 870         struct fs_node *parent;
 871
 872         parent = prio->node.parent;
 873         curr_node = &prio->node;
 874         while (!ft && parent) {
 875                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 876                 curr_node = parent;
 877                 parent = curr_node->parent;
 878         }
 879         return ft;
 880 }
 881
 882 /* Assuming all the tree is locked by mutex chain lock */
 883 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 884 {
 885         return find_closest_ft(prio, false);
 886 }
 887
 888 /* Assuming all the tree is locked by mutex chain lock */
 889 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 890 {
 891         return find_closest_ft(prio, true);
 892 }
 893
 894 static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft,
 895                                                 struct mlx5_flow_act *flow_act)
 896 {
 897         struct fs_prio *prio;
 898         bool next_ns;
 899
 900         next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
 901         fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent);
 902
 903         return find_next_chained_ft(prio);
 904 }
 905
 906 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 907                                struct fs_prio *prio,
 908                                struct mlx5_flow_table *ft)
 909 {
 910         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 911         struct mlx5_flow_table *iter;
 912         int err;
 913
 914         fs_for_each_ft(iter, prio) {
 915                 err = root->cmds->modify_flow_table(root, iter, ft);
 916                 if (err) {
 917                         mlx5_core_err(dev,
 918                                       "Failed to modify flow table id %d, type %d, err %d\n",
 919                                       iter->id, iter->type, err);
 920                         /* The driver is out of sync with the FW */
 921                         return err;
 922                 }
 923         }
 924         return 0;
 925 }
 926
 927 /* Connect flow tables from previous priority of prio to ft */
 928 static int connect_prev_fts(struct mlx5_core_dev *dev,
 929                             struct mlx5_flow_table *ft,
 930                             struct fs_prio *prio)
 931 {
 932         struct mlx5_flow_table *prev_ft;
 933
 934         prev_ft = find_prev_chained_ft(prio);
 935         if (prev_ft) {
 936                 struct fs_prio *prev_prio;
 937
 938                 fs_get_obj(prev_prio, prev_ft->node.parent);
 939                 return connect_fts_in_prio(dev, prev_prio, ft);
 940         }
 941         return 0;
 942 }
 943
 944 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 945                                  *prio)
 946 {
 947         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 948         struct mlx5_ft_underlay_qp *uqp;
 949         int min_level = INT_MAX;
 950         int err = 0;
 951         u32 qpn;
 952
 953         if (root->root_ft)
 954                 min_level = root->root_ft->level;
 955
 956         if (ft->level >= min_level)
 957                 return 0;
 958
 959         if (list_empty(&root->underlay_qpns)) {
 960                 /* Don't set any QPN (zero) in case QPN list is empty */
 961                 qpn = 0;
 962                 err = root->cmds->update_root_ft(root, ft, qpn, false);
 963         } else {
 964                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
 965                         qpn = uqp->qpn;
 966                         err = root->cmds->update_root_ft(root, ft,
 967                                                          qpn, false);
 968                         if (err)
 969                                 break;
 970                 }
 971         }
 972
 973         if (err)
 974                 mlx5_core_warn(root->dev,
 975                                "Update root flow table of id(%u) qpn(%d) failed\n",
 976                                ft->id, qpn);
 977         else
 978                 root->root_ft = ft;
 979
 980         return err;
 981 }
 982
 983 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 984                                          struct mlx5_flow_destination *dest)
 985 {
 986         struct mlx5_flow_root_namespace *root;
 987         struct mlx5_flow_table *ft;
 988         struct mlx5_flow_group *fg;
 989         struct fs_fte *fte;
 990         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 991         int err = 0;
 992
 993         fs_get_obj(fte, rule->node.parent);
 994         if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 995                 return -EINVAL;
 996         down_write_ref_node(&fte->node, false);
 997         fs_get_obj(fg, fte->node.parent);
 998         fs_get_obj(ft, fg->node.parent);
 999
1000         memcpy(&rule->dest_attr, dest, sizeof(*dest));
1001         root = find_root(&ft->node);
1002         err = root->cmds->update_fte(root, ft, fg,
1003                                      modify_mask, fte);
1004         up_write_ref_node(&fte->node, false);
1005
1006         return err;
1007 }
1008
1009 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
1010                                  struct mlx5_flow_destination *new_dest,
1011                                  struct mlx5_flow_destination *old_dest)
1012 {
1013         int i;
1014
1015         if (!old_dest) {
1016                 if (handle->num_rules != 1)
1017                         return -EINVAL;
1018                 return _mlx5_modify_rule_destination(handle->rule[0],
1019                                                      new_dest);
1020         }
1021
1022         for (i = 0; i < handle->num_rules; i++) {
1023                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
1024                         return _mlx5_modify_rule_destination(handle->rule[i],
1025                                                              new_dest);
1026         }
1027
1028         return -EINVAL;
1029 }
1030
1031 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
1032 static int connect_fwd_rules(struct mlx5_core_dev *dev,
1033                              struct mlx5_flow_table *new_next_ft,
1034                              struct mlx5_flow_table *old_next_ft)
1035 {
1036         struct mlx5_flow_destination dest = {};
1037         struct mlx5_flow_rule *iter;
1038         int err = 0;
1039
1040         /* new_next_ft and old_next_ft could be NULL only
1041          * when we create/destroy the anchor flow table.
1042          */
1043         if (!new_next_ft || !old_next_ft)
1044                 return 0;
1045
1046         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1047         dest.ft = new_next_ft;
1048
1049         mutex_lock(&old_next_ft->lock);
1050         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
1051         mutex_unlock(&old_next_ft->lock);
1052         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
1053                 if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) &&
1054                     iter->ft->ns == new_next_ft->ns)
1055                         continue;
1056
1057                 err = _mlx5_modify_rule_destination(iter, &dest);
1058                 if (err)
1059                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
1060                                new_next_ft->id);
1061         }
1062         return 0;
1063 }
1064
1065 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
1066                               struct fs_prio *prio)
1067 {
1068         struct mlx5_flow_table *next_ft, *first_ft;
1069         int err = 0;
1070
1071         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
1072
1073         first_ft = list_first_entry_or_null(&prio->node.children,
1074                                             struct mlx5_flow_table, node.list);
1075         if (!first_ft || first_ft->level > ft->level) {
1076                 err = connect_prev_fts(dev, ft, prio);
1077                 if (err)
1078                         return err;
1079
1080                 next_ft = first_ft ? first_ft : find_next_chained_ft(prio);
1081                 err = connect_fwd_rules(dev, ft, next_ft);
1082                 if (err)
1083                         return err;
1084         }
1085
1086         if (MLX5_CAP_FLOWTABLE(dev,
1087                                flow_table_properties_nic_receive.modify_root))
1088                 err = update_root_ft_create(ft, prio);
1089         return err;
1090 }
1091
1092 static void list_add_flow_table(struct mlx5_flow_table *ft,
1093                                 struct fs_prio *prio)
1094 {
1095         struct list_head *prev = &prio->node.children;
1096         struct mlx5_flow_table *iter;
1097
1098         fs_for_each_ft(iter, prio) {
1099                 if (iter->level > ft->level)
1100                         break;
1101                 prev = &iter->node.list;
1102         }
1103         list_add(&ft->node.list, prev);
1104 }
1105
1106 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1107                                                         struct mlx5_flow_table_attr *ft_attr,
1108                                                         enum fs_flow_table_op_mod op_mod,
1109                                                         u16 vport)
1110 {
1111         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1112         bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED;
1113         struct mlx5_flow_table *next_ft;
1114         struct fs_prio *fs_prio = NULL;
1115         struct mlx5_flow_table *ft;
1116         int err;
1117
1118         if (!root) {
1119                 pr_err("mlx5: flow steering failed to find root of namespace\n");
1120                 return ERR_PTR(-ENODEV);
1121         }
1122
1123         mutex_lock(&root->chain_lock);
1124         fs_prio = find_prio(ns, ft_attr->prio);
1125         if (!fs_prio) {
1126                 err = -EINVAL;
1127                 goto unlock_root;
1128         }
1129         if (!unmanaged) {
1130                 /* The level is related to the
1131                  * priority level range.
1132                  */
1133                 if (ft_attr->level >= fs_prio->num_levels) {
1134                         err = -ENOSPC;
1135                         goto unlock_root;
1136                 }
1137
1138                 ft_attr->level += fs_prio->start_level;
1139         }
1140
1141         /* The level is related to the
1142          * priority level range.
1143          */
1144         ft = alloc_flow_table(ft_attr->level,
1145                               vport,
1146                               root->table_type,
1147                               op_mod, ft_attr->flags);
1148         if (IS_ERR(ft)) {
1149                 err = PTR_ERR(ft);
1150                 goto unlock_root;
1151         }
1152
1153         tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1154         next_ft = unmanaged ? ft_attr->next_ft :
1155                               find_next_chained_ft(fs_prio);
1156         ft->def_miss_action = ns->def_miss_action;
1157         ft->ns = ns;
1158         err = root->cmds->create_flow_table(root, ft, ft_attr->max_fte, next_ft);
1159         if (err)
1160                 goto free_ft;
1161
1162         if (!unmanaged) {
1163                 err = connect_flow_table(root->dev, ft, fs_prio);
1164                 if (err)
1165                         goto destroy_ft;
1166         }
1167
1168         ft->node.active = true;
1169         down_write_ref_node(&fs_prio->node, false);
1170         if (!unmanaged) {
1171                 tree_add_node(&ft->node, &fs_prio->node);
1172                 list_add_flow_table(ft, fs_prio);
1173         } else {
1174                 ft->node.root = fs_prio->node.root;
1175         }
1176         fs_prio->num_ft++;
1177         up_write_ref_node(&fs_prio->node, false);
1178         mutex_unlock(&root->chain_lock);
1179         trace_mlx5_fs_add_ft(ft);
1180         return ft;
1181 destroy_ft:
1182         root->cmds->destroy_flow_table(root, ft);
1183 free_ft:
1184         rhltable_destroy(&ft->fgs_hash);
1185         kfree(ft);
1186 unlock_root:
1187         mutex_unlock(&root->chain_lock);
1188         return ERR_PTR(err);
1189 }
1190
1191 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1192                                                struct mlx5_flow_table_attr *ft_attr)
1193 {
1194         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1195 }
1196 EXPORT_SYMBOL(mlx5_create_flow_table);
1197
1198 struct mlx5_flow_table *
1199 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1200                              struct mlx5_flow_table_attr *ft_attr, u16 vport)
1201 {
1202         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1203 }
1204
1205 struct mlx5_flow_table*
1206 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1207                                  int prio, u32 level)
1208 {
1209         struct mlx5_flow_table_attr ft_attr = {};
1210
1211         ft_attr.level = level;
1212         ft_attr.prio  = prio;
1213         ft_attr.max_fte = 1;
1214
1215         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1216 }
1217 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1218
1219 #define MAX_FLOW_GROUP_SIZE BIT(24)
1220 struct mlx5_flow_table*
1221 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1222                                     struct mlx5_flow_table_attr *ft_attr)
1223 {
1224         int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1225         int max_num_groups = ft_attr->autogroup.max_num_groups;
1226         struct mlx5_flow_table *ft;
1227         int autogroups_max_fte;
1228
1229         ft = mlx5_create_flow_table(ns, ft_attr);
1230         if (IS_ERR(ft))
1231                 return ft;
1232
1233         autogroups_max_fte = ft->max_fte - num_reserved_entries;
1234         if (max_num_groups > autogroups_max_fte)
1235                 goto err_validate;
1236         if (num_reserved_entries > ft->max_fte)
1237                 goto err_validate;
1238
1239         /* Align the number of groups according to the largest group size */
1240         if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE)
1241                 max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1;
1242
1243         ft->autogroup.active = true;
1244         ft->autogroup.required_groups = max_num_groups;
1245         ft->autogroup.max_fte = autogroups_max_fte;
1246         /* We save place for flow groups in addition to max types */
1247         ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1248
1249         return ft;
1250
1251 err_validate:
1252         mlx5_destroy_flow_table(ft);
1253         return ERR_PTR(-ENOSPC);
1254 }
1255 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1256
1257 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1258                                                u32 *fg_in)
1259 {
1260         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1261         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1262                                             fg_in, match_criteria);
1263         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1264                                             fg_in,
1265                                             match_criteria_enable);
1266         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1267                                    start_flow_index);
1268         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1269                                  end_flow_index);
1270         struct mlx5_flow_group *fg;
1271         int err;
1272
1273         if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1274                 return ERR_PTR(-EPERM);
1275
1276         down_write_ref_node(&ft->node, false);
1277         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1278                                      start_index, end_index,
1279                                      ft->node.children.prev);
1280         up_write_ref_node(&ft->node, false);
1281         if (IS_ERR(fg))
1282                 return fg;
1283
1284         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1285         if (err) {
1286                 tree_put_node(&fg->node, false);
1287                 return ERR_PTR(err);
1288         }
1289         trace_mlx5_fs_add_fg(fg);
1290         fg->node.active = true;
1291
1292         return fg;
1293 }
1294 EXPORT_SYMBOL(mlx5_create_flow_group);
1295
1296 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1297 {
1298         struct mlx5_flow_rule *rule;
1299
1300         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1301         if (!rule)
1302                 return NULL;
1303
1304         INIT_LIST_HEAD(&rule->next_ft);
1305         rule->node.type = FS_TYPE_FLOW_DEST;
1306         if (dest)
1307                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1308         else
1309                 rule->dest_attr.type = MLX5_FLOW_DESTINATION_TYPE_NONE;
1310
1311         return rule;
1312 }
1313
1314 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1315 {
1316         struct mlx5_flow_handle *handle;
1317
1318         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1319         if (!handle)
1320                 return NULL;
1321
1322         handle->num_rules = num_rules;
1323
1324         return handle;
1325 }
1326
1327 static void destroy_flow_handle(struct fs_fte *fte,
1328                                 struct mlx5_flow_handle *handle,
1329                                 struct mlx5_flow_destination *dest,
1330                                 int i)
1331 {
1332         for (; --i >= 0;) {
1333                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1334                         fte->dests_size--;
1335                         list_del(&handle->rule[i]->node.list);
1336                         kfree(handle->rule[i]);
1337                 }
1338         }
1339         kfree(handle);
1340 }
1341
1342 static struct mlx5_flow_handle *
1343 create_flow_handle(struct fs_fte *fte,
1344                    struct mlx5_flow_destination *dest,
1345                    int dest_num,
1346                    int *modify_mask,
1347                    bool *new_rule)
1348 {
1349         struct mlx5_flow_handle *handle;
1350         struct mlx5_flow_rule *rule = NULL;
1351         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1352         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1353         int type;
1354         int i = 0;
1355
1356         handle = alloc_handle((dest_num) ? dest_num : 1);
1357         if (!handle)
1358                 return ERR_PTR(-ENOMEM);
1359
1360         do {
1361                 if (dest) {
1362                         rule = find_flow_rule(fte, dest + i);
1363                         if (rule) {
1364                                 refcount_inc(&rule->node.refcount);
1365                                 goto rule_found;
1366                         }
1367                 }
1368
1369                 *new_rule = true;
1370                 rule = alloc_rule(dest + i);
1371                 if (!rule)
1372                         goto free_rules;
1373
1374                 /* Add dest to dests list- we need flow tables to be in the
1375                  * end of the list for forward to next prio rules.
1376                  */
1377                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1378                 if (dest &&
1379                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1380                         list_add(&rule->node.list, &fte->node.children);
1381                 else
1382                         list_add_tail(&rule->node.list, &fte->node.children);
1383                 if (dest) {
1384                         fte->dests_size++;
1385
1386                         if (is_fwd_dest_type(dest[i].type))
1387                                 fte->fwd_dests++;
1388
1389                         type = dest[i].type ==
1390                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1391                         *modify_mask |= type ? count : dst;
1392                 }
1393 rule_found:
1394                 handle->rule[i] = rule;
1395         } while (++i < dest_num);
1396
1397         return handle;
1398
1399 free_rules:
1400         destroy_flow_handle(fte, handle, dest, i);
1401         return ERR_PTR(-ENOMEM);
1402 }
1403
1404 /* fte should not be deleted while calling this function */
1405 static struct mlx5_flow_handle *
1406 add_rule_fte(struct fs_fte *fte,
1407              struct mlx5_flow_group *fg,
1408              struct mlx5_flow_destination *dest,
1409              int dest_num,
1410              bool update_action)
1411 {
1412         struct mlx5_flow_root_namespace *root;
1413         struct mlx5_flow_handle *handle;
1414         struct mlx5_flow_table *ft;
1415         int modify_mask = 0;
1416         int err;
1417         bool new_rule = false;
1418
1419         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1420                                     &new_rule);
1421         if (IS_ERR(handle) || !new_rule)
1422                 goto out;
1423
1424         if (update_action)
1425                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1426
1427         fs_get_obj(ft, fg->node.parent);
1428         root = find_root(&fg->node);
1429         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1430                 err = root->cmds->create_fte(root, ft, fg, fte);
1431         else
1432                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1433         if (err)
1434                 goto free_handle;
1435
1436         fte->node.active = true;
1437         fte->status |= FS_FTE_STATUS_EXISTING;
1438         atomic_inc(&fg->node.version);
1439
1440 out:
1441         return handle;
1442
1443 free_handle:
1444         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1445         return ERR_PTR(err);
1446 }
1447
1448 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1449                                                      const struct mlx5_flow_spec *spec)
1450 {
1451         struct list_head *prev = &ft->node.children;
1452         u32 max_fte = ft->autogroup.max_fte;
1453         unsigned int candidate_index = 0;
1454         unsigned int group_size = 0;
1455         struct mlx5_flow_group *fg;
1456
1457         if (!ft->autogroup.active)
1458                 return ERR_PTR(-ENOENT);
1459
1460         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1461                 group_size = ft->autogroup.group_size;
1462
1463         /*  max_fte == ft->autogroup.max_types */
1464         if (group_size == 0)
1465                 group_size = 1;
1466
1467         /* sorted by start_index */
1468         fs_for_each_fg(fg, ft) {
1469                 if (candidate_index + group_size > fg->start_index)
1470                         candidate_index = fg->start_index + fg->max_ftes;
1471                 else
1472                         break;
1473                 prev = &fg->node.list;
1474         }
1475
1476         if (candidate_index + group_size > max_fte)
1477                 return ERR_PTR(-ENOSPC);
1478
1479         fg = alloc_insert_flow_group(ft,
1480                                      spec->match_criteria_enable,
1481                                      spec->match_criteria,
1482                                      candidate_index,
1483                                      candidate_index + group_size - 1,
1484                                      prev);
1485         if (IS_ERR(fg))
1486                 goto out;
1487
1488         if (group_size == ft->autogroup.group_size)
1489                 ft->autogroup.num_groups++;
1490
1491 out:
1492         return fg;
1493 }
1494
1495 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1496                                   struct mlx5_flow_group *fg)
1497 {
1498         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1499         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1500         void *match_criteria_addr;
1501         u8 src_esw_owner_mask_on;
1502         void *misc;
1503         int err;
1504         u32 *in;
1505
1506         in = kvzalloc(inlen, GFP_KERNEL);
1507         if (!in)
1508                 return -ENOMEM;
1509
1510         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1511                  fg->mask.match_criteria_enable);
1512         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1513         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1514                  fg->max_ftes - 1);
1515
1516         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1517                             misc_parameters);
1518         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1519                                          source_eswitch_owner_vhca_id);
1520         MLX5_SET(create_flow_group_in, in,
1521                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1522
1523         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1524                                            in, match_criteria);
1525         memcpy(match_criteria_addr, fg->mask.match_criteria,
1526                sizeof(fg->mask.match_criteria));
1527
1528         err = root->cmds->create_flow_group(root, ft, in, fg);
1529         if (!err) {
1530                 fg->node.active = true;
1531                 trace_mlx5_fs_add_fg(fg);
1532         }
1533
1534         kvfree(in);
1535         return err;
1536 }
1537
1538 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1539                                 struct mlx5_flow_destination *d2)
1540 {
1541         if (d1->type == d2->type) {
1542                 if (((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT ||
1543                       d1->type == MLX5_FLOW_DESTINATION_TYPE_UPLINK) &&
1544                      d1->vport.num == d2->vport.num &&
1545                      d1->vport.flags == d2->vport.flags &&
1546                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1547                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1548                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1549                       (d1->vport.pkt_reformat->id ==
1550                        d2->vport.pkt_reformat->id) : true)) ||
1551                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1552                      d1->ft == d2->ft) ||
1553                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1554                      d1->tir_num == d2->tir_num) ||
1555                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1556                      d1->ft_num == d2->ft_num) ||
1557                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER &&
1558                      d1->sampler_id == d2->sampler_id))
1559                         return true;
1560         }
1561
1562         return false;
1563 }
1564
1565 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1566                                              struct mlx5_flow_destination *dest)
1567 {
1568         struct mlx5_flow_rule *rule;
1569
1570         list_for_each_entry(rule, &fte->node.children, node.list) {
1571                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1572                         return rule;
1573         }
1574         return NULL;
1575 }
1576
1577 static bool check_conflicting_actions(u32 action1, u32 action2)
1578 {
1579         u32 xored_actions = action1 ^ action2;
1580
1581         /* if one rule only wants to count, it's ok */
1582         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1583             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1584                 return false;
1585
1586         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1587                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1588                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1589                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1590                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1591                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1592                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1593                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1594                 return true;
1595
1596         return false;
1597 }
1598
1599 static int check_conflicting_ftes(struct fs_fte *fte,
1600                                   const struct mlx5_flow_context *flow_context,
1601                                   const struct mlx5_flow_act *flow_act)
1602 {
1603         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1604                 mlx5_core_warn(get_dev(&fte->node),
1605                                "Found two FTEs with conflicting actions\n");
1606                 return -EEXIST;
1607         }
1608
1609         if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1610             fte->flow_context.flow_tag != flow_context->flow_tag) {
1611                 mlx5_core_warn(get_dev(&fte->node),
1612                                "FTE flow tag %u already exists with different flow tag %u\n",
1613                                fte->flow_context.flow_tag,
1614                                flow_context->flow_tag);
1615                 return -EEXIST;
1616         }
1617
1618         return 0;
1619 }
1620
1621 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1622                                             const struct mlx5_flow_spec *spec,
1623                                             struct mlx5_flow_act *flow_act,
1624                                             struct mlx5_flow_destination *dest,
1625                                             int dest_num,
1626                                             struct fs_fte *fte)
1627 {
1628         struct mlx5_flow_handle *handle;
1629         int old_action;
1630         int i;
1631         int ret;
1632
1633         ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1634         if (ret)
1635                 return ERR_PTR(ret);
1636
1637         old_action = fte->action.action;
1638         fte->action.action |= flow_act->action;
1639         handle = add_rule_fte(fte, fg, dest, dest_num,
1640                               old_action != flow_act->action);
1641         if (IS_ERR(handle)) {
1642                 fte->action.action = old_action;
1643                 return handle;
1644         }
1645         trace_mlx5_fs_set_fte(fte, false);
1646
1647         for (i = 0; i < handle->num_rules; i++) {
1648                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1649                         tree_add_node(&handle->rule[i]->node, &fte->node);
1650                         trace_mlx5_fs_add_rule(handle->rule[i]);
1651                 }
1652         }
1653         return handle;
1654 }
1655
1656 static bool counter_is_valid(u32 action)
1657 {
1658         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1659                           MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1660                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1661 }
1662
1663 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1664                           struct mlx5_flow_act *flow_act,
1665                           struct mlx5_flow_table *ft)
1666 {
1667         bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1668         u32 action = flow_act->action;
1669
1670         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1671                 return counter_is_valid(action);
1672
1673         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1674                 return true;
1675
1676         if (ignore_level) {
1677                 if (ft->type != FS_FT_FDB &&
1678                     ft->type != FS_FT_NIC_RX)
1679                         return false;
1680
1681                 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1682                     ft->type != dest->ft->type)
1683                         return false;
1684         }
1685
1686         if (!dest || ((dest->type ==
1687             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1688             (dest->ft->level <= ft->level && !ignore_level)))
1689                 return false;
1690         return true;
1691 }
1692
1693 struct match_list {
1694         struct list_head        list;
1695         struct mlx5_flow_group *g;
1696 };
1697
1698 static void free_match_list(struct match_list *head, bool ft_locked)
1699 {
1700         struct match_list *iter, *match_tmp;
1701
1702         list_for_each_entry_safe(iter, match_tmp, &head->list,
1703                                  list) {
1704                 tree_put_node(&iter->g->node, ft_locked);
1705                 list_del(&iter->list);
1706                 kfree(iter);
1707         }
1708 }
1709
1710 static int build_match_list(struct match_list *match_head,
1711                             struct mlx5_flow_table *ft,
1712                             const struct mlx5_flow_spec *spec,
1713                             struct mlx5_flow_group *fg,
1714                             bool ft_locked)
1715 {
1716         struct rhlist_head *tmp, *list;
1717         struct mlx5_flow_group *g;
1718         int err = 0;
1719
1720         rcu_read_lock();
1721         INIT_LIST_HEAD(&match_head->list);
1722         /* Collect all fgs which has a matching match_criteria */
1723         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1724         /* RCU is atomic, we can't execute FW commands here */
1725         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1726                 struct match_list *curr_match;
1727
1728                 if (fg && fg != g)
1729                         continue;
1730
1731                 if (unlikely(!tree_get_node(&g->node)))
1732                         continue;
1733
1734                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1735                 if (!curr_match) {
1736                         rcu_read_unlock();
1737                         free_match_list(match_head, ft_locked);
1738                         return -ENOMEM;
1739                 }
1740                 curr_match->g = g;
1741                 list_add_tail(&curr_match->list, &match_head->list);
1742         }
1743         rcu_read_unlock();
1744         return err;
1745 }
1746
1747 static u64 matched_fgs_get_version(struct list_head *match_head)
1748 {
1749         struct match_list *iter;
1750         u64 version = 0;
1751
1752         list_for_each_entry(iter, match_head, list)
1753                 version += (u64)atomic_read(&iter->g->node.version);
1754         return version;
1755 }
1756
1757 static struct fs_fte *
1758 lookup_fte_locked(struct mlx5_flow_group *g,
1759                   const u32 *match_value,
1760                   bool take_write)
1761 {
1762         struct fs_fte *fte_tmp;
1763
1764         if (take_write)
1765                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1766         else
1767                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1768         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1769                                          rhash_fte);
1770         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1771                 fte_tmp = NULL;
1772                 goto out;
1773         }
1774         if (!fte_tmp->node.active) {
1775                 tree_put_node(&fte_tmp->node, false);
1776                 fte_tmp = NULL;
1777                 goto out;
1778         }
1779
1780         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1781 out:
1782         if (take_write)
1783                 up_write_ref_node(&g->node, false);
1784         else
1785                 up_read_ref_node(&g->node);
1786         return fte_tmp;
1787 }
1788
1789 static struct mlx5_flow_handle *
1790 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1791                        struct list_head *match_head,
1792                        const struct mlx5_flow_spec *spec,
1793                        struct mlx5_flow_act *flow_act,
1794                        struct mlx5_flow_destination *dest,
1795                        int dest_num,
1796                        int ft_version)
1797 {
1798         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1799         struct mlx5_flow_group *g;
1800         struct mlx5_flow_handle *rule;
1801         struct match_list *iter;
1802         bool take_write = false;
1803         struct fs_fte *fte;
1804         u64  version = 0;
1805         int err;
1806
1807         fte = alloc_fte(ft, spec, flow_act);
1808         if (IS_ERR(fte))
1809                 return  ERR_PTR(-ENOMEM);
1810
1811 search_again_locked:
1812         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1813                 goto skip_search;
1814         version = matched_fgs_get_version(match_head);
1815         /* Try to find an fte with identical match value and attempt update its
1816          * action.
1817          */
1818         list_for_each_entry(iter, match_head, list) {
1819                 struct fs_fte *fte_tmp;
1820
1821                 g = iter->g;
1822                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1823                 if (!fte_tmp)
1824                         continue;
1825                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1826                 /* No error check needed here, because insert_fte() is not called */
1827                 up_write_ref_node(&fte_tmp->node, false);
1828                 tree_put_node(&fte_tmp->node, false);
1829                 kmem_cache_free(steering->ftes_cache, fte);
1830                 return rule;
1831         }
1832
1833 skip_search:
1834         /* No group with matching fte found, or we skipped the search.
1835          * Try to add a new fte to any matching fg.
1836          */
1837
1838         /* Check the ft version, for case that new flow group
1839          * was added while the fgs weren't locked
1840          */
1841         if (atomic_read(&ft->node.version) != ft_version) {
1842                 rule = ERR_PTR(-EAGAIN);
1843                 goto out;
1844         }
1845
1846         /* Check the fgs version. If version have changed it could be that an
1847          * FTE with the same match value was added while the fgs weren't
1848          * locked.
1849          */
1850         if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1851             version != matched_fgs_get_version(match_head)) {
1852                 take_write = true;
1853                 goto search_again_locked;
1854         }
1855
1856         list_for_each_entry(iter, match_head, list) {
1857                 g = iter->g;
1858
1859                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1860
1861                 if (!g->node.active) {
1862                         up_write_ref_node(&g->node, false);
1863                         continue;
1864                 }
1865
1866                 err = insert_fte(g, fte);
1867                 if (err) {
1868                         up_write_ref_node(&g->node, false);
1869                         if (err == -ENOSPC)
1870                                 continue;
1871                         kmem_cache_free(steering->ftes_cache, fte);
1872                         return ERR_PTR(err);
1873                 }
1874
1875                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1876                 up_write_ref_node(&g->node, false);
1877                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1878                 up_write_ref_node(&fte->node, false);
1879                 if (IS_ERR(rule))
1880                         tree_put_node(&fte->node, false);
1881                 return rule;
1882         }
1883         rule = ERR_PTR(-ENOENT);
1884 out:
1885         kmem_cache_free(steering->ftes_cache, fte);
1886         return rule;
1887 }
1888
1889 static struct mlx5_flow_handle *
1890 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1891                      const struct mlx5_flow_spec *spec,
1892                      struct mlx5_flow_act *flow_act,
1893                      struct mlx5_flow_destination *dest,
1894                      int dest_num)
1895
1896 {
1897         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1898         struct mlx5_flow_handle *rule;
1899         struct match_list match_head;
1900         struct mlx5_flow_group *g;
1901         bool take_write = false;
1902         struct fs_fte *fte;
1903         int version;
1904         int err;
1905         int i;
1906
1907         if (!check_valid_spec(spec))
1908                 return ERR_PTR(-EINVAL);
1909
1910         if (flow_act->fg && ft->autogroup.active)
1911                 return ERR_PTR(-EINVAL);
1912
1913         for (i = 0; i < dest_num; i++) {
1914                 if (!dest_is_valid(&dest[i], flow_act, ft))
1915                         return ERR_PTR(-EINVAL);
1916         }
1917         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1918 search_again_locked:
1919         version = atomic_read(&ft->node.version);
1920
1921         /* Collect all fgs which has a matching match_criteria */
1922         err = build_match_list(&match_head, ft, spec, flow_act->fg, take_write);
1923         if (err) {
1924                 if (take_write)
1925                         up_write_ref_node(&ft->node, false);
1926                 else
1927                         up_read_ref_node(&ft->node);
1928                 return ERR_PTR(err);
1929         }
1930
1931         if (!take_write)
1932                 up_read_ref_node(&ft->node);
1933
1934         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1935                                       dest_num, version);
1936         free_match_list(&match_head, take_write);
1937         if (!IS_ERR(rule) ||
1938             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1939                 if (take_write)
1940                         up_write_ref_node(&ft->node, false);
1941                 return rule;
1942         }
1943
1944         if (!take_write) {
1945                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1946                 take_write = true;
1947         }
1948
1949         if (PTR_ERR(rule) == -EAGAIN ||
1950             version != atomic_read(&ft->node.version))
1951                 goto search_again_locked;
1952
1953         g = alloc_auto_flow_group(ft, spec);
1954         if (IS_ERR(g)) {
1955                 rule = ERR_CAST(g);
1956                 up_write_ref_node(&ft->node, false);
1957                 return rule;
1958         }
1959
1960         fte = alloc_fte(ft, spec, flow_act);
1961         if (IS_ERR(fte)) {
1962                 up_write_ref_node(&ft->node, false);
1963                 err = PTR_ERR(fte);
1964                 goto err_alloc_fte;
1965         }
1966
1967         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1968         up_write_ref_node(&ft->node, false);
1969
1970         err = create_auto_flow_group(ft, g);
1971         if (err)
1972                 goto err_release_fg;
1973
1974         err = insert_fte(g, fte);
1975         if (err)
1976                 goto err_release_fg;
1977
1978         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1979         up_write_ref_node(&g->node, false);
1980         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1981         up_write_ref_node(&fte->node, false);
1982         if (IS_ERR(rule))
1983                 tree_put_node(&fte->node, false);
1984         tree_put_node(&g->node, false);
1985         return rule;
1986
1987 err_release_fg:
1988         up_write_ref_node(&g->node, false);
1989         kmem_cache_free(steering->ftes_cache, fte);
1990 err_alloc_fte:
1991         tree_put_node(&g->node, false);
1992         return ERR_PTR(err);
1993 }
1994
1995 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1996 {
1997         return ((ft->type == FS_FT_NIC_RX) &&
1998                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1999 }
2000
2001 struct mlx5_flow_handle *
2002 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
2003                     const struct mlx5_flow_spec *spec,
2004                     struct mlx5_flow_act *flow_act,
2005                     struct mlx5_flow_destination *dest,
2006                     int num_dest)
2007 {
2008         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2009         static const struct mlx5_flow_spec zero_spec = {};
2010         struct mlx5_flow_destination *gen_dest = NULL;
2011         struct mlx5_flow_table *next_ft = NULL;
2012         struct mlx5_flow_handle *handle = NULL;
2013         u32 sw_action = flow_act->action;
2014         int i;
2015
2016         if (!spec)
2017                 spec = &zero_spec;
2018
2019         if (!is_fwd_next_action(sw_action))
2020                 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2021
2022         if (!fwd_next_prio_supported(ft))
2023                 return ERR_PTR(-EOPNOTSUPP);
2024
2025         mutex_lock(&root->chain_lock);
2026         next_ft = find_next_fwd_ft(ft, flow_act);
2027         if (!next_ft) {
2028                 handle = ERR_PTR(-EOPNOTSUPP);
2029                 goto unlock;
2030         }
2031
2032         gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
2033                            GFP_KERNEL);
2034         if (!gen_dest) {
2035                 handle = ERR_PTR(-ENOMEM);
2036                 goto unlock;
2037         }
2038         for (i = 0; i < num_dest; i++)
2039                 gen_dest[i] = dest[i];
2040         gen_dest[i].type =
2041                 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
2042         gen_dest[i].ft = next_ft;
2043         dest = gen_dest;
2044         num_dest++;
2045         flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
2046                               MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
2047         flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2048         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2049         if (IS_ERR(handle))
2050                 goto unlock;
2051
2052         if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
2053                 mutex_lock(&next_ft->lock);
2054                 list_add(&handle->rule[num_dest - 1]->next_ft,
2055                          &next_ft->fwd_rules);
2056                 mutex_unlock(&next_ft->lock);
2057                 handle->rule[num_dest - 1]->sw_action = sw_action;
2058                 handle->rule[num_dest - 1]->ft = ft;
2059         }
2060 unlock:
2061         mutex_unlock(&root->chain_lock);
2062         kfree(gen_dest);
2063         return handle;
2064 }
2065 EXPORT_SYMBOL(mlx5_add_flow_rules);
2066
2067 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2068 {
2069         struct fs_fte *fte;
2070         int i;
2071
2072         /* In order to consolidate the HW changes we lock the FTE for other
2073          * changes, and increase its refcount, in order not to perform the
2074          * "del" functions of the FTE. Will handle them here.
2075          * The removal of the rules is done under locked FTE.
2076          * After removing all the handle's rules, if there are remaining
2077          * rules, it means we just need to modify the FTE in FW, and
2078          * unlock/decrease the refcount we increased before.
2079          * Otherwise, it means the FTE should be deleted. First delete the
2080          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2081          * the FTE, which will handle the last decrease of the refcount, as
2082          * well as required handling of its parent.
2083          */
2084         fs_get_obj(fte, handle->rule[0]->node.parent);
2085         down_write_ref_node(&fte->node, false);
2086         for (i = handle->num_rules - 1; i >= 0; i--)
2087                 tree_remove_node(&handle->rule[i]->node, true);
2088         if (list_empty(&fte->node.children)) {
2089                 fte->node.del_hw_func(&fte->node);
2090                 /* Avoid double call to del_hw_fte */
2091                 fte->node.del_hw_func = NULL;
2092                 up_write_ref_node(&fte->node, false);
2093                 tree_put_node(&fte->node, false);
2094         } else if (fte->dests_size) {
2095                 if (fte->modify_mask)
2096                         modify_fte(fte);
2097                 up_write_ref_node(&fte->node, false);
2098         } else {
2099                 up_write_ref_node(&fte->node, false);
2100         }
2101         kfree(handle);
2102 }
2103 EXPORT_SYMBOL(mlx5_del_flow_rules);
2104
2105 /* Assuming prio->node.children(flow tables) is sorted by level */
2106 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2107 {
2108         struct fs_prio *prio;
2109
2110         fs_get_obj(prio, ft->node.parent);
2111
2112         if (!list_is_last(&ft->node.list, &prio->node.children))
2113                 return list_next_entry(ft, node.list);
2114         return find_next_chained_ft(prio);
2115 }
2116
2117 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2118 {
2119         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2120         struct mlx5_ft_underlay_qp *uqp;
2121         struct mlx5_flow_table *new_root_ft = NULL;
2122         int err = 0;
2123         u32 qpn;
2124
2125         if (root->root_ft != ft)
2126                 return 0;
2127
2128         new_root_ft = find_next_ft(ft);
2129         if (!new_root_ft) {
2130                 root->root_ft = NULL;
2131                 return 0;
2132         }
2133
2134         if (list_empty(&root->underlay_qpns)) {
2135                 /* Don't set any QPN (zero) in case QPN list is empty */
2136                 qpn = 0;
2137                 err = root->cmds->update_root_ft(root, new_root_ft,
2138                                                  qpn, false);
2139         } else {
2140                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2141                         qpn = uqp->qpn;
2142                         err = root->cmds->update_root_ft(root,
2143                                                          new_root_ft, qpn,
2144                                                          false);
2145                         if (err)
2146                                 break;
2147                 }
2148         }
2149
2150         if (err)
2151                 mlx5_core_warn(root->dev,
2152                                "Update root flow table of id(%u) qpn(%d) failed\n",
2153                                ft->id, qpn);
2154         else
2155                 root->root_ft = new_root_ft;
2156
2157         return 0;
2158 }
2159
2160 /* Connect flow table from previous priority to
2161  * the next flow table.
2162  */
2163 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2164 {
2165         struct mlx5_core_dev *dev = get_dev(&ft->node);
2166         struct mlx5_flow_table *next_ft;
2167         struct fs_prio *prio;
2168         int err = 0;
2169
2170         err = update_root_ft_destroy(ft);
2171         if (err)
2172                 return err;
2173
2174         fs_get_obj(prio, ft->node.parent);
2175         if  (!(list_first_entry(&prio->node.children,
2176                                 struct mlx5_flow_table,
2177                                 node.list) == ft))
2178                 return 0;
2179
2180         next_ft = find_next_ft(ft);
2181         err = connect_fwd_rules(dev, next_ft, ft);
2182         if (err)
2183                 return err;
2184
2185         err = connect_prev_fts(dev, next_ft, prio);
2186         if (err)
2187                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2188                                ft->id);
2189         return err;
2190 }
2191
2192 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2193 {
2194         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2195         int err = 0;
2196
2197         mutex_lock(&root->chain_lock);
2198         if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2199                 err = disconnect_flow_table(ft);
2200         if (err) {
2201                 mutex_unlock(&root->chain_lock);
2202                 return err;
2203         }
2204         if (tree_remove_node(&ft->node, false))
2205                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2206                                ft->id);
2207         mutex_unlock(&root->chain_lock);
2208
2209         return err;
2210 }
2211 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2212
2213 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2214 {
2215         if (tree_remove_node(&fg->node, false))
2216                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2217                                fg->id);
2218 }
2219 EXPORT_SYMBOL(mlx5_destroy_flow_group);
2220
2221 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2222                                                 int n)
2223 {
2224         struct mlx5_flow_steering *steering = dev->priv.steering;
2225
2226         if (!steering || !steering->fdb_sub_ns)
2227                 return NULL;
2228
2229         return steering->fdb_sub_ns[n];
2230 }
2231 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2232
2233 static bool is_nic_rx_ns(enum mlx5_flow_namespace_type type)
2234 {
2235         switch (type) {
2236         case MLX5_FLOW_NAMESPACE_BYPASS:
2237         case MLX5_FLOW_NAMESPACE_LAG:
2238         case MLX5_FLOW_NAMESPACE_OFFLOADS:
2239         case MLX5_FLOW_NAMESPACE_ETHTOOL:
2240         case MLX5_FLOW_NAMESPACE_KERNEL:
2241         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
2242         case MLX5_FLOW_NAMESPACE_ANCHOR:
2243                 return true;
2244         default:
2245                 return false;
2246         }
2247 }
2248
2249 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2250                                                     enum mlx5_flow_namespace_type type)
2251 {
2252         struct mlx5_flow_steering *steering = dev->priv.steering;
2253         struct mlx5_flow_root_namespace *root_ns;
2254         int prio = 0;
2255         struct fs_prio *fs_prio;
2256         struct mlx5_flow_namespace *ns;
2257
2258         if (!steering)
2259                 return NULL;
2260
2261         switch (type) {
2262         case MLX5_FLOW_NAMESPACE_FDB:
2263                 if (steering->fdb_root_ns)
2264                         return &steering->fdb_root_ns->ns;
2265                 return NULL;
2266         case MLX5_FLOW_NAMESPACE_PORT_SEL:
2267                 if (steering->port_sel_root_ns)
2268                         return &steering->port_sel_root_ns->ns;
2269                 return NULL;
2270         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2271                 if (steering->sniffer_rx_root_ns)
2272                         return &steering->sniffer_rx_root_ns->ns;
2273                 return NULL;
2274         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2275                 if (steering->sniffer_tx_root_ns)
2276                         return &steering->sniffer_tx_root_ns->ns;
2277                 return NULL;
2278         case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
2279                 root_ns = steering->fdb_root_ns;
2280                 prio =  FDB_BYPASS_PATH;
2281                 break;
2282         case MLX5_FLOW_NAMESPACE_EGRESS:
2283         case MLX5_FLOW_NAMESPACE_EGRESS_KERNEL:
2284                 root_ns = steering->egress_root_ns;
2285                 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2286                 break;
2287         case MLX5_FLOW_NAMESPACE_RDMA_RX:
2288                 root_ns = steering->rdma_rx_root_ns;
2289                 prio = RDMA_RX_BYPASS_PRIO;
2290                 break;
2291         case MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL:
2292                 root_ns = steering->rdma_rx_root_ns;
2293                 prio = RDMA_RX_KERNEL_PRIO;
2294                 break;
2295         case MLX5_FLOW_NAMESPACE_RDMA_TX:
2296                 root_ns = steering->rdma_tx_root_ns;
2297                 break;
2298         case MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS:
2299                 root_ns = steering->rdma_rx_root_ns;
2300                 prio = RDMA_RX_COUNTERS_PRIO;
2301                 break;
2302         case MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS:
2303                 root_ns = steering->rdma_tx_root_ns;
2304                 prio = RDMA_TX_COUNTERS_PRIO;
2305                 break;
2306         default: /* Must be NIC RX */
2307                 WARN_ON(!is_nic_rx_ns(type));
2308                 root_ns = steering->root_ns;
2309                 prio = type;
2310                 break;
2311         }
2312
2313         if (!root_ns)
2314                 return NULL;
2315
2316         fs_prio = find_prio(&root_ns->ns, prio);
2317         if (!fs_prio)
2318                 return NULL;
2319
2320         ns = list_first_entry(&fs_prio->node.children,
2321                               typeof(*ns),
2322                               node.list);
2323
2324         return ns;
2325 }
2326 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2327
2328 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2329                                                               enum mlx5_flow_namespace_type type,
2330                                                               int vport)
2331 {
2332         struct mlx5_flow_steering *steering = dev->priv.steering;
2333
2334         if (!steering)
2335                 return NULL;
2336
2337         switch (type) {
2338         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2339                 if (vport >= steering->esw_egress_acl_vports)
2340                         return NULL;
2341                 if (steering->esw_egress_root_ns &&
2342                     steering->esw_egress_root_ns[vport])
2343                         return &steering->esw_egress_root_ns[vport]->ns;
2344                 else
2345                         return NULL;
2346         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2347                 if (vport >= steering->esw_ingress_acl_vports)
2348                         return NULL;
2349                 if (steering->esw_ingress_root_ns &&
2350                     steering->esw_ingress_root_ns[vport])
2351                         return &steering->esw_ingress_root_ns[vport]->ns;
2352                 else
2353                         return NULL;
2354         default:
2355                 return NULL;
2356         }
2357 }
2358
2359 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2360                                        unsigned int prio,
2361                                        int num_levels,
2362                                        enum fs_node_type type)
2363 {
2364         struct fs_prio *fs_prio;
2365
2366         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2367         if (!fs_prio)
2368                 return ERR_PTR(-ENOMEM);
2369
2370         fs_prio->node.type = type;
2371         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2372         tree_add_node(&fs_prio->node, &ns->node);
2373         fs_prio->num_levels = num_levels;
2374         fs_prio->prio = prio;
2375         list_add_tail(&fs_prio->node.list, &ns->node.children);
2376
2377         return fs_prio;
2378 }
2379
2380 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2381                                               unsigned int prio,
2382                                               int num_levels)
2383 {
2384         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2385 }
2386
2387 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2388                                       unsigned int prio, int num_levels)
2389 {
2390         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2391 }
2392
2393 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2394                                                      *ns)
2395 {
2396         ns->node.type = FS_TYPE_NAMESPACE;
2397
2398         return ns;
2399 }
2400
2401 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2402                                                        int def_miss_act)
2403 {
2404         struct mlx5_flow_namespace      *ns;
2405
2406         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2407         if (!ns)
2408                 return ERR_PTR(-ENOMEM);
2409
2410         fs_init_namespace(ns);
2411         ns->def_miss_action = def_miss_act;
2412         tree_init_node(&ns->node, NULL, del_sw_ns);
2413         tree_add_node(&ns->node, &prio->node);
2414         list_add_tail(&ns->node.list, &prio->node.children);
2415
2416         return ns;
2417 }
2418
2419 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2420                              struct init_tree_node *prio_metadata)
2421 {
2422         struct fs_prio *fs_prio;
2423         int i;
2424
2425         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2426                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2427                 if (IS_ERR(fs_prio))
2428                         return PTR_ERR(fs_prio);
2429         }
2430         return 0;
2431 }
2432
2433 #define FLOW_TABLE_BIT_SZ 1
2434 #define GET_FLOW_TABLE_CAP(dev, offset) \
2435         ((be32_to_cpu(*((__be32 *)(dev->caps.hca[MLX5_CAP_FLOW_TABLE]->cur) +   \
2436                         offset / 32)) >>                                        \
2437           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2438 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2439 {
2440         int i;
2441
2442         for (i = 0; i < caps->arr_sz; i++) {
2443                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2444                         return false;
2445         }
2446         return true;
2447 }
2448
2449 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2450                                     struct init_tree_node *init_node,
2451                                     struct fs_node *fs_parent_node,
2452                                     struct init_tree_node *init_parent_node,
2453                                     int prio)
2454 {
2455         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2456                                               flow_table_properties_nic_receive.
2457                                               max_ft_level);
2458         struct mlx5_flow_namespace *fs_ns;
2459         struct fs_prio *fs_prio;
2460         struct fs_node *base;
2461         int i;
2462         int err;
2463
2464         if (init_node->type == FS_TYPE_PRIO) {
2465                 if ((init_node->min_ft_level > max_ft_level) ||
2466                     !has_required_caps(steering->dev, &init_node->caps))
2467                         return 0;
2468
2469                 fs_get_obj(fs_ns, fs_parent_node);
2470                 if (init_node->num_leaf_prios)
2471                         return create_leaf_prios(fs_ns, prio, init_node);
2472                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2473                 if (IS_ERR(fs_prio))
2474                         return PTR_ERR(fs_prio);
2475                 base = &fs_prio->node;
2476         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2477                 fs_get_obj(fs_prio, fs_parent_node);
2478                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2479                 if (IS_ERR(fs_ns))
2480                         return PTR_ERR(fs_ns);
2481                 base = &fs_ns->node;
2482         } else {
2483                 return -EINVAL;
2484         }
2485         prio = 0;
2486         for (i = 0; i < init_node->ar_size; i++) {
2487                 err = init_root_tree_recursive(steering, &init_node->children[i],
2488                                                base, init_node, prio);
2489                 if (err)
2490                         return err;
2491                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2492                     init_node->children[i].num_leaf_prios) {
2493                         prio += init_node->children[i].num_leaf_prios;
2494                 }
2495         }
2496
2497         return 0;
2498 }
2499
2500 static int init_root_tree(struct mlx5_flow_steering *steering,
2501                           struct init_tree_node *init_node,
2502                           struct fs_node *fs_parent_node)
2503 {
2504         int err;
2505         int i;
2506
2507         for (i = 0; i < init_node->ar_size; i++) {
2508                 err = init_root_tree_recursive(steering, &init_node->children[i],
2509                                                fs_parent_node,
2510                                                init_node, i);
2511                 if (err)
2512                         return err;
2513         }
2514         return 0;
2515 }
2516
2517 static void del_sw_root_ns(struct fs_node *node)
2518 {
2519         struct mlx5_flow_root_namespace *root_ns;
2520         struct mlx5_flow_namespace *ns;
2521
2522         fs_get_obj(ns, node);
2523         root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2524         mutex_destroy(&root_ns->chain_lock);
2525         kfree(node);
2526 }
2527
2528 static struct mlx5_flow_root_namespace
2529 *create_root_ns(struct mlx5_flow_steering *steering,
2530                 enum fs_flow_table_type table_type)
2531 {
2532         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2533         struct mlx5_flow_root_namespace *root_ns;
2534         struct mlx5_flow_namespace *ns;
2535
2536         /* Create the root namespace */
2537         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2538         if (!root_ns)
2539                 return NULL;
2540
2541         root_ns->dev = steering->dev;
2542         root_ns->table_type = table_type;
2543         root_ns->cmds = cmds;
2544
2545         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2546
2547         ns = &root_ns->ns;
2548         fs_init_namespace(ns);
2549         mutex_init(&root_ns->chain_lock);
2550         tree_init_node(&ns->node, NULL, del_sw_root_ns);
2551         tree_add_node(&ns->node, NULL);
2552
2553         return root_ns;
2554 }
2555
2556 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2557
2558 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2559 {
2560         struct fs_prio *prio;
2561
2562         fs_for_each_prio(prio, ns) {
2563                  /* This updates prio start_level and num_levels */
2564                 set_prio_attrs_in_prio(prio, acc_level);
2565                 acc_level += prio->num_levels;
2566         }
2567         return acc_level;
2568 }
2569
2570 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2571 {
2572         struct mlx5_flow_namespace *ns;
2573         int acc_level_ns = acc_level;
2574
2575         prio->start_level = acc_level;
2576         fs_for_each_ns(ns, prio) {
2577                 /* This updates start_level and num_levels of ns's priority descendants */
2578                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2579
2580                 /* If this a prio with chains, and we can jump from one chain
2581                  * (namespace) to another, so we accumulate the levels
2582                  */
2583                 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2584                         acc_level = acc_level_ns;
2585         }
2586
2587         if (!prio->num_levels)
2588                 prio->num_levels = acc_level_ns - prio->start_level;
2589         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2590 }
2591
2592 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2593 {
2594         struct mlx5_flow_namespace *ns = &root_ns->ns;
2595         struct fs_prio *prio;
2596         int start_level = 0;
2597
2598         fs_for_each_prio(prio, ns) {
2599                 set_prio_attrs_in_prio(prio, start_level);
2600                 start_level += prio->num_levels;
2601         }
2602 }
2603
2604 #define ANCHOR_PRIO 0
2605 #define ANCHOR_SIZE 1
2606 #define ANCHOR_LEVEL 0
2607 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2608 {
2609         struct mlx5_flow_namespace *ns = NULL;
2610         struct mlx5_flow_table_attr ft_attr = {};
2611         struct mlx5_flow_table *ft;
2612
2613         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2614         if (WARN_ON(!ns))
2615                 return -EINVAL;
2616
2617         ft_attr.max_fte = ANCHOR_SIZE;
2618         ft_attr.level   = ANCHOR_LEVEL;
2619         ft_attr.prio    = ANCHOR_PRIO;
2620
2621         ft = mlx5_create_flow_table(ns, &ft_attr);
2622         if (IS_ERR(ft)) {
2623                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2624                 return PTR_ERR(ft);
2625         }
2626         return 0;
2627 }
2628
2629 static int init_root_ns(struct mlx5_flow_steering *steering)
2630 {
2631         int err;
2632
2633         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2634         if (!steering->root_ns)
2635                 return -ENOMEM;
2636
2637         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2638         if (err)
2639                 goto out_err;
2640
2641         set_prio_attrs(steering->root_ns);
2642         err = create_anchor_flow_table(steering);
2643         if (err)
2644                 goto out_err;
2645
2646         return 0;
2647
2648 out_err:
2649         cleanup_root_ns(steering->root_ns);
2650         steering->root_ns = NULL;
2651         return err;
2652 }
2653
2654 static void clean_tree(struct fs_node *node)
2655 {
2656         if (node) {
2657                 struct fs_node *iter;
2658                 struct fs_node *temp;
2659
2660                 tree_get_node(node);
2661                 list_for_each_entry_safe(iter, temp, &node->children, list)
2662                         clean_tree(iter);
2663                 tree_put_node(node, false);
2664                 tree_remove_node(node, false);
2665         }
2666 }
2667
2668 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2669 {
2670         if (!root_ns)
2671                 return;
2672
2673         clean_tree(&root_ns->ns.node);
2674 }
2675
2676 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2677 {
2678         struct fs_prio *prio;
2679
2680         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2681         if (!steering->sniffer_tx_root_ns)
2682                 return -ENOMEM;
2683
2684         /* Create single prio */
2685         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2686         return PTR_ERR_OR_ZERO(prio);
2687 }
2688
2689 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2690 {
2691         struct fs_prio *prio;
2692
2693         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2694         if (!steering->sniffer_rx_root_ns)
2695                 return -ENOMEM;
2696
2697         /* Create single prio */
2698         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2699         return PTR_ERR_OR_ZERO(prio);
2700 }
2701
2702 #define PORT_SEL_NUM_LEVELS 3
2703 static int init_port_sel_root_ns(struct mlx5_flow_steering *steering)
2704 {
2705         struct fs_prio *prio;
2706
2707         steering->port_sel_root_ns = create_root_ns(steering, FS_FT_PORT_SEL);
2708         if (!steering->port_sel_root_ns)
2709                 return -ENOMEM;
2710
2711         /* Create single prio */
2712         prio = fs_create_prio(&steering->port_sel_root_ns->ns, 0,
2713                               PORT_SEL_NUM_LEVELS);
2714         return PTR_ERR_OR_ZERO(prio);
2715 }
2716
2717 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2718 {
2719         int err;
2720
2721         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2722         if (!steering->rdma_rx_root_ns)
2723                 return -ENOMEM;
2724
2725         err = init_root_tree(steering, &rdma_rx_root_fs,
2726                              &steering->rdma_rx_root_ns->ns.node);
2727         if (err)
2728                 goto out_err;
2729
2730         set_prio_attrs(steering->rdma_rx_root_ns);
2731
2732         return 0;
2733
2734 out_err:
2735         cleanup_root_ns(steering->rdma_rx_root_ns);
2736         steering->rdma_rx_root_ns = NULL;
2737         return err;
2738 }
2739
2740 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2741 {
2742         int err;
2743
2744         steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2745         if (!steering->rdma_tx_root_ns)
2746                 return -ENOMEM;
2747
2748         err = init_root_tree(steering, &rdma_tx_root_fs,
2749                              &steering->rdma_tx_root_ns->ns.node);
2750         if (err)
2751                 goto out_err;
2752
2753         set_prio_attrs(steering->rdma_tx_root_ns);
2754
2755         return 0;
2756
2757 out_err:
2758         cleanup_root_ns(steering->rdma_tx_root_ns);
2759         steering->rdma_tx_root_ns = NULL;
2760         return err;
2761 }
2762
2763 /* FT and tc chains are stored in the same array so we can re-use the
2764  * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2765  * When creating a new ns for each chain store it in the first available slot.
2766  * Assume tc chains are created and stored first and only then the FT chain.
2767  */
2768 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2769                                         struct mlx5_flow_namespace *ns)
2770 {
2771         int chain = 0;
2772
2773         while (steering->fdb_sub_ns[chain])
2774                 ++chain;
2775
2776         steering->fdb_sub_ns[chain] = ns;
2777 }
2778
2779 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2780                                         struct fs_prio *maj_prio)
2781 {
2782         struct mlx5_flow_namespace *ns;
2783         struct fs_prio *min_prio;
2784         int prio;
2785
2786         ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2787         if (IS_ERR(ns))
2788                 return PTR_ERR(ns);
2789
2790         for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2791                 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2792                 if (IS_ERR(min_prio))
2793                         return PTR_ERR(min_prio);
2794         }
2795
2796         store_fdb_sub_ns_prio_chain(steering, ns);
2797
2798         return 0;
2799 }
2800
2801 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2802                              int fs_prio,
2803                              int chains)
2804 {
2805         struct fs_prio *maj_prio;
2806         int levels;
2807         int chain;
2808         int err;
2809
2810         levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2811         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2812                                           fs_prio,
2813                                           levels);
2814         if (IS_ERR(maj_prio))
2815                 return PTR_ERR(maj_prio);
2816
2817         for (chain = 0; chain < chains; chain++) {
2818                 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2819                 if (err)
2820                         return err;
2821         }
2822
2823         return 0;
2824 }
2825
2826 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2827 {
2828         int err;
2829
2830         steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2831                                        sizeof(*steering->fdb_sub_ns),
2832                                        GFP_KERNEL);
2833         if (!steering->fdb_sub_ns)
2834                 return -ENOMEM;
2835
2836         err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
2837         if (err)
2838                 return err;
2839
2840         err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
2841         if (err)
2842                 return err;
2843
2844         return 0;
2845 }
2846
2847 static int create_fdb_bypass(struct mlx5_flow_steering *steering)
2848 {
2849         struct mlx5_flow_namespace *ns;
2850         struct fs_prio *prio;
2851         int i;
2852
2853         prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH, 0);
2854         if (IS_ERR(prio))
2855                 return PTR_ERR(prio);
2856
2857         ns = fs_create_namespace(prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2858         if (IS_ERR(ns))
2859                 return PTR_ERR(ns);
2860
2861         for (i = 0; i < MLX5_BY_PASS_NUM_REGULAR_PRIOS; i++) {
2862                 prio = fs_create_prio(ns, i, 1);
2863                 if (IS_ERR(prio))
2864                         return PTR_ERR(prio);
2865         }
2866         return 0;
2867 }
2868
2869 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2870 {
2871         struct fs_prio *maj_prio;
2872         int err;
2873
2874         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2875         if (!steering->fdb_root_ns)
2876                 return -ENOMEM;
2877
2878         err = create_fdb_bypass(steering);
2879         if (err)
2880                 goto out_err;
2881
2882         err = create_fdb_fast_path(steering);
2883         if (err)
2884                 goto out_err;
2885
2886         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1);
2887         if (IS_ERR(maj_prio)) {
2888                 err = PTR_ERR(maj_prio);
2889                 goto out_err;
2890         }
2891
2892         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 3);
2893         if (IS_ERR(maj_prio)) {
2894                 err = PTR_ERR(maj_prio);
2895                 goto out_err;
2896         }
2897
2898         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2899         if (IS_ERR(maj_prio)) {
2900                 err = PTR_ERR(maj_prio);
2901                 goto out_err;
2902         }
2903
2904         /* We put this priority last, knowing that nothing will get here
2905          * unless explicitly forwarded to. This is possible because the
2906          * slow path tables have catch all rules and nothing gets passed
2907          * those tables.
2908          */
2909         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
2910         if (IS_ERR(maj_prio)) {
2911                 err = PTR_ERR(maj_prio);
2912                 goto out_err;
2913         }
2914
2915         set_prio_attrs(steering->fdb_root_ns);
2916         return 0;
2917
2918 out_err:
2919         cleanup_root_ns(steering->fdb_root_ns);
2920         kfree(steering->fdb_sub_ns);
2921         steering->fdb_sub_ns = NULL;
2922         steering->fdb_root_ns = NULL;
2923         return err;
2924 }
2925
2926 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2927 {
2928         struct fs_prio *prio;
2929
2930         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2931         if (!steering->esw_egress_root_ns[vport])
2932                 return -ENOMEM;
2933
2934         /* create 1 prio*/
2935         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2936         return PTR_ERR_OR_ZERO(prio);
2937 }
2938
2939 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2940 {
2941         struct fs_prio *prio;
2942
2943         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2944         if (!steering->esw_ingress_root_ns[vport])
2945                 return -ENOMEM;
2946
2947         /* create 1 prio*/
2948         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2949         return PTR_ERR_OR_ZERO(prio);
2950 }
2951
2952 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2953 {
2954         struct mlx5_flow_steering *steering = dev->priv.steering;
2955         int err;
2956         int i;
2957
2958         steering->esw_egress_root_ns =
2959                         kcalloc(total_vports,
2960                                 sizeof(*steering->esw_egress_root_ns),
2961                                 GFP_KERNEL);
2962         if (!steering->esw_egress_root_ns)
2963                 return -ENOMEM;
2964
2965         for (i = 0; i < total_vports; i++) {
2966                 err = init_egress_acl_root_ns(steering, i);
2967                 if (err)
2968                         goto cleanup_root_ns;
2969         }
2970         steering->esw_egress_acl_vports = total_vports;
2971         return 0;
2972
2973 cleanup_root_ns:
2974         for (i--; i >= 0; i--)
2975                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2976         kfree(steering->esw_egress_root_ns);
2977         steering->esw_egress_root_ns = NULL;
2978         return err;
2979 }
2980
2981 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev)
2982 {
2983         struct mlx5_flow_steering *steering = dev->priv.steering;
2984         int i;
2985
2986         if (!steering->esw_egress_root_ns)
2987                 return;
2988
2989         for (i = 0; i < steering->esw_egress_acl_vports; i++)
2990                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2991
2992         kfree(steering->esw_egress_root_ns);
2993         steering->esw_egress_root_ns = NULL;
2994 }
2995
2996 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2997 {
2998         struct mlx5_flow_steering *steering = dev->priv.steering;
2999         int err;
3000         int i;
3001
3002         steering->esw_ingress_root_ns =
3003                         kcalloc(total_vports,
3004                                 sizeof(*steering->esw_ingress_root_ns),
3005                                 GFP_KERNEL);
3006         if (!steering->esw_ingress_root_ns)
3007                 return -ENOMEM;
3008
3009         for (i = 0; i < total_vports; i++) {
3010                 err = init_ingress_acl_root_ns(steering, i);
3011                 if (err)
3012                         goto cleanup_root_ns;
3013         }
3014         steering->esw_ingress_acl_vports = total_vports;
3015         return 0;
3016
3017 cleanup_root_ns:
3018         for (i--; i >= 0; i--)
3019                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
3020         kfree(steering->esw_ingress_root_ns);
3021         steering->esw_ingress_root_ns = NULL;
3022         return err;
3023 }
3024
3025 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev)
3026 {
3027         struct mlx5_flow_steering *steering = dev->priv.steering;
3028         int i;
3029
3030         if (!steering->esw_ingress_root_ns)
3031                 return;
3032
3033         for (i = 0; i < steering->esw_ingress_acl_vports; i++)
3034                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
3035
3036         kfree(steering->esw_ingress_root_ns);
3037         steering->esw_ingress_root_ns = NULL;
3038 }
3039
3040 u32 mlx5_fs_get_capabilities(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type type)
3041 {
3042         struct mlx5_flow_root_namespace *root;
3043         struct mlx5_flow_namespace *ns;
3044
3045         ns = mlx5_get_flow_namespace(dev, type);
3046         if (!ns)
3047                 return 0;
3048
3049         root = find_root(&ns->node);
3050         if (!root)
3051                 return 0;
3052
3053         return root->cmds->get_capabilities(root, root->table_type);
3054 }
3055
3056 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
3057 {
3058         int err;
3059
3060         steering->egress_root_ns = create_root_ns(steering,
3061                                                   FS_FT_NIC_TX);
3062         if (!steering->egress_root_ns)
3063                 return -ENOMEM;
3064
3065         err = init_root_tree(steering, &egress_root_fs,
3066                              &steering->egress_root_ns->ns.node);
3067         if (err)
3068                 goto cleanup;
3069         set_prio_attrs(steering->egress_root_ns);
3070         return 0;
3071 cleanup:
3072         cleanup_root_ns(steering->egress_root_ns);
3073         steering->egress_root_ns = NULL;
3074         return err;
3075 }
3076
3077 void mlx5_fs_core_cleanup(struct mlx5_core_dev *dev)
3078 {
3079         struct mlx5_flow_steering *steering = dev->priv.steering;
3080
3081         cleanup_root_ns(steering->root_ns);
3082         cleanup_root_ns(steering->fdb_root_ns);
3083         steering->fdb_root_ns = NULL;
3084         kfree(steering->fdb_sub_ns);
3085         steering->fdb_sub_ns = NULL;
3086         cleanup_root_ns(steering->port_sel_root_ns);
3087         cleanup_root_ns(steering->sniffer_rx_root_ns);
3088         cleanup_root_ns(steering->sniffer_tx_root_ns);
3089         cleanup_root_ns(steering->rdma_rx_root_ns);
3090         cleanup_root_ns(steering->rdma_tx_root_ns);
3091         cleanup_root_ns(steering->egress_root_ns);
3092 }
3093
3094 int mlx5_fs_core_init(struct mlx5_core_dev *dev)
3095 {
3096         struct mlx5_flow_steering *steering = dev->priv.steering;
3097         int err = 0;
3098
3099         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
3100               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
3101              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
3102               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
3103             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
3104                 err = init_root_ns(steering);
3105                 if (err)
3106                         goto err;
3107         }
3108
3109         if (MLX5_ESWITCH_MANAGER(dev)) {
3110                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
3111                         err = init_fdb_root_ns(steering);
3112                         if (err)
3113                                 goto err;
3114                 }
3115         }
3116
3117         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
3118                 err = init_sniffer_rx_root_ns(steering);
3119                 if (err)
3120                         goto err;
3121         }
3122
3123         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
3124                 err = init_sniffer_tx_root_ns(steering);
3125                 if (err)
3126                         goto err;
3127         }
3128
3129         if (MLX5_CAP_FLOWTABLE_PORT_SELECTION(dev, ft_support)) {
3130                 err = init_port_sel_root_ns(steering);
3131                 if (err)
3132                         goto err;
3133         }
3134
3135         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
3136             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3137                 err = init_rdma_rx_root_ns(steering);
3138                 if (err)
3139                         goto err;
3140         }
3141
3142         if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3143                 err = init_rdma_tx_root_ns(steering);
3144                 if (err)
3145                         goto err;
3146         }
3147
3148         if (MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3149                 err = init_egress_root_ns(steering);
3150                 if (err)
3151                         goto err;
3152         }
3153
3154         return 0;
3155
3156 err:
3157         mlx5_fs_core_cleanup(dev);
3158         return err;
3159 }
3160
3161 void mlx5_fs_core_free(struct mlx5_core_dev *dev)
3162 {
3163         struct mlx5_flow_steering *steering = dev->priv.steering;
3164
3165         kmem_cache_destroy(steering->ftes_cache);
3166         kmem_cache_destroy(steering->fgs_cache);
3167         kfree(steering);
3168         mlx5_ft_pool_destroy(dev);
3169         mlx5_cleanup_fc_stats(dev);
3170 }
3171
3172 int mlx5_fs_core_alloc(struct mlx5_core_dev *dev)
3173 {
3174         struct mlx5_flow_steering *steering;
3175         int err = 0;
3176
3177         err = mlx5_init_fc_stats(dev);
3178         if (err)
3179                 return err;
3180
3181         err = mlx5_ft_pool_init(dev);
3182         if (err)
3183                 goto err;
3184
3185         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
3186         if (!steering) {
3187                 err = -ENOMEM;
3188                 goto err;
3189         }
3190
3191         steering->dev = dev;
3192         dev->priv.steering = steering;
3193
3194         if (mlx5_fs_dr_is_supported(dev))
3195                 steering->mode = MLX5_FLOW_STEERING_MODE_SMFS;
3196         else
3197                 steering->mode = MLX5_FLOW_STEERING_MODE_DMFS;
3198
3199         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
3200                                                 sizeof(struct mlx5_flow_group), 0,
3201                                                 0, NULL);
3202         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
3203                                                  0, NULL);
3204         if (!steering->ftes_cache || !steering->fgs_cache) {
3205                 err = -ENOMEM;
3206                 goto err;
3207         }
3208
3209         return 0;
3210
3211 err:
3212         mlx5_fs_core_free(dev);
3213         return err;
3214 }
3215
3216 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3217 {
3218         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3219         struct mlx5_ft_underlay_qp *new_uqp;
3220         int err = 0;
3221
3222         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3223         if (!new_uqp)
3224                 return -ENOMEM;
3225
3226         mutex_lock(&root->chain_lock);
3227
3228         if (!root->root_ft) {
3229                 err = -EINVAL;
3230                 goto update_ft_fail;
3231         }
3232
3233         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3234                                          false);
3235         if (err) {
3236                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3237                                underlay_qpn, err);
3238                 goto update_ft_fail;
3239         }
3240
3241         new_uqp->qpn = underlay_qpn;
3242         list_add_tail(&new_uqp->list, &root->underlay_qpns);
3243
3244         mutex_unlock(&root->chain_lock);
3245
3246         return 0;
3247
3248 update_ft_fail:
3249         mutex_unlock(&root->chain_lock);
3250         kfree(new_uqp);
3251         return err;
3252 }
3253 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3254
3255 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3256 {
3257         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3258         struct mlx5_ft_underlay_qp *uqp;
3259         bool found = false;
3260         int err = 0;
3261
3262         mutex_lock(&root->chain_lock);
3263         list_for_each_entry(uqp, &root->underlay_qpns, list) {
3264                 if (uqp->qpn == underlay_qpn) {
3265                         found = true;
3266                         break;
3267                 }
3268         }
3269
3270         if (!found) {
3271                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3272                                underlay_qpn);
3273                 err = -EINVAL;
3274                 goto out;
3275         }
3276
3277         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3278                                          true);
3279         if (err)
3280                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3281                                underlay_qpn, err);
3282
3283         list_del(&uqp->list);
3284         mutex_unlock(&root->chain_lock);
3285         kfree(uqp);
3286
3287         return 0;
3288
3289 out:
3290         mutex_unlock(&root->chain_lock);
3291         return err;
3292 }
3293 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3294
3295 static struct mlx5_flow_root_namespace
3296 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3297 {
3298         struct mlx5_flow_namespace *ns;
3299
3300         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3301             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3302                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3303         else
3304                 ns = mlx5_get_flow_namespace(dev, ns_type);
3305         if (!ns)
3306                 return NULL;
3307
3308         return find_root(&ns->node);
3309 }
3310
3311 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3312                                                  u8 ns_type, u8 num_actions,
3313                                                  void *modify_actions)
3314 {
3315         struct mlx5_flow_root_namespace *root;
3316         struct mlx5_modify_hdr *modify_hdr;
3317         int err;
3318
3319         root = get_root_namespace(dev, ns_type);
3320         if (!root)
3321                 return ERR_PTR(-EOPNOTSUPP);
3322
3323         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3324         if (!modify_hdr)
3325                 return ERR_PTR(-ENOMEM);
3326
3327         modify_hdr->ns_type = ns_type;
3328         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3329                                               modify_actions, modify_hdr);
3330         if (err) {
3331                 kfree(modify_hdr);
3332                 return ERR_PTR(err);
3333         }
3334
3335         return modify_hdr;
3336 }
3337 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3338
3339 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3340                                 struct mlx5_modify_hdr *modify_hdr)
3341 {
3342         struct mlx5_flow_root_namespace *root;
3343
3344         root = get_root_namespace(dev, modify_hdr->ns_type);
3345         if (WARN_ON(!root))
3346                 return;
3347         root->cmds->modify_header_dealloc(root, modify_hdr);
3348         kfree(modify_hdr);
3349 }
3350 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3351
3352 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3353                                                      struct mlx5_pkt_reformat_params *params,
3354                                                      enum mlx5_flow_namespace_type ns_type)
3355 {
3356         struct mlx5_pkt_reformat *pkt_reformat;
3357         struct mlx5_flow_root_namespace *root;
3358         int err;
3359
3360         root = get_root_namespace(dev, ns_type);
3361         if (!root)
3362                 return ERR_PTR(-EOPNOTSUPP);
3363
3364         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3365         if (!pkt_reformat)
3366                 return ERR_PTR(-ENOMEM);
3367
3368         pkt_reformat->ns_type = ns_type;
3369         pkt_reformat->reformat_type = params->type;
3370         err = root->cmds->packet_reformat_alloc(root, params, ns_type,
3371                                                 pkt_reformat);
3372         if (err) {
3373                 kfree(pkt_reformat);
3374                 return ERR_PTR(err);
3375         }
3376
3377         return pkt_reformat;
3378 }
3379 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3380
3381 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3382                                   struct mlx5_pkt_reformat *pkt_reformat)
3383 {
3384         struct mlx5_flow_root_namespace *root;
3385
3386         root = get_root_namespace(dev, pkt_reformat->ns_type);
3387         if (WARN_ON(!root))
3388                 return;
3389         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3390         kfree(pkt_reformat);
3391 }
3392 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3393
3394 int mlx5_get_match_definer_id(struct mlx5_flow_definer *definer)
3395 {
3396         return definer->id;
3397 }
3398
3399 struct mlx5_flow_definer *
3400 mlx5_create_match_definer(struct mlx5_core_dev *dev,
3401                           enum mlx5_flow_namespace_type ns_type, u16 format_id,
3402                           u32 *match_mask)
3403 {
3404         struct mlx5_flow_root_namespace *root;
3405         struct mlx5_flow_definer *definer;
3406         int id;
3407
3408         root = get_root_namespace(dev, ns_type);
3409         if (!root)
3410                 return ERR_PTR(-EOPNOTSUPP);
3411
3412         definer = kzalloc(sizeof(*definer), GFP_KERNEL);
3413         if (!definer)
3414                 return ERR_PTR(-ENOMEM);
3415
3416         definer->ns_type = ns_type;
3417         id = root->cmds->create_match_definer(root, format_id, match_mask);
3418         if (id < 0) {
3419                 mlx5_core_warn(root->dev, "Failed to create match definer (%d)\n", id);
3420                 kfree(definer);
3421                 return ERR_PTR(id);
3422         }
3423         definer->id = id;
3424         return definer;
3425 }
3426
3427 void mlx5_destroy_match_definer(struct mlx5_core_dev *dev,
3428                                 struct mlx5_flow_definer *definer)
3429 {
3430         struct mlx5_flow_root_namespace *root;
3431
3432         root = get_root_namespace(dev, definer->ns_type);
3433         if (WARN_ON(!root))
3434                 return;
3435
3436         root->cmds->destroy_match_definer(root, definer->id);
3437         kfree(definer);
3438 }
3439
3440 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3441                                  struct mlx5_flow_root_namespace *peer_ns)
3442 {
3443         if (peer_ns && ns->mode != peer_ns->mode) {
3444                 mlx5_core_err(ns->dev,
3445                               "Can't peer namespace of different steering mode\n");
3446                 return -EINVAL;
3447         }
3448
3449         return ns->cmds->set_peer(ns, peer_ns);
3450 }
3451
3452 /* This function should be called only at init stage of the namespace.
3453  * It is not safe to call this function while steering operations
3454  * are executed in the namespace.
3455  */
3456 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3457                                  enum mlx5_flow_steering_mode mode)
3458 {
3459         struct mlx5_flow_root_namespace *root;
3460         const struct mlx5_flow_cmds *cmds;
3461         int err;
3462
3463         root = find_root(&ns->node);
3464         if (&root->ns != ns)
3465         /* Can't set cmds to non root namespace */
3466                 return -EINVAL;
3467
3468         if (root->table_type != FS_FT_FDB)
3469                 return -EOPNOTSUPP;
3470
3471         if (root->mode == mode)
3472                 return 0;
3473
3474         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3475                 cmds = mlx5_fs_cmd_get_dr_cmds();
3476         else
3477                 cmds = mlx5_fs_cmd_get_fw_cmds();
3478         if (!cmds)
3479                 return -EOPNOTSUPP;
3480
3481         err = cmds->create_ns(root);
3482         if (err) {
3483                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3484                               err);
3485                 return err;
3486         }
3487
3488         root->cmds->destroy_ns(root);
3489         root->cmds = cmds;
3490         root->mode = mode;
3491
3492         return 0;
3493 }