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, 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 u32 mlx5_flow_table_id(struct mlx5_flow_table *ft)
1199 {
1200         return ft->id;
1201 }
1202 EXPORT_SYMBOL(mlx5_flow_table_id);
1203
1204 struct mlx5_flow_table *
1205 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1206                              struct mlx5_flow_table_attr *ft_attr, u16 vport)
1207 {
1208         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1209 }
1210
1211 struct mlx5_flow_table*
1212 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1213                                  int prio, u32 level)
1214 {
1215         struct mlx5_flow_table_attr ft_attr = {};
1216
1217         ft_attr.level = level;
1218         ft_attr.prio  = prio;
1219         ft_attr.max_fte = 1;
1220
1221         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1222 }
1223 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1224
1225 #define MAX_FLOW_GROUP_SIZE BIT(24)
1226 struct mlx5_flow_table*
1227 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1228                                     struct mlx5_flow_table_attr *ft_attr)
1229 {
1230         int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1231         int max_num_groups = ft_attr->autogroup.max_num_groups;
1232         struct mlx5_flow_table *ft;
1233         int autogroups_max_fte;
1234
1235         ft = mlx5_create_flow_table(ns, ft_attr);
1236         if (IS_ERR(ft))
1237                 return ft;
1238
1239         autogroups_max_fte = ft->max_fte - num_reserved_entries;
1240         if (max_num_groups > autogroups_max_fte)
1241                 goto err_validate;
1242         if (num_reserved_entries > ft->max_fte)
1243                 goto err_validate;
1244
1245         /* Align the number of groups according to the largest group size */
1246         if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE)
1247                 max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1;
1248
1249         ft->autogroup.active = true;
1250         ft->autogroup.required_groups = max_num_groups;
1251         ft->autogroup.max_fte = autogroups_max_fte;
1252         /* We save place for flow groups in addition to max types */
1253         ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1254
1255         return ft;
1256
1257 err_validate:
1258         mlx5_destroy_flow_table(ft);
1259         return ERR_PTR(-ENOSPC);
1260 }
1261 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1262
1263 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1264                                                u32 *fg_in)
1265 {
1266         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1267         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1268                                             fg_in, match_criteria);
1269         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1270                                             fg_in,
1271                                             match_criteria_enable);
1272         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1273                                    start_flow_index);
1274         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1275                                  end_flow_index);
1276         struct mlx5_flow_group *fg;
1277         int err;
1278
1279         if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1280                 return ERR_PTR(-EPERM);
1281
1282         down_write_ref_node(&ft->node, false);
1283         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1284                                      start_index, end_index,
1285                                      ft->node.children.prev);
1286         up_write_ref_node(&ft->node, false);
1287         if (IS_ERR(fg))
1288                 return fg;
1289
1290         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1291         if (err) {
1292                 tree_put_node(&fg->node, false);
1293                 return ERR_PTR(err);
1294         }
1295         trace_mlx5_fs_add_fg(fg);
1296         fg->node.active = true;
1297
1298         return fg;
1299 }
1300 EXPORT_SYMBOL(mlx5_create_flow_group);
1301
1302 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1303 {
1304         struct mlx5_flow_rule *rule;
1305
1306         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1307         if (!rule)
1308                 return NULL;
1309
1310         INIT_LIST_HEAD(&rule->next_ft);
1311         rule->node.type = FS_TYPE_FLOW_DEST;
1312         if (dest)
1313                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1314         else
1315                 rule->dest_attr.type = MLX5_FLOW_DESTINATION_TYPE_NONE;
1316
1317         return rule;
1318 }
1319
1320 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1321 {
1322         struct mlx5_flow_handle *handle;
1323
1324         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1325         if (!handle)
1326                 return NULL;
1327
1328         handle->num_rules = num_rules;
1329
1330         return handle;
1331 }
1332
1333 static void destroy_flow_handle(struct fs_fte *fte,
1334                                 struct mlx5_flow_handle *handle,
1335                                 struct mlx5_flow_destination *dest,
1336                                 int i)
1337 {
1338         for (; --i >= 0;) {
1339                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1340                         fte->dests_size--;
1341                         list_del(&handle->rule[i]->node.list);
1342                         kfree(handle->rule[i]);
1343                 }
1344         }
1345         kfree(handle);
1346 }
1347
1348 static struct mlx5_flow_handle *
1349 create_flow_handle(struct fs_fte *fte,
1350                    struct mlx5_flow_destination *dest,
1351                    int dest_num,
1352                    int *modify_mask,
1353                    bool *new_rule)
1354 {
1355         struct mlx5_flow_handle *handle;
1356         struct mlx5_flow_rule *rule = NULL;
1357         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1358         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1359         int type;
1360         int i = 0;
1361
1362         handle = alloc_handle((dest_num) ? dest_num : 1);
1363         if (!handle)
1364                 return ERR_PTR(-ENOMEM);
1365
1366         do {
1367                 if (dest) {
1368                         rule = find_flow_rule(fte, dest + i);
1369                         if (rule) {
1370                                 refcount_inc(&rule->node.refcount);
1371                                 goto rule_found;
1372                         }
1373                 }
1374
1375                 *new_rule = true;
1376                 rule = alloc_rule(dest + i);
1377                 if (!rule)
1378                         goto free_rules;
1379
1380                 /* Add dest to dests list- we need flow tables to be in the
1381                  * end of the list for forward to next prio rules.
1382                  */
1383                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1384                 if (dest &&
1385                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1386                         list_add(&rule->node.list, &fte->node.children);
1387                 else
1388                         list_add_tail(&rule->node.list, &fte->node.children);
1389                 if (dest) {
1390                         fte->dests_size++;
1391
1392                         if (is_fwd_dest_type(dest[i].type))
1393                                 fte->fwd_dests++;
1394
1395                         type = dest[i].type ==
1396                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1397                         *modify_mask |= type ? count : dst;
1398                 }
1399 rule_found:
1400                 handle->rule[i] = rule;
1401         } while (++i < dest_num);
1402
1403         return handle;
1404
1405 free_rules:
1406         destroy_flow_handle(fte, handle, dest, i);
1407         return ERR_PTR(-ENOMEM);
1408 }
1409
1410 /* fte should not be deleted while calling this function */
1411 static struct mlx5_flow_handle *
1412 add_rule_fte(struct fs_fte *fte,
1413              struct mlx5_flow_group *fg,
1414              struct mlx5_flow_destination *dest,
1415              int dest_num,
1416              bool update_action)
1417 {
1418         struct mlx5_flow_root_namespace *root;
1419         struct mlx5_flow_handle *handle;
1420         struct mlx5_flow_table *ft;
1421         int modify_mask = 0;
1422         int err;
1423         bool new_rule = false;
1424
1425         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1426                                     &new_rule);
1427         if (IS_ERR(handle) || !new_rule)
1428                 goto out;
1429
1430         if (update_action)
1431                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1432
1433         fs_get_obj(ft, fg->node.parent);
1434         root = find_root(&fg->node);
1435         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1436                 err = root->cmds->create_fte(root, ft, fg, fte);
1437         else
1438                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1439         if (err)
1440                 goto free_handle;
1441
1442         fte->node.active = true;
1443         fte->status |= FS_FTE_STATUS_EXISTING;
1444         atomic_inc(&fg->node.version);
1445
1446 out:
1447         return handle;
1448
1449 free_handle:
1450         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1451         return ERR_PTR(err);
1452 }
1453
1454 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1455                                                      const struct mlx5_flow_spec *spec)
1456 {
1457         struct list_head *prev = &ft->node.children;
1458         u32 max_fte = ft->autogroup.max_fte;
1459         unsigned int candidate_index = 0;
1460         unsigned int group_size = 0;
1461         struct mlx5_flow_group *fg;
1462
1463         if (!ft->autogroup.active)
1464                 return ERR_PTR(-ENOENT);
1465
1466         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1467                 group_size = ft->autogroup.group_size;
1468
1469         /*  max_fte == ft->autogroup.max_types */
1470         if (group_size == 0)
1471                 group_size = 1;
1472
1473         /* sorted by start_index */
1474         fs_for_each_fg(fg, ft) {
1475                 if (candidate_index + group_size > fg->start_index)
1476                         candidate_index = fg->start_index + fg->max_ftes;
1477                 else
1478                         break;
1479                 prev = &fg->node.list;
1480         }
1481
1482         if (candidate_index + group_size > max_fte)
1483                 return ERR_PTR(-ENOSPC);
1484
1485         fg = alloc_insert_flow_group(ft,
1486                                      spec->match_criteria_enable,
1487                                      spec->match_criteria,
1488                                      candidate_index,
1489                                      candidate_index + group_size - 1,
1490                                      prev);
1491         if (IS_ERR(fg))
1492                 goto out;
1493
1494         if (group_size == ft->autogroup.group_size)
1495                 ft->autogroup.num_groups++;
1496
1497 out:
1498         return fg;
1499 }
1500
1501 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1502                                   struct mlx5_flow_group *fg)
1503 {
1504         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1505         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1506         void *match_criteria_addr;
1507         u8 src_esw_owner_mask_on;
1508         void *misc;
1509         int err;
1510         u32 *in;
1511
1512         in = kvzalloc(inlen, GFP_KERNEL);
1513         if (!in)
1514                 return -ENOMEM;
1515
1516         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1517                  fg->mask.match_criteria_enable);
1518         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1519         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1520                  fg->max_ftes - 1);
1521
1522         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1523                             misc_parameters);
1524         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1525                                          source_eswitch_owner_vhca_id);
1526         MLX5_SET(create_flow_group_in, in,
1527                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1528
1529         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1530                                            in, match_criteria);
1531         memcpy(match_criteria_addr, fg->mask.match_criteria,
1532                sizeof(fg->mask.match_criteria));
1533
1534         err = root->cmds->create_flow_group(root, ft, in, fg);
1535         if (!err) {
1536                 fg->node.active = true;
1537                 trace_mlx5_fs_add_fg(fg);
1538         }
1539
1540         kvfree(in);
1541         return err;
1542 }
1543
1544 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1545                                 struct mlx5_flow_destination *d2)
1546 {
1547         if (d1->type == d2->type) {
1548                 if (((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT ||
1549                       d1->type == MLX5_FLOW_DESTINATION_TYPE_UPLINK) &&
1550                      d1->vport.num == d2->vport.num &&
1551                      d1->vport.flags == d2->vport.flags &&
1552                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1553                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1554                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1555                       (d1->vport.pkt_reformat->id ==
1556                        d2->vport.pkt_reformat->id) : true)) ||
1557                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1558                      d1->ft == d2->ft) ||
1559                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1560                      d1->tir_num == d2->tir_num) ||
1561                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1562                      d1->ft_num == d2->ft_num) ||
1563                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER &&
1564                      d1->sampler_id == d2->sampler_id))
1565                         return true;
1566         }
1567
1568         return false;
1569 }
1570
1571 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1572                                              struct mlx5_flow_destination *dest)
1573 {
1574         struct mlx5_flow_rule *rule;
1575
1576         list_for_each_entry(rule, &fte->node.children, node.list) {
1577                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1578                         return rule;
1579         }
1580         return NULL;
1581 }
1582
1583 static bool check_conflicting_actions(u32 action1, u32 action2)
1584 {
1585         u32 xored_actions = action1 ^ action2;
1586
1587         /* if one rule only wants to count, it's ok */
1588         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1589             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1590                 return false;
1591
1592         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1593                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1594                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1595                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1596                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1597                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1598                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1599                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1600                 return true;
1601
1602         return false;
1603 }
1604
1605 static int check_conflicting_ftes(struct fs_fte *fte,
1606                                   const struct mlx5_flow_context *flow_context,
1607                                   const struct mlx5_flow_act *flow_act)
1608 {
1609         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1610                 mlx5_core_warn(get_dev(&fte->node),
1611                                "Found two FTEs with conflicting actions\n");
1612                 return -EEXIST;
1613         }
1614
1615         if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1616             fte->flow_context.flow_tag != flow_context->flow_tag) {
1617                 mlx5_core_warn(get_dev(&fte->node),
1618                                "FTE flow tag %u already exists with different flow tag %u\n",
1619                                fte->flow_context.flow_tag,
1620                                flow_context->flow_tag);
1621                 return -EEXIST;
1622         }
1623
1624         return 0;
1625 }
1626
1627 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1628                                             const struct mlx5_flow_spec *spec,
1629                                             struct mlx5_flow_act *flow_act,
1630                                             struct mlx5_flow_destination *dest,
1631                                             int dest_num,
1632                                             struct fs_fte *fte)
1633 {
1634         struct mlx5_flow_handle *handle;
1635         int old_action;
1636         int i;
1637         int ret;
1638
1639         ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1640         if (ret)
1641                 return ERR_PTR(ret);
1642
1643         old_action = fte->action.action;
1644         fte->action.action |= flow_act->action;
1645         handle = add_rule_fte(fte, fg, dest, dest_num,
1646                               old_action != flow_act->action);
1647         if (IS_ERR(handle)) {
1648                 fte->action.action = old_action;
1649                 return handle;
1650         }
1651         trace_mlx5_fs_set_fte(fte, false);
1652
1653         for (i = 0; i < handle->num_rules; i++) {
1654                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1655                         tree_add_node(&handle->rule[i]->node, &fte->node);
1656                         trace_mlx5_fs_add_rule(handle->rule[i]);
1657                 }
1658         }
1659         return handle;
1660 }
1661
1662 static bool counter_is_valid(u32 action)
1663 {
1664         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1665                           MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1666                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1667 }
1668
1669 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1670                           struct mlx5_flow_act *flow_act,
1671                           struct mlx5_flow_table *ft)
1672 {
1673         bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1674         u32 action = flow_act->action;
1675
1676         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1677                 return counter_is_valid(action);
1678
1679         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1680                 return true;
1681
1682         if (ignore_level) {
1683                 if (ft->type != FS_FT_FDB &&
1684                     ft->type != FS_FT_NIC_RX)
1685                         return false;
1686
1687                 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1688                     ft->type != dest->ft->type)
1689                         return false;
1690         }
1691
1692         if (!dest || ((dest->type ==
1693             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1694             (dest->ft->level <= ft->level && !ignore_level)))
1695                 return false;
1696         return true;
1697 }
1698
1699 struct match_list {
1700         struct list_head        list;
1701         struct mlx5_flow_group *g;
1702 };
1703
1704 static void free_match_list(struct match_list *head, bool ft_locked)
1705 {
1706         struct match_list *iter, *match_tmp;
1707
1708         list_for_each_entry_safe(iter, match_tmp, &head->list,
1709                                  list) {
1710                 tree_put_node(&iter->g->node, ft_locked);
1711                 list_del(&iter->list);
1712                 kfree(iter);
1713         }
1714 }
1715
1716 static int build_match_list(struct match_list *match_head,
1717                             struct mlx5_flow_table *ft,
1718                             const struct mlx5_flow_spec *spec,
1719                             struct mlx5_flow_group *fg,
1720                             bool ft_locked)
1721 {
1722         struct rhlist_head *tmp, *list;
1723         struct mlx5_flow_group *g;
1724         int err = 0;
1725
1726         rcu_read_lock();
1727         INIT_LIST_HEAD(&match_head->list);
1728         /* Collect all fgs which has a matching match_criteria */
1729         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1730         /* RCU is atomic, we can't execute FW commands here */
1731         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1732                 struct match_list *curr_match;
1733
1734                 if (fg && fg != g)
1735                         continue;
1736
1737                 if (unlikely(!tree_get_node(&g->node)))
1738                         continue;
1739
1740                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1741                 if (!curr_match) {
1742                         rcu_read_unlock();
1743                         free_match_list(match_head, ft_locked);
1744                         return -ENOMEM;
1745                 }
1746                 curr_match->g = g;
1747                 list_add_tail(&curr_match->list, &match_head->list);
1748         }
1749         rcu_read_unlock();
1750         return err;
1751 }
1752
1753 static u64 matched_fgs_get_version(struct list_head *match_head)
1754 {
1755         struct match_list *iter;
1756         u64 version = 0;
1757
1758         list_for_each_entry(iter, match_head, list)
1759                 version += (u64)atomic_read(&iter->g->node.version);
1760         return version;
1761 }
1762
1763 static struct fs_fte *
1764 lookup_fte_locked(struct mlx5_flow_group *g,
1765                   const u32 *match_value,
1766                   bool take_write)
1767 {
1768         struct fs_fte *fte_tmp;
1769
1770         if (take_write)
1771                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1772         else
1773                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1774         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1775                                          rhash_fte);
1776         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1777                 fte_tmp = NULL;
1778                 goto out;
1779         }
1780         if (!fte_tmp->node.active) {
1781                 tree_put_node(&fte_tmp->node, false);
1782                 fte_tmp = NULL;
1783                 goto out;
1784         }
1785
1786         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1787 out:
1788         if (take_write)
1789                 up_write_ref_node(&g->node, false);
1790         else
1791                 up_read_ref_node(&g->node);
1792         return fte_tmp;
1793 }
1794
1795 static struct mlx5_flow_handle *
1796 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1797                        struct list_head *match_head,
1798                        const struct mlx5_flow_spec *spec,
1799                        struct mlx5_flow_act *flow_act,
1800                        struct mlx5_flow_destination *dest,
1801                        int dest_num,
1802                        int ft_version)
1803 {
1804         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1805         struct mlx5_flow_group *g;
1806         struct mlx5_flow_handle *rule;
1807         struct match_list *iter;
1808         bool take_write = false;
1809         struct fs_fte *fte;
1810         u64  version = 0;
1811         int err;
1812
1813         fte = alloc_fte(ft, spec, flow_act);
1814         if (IS_ERR(fte))
1815                 return  ERR_PTR(-ENOMEM);
1816
1817 search_again_locked:
1818         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1819                 goto skip_search;
1820         version = matched_fgs_get_version(match_head);
1821         /* Try to find an fte with identical match value and attempt update its
1822          * action.
1823          */
1824         list_for_each_entry(iter, match_head, list) {
1825                 struct fs_fte *fte_tmp;
1826
1827                 g = iter->g;
1828                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1829                 if (!fte_tmp)
1830                         continue;
1831                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1832                 /* No error check needed here, because insert_fte() is not called */
1833                 up_write_ref_node(&fte_tmp->node, false);
1834                 tree_put_node(&fte_tmp->node, false);
1835                 kmem_cache_free(steering->ftes_cache, fte);
1836                 return rule;
1837         }
1838
1839 skip_search:
1840         /* No group with matching fte found, or we skipped the search.
1841          * Try to add a new fte to any matching fg.
1842          */
1843
1844         /* Check the ft version, for case that new flow group
1845          * was added while the fgs weren't locked
1846          */
1847         if (atomic_read(&ft->node.version) != ft_version) {
1848                 rule = ERR_PTR(-EAGAIN);
1849                 goto out;
1850         }
1851
1852         /* Check the fgs version. If version have changed it could be that an
1853          * FTE with the same match value was added while the fgs weren't
1854          * locked.
1855          */
1856         if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1857             version != matched_fgs_get_version(match_head)) {
1858                 take_write = true;
1859                 goto search_again_locked;
1860         }
1861
1862         list_for_each_entry(iter, match_head, list) {
1863                 g = iter->g;
1864
1865                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1866
1867                 if (!g->node.active) {
1868                         up_write_ref_node(&g->node, false);
1869                         continue;
1870                 }
1871
1872                 err = insert_fte(g, fte);
1873                 if (err) {
1874                         up_write_ref_node(&g->node, false);
1875                         if (err == -ENOSPC)
1876                                 continue;
1877                         kmem_cache_free(steering->ftes_cache, fte);
1878                         return ERR_PTR(err);
1879                 }
1880
1881                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1882                 up_write_ref_node(&g->node, false);
1883                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1884                 up_write_ref_node(&fte->node, false);
1885                 if (IS_ERR(rule))
1886                         tree_put_node(&fte->node, false);
1887                 return rule;
1888         }
1889         rule = ERR_PTR(-ENOENT);
1890 out:
1891         kmem_cache_free(steering->ftes_cache, fte);
1892         return rule;
1893 }
1894
1895 static struct mlx5_flow_handle *
1896 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1897                      const struct mlx5_flow_spec *spec,
1898                      struct mlx5_flow_act *flow_act,
1899                      struct mlx5_flow_destination *dest,
1900                      int dest_num)
1901
1902 {
1903         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1904         struct mlx5_flow_handle *rule;
1905         struct match_list match_head;
1906         struct mlx5_flow_group *g;
1907         bool take_write = false;
1908         struct fs_fte *fte;
1909         int version;
1910         int err;
1911         int i;
1912
1913         if (!check_valid_spec(spec))
1914                 return ERR_PTR(-EINVAL);
1915
1916         if (flow_act->fg && ft->autogroup.active)
1917                 return ERR_PTR(-EINVAL);
1918
1919         for (i = 0; i < dest_num; i++) {
1920                 if (!dest_is_valid(&dest[i], flow_act, ft))
1921                         return ERR_PTR(-EINVAL);
1922         }
1923         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1924 search_again_locked:
1925         version = atomic_read(&ft->node.version);
1926
1927         /* Collect all fgs which has a matching match_criteria */
1928         err = build_match_list(&match_head, ft, spec, flow_act->fg, take_write);
1929         if (err) {
1930                 if (take_write)
1931                         up_write_ref_node(&ft->node, false);
1932                 else
1933                         up_read_ref_node(&ft->node);
1934                 return ERR_PTR(err);
1935         }
1936
1937         if (!take_write)
1938                 up_read_ref_node(&ft->node);
1939
1940         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1941                                       dest_num, version);
1942         free_match_list(&match_head, take_write);
1943         if (!IS_ERR(rule) ||
1944             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1945                 if (take_write)
1946                         up_write_ref_node(&ft->node, false);
1947                 return rule;
1948         }
1949
1950         if (!take_write) {
1951                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1952                 take_write = true;
1953         }
1954
1955         if (PTR_ERR(rule) == -EAGAIN ||
1956             version != atomic_read(&ft->node.version))
1957                 goto search_again_locked;
1958
1959         g = alloc_auto_flow_group(ft, spec);
1960         if (IS_ERR(g)) {
1961                 rule = ERR_CAST(g);
1962                 up_write_ref_node(&ft->node, false);
1963                 return rule;
1964         }
1965
1966         fte = alloc_fte(ft, spec, flow_act);
1967         if (IS_ERR(fte)) {
1968                 up_write_ref_node(&ft->node, false);
1969                 err = PTR_ERR(fte);
1970                 goto err_alloc_fte;
1971         }
1972
1973         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1974         up_write_ref_node(&ft->node, false);
1975
1976         err = create_auto_flow_group(ft, g);
1977         if (err)
1978                 goto err_release_fg;
1979
1980         err = insert_fte(g, fte);
1981         if (err)
1982                 goto err_release_fg;
1983
1984         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1985         up_write_ref_node(&g->node, false);
1986         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1987         up_write_ref_node(&fte->node, false);
1988         if (IS_ERR(rule))
1989                 tree_put_node(&fte->node, false);
1990         tree_put_node(&g->node, false);
1991         return rule;
1992
1993 err_release_fg:
1994         up_write_ref_node(&g->node, false);
1995         kmem_cache_free(steering->ftes_cache, fte);
1996 err_alloc_fte:
1997         tree_put_node(&g->node, false);
1998         return ERR_PTR(err);
1999 }
2000
2001 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
2002 {
2003         return ((ft->type == FS_FT_NIC_RX) &&
2004                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
2005 }
2006
2007 struct mlx5_flow_handle *
2008 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
2009                     const struct mlx5_flow_spec *spec,
2010                     struct mlx5_flow_act *flow_act,
2011                     struct mlx5_flow_destination *dest,
2012                     int num_dest)
2013 {
2014         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2015         static const struct mlx5_flow_spec zero_spec = {};
2016         struct mlx5_flow_destination *gen_dest = NULL;
2017         struct mlx5_flow_table *next_ft = NULL;
2018         struct mlx5_flow_handle *handle = NULL;
2019         u32 sw_action = flow_act->action;
2020         int i;
2021
2022         if (!spec)
2023                 spec = &zero_spec;
2024
2025         if (!is_fwd_next_action(sw_action))
2026                 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2027
2028         if (!fwd_next_prio_supported(ft))
2029                 return ERR_PTR(-EOPNOTSUPP);
2030
2031         mutex_lock(&root->chain_lock);
2032         next_ft = find_next_fwd_ft(ft, flow_act);
2033         if (!next_ft) {
2034                 handle = ERR_PTR(-EOPNOTSUPP);
2035                 goto unlock;
2036         }
2037
2038         gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
2039                            GFP_KERNEL);
2040         if (!gen_dest) {
2041                 handle = ERR_PTR(-ENOMEM);
2042                 goto unlock;
2043         }
2044         for (i = 0; i < num_dest; i++)
2045                 gen_dest[i] = dest[i];
2046         gen_dest[i].type =
2047                 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
2048         gen_dest[i].ft = next_ft;
2049         dest = gen_dest;
2050         num_dest++;
2051         flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
2052                               MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
2053         flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
2054         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
2055         if (IS_ERR(handle))
2056                 goto unlock;
2057
2058         if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
2059                 mutex_lock(&next_ft->lock);
2060                 list_add(&handle->rule[num_dest - 1]->next_ft,
2061                          &next_ft->fwd_rules);
2062                 mutex_unlock(&next_ft->lock);
2063                 handle->rule[num_dest - 1]->sw_action = sw_action;
2064                 handle->rule[num_dest - 1]->ft = ft;
2065         }
2066 unlock:
2067         mutex_unlock(&root->chain_lock);
2068         kfree(gen_dest);
2069         return handle;
2070 }
2071 EXPORT_SYMBOL(mlx5_add_flow_rules);
2072
2073 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2074 {
2075         struct fs_fte *fte;
2076         int i;
2077
2078         /* In order to consolidate the HW changes we lock the FTE for other
2079          * changes, and increase its refcount, in order not to perform the
2080          * "del" functions of the FTE. Will handle them here.
2081          * The removal of the rules is done under locked FTE.
2082          * After removing all the handle's rules, if there are remaining
2083          * rules, it means we just need to modify the FTE in FW, and
2084          * unlock/decrease the refcount we increased before.
2085          * Otherwise, it means the FTE should be deleted. First delete the
2086          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2087          * the FTE, which will handle the last decrease of the refcount, as
2088          * well as required handling of its parent.
2089          */
2090         fs_get_obj(fte, handle->rule[0]->node.parent);
2091         down_write_ref_node(&fte->node, false);
2092         for (i = handle->num_rules - 1; i >= 0; i--)
2093                 tree_remove_node(&handle->rule[i]->node, true);
2094         if (list_empty(&fte->node.children)) {
2095                 fte->node.del_hw_func(&fte->node);
2096                 /* Avoid double call to del_hw_fte */
2097                 fte->node.del_hw_func = NULL;
2098                 up_write_ref_node(&fte->node, false);
2099                 tree_put_node(&fte->node, false);
2100         } else if (fte->dests_size) {
2101                 if (fte->modify_mask)
2102                         modify_fte(fte);
2103                 up_write_ref_node(&fte->node, false);
2104         } else {
2105                 up_write_ref_node(&fte->node, false);
2106         }
2107         kfree(handle);
2108 }
2109 EXPORT_SYMBOL(mlx5_del_flow_rules);
2110
2111 /* Assuming prio->node.children(flow tables) is sorted by level */
2112 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2113 {
2114         struct fs_prio *prio;
2115
2116         fs_get_obj(prio, ft->node.parent);
2117
2118         if (!list_is_last(&ft->node.list, &prio->node.children))
2119                 return list_next_entry(ft, node.list);
2120         return find_next_chained_ft(prio);
2121 }
2122
2123 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2124 {
2125         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2126         struct mlx5_ft_underlay_qp *uqp;
2127         struct mlx5_flow_table *new_root_ft = NULL;
2128         int err = 0;
2129         u32 qpn;
2130
2131         if (root->root_ft != ft)
2132                 return 0;
2133
2134         new_root_ft = find_next_ft(ft);
2135         if (!new_root_ft) {
2136                 root->root_ft = NULL;
2137                 return 0;
2138         }
2139
2140         if (list_empty(&root->underlay_qpns)) {
2141                 /* Don't set any QPN (zero) in case QPN list is empty */
2142                 qpn = 0;
2143                 err = root->cmds->update_root_ft(root, new_root_ft,
2144                                                  qpn, false);
2145         } else {
2146                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2147                         qpn = uqp->qpn;
2148                         err = root->cmds->update_root_ft(root,
2149                                                          new_root_ft, qpn,
2150                                                          false);
2151                         if (err)
2152                                 break;
2153                 }
2154         }
2155
2156         if (err)
2157                 mlx5_core_warn(root->dev,
2158                                "Update root flow table of id(%u) qpn(%d) failed\n",
2159                                ft->id, qpn);
2160         else
2161                 root->root_ft = new_root_ft;
2162
2163         return 0;
2164 }
2165
2166 /* Connect flow table from previous priority to
2167  * the next flow table.
2168  */
2169 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2170 {
2171         struct mlx5_core_dev *dev = get_dev(&ft->node);
2172         struct mlx5_flow_table *next_ft;
2173         struct fs_prio *prio;
2174         int err = 0;
2175
2176         err = update_root_ft_destroy(ft);
2177         if (err)
2178                 return err;
2179
2180         fs_get_obj(prio, ft->node.parent);
2181         if  (!(list_first_entry(&prio->node.children,
2182                                 struct mlx5_flow_table,
2183                                 node.list) == ft))
2184                 return 0;
2185
2186         next_ft = find_next_ft(ft);
2187         err = connect_fwd_rules(dev, next_ft, ft);
2188         if (err)
2189                 return err;
2190
2191         err = connect_prev_fts(dev, next_ft, prio);
2192         if (err)
2193                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2194                                ft->id);
2195         return err;
2196 }
2197
2198 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2199 {
2200         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2201         int err = 0;
2202
2203         mutex_lock(&root->chain_lock);
2204         if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2205                 err = disconnect_flow_table(ft);
2206         if (err) {
2207                 mutex_unlock(&root->chain_lock);
2208                 return err;
2209         }
2210         if (tree_remove_node(&ft->node, false))
2211                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2212                                ft->id);
2213         mutex_unlock(&root->chain_lock);
2214
2215         return err;
2216 }
2217 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2218
2219 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2220 {
2221         if (tree_remove_node(&fg->node, false))
2222                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2223                                fg->id);
2224 }
2225 EXPORT_SYMBOL(mlx5_destroy_flow_group);
2226
2227 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2228                                                 int n)
2229 {
2230         struct mlx5_flow_steering *steering = dev->priv.steering;
2231
2232         if (!steering || !steering->fdb_sub_ns)
2233                 return NULL;
2234
2235         return steering->fdb_sub_ns[n];
2236 }
2237 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2238
2239 static bool is_nic_rx_ns(enum mlx5_flow_namespace_type type)
2240 {
2241         switch (type) {
2242         case MLX5_FLOW_NAMESPACE_BYPASS:
2243         case MLX5_FLOW_NAMESPACE_LAG:
2244         case MLX5_FLOW_NAMESPACE_OFFLOADS:
2245         case MLX5_FLOW_NAMESPACE_ETHTOOL:
2246         case MLX5_FLOW_NAMESPACE_KERNEL:
2247         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
2248         case MLX5_FLOW_NAMESPACE_ANCHOR:
2249                 return true;
2250         default:
2251                 return false;
2252         }
2253 }
2254
2255 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2256                                                     enum mlx5_flow_namespace_type type)
2257 {
2258         struct mlx5_flow_steering *steering = dev->priv.steering;
2259         struct mlx5_flow_root_namespace *root_ns;
2260         int prio = 0;
2261         struct fs_prio *fs_prio;
2262         struct mlx5_flow_namespace *ns;
2263
2264         if (!steering)
2265                 return NULL;
2266
2267         switch (type) {
2268         case MLX5_FLOW_NAMESPACE_FDB:
2269                 if (steering->fdb_root_ns)
2270                         return &steering->fdb_root_ns->ns;
2271                 return NULL;
2272         case MLX5_FLOW_NAMESPACE_PORT_SEL:
2273                 if (steering->port_sel_root_ns)
2274                         return &steering->port_sel_root_ns->ns;
2275                 return NULL;
2276         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2277                 if (steering->sniffer_rx_root_ns)
2278                         return &steering->sniffer_rx_root_ns->ns;
2279                 return NULL;
2280         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2281                 if (steering->sniffer_tx_root_ns)
2282                         return &steering->sniffer_tx_root_ns->ns;
2283                 return NULL;
2284         case MLX5_FLOW_NAMESPACE_FDB_BYPASS:
2285                 root_ns = steering->fdb_root_ns;
2286                 prio =  FDB_BYPASS_PATH;
2287                 break;
2288         case MLX5_FLOW_NAMESPACE_EGRESS:
2289         case MLX5_FLOW_NAMESPACE_EGRESS_KERNEL:
2290                 root_ns = steering->egress_root_ns;
2291                 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2292                 break;
2293         case MLX5_FLOW_NAMESPACE_RDMA_RX:
2294                 root_ns = steering->rdma_rx_root_ns;
2295                 prio = RDMA_RX_BYPASS_PRIO;
2296                 break;
2297         case MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL:
2298                 root_ns = steering->rdma_rx_root_ns;
2299                 prio = RDMA_RX_KERNEL_PRIO;
2300                 break;
2301         case MLX5_FLOW_NAMESPACE_RDMA_TX:
2302                 root_ns = steering->rdma_tx_root_ns;
2303                 break;
2304         case MLX5_FLOW_NAMESPACE_RDMA_RX_COUNTERS:
2305                 root_ns = steering->rdma_rx_root_ns;
2306                 prio = RDMA_RX_COUNTERS_PRIO;
2307                 break;
2308         case MLX5_FLOW_NAMESPACE_RDMA_TX_COUNTERS:
2309                 root_ns = steering->rdma_tx_root_ns;
2310                 prio = RDMA_TX_COUNTERS_PRIO;
2311                 break;
2312         default: /* Must be NIC RX */
2313                 WARN_ON(!is_nic_rx_ns(type));
2314                 root_ns = steering->root_ns;
2315                 prio = type;
2316                 break;
2317         }
2318
2319         if (!root_ns)
2320                 return NULL;
2321
2322         fs_prio = find_prio(&root_ns->ns, prio);
2323         if (!fs_prio)
2324                 return NULL;
2325
2326         ns = list_first_entry(&fs_prio->node.children,
2327                               typeof(*ns),
2328                               node.list);
2329
2330         return ns;
2331 }
2332 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2333
2334 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2335                                                               enum mlx5_flow_namespace_type type,
2336                                                               int vport)
2337 {
2338         struct mlx5_flow_steering *steering = dev->priv.steering;
2339
2340         if (!steering)
2341                 return NULL;
2342
2343         switch (type) {
2344         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2345                 if (vport >= steering->esw_egress_acl_vports)
2346                         return NULL;
2347                 if (steering->esw_egress_root_ns &&
2348                     steering->esw_egress_root_ns[vport])
2349                         return &steering->esw_egress_root_ns[vport]->ns;
2350                 else
2351                         return NULL;
2352         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2353                 if (vport >= steering->esw_ingress_acl_vports)
2354                         return NULL;
2355                 if (steering->esw_ingress_root_ns &&
2356                     steering->esw_ingress_root_ns[vport])
2357                         return &steering->esw_ingress_root_ns[vport]->ns;
2358                 else
2359                         return NULL;
2360         default:
2361                 return NULL;
2362         }
2363 }
2364
2365 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2366                                        unsigned int prio,
2367                                        int num_levels,
2368                                        enum fs_node_type type)
2369 {
2370         struct fs_prio *fs_prio;
2371
2372         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2373         if (!fs_prio)
2374                 return ERR_PTR(-ENOMEM);
2375
2376         fs_prio->node.type = type;
2377         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2378         tree_add_node(&fs_prio->node, &ns->node);
2379         fs_prio->num_levels = num_levels;
2380         fs_prio->prio = prio;
2381         list_add_tail(&fs_prio->node.list, &ns->node.children);
2382
2383         return fs_prio;
2384 }
2385
2386 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2387                                               unsigned int prio,
2388                                               int num_levels)
2389 {
2390         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2391 }
2392
2393 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2394                                       unsigned int prio, int num_levels)
2395 {
2396         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2397 }
2398
2399 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2400                                                      *ns)
2401 {
2402         ns->node.type = FS_TYPE_NAMESPACE;
2403
2404         return ns;
2405 }
2406
2407 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2408                                                        int def_miss_act)
2409 {
2410         struct mlx5_flow_namespace      *ns;
2411
2412         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2413         if (!ns)
2414                 return ERR_PTR(-ENOMEM);
2415
2416         fs_init_namespace(ns);
2417         ns->def_miss_action = def_miss_act;
2418         tree_init_node(&ns->node, NULL, del_sw_ns);
2419         tree_add_node(&ns->node, &prio->node);
2420         list_add_tail(&ns->node.list, &prio->node.children);
2421
2422         return ns;
2423 }
2424
2425 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2426                              struct init_tree_node *prio_metadata)
2427 {
2428         struct fs_prio *fs_prio;
2429         int i;
2430
2431         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2432                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2433                 if (IS_ERR(fs_prio))
2434                         return PTR_ERR(fs_prio);
2435         }
2436         return 0;
2437 }
2438
2439 #define FLOW_TABLE_BIT_SZ 1
2440 #define GET_FLOW_TABLE_CAP(dev, offset) \
2441         ((be32_to_cpu(*((__be32 *)(dev->caps.hca[MLX5_CAP_FLOW_TABLE]->cur) +   \
2442                         offset / 32)) >>                                        \
2443           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2444 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2445 {
2446         int i;
2447
2448         for (i = 0; i < caps->arr_sz; i++) {
2449                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2450                         return false;
2451         }
2452         return true;
2453 }
2454
2455 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2456                                     struct init_tree_node *init_node,
2457                                     struct fs_node *fs_parent_node,
2458                                     struct init_tree_node *init_parent_node,
2459                                     int prio)
2460 {
2461         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2462                                               flow_table_properties_nic_receive.
2463                                               max_ft_level);
2464         struct mlx5_flow_namespace *fs_ns;
2465         struct fs_prio *fs_prio;
2466         struct fs_node *base;
2467         int i;
2468         int err;
2469
2470         if (init_node->type == FS_TYPE_PRIO) {
2471                 if ((init_node->min_ft_level > max_ft_level) ||
2472                     !has_required_caps(steering->dev, &init_node->caps))
2473                         return 0;
2474
2475                 fs_get_obj(fs_ns, fs_parent_node);
2476                 if (init_node->num_leaf_prios)
2477                         return create_leaf_prios(fs_ns, prio, init_node);
2478                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2479                 if (IS_ERR(fs_prio))
2480                         return PTR_ERR(fs_prio);
2481                 base = &fs_prio->node;
2482         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2483                 fs_get_obj(fs_prio, fs_parent_node);
2484                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2485                 if (IS_ERR(fs_ns))
2486                         return PTR_ERR(fs_ns);
2487                 base = &fs_ns->node;
2488         } else {
2489                 return -EINVAL;
2490         }
2491         prio = 0;
2492         for (i = 0; i < init_node->ar_size; i++) {
2493                 err = init_root_tree_recursive(steering, &init_node->children[i],
2494                                                base, init_node, prio);
2495                 if (err)
2496                         return err;
2497                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2498                     init_node->children[i].num_leaf_prios) {
2499                         prio += init_node->children[i].num_leaf_prios;
2500                 }
2501         }
2502
2503         return 0;
2504 }
2505
2506 static int init_root_tree(struct mlx5_flow_steering *steering,
2507                           struct init_tree_node *init_node,
2508                           struct fs_node *fs_parent_node)
2509 {
2510         int err;
2511         int i;
2512
2513         for (i = 0; i < init_node->ar_size; i++) {
2514                 err = init_root_tree_recursive(steering, &init_node->children[i],
2515                                                fs_parent_node,
2516                                                init_node, i);
2517                 if (err)
2518                         return err;
2519         }
2520         return 0;
2521 }
2522
2523 static void del_sw_root_ns(struct fs_node *node)
2524 {
2525         struct mlx5_flow_root_namespace *root_ns;
2526         struct mlx5_flow_namespace *ns;
2527
2528         fs_get_obj(ns, node);
2529         root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2530         mutex_destroy(&root_ns->chain_lock);
2531         kfree(node);
2532 }
2533
2534 static struct mlx5_flow_root_namespace
2535 *create_root_ns(struct mlx5_flow_steering *steering,
2536                 enum fs_flow_table_type table_type)
2537 {
2538         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2539         struct mlx5_flow_root_namespace *root_ns;
2540         struct mlx5_flow_namespace *ns;
2541
2542         /* Create the root namespace */
2543         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2544         if (!root_ns)
2545                 return NULL;
2546
2547         root_ns->dev = steering->dev;
2548         root_ns->table_type = table_type;
2549         root_ns->cmds = cmds;
2550
2551         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2552
2553         ns = &root_ns->ns;
2554         fs_init_namespace(ns);
2555         mutex_init(&root_ns->chain_lock);
2556         tree_init_node(&ns->node, NULL, del_sw_root_ns);
2557         tree_add_node(&ns->node, NULL);
2558
2559         return root_ns;
2560 }
2561
2562 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2563
2564 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2565 {
2566         struct fs_prio *prio;
2567
2568         fs_for_each_prio(prio, ns) {
2569                  /* This updates prio start_level and num_levels */
2570                 set_prio_attrs_in_prio(prio, acc_level);
2571                 acc_level += prio->num_levels;
2572         }
2573         return acc_level;
2574 }
2575
2576 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2577 {
2578         struct mlx5_flow_namespace *ns;
2579         int acc_level_ns = acc_level;
2580
2581         prio->start_level = acc_level;
2582         fs_for_each_ns(ns, prio) {
2583                 /* This updates start_level and num_levels of ns's priority descendants */
2584                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2585
2586                 /* If this a prio with chains, and we can jump from one chain
2587                  * (namespace) to another, so we accumulate the levels
2588                  */
2589                 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2590                         acc_level = acc_level_ns;
2591         }
2592
2593         if (!prio->num_levels)
2594                 prio->num_levels = acc_level_ns - prio->start_level;
2595         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2596 }
2597
2598 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2599 {
2600         struct mlx5_flow_namespace *ns = &root_ns->ns;
2601         struct fs_prio *prio;
2602         int start_level = 0;
2603
2604         fs_for_each_prio(prio, ns) {
2605                 set_prio_attrs_in_prio(prio, start_level);
2606                 start_level += prio->num_levels;
2607         }
2608 }
2609
2610 #define ANCHOR_PRIO 0
2611 #define ANCHOR_SIZE 1
2612 #define ANCHOR_LEVEL 0
2613 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2614 {
2615         struct mlx5_flow_namespace *ns = NULL;
2616         struct mlx5_flow_table_attr ft_attr = {};
2617         struct mlx5_flow_table *ft;
2618
2619         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2620         if (WARN_ON(!ns))
2621                 return -EINVAL;
2622
2623         ft_attr.max_fte = ANCHOR_SIZE;
2624         ft_attr.level   = ANCHOR_LEVEL;
2625         ft_attr.prio    = ANCHOR_PRIO;
2626
2627         ft = mlx5_create_flow_table(ns, &ft_attr);
2628         if (IS_ERR(ft)) {
2629                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2630                 return PTR_ERR(ft);
2631         }
2632         return 0;
2633 }
2634
2635 static int init_root_ns(struct mlx5_flow_steering *steering)
2636 {
2637         int err;
2638
2639         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2640         if (!steering->root_ns)
2641                 return -ENOMEM;
2642
2643         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2644         if (err)
2645                 goto out_err;
2646
2647         set_prio_attrs(steering->root_ns);
2648         err = create_anchor_flow_table(steering);
2649         if (err)
2650                 goto out_err;
2651
2652         return 0;
2653
2654 out_err:
2655         cleanup_root_ns(steering->root_ns);
2656         steering->root_ns = NULL;
2657         return err;
2658 }
2659
2660 static void clean_tree(struct fs_node *node)
2661 {
2662         if (node) {
2663                 struct fs_node *iter;
2664                 struct fs_node *temp;
2665
2666                 tree_get_node(node);
2667                 list_for_each_entry_safe(iter, temp, &node->children, list)
2668                         clean_tree(iter);
2669                 tree_put_node(node, false);
2670                 tree_remove_node(node, false);
2671         }
2672 }
2673
2674 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2675 {
2676         if (!root_ns)
2677                 return;
2678
2679         clean_tree(&root_ns->ns.node);
2680 }
2681
2682 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2683 {
2684         struct fs_prio *prio;
2685
2686         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2687         if (!steering->sniffer_tx_root_ns)
2688                 return -ENOMEM;
2689
2690         /* Create single prio */
2691         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2692         return PTR_ERR_OR_ZERO(prio);
2693 }
2694
2695 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2696 {
2697         struct fs_prio *prio;
2698
2699         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2700         if (!steering->sniffer_rx_root_ns)
2701                 return -ENOMEM;
2702
2703         /* Create single prio */
2704         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2705         return PTR_ERR_OR_ZERO(prio);
2706 }
2707
2708 #define PORT_SEL_NUM_LEVELS 3
2709 static int init_port_sel_root_ns(struct mlx5_flow_steering *steering)
2710 {
2711         struct fs_prio *prio;
2712
2713         steering->port_sel_root_ns = create_root_ns(steering, FS_FT_PORT_SEL);
2714         if (!steering->port_sel_root_ns)
2715                 return -ENOMEM;
2716
2717         /* Create single prio */
2718         prio = fs_create_prio(&steering->port_sel_root_ns->ns, 0,
2719                               PORT_SEL_NUM_LEVELS);
2720         return PTR_ERR_OR_ZERO(prio);
2721 }
2722
2723 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2724 {
2725         int err;
2726
2727         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2728         if (!steering->rdma_rx_root_ns)
2729                 return -ENOMEM;
2730
2731         err = init_root_tree(steering, &rdma_rx_root_fs,
2732                              &steering->rdma_rx_root_ns->ns.node);
2733         if (err)
2734                 goto out_err;
2735
2736         set_prio_attrs(steering->rdma_rx_root_ns);
2737
2738         return 0;
2739
2740 out_err:
2741         cleanup_root_ns(steering->rdma_rx_root_ns);
2742         steering->rdma_rx_root_ns = NULL;
2743         return err;
2744 }
2745
2746 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2747 {
2748         int err;
2749
2750         steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2751         if (!steering->rdma_tx_root_ns)
2752                 return -ENOMEM;
2753
2754         err = init_root_tree(steering, &rdma_tx_root_fs,
2755                              &steering->rdma_tx_root_ns->ns.node);
2756         if (err)
2757                 goto out_err;
2758
2759         set_prio_attrs(steering->rdma_tx_root_ns);
2760
2761         return 0;
2762
2763 out_err:
2764         cleanup_root_ns(steering->rdma_tx_root_ns);
2765         steering->rdma_tx_root_ns = NULL;
2766         return err;
2767 }
2768
2769 /* FT and tc chains are stored in the same array so we can re-use the
2770  * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2771  * When creating a new ns for each chain store it in the first available slot.
2772  * Assume tc chains are created and stored first and only then the FT chain.
2773  */
2774 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2775                                         struct mlx5_flow_namespace *ns)
2776 {
2777         int chain = 0;
2778
2779         while (steering->fdb_sub_ns[chain])
2780                 ++chain;
2781
2782         steering->fdb_sub_ns[chain] = ns;
2783 }
2784
2785 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2786                                         struct fs_prio *maj_prio)
2787 {
2788         struct mlx5_flow_namespace *ns;
2789         struct fs_prio *min_prio;
2790         int prio;
2791
2792         ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2793         if (IS_ERR(ns))
2794                 return PTR_ERR(ns);
2795
2796         for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2797                 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2798                 if (IS_ERR(min_prio))
2799                         return PTR_ERR(min_prio);
2800         }
2801
2802         store_fdb_sub_ns_prio_chain(steering, ns);
2803
2804         return 0;
2805 }
2806
2807 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2808                              int fs_prio,
2809                              int chains)
2810 {
2811         struct fs_prio *maj_prio;
2812         int levels;
2813         int chain;
2814         int err;
2815
2816         levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2817         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2818                                           fs_prio,
2819                                           levels);
2820         if (IS_ERR(maj_prio))
2821                 return PTR_ERR(maj_prio);
2822
2823         for (chain = 0; chain < chains; chain++) {
2824                 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2825                 if (err)
2826                         return err;
2827         }
2828
2829         return 0;
2830 }
2831
2832 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2833 {
2834         int err;
2835
2836         steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2837                                        sizeof(*steering->fdb_sub_ns),
2838                                        GFP_KERNEL);
2839         if (!steering->fdb_sub_ns)
2840                 return -ENOMEM;
2841
2842         err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
2843         if (err)
2844                 return err;
2845
2846         err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
2847         if (err)
2848                 return err;
2849
2850         return 0;
2851 }
2852
2853 static int create_fdb_bypass(struct mlx5_flow_steering *steering)
2854 {
2855         struct mlx5_flow_namespace *ns;
2856         struct fs_prio *prio;
2857         int i;
2858
2859         prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH, 0);
2860         if (IS_ERR(prio))
2861                 return PTR_ERR(prio);
2862
2863         ns = fs_create_namespace(prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2864         if (IS_ERR(ns))
2865                 return PTR_ERR(ns);
2866
2867         for (i = 0; i < MLX5_BY_PASS_NUM_REGULAR_PRIOS; i++) {
2868                 prio = fs_create_prio(ns, i, 1);
2869                 if (IS_ERR(prio))
2870                         return PTR_ERR(prio);
2871         }
2872         return 0;
2873 }
2874
2875 static void cleanup_fdb_root_ns(struct mlx5_flow_steering *steering)
2876 {
2877         cleanup_root_ns(steering->fdb_root_ns);
2878         steering->fdb_root_ns = NULL;
2879         kfree(steering->fdb_sub_ns);
2880         steering->fdb_sub_ns = NULL;
2881 }
2882
2883 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2884 {
2885         struct fs_prio *maj_prio;
2886         int err;
2887
2888         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2889         if (!steering->fdb_root_ns)
2890                 return -ENOMEM;
2891
2892         err = create_fdb_bypass(steering);
2893         if (err)
2894                 goto out_err;
2895
2896         err = create_fdb_fast_path(steering);
2897         if (err)
2898                 goto out_err;
2899
2900         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1);
2901         if (IS_ERR(maj_prio)) {
2902                 err = PTR_ERR(maj_prio);
2903                 goto out_err;
2904         }
2905
2906         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 3);
2907         if (IS_ERR(maj_prio)) {
2908                 err = PTR_ERR(maj_prio);
2909                 goto out_err;
2910         }
2911
2912         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2913         if (IS_ERR(maj_prio)) {
2914                 err = PTR_ERR(maj_prio);
2915                 goto out_err;
2916         }
2917
2918         /* We put this priority last, knowing that nothing will get here
2919          * unless explicitly forwarded to. This is possible because the
2920          * slow path tables have catch all rules and nothing gets passed
2921          * those tables.
2922          */
2923         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
2924         if (IS_ERR(maj_prio)) {
2925                 err = PTR_ERR(maj_prio);
2926                 goto out_err;
2927         }
2928
2929         set_prio_attrs(steering->fdb_root_ns);
2930         return 0;
2931
2932 out_err:
2933         cleanup_fdb_root_ns(steering);
2934         return err;
2935 }
2936
2937 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2938 {
2939         struct fs_prio *prio;
2940
2941         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2942         if (!steering->esw_egress_root_ns[vport])
2943                 return -ENOMEM;
2944
2945         /* create 1 prio*/
2946         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2947         return PTR_ERR_OR_ZERO(prio);
2948 }
2949
2950 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2951 {
2952         struct fs_prio *prio;
2953
2954         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2955         if (!steering->esw_ingress_root_ns[vport])
2956                 return -ENOMEM;
2957
2958         /* create 1 prio*/
2959         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2960         return PTR_ERR_OR_ZERO(prio);
2961 }
2962
2963 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2964 {
2965         struct mlx5_flow_steering *steering = dev->priv.steering;
2966         int err;
2967         int i;
2968
2969         steering->esw_egress_root_ns =
2970                         kcalloc(total_vports,
2971                                 sizeof(*steering->esw_egress_root_ns),
2972                                 GFP_KERNEL);
2973         if (!steering->esw_egress_root_ns)
2974                 return -ENOMEM;
2975
2976         for (i = 0; i < total_vports; i++) {
2977                 err = init_egress_acl_root_ns(steering, i);
2978                 if (err)
2979                         goto cleanup_root_ns;
2980         }
2981         steering->esw_egress_acl_vports = total_vports;
2982         return 0;
2983
2984 cleanup_root_ns:
2985         for (i--; i >= 0; i--)
2986                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2987         kfree(steering->esw_egress_root_ns);
2988         steering->esw_egress_root_ns = NULL;
2989         return err;
2990 }
2991
2992 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev)
2993 {
2994         struct mlx5_flow_steering *steering = dev->priv.steering;
2995         int i;
2996
2997         if (!steering->esw_egress_root_ns)
2998                 return;
2999
3000         for (i = 0; i < steering->esw_egress_acl_vports; i++)
3001                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
3002
3003         kfree(steering->esw_egress_root_ns);
3004         steering->esw_egress_root_ns = NULL;
3005 }
3006
3007 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports)
3008 {
3009         struct mlx5_flow_steering *steering = dev->priv.steering;
3010         int err;
3011         int i;
3012
3013         steering->esw_ingress_root_ns =
3014                         kcalloc(total_vports,
3015                                 sizeof(*steering->esw_ingress_root_ns),
3016                                 GFP_KERNEL);
3017         if (!steering->esw_ingress_root_ns)
3018                 return -ENOMEM;
3019
3020         for (i = 0; i < total_vports; i++) {
3021                 err = init_ingress_acl_root_ns(steering, i);
3022                 if (err)
3023                         goto cleanup_root_ns;
3024         }
3025         steering->esw_ingress_acl_vports = total_vports;
3026         return 0;
3027
3028 cleanup_root_ns:
3029         for (i--; i >= 0; i--)
3030                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
3031         kfree(steering->esw_ingress_root_ns);
3032         steering->esw_ingress_root_ns = NULL;
3033         return err;
3034 }
3035
3036 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev)
3037 {
3038         struct mlx5_flow_steering *steering = dev->priv.steering;
3039         int i;
3040
3041         if (!steering->esw_ingress_root_ns)
3042                 return;
3043
3044         for (i = 0; i < steering->esw_ingress_acl_vports; i++)
3045                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
3046
3047         kfree(steering->esw_ingress_root_ns);
3048         steering->esw_ingress_root_ns = NULL;
3049 }
3050
3051 u32 mlx5_fs_get_capabilities(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type type)
3052 {
3053         struct mlx5_flow_root_namespace *root;
3054         struct mlx5_flow_namespace *ns;
3055
3056         ns = mlx5_get_flow_namespace(dev, type);
3057         if (!ns)
3058                 return 0;
3059
3060         root = find_root(&ns->node);
3061         if (!root)
3062                 return 0;
3063
3064         return root->cmds->get_capabilities(root, root->table_type);
3065 }
3066
3067 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
3068 {
3069         int err;
3070
3071         steering->egress_root_ns = create_root_ns(steering,
3072                                                   FS_FT_NIC_TX);
3073         if (!steering->egress_root_ns)
3074                 return -ENOMEM;
3075
3076         err = init_root_tree(steering, &egress_root_fs,
3077                              &steering->egress_root_ns->ns.node);
3078         if (err)
3079                 goto cleanup;
3080         set_prio_attrs(steering->egress_root_ns);
3081         return 0;
3082 cleanup:
3083         cleanup_root_ns(steering->egress_root_ns);
3084         steering->egress_root_ns = NULL;
3085         return err;
3086 }
3087
3088 void mlx5_fs_core_cleanup(struct mlx5_core_dev *dev)
3089 {
3090         struct mlx5_flow_steering *steering = dev->priv.steering;
3091
3092         cleanup_root_ns(steering->root_ns);
3093         cleanup_fdb_root_ns(steering);
3094         cleanup_root_ns(steering->port_sel_root_ns);
3095         cleanup_root_ns(steering->sniffer_rx_root_ns);
3096         cleanup_root_ns(steering->sniffer_tx_root_ns);
3097         cleanup_root_ns(steering->rdma_rx_root_ns);
3098         cleanup_root_ns(steering->rdma_tx_root_ns);
3099         cleanup_root_ns(steering->egress_root_ns);
3100 }
3101
3102 int mlx5_fs_core_init(struct mlx5_core_dev *dev)
3103 {
3104         struct mlx5_flow_steering *steering = dev->priv.steering;
3105         int err = 0;
3106
3107         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
3108               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
3109              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
3110               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
3111             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
3112                 err = init_root_ns(steering);
3113                 if (err)
3114                         goto err;
3115         }
3116
3117         if (MLX5_ESWITCH_MANAGER(dev)) {
3118                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
3119                         err = init_fdb_root_ns(steering);
3120                         if (err)
3121                                 goto err;
3122                 }
3123         }
3124
3125         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
3126                 err = init_sniffer_rx_root_ns(steering);
3127                 if (err)
3128                         goto err;
3129         }
3130
3131         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
3132                 err = init_sniffer_tx_root_ns(steering);
3133                 if (err)
3134                         goto err;
3135         }
3136
3137         if (MLX5_CAP_FLOWTABLE_PORT_SELECTION(dev, ft_support)) {
3138                 err = init_port_sel_root_ns(steering);
3139                 if (err)
3140                         goto err;
3141         }
3142
3143         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
3144             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3145                 err = init_rdma_rx_root_ns(steering);
3146                 if (err)
3147                         goto err;
3148         }
3149
3150         if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3151                 err = init_rdma_tx_root_ns(steering);
3152                 if (err)
3153                         goto err;
3154         }
3155
3156         if (MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3157                 err = init_egress_root_ns(steering);
3158                 if (err)
3159                         goto err;
3160         }
3161
3162         return 0;
3163
3164 err:
3165         mlx5_fs_core_cleanup(dev);
3166         return err;
3167 }
3168
3169 void mlx5_fs_core_free(struct mlx5_core_dev *dev)
3170 {
3171         struct mlx5_flow_steering *steering = dev->priv.steering;
3172
3173         kmem_cache_destroy(steering->ftes_cache);
3174         kmem_cache_destroy(steering->fgs_cache);
3175         kfree(steering);
3176         mlx5_ft_pool_destroy(dev);
3177         mlx5_cleanup_fc_stats(dev);
3178 }
3179
3180 int mlx5_fs_core_alloc(struct mlx5_core_dev *dev)
3181 {
3182         struct mlx5_flow_steering *steering;
3183         int err = 0;
3184
3185         err = mlx5_init_fc_stats(dev);
3186         if (err)
3187                 return err;
3188
3189         err = mlx5_ft_pool_init(dev);
3190         if (err)
3191                 goto err;
3192
3193         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
3194         if (!steering) {
3195                 err = -ENOMEM;
3196                 goto err;
3197         }
3198
3199         steering->dev = dev;
3200         dev->priv.steering = steering;
3201
3202         if (mlx5_fs_dr_is_supported(dev))
3203                 steering->mode = MLX5_FLOW_STEERING_MODE_SMFS;
3204         else
3205                 steering->mode = MLX5_FLOW_STEERING_MODE_DMFS;
3206
3207         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
3208                                                 sizeof(struct mlx5_flow_group), 0,
3209                                                 0, NULL);
3210         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
3211                                                  0, NULL);
3212         if (!steering->ftes_cache || !steering->fgs_cache) {
3213                 err = -ENOMEM;
3214                 goto err;
3215         }
3216
3217         return 0;
3218
3219 err:
3220         mlx5_fs_core_free(dev);
3221         return err;
3222 }
3223
3224 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3225 {
3226         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3227         struct mlx5_ft_underlay_qp *new_uqp;
3228         int err = 0;
3229
3230         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3231         if (!new_uqp)
3232                 return -ENOMEM;
3233
3234         mutex_lock(&root->chain_lock);
3235
3236         if (!root->root_ft) {
3237                 err = -EINVAL;
3238                 goto update_ft_fail;
3239         }
3240
3241         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3242                                          false);
3243         if (err) {
3244                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3245                                underlay_qpn, err);
3246                 goto update_ft_fail;
3247         }
3248
3249         new_uqp->qpn = underlay_qpn;
3250         list_add_tail(&new_uqp->list, &root->underlay_qpns);
3251
3252         mutex_unlock(&root->chain_lock);
3253
3254         return 0;
3255
3256 update_ft_fail:
3257         mutex_unlock(&root->chain_lock);
3258         kfree(new_uqp);
3259         return err;
3260 }
3261 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3262
3263 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3264 {
3265         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3266         struct mlx5_ft_underlay_qp *uqp;
3267         bool found = false;
3268         int err = 0;
3269
3270         mutex_lock(&root->chain_lock);
3271         list_for_each_entry(uqp, &root->underlay_qpns, list) {
3272                 if (uqp->qpn == underlay_qpn) {
3273                         found = true;
3274                         break;
3275                 }
3276         }
3277
3278         if (!found) {
3279                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3280                                underlay_qpn);
3281                 err = -EINVAL;
3282                 goto out;
3283         }
3284
3285         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3286                                          true);
3287         if (err)
3288                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3289                                underlay_qpn, err);
3290
3291         list_del(&uqp->list);
3292         mutex_unlock(&root->chain_lock);
3293         kfree(uqp);
3294
3295         return 0;
3296
3297 out:
3298         mutex_unlock(&root->chain_lock);
3299         return err;
3300 }
3301 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3302
3303 static struct mlx5_flow_root_namespace
3304 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3305 {
3306         struct mlx5_flow_namespace *ns;
3307
3308         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3309             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3310                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3311         else
3312                 ns = mlx5_get_flow_namespace(dev, ns_type);
3313         if (!ns)
3314                 return NULL;
3315
3316         return find_root(&ns->node);
3317 }
3318
3319 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3320                                                  u8 ns_type, u8 num_actions,
3321                                                  void *modify_actions)
3322 {
3323         struct mlx5_flow_root_namespace *root;
3324         struct mlx5_modify_hdr *modify_hdr;
3325         int err;
3326
3327         root = get_root_namespace(dev, ns_type);
3328         if (!root)
3329                 return ERR_PTR(-EOPNOTSUPP);
3330
3331         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3332         if (!modify_hdr)
3333                 return ERR_PTR(-ENOMEM);
3334
3335         modify_hdr->ns_type = ns_type;
3336         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3337                                               modify_actions, modify_hdr);
3338         if (err) {
3339                 kfree(modify_hdr);
3340                 return ERR_PTR(err);
3341         }
3342
3343         return modify_hdr;
3344 }
3345 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3346
3347 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3348                                 struct mlx5_modify_hdr *modify_hdr)
3349 {
3350         struct mlx5_flow_root_namespace *root;
3351
3352         root = get_root_namespace(dev, modify_hdr->ns_type);
3353         if (WARN_ON(!root))
3354                 return;
3355         root->cmds->modify_header_dealloc(root, modify_hdr);
3356         kfree(modify_hdr);
3357 }
3358 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3359
3360 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3361                                                      struct mlx5_pkt_reformat_params *params,
3362                                                      enum mlx5_flow_namespace_type ns_type)
3363 {
3364         struct mlx5_pkt_reformat *pkt_reformat;
3365         struct mlx5_flow_root_namespace *root;
3366         int err;
3367
3368         root = get_root_namespace(dev, ns_type);
3369         if (!root)
3370                 return ERR_PTR(-EOPNOTSUPP);
3371
3372         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3373         if (!pkt_reformat)
3374                 return ERR_PTR(-ENOMEM);
3375
3376         pkt_reformat->ns_type = ns_type;
3377         pkt_reformat->reformat_type = params->type;
3378         err = root->cmds->packet_reformat_alloc(root, params, ns_type,
3379                                                 pkt_reformat);
3380         if (err) {
3381                 kfree(pkt_reformat);
3382                 return ERR_PTR(err);
3383         }
3384
3385         return pkt_reformat;
3386 }
3387 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3388
3389 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3390                                   struct mlx5_pkt_reformat *pkt_reformat)
3391 {
3392         struct mlx5_flow_root_namespace *root;
3393
3394         root = get_root_namespace(dev, pkt_reformat->ns_type);
3395         if (WARN_ON(!root))
3396                 return;
3397         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3398         kfree(pkt_reformat);
3399 }
3400 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3401
3402 int mlx5_get_match_definer_id(struct mlx5_flow_definer *definer)
3403 {
3404         return definer->id;
3405 }
3406
3407 struct mlx5_flow_definer *
3408 mlx5_create_match_definer(struct mlx5_core_dev *dev,
3409                           enum mlx5_flow_namespace_type ns_type, u16 format_id,
3410                           u32 *match_mask)
3411 {
3412         struct mlx5_flow_root_namespace *root;
3413         struct mlx5_flow_definer *definer;
3414         int id;
3415
3416         root = get_root_namespace(dev, ns_type);
3417         if (!root)
3418                 return ERR_PTR(-EOPNOTSUPP);
3419
3420         definer = kzalloc(sizeof(*definer), GFP_KERNEL);
3421         if (!definer)
3422                 return ERR_PTR(-ENOMEM);
3423
3424         definer->ns_type = ns_type;
3425         id = root->cmds->create_match_definer(root, format_id, match_mask);
3426         if (id < 0) {
3427                 mlx5_core_warn(root->dev, "Failed to create match definer (%d)\n", id);
3428                 kfree(definer);
3429                 return ERR_PTR(id);
3430         }
3431         definer->id = id;
3432         return definer;
3433 }
3434
3435 void mlx5_destroy_match_definer(struct mlx5_core_dev *dev,
3436                                 struct mlx5_flow_definer *definer)
3437 {
3438         struct mlx5_flow_root_namespace *root;
3439
3440         root = get_root_namespace(dev, definer->ns_type);
3441         if (WARN_ON(!root))
3442                 return;
3443
3444         root->cmds->destroy_match_definer(root, definer->id);
3445         kfree(definer);
3446 }
3447
3448 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3449                                  struct mlx5_flow_root_namespace *peer_ns)
3450 {
3451         if (peer_ns && ns->mode != peer_ns->mode) {
3452                 mlx5_core_err(ns->dev,
3453                               "Can't peer namespace of different steering mode\n");
3454                 return -EINVAL;
3455         }
3456
3457         return ns->cmds->set_peer(ns, peer_ns);
3458 }
3459
3460 /* This function should be called only at init stage of the namespace.
3461  * It is not safe to call this function while steering operations
3462  * are executed in the namespace.
3463  */
3464 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3465                                  enum mlx5_flow_steering_mode mode)
3466 {
3467         struct mlx5_flow_root_namespace *root;
3468         const struct mlx5_flow_cmds *cmds;
3469         int err;
3470
3471         root = find_root(&ns->node);
3472         if (&root->ns != ns)
3473         /* Can't set cmds to non root namespace */
3474                 return -EINVAL;
3475
3476         if (root->table_type != FS_FT_FDB)
3477                 return -EOPNOTSUPP;
3478
3479         if (root->mode == mode)
3480                 return 0;
3481
3482         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3483                 cmds = mlx5_fs_cmd_get_dr_cmds();
3484         else
3485                 cmds = mlx5_fs_cmd_get_fw_cmds();
3486         if (!cmds)
3487                 return -EOPNOTSUPP;
3488
3489         err = cmds->create_ns(root);
3490         if (err) {
3491                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3492                               err);
3493                 return err;
3494         }
3495
3496         root->cmds->destroy_ns(root);
3497         root->cmds = cmds;
3498         root->mode = mode;
3499
3500         return 0;
3501 }