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 "diag/fs_tracepoint.h"
  42 #include "accel/ipsec.h"
  43 #include "fpga/ipsec.h"
  44
  45 #define INIT_TREE_NODE_ARRAY_SIZE(...)  (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
  46                                          sizeof(struct init_tree_node))
  47
  48 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
  49                  ...) {.type = FS_TYPE_PRIO,\
  50         .min_ft_level = min_level_val,\
  51         .num_levels = num_levels_val,\
  52         .num_leaf_prios = num_prios_val,\
  53         .caps = caps_val,\
  54         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  55         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  56 }
  57
  58 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
  59         ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
  60                  __VA_ARGS__)\
  61
  62 #define ADD_NS(def_miss_act, ...) {.type = FS_TYPE_NAMESPACE,   \
  63         .def_miss_action = def_miss_act,\
  64         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  65         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  66 }
  67
  68 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  69                                    sizeof(long))
  70
  71 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  72
  73 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  74                                .caps = (long[]) {__VA_ARGS__} }
  75
  76 #define FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  77                                            FS_CAP(flow_table_properties_nic_receive.modify_root), \
  78                                            FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  79                                            FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  80
  81 #define FS_CHAINING_CAPS_EGRESS                                                \
  82         FS_REQUIRED_CAPS(                                                      \
  83                 FS_CAP(flow_table_properties_nic_transmit.flow_modify_en),     \
  84                 FS_CAP(flow_table_properties_nic_transmit.modify_root),        \
  85                 FS_CAP(flow_table_properties_nic_transmit                      \
  86                                .identified_miss_table_mode),                   \
  87                 FS_CAP(flow_table_properties_nic_transmit.flow_table_modify))
  88
  89 #define FS_CHAINING_CAPS_RDMA_TX                                                \
  90         FS_REQUIRED_CAPS(                                                       \
  91                 FS_CAP(flow_table_properties_nic_transmit_rdma.flow_modify_en), \
  92                 FS_CAP(flow_table_properties_nic_transmit_rdma.modify_root),    \
  93                 FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  94                                .identified_miss_table_mode),                    \
  95                 FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  96                                .flow_table_modify))
  97
  98 #define LEFTOVERS_NUM_LEVELS 1
  99 #define LEFTOVERS_NUM_PRIOS 1
 100
 101 #define BY_PASS_PRIO_NUM_LEVELS 1
 102 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
 103                            LEFTOVERS_NUM_PRIOS)
 104
 105 #define ETHTOOL_PRIO_NUM_LEVELS 1
 106 #define ETHTOOL_NUM_PRIOS 11
 107 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
 108 /* Vlan, mac, ttc, inner ttc, {aRFS/accel and esp/esp_err} */
 109 #define KERNEL_NIC_PRIO_NUM_LEVELS 6
 110 #define KERNEL_NIC_NUM_PRIOS 1
 111 /* One more level for tc */
 112 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
 113
 114 #define KERNEL_NIC_TC_NUM_PRIOS  1
 115 #define KERNEL_NIC_TC_NUM_LEVELS 2
 116
 117 #define ANCHOR_NUM_LEVELS 1
 118 #define ANCHOR_NUM_PRIOS 1
 119 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 120
 121 #define OFFLOADS_MAX_FT 2
 122 #define OFFLOADS_NUM_PRIOS 2
 123 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS)
 124
 125 #define LAG_PRIO_NUM_LEVELS 1
 126 #define LAG_NUM_PRIOS 1
 127 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 128
 129 #define KERNEL_TX_IPSEC_NUM_PRIOS  1
 130 #define KERNEL_TX_IPSEC_NUM_LEVELS 1
 131 #define KERNEL_TX_MIN_LEVEL        (KERNEL_TX_IPSEC_NUM_LEVELS)
 132
 133 struct node_caps {
 134         size_t  arr_sz;
 135         long    *caps;
 136 };
 137
 138 static struct init_tree_node {
 139         enum fs_node_type       type;
 140         struct init_tree_node *children;
 141         int ar_size;
 142         struct node_caps caps;
 143         int min_ft_level;
 144         int num_leaf_prios;
 145         int prio;
 146         int num_levels;
 147         enum mlx5_flow_table_miss_action def_miss_action;
 148 } root_fs = {
 149         .type = FS_TYPE_NAMESPACE,
 150         .ar_size = 7,
 151           .children = (struct init_tree_node[]){
 152                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 153                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 154                                   ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 155                                                     BY_PASS_PRIO_NUM_LEVELS))),
 156                   ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 157                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 158                                   ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 159                                                     LAG_PRIO_NUM_LEVELS))),
 160                   ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 161                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 162                                   ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS,
 163                                                     OFFLOADS_MAX_FT))),
 164                   ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 165                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 166                                   ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 167                                                     ETHTOOL_PRIO_NUM_LEVELS))),
 168                   ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 169                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 170                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS,
 171                                                     KERNEL_NIC_TC_NUM_LEVELS),
 172                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 173                                                     KERNEL_NIC_PRIO_NUM_LEVELS))),
 174                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 175                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 176                                   ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS,
 177                                                     LEFTOVERS_NUM_LEVELS))),
 178                   ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 179                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 180                                   ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS,
 181                                                     ANCHOR_NUM_LEVELS))),
 182         }
 183 };
 184
 185 static struct init_tree_node egress_root_fs = {
 186         .type = FS_TYPE_NAMESPACE,
 187 #ifdef CONFIG_MLX5_IPSEC
 188         .ar_size = 2,
 189 #else
 190         .ar_size = 1,
 191 #endif
 192         .children = (struct init_tree_node[]) {
 193                 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 194                          FS_CHAINING_CAPS_EGRESS,
 195                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 196                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 197                                                   BY_PASS_PRIO_NUM_LEVELS))),
 198 #ifdef CONFIG_MLX5_IPSEC
 199                 ADD_PRIO(0, KERNEL_TX_MIN_LEVEL, 0,
 200                          FS_CHAINING_CAPS_EGRESS,
 201                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 202                                 ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS,
 203                                                   KERNEL_TX_IPSEC_NUM_LEVELS))),
 204 #endif
 205         }
 206 };
 207
 208 #define RDMA_RX_BYPASS_PRIO 0
 209 #define RDMA_RX_KERNEL_PRIO 1
 210 static struct init_tree_node rdma_rx_root_fs = {
 211         .type = FS_TYPE_NAMESPACE,
 212         .ar_size = 2,
 213         .children = (struct init_tree_node[]) {
 214                 [RDMA_RX_BYPASS_PRIO] =
 215                 ADD_PRIO(0, MLX5_BY_PASS_NUM_REGULAR_PRIOS, 0,
 216                          FS_CHAINING_CAPS,
 217                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 218                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS,
 219                                                   BY_PASS_PRIO_NUM_LEVELS))),
 220                 [RDMA_RX_KERNEL_PRIO] =
 221                 ADD_PRIO(0, MLX5_BY_PASS_NUM_REGULAR_PRIOS + 1, 0,
 222                          FS_CHAINING_CAPS,
 223                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN,
 224                                 ADD_MULTIPLE_PRIO(1, 1))),
 225         }
 226 };
 227
 228 static struct init_tree_node rdma_tx_root_fs = {
 229         .type = FS_TYPE_NAMESPACE,
 230         .ar_size = 1,
 231         .children = (struct init_tree_node[]) {
 232                 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 233                          FS_CHAINING_CAPS_RDMA_TX,
 234                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 235                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 236                                                   BY_PASS_PRIO_NUM_LEVELS))),
 237         }
 238 };
 239
 240 enum fs_i_lock_class {
 241         FS_LOCK_GRANDPARENT,
 242         FS_LOCK_PARENT,
 243         FS_LOCK_CHILD
 244 };
 245
 246 static const struct rhashtable_params rhash_fte = {
 247         .key_len = sizeof_field(struct fs_fte, val),
 248         .key_offset = offsetof(struct fs_fte, val),
 249         .head_offset = offsetof(struct fs_fte, hash),
 250         .automatic_shrinking = true,
 251         .min_size = 1,
 252 };
 253
 254 static const struct rhashtable_params rhash_fg = {
 255         .key_len = sizeof_field(struct mlx5_flow_group, mask),
 256         .key_offset = offsetof(struct mlx5_flow_group, mask),
 257         .head_offset = offsetof(struct mlx5_flow_group, hash),
 258         .automatic_shrinking = true,
 259         .min_size = 1,
 260
 261 };
 262
 263 static void del_hw_flow_table(struct fs_node *node);
 264 static void del_hw_flow_group(struct fs_node *node);
 265 static void del_hw_fte(struct fs_node *node);
 266 static void del_sw_flow_table(struct fs_node *node);
 267 static void del_sw_flow_group(struct fs_node *node);
 268 static void del_sw_fte(struct fs_node *node);
 269 static void del_sw_prio(struct fs_node *node);
 270 static void del_sw_ns(struct fs_node *node);
 271 /* Delete rule (destination) is special case that
 272  * requires to lock the FTE for all the deletion process.
 273  */
 274 static void del_sw_hw_rule(struct fs_node *node);
 275 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 276                                 struct mlx5_flow_destination *d2);
 277 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 278 static struct mlx5_flow_rule *
 279 find_flow_rule(struct fs_fte *fte,
 280                struct mlx5_flow_destination *dest);
 281
 282 static void tree_init_node(struct fs_node *node,
 283                            void (*del_hw_func)(struct fs_node *),
 284                            void (*del_sw_func)(struct fs_node *))
 285 {
 286         refcount_set(&node->refcount, 1);
 287         INIT_LIST_HEAD(&node->list);
 288         INIT_LIST_HEAD(&node->children);
 289         init_rwsem(&node->lock);
 290         node->del_hw_func = del_hw_func;
 291         node->del_sw_func = del_sw_func;
 292         node->active = false;
 293 }
 294
 295 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 296 {
 297         if (parent)
 298                 refcount_inc(&parent->refcount);
 299         node->parent = parent;
 300
 301         /* Parent is the root */
 302         if (!parent)
 303                 node->root = node;
 304         else
 305                 node->root = parent->root;
 306 }
 307
 308 static int tree_get_node(struct fs_node *node)
 309 {
 310         return refcount_inc_not_zero(&node->refcount);
 311 }
 312
 313 static void nested_down_read_ref_node(struct fs_node *node,
 314                                       enum fs_i_lock_class class)
 315 {
 316         if (node) {
 317                 down_read_nested(&node->lock, class);
 318                 refcount_inc(&node->refcount);
 319         }
 320 }
 321
 322 static void nested_down_write_ref_node(struct fs_node *node,
 323                                        enum fs_i_lock_class class)
 324 {
 325         if (node) {
 326                 down_write_nested(&node->lock, class);
 327                 refcount_inc(&node->refcount);
 328         }
 329 }
 330
 331 static void down_write_ref_node(struct fs_node *node, bool locked)
 332 {
 333         if (node) {
 334                 if (!locked)
 335                         down_write(&node->lock);
 336                 refcount_inc(&node->refcount);
 337         }
 338 }
 339
 340 static void up_read_ref_node(struct fs_node *node)
 341 {
 342         refcount_dec(&node->refcount);
 343         up_read(&node->lock);
 344 }
 345
 346 static void up_write_ref_node(struct fs_node *node, bool locked)
 347 {
 348         refcount_dec(&node->refcount);
 349         if (!locked)
 350                 up_write(&node->lock);
 351 }
 352
 353 static void tree_put_node(struct fs_node *node, bool locked)
 354 {
 355         struct fs_node *parent_node = node->parent;
 356
 357         if (refcount_dec_and_test(&node->refcount)) {
 358                 if (node->del_hw_func)
 359                         node->del_hw_func(node);
 360                 if (parent_node) {
 361                         down_write_ref_node(parent_node, locked);
 362                         list_del_init(&node->list);
 363                 }
 364                 node->del_sw_func(node);
 365                 if (parent_node)
 366                         up_write_ref_node(parent_node, locked);
 367                 node = NULL;
 368         }
 369         if (!node && parent_node)
 370                 tree_put_node(parent_node, locked);
 371 }
 372
 373 static int tree_remove_node(struct fs_node *node, bool locked)
 374 {
 375         if (refcount_read(&node->refcount) > 1) {
 376                 refcount_dec(&node->refcount);
 377                 return -EEXIST;
 378         }
 379         tree_put_node(node, locked);
 380         return 0;
 381 }
 382
 383 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 384                                  unsigned int prio)
 385 {
 386         struct fs_prio *iter_prio;
 387
 388         fs_for_each_prio(iter_prio, ns) {
 389                 if (iter_prio->prio == prio)
 390                         return iter_prio;
 391         }
 392
 393         return NULL;
 394 }
 395
 396 static bool is_fwd_next_action(u32 action)
 397 {
 398         return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
 399                          MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
 400 }
 401
 402 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 403 {
 404         int i;
 405
 406         for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 407                 if (spec->match_value[i] & ~spec->match_criteria[i]) {
 408                         pr_warn("mlx5_core: match_value differs from match_criteria\n");
 409                         return false;
 410                 }
 411
 412         return true;
 413 }
 414
 415 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 416 {
 417         struct fs_node *root;
 418         struct mlx5_flow_namespace *ns;
 419
 420         root = node->root;
 421
 422         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 423                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 424                 return NULL;
 425         }
 426
 427         ns = container_of(root, struct mlx5_flow_namespace, node);
 428         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 429 }
 430
 431 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 432 {
 433         struct mlx5_flow_root_namespace *root = find_root(node);
 434
 435         if (root)
 436                 return root->dev->priv.steering;
 437         return NULL;
 438 }
 439
 440 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 441 {
 442         struct mlx5_flow_root_namespace *root = find_root(node);
 443
 444         if (root)
 445                 return root->dev;
 446         return NULL;
 447 }
 448
 449 static void del_sw_ns(struct fs_node *node)
 450 {
 451         kfree(node);
 452 }
 453
 454 static void del_sw_prio(struct fs_node *node)
 455 {
 456         kfree(node);
 457 }
 458
 459 static void del_hw_flow_table(struct fs_node *node)
 460 {
 461         struct mlx5_flow_root_namespace *root;
 462         struct mlx5_flow_table *ft;
 463         struct mlx5_core_dev *dev;
 464         int err;
 465
 466         fs_get_obj(ft, node);
 467         dev = get_dev(&ft->node);
 468         root = find_root(&ft->node);
 469         trace_mlx5_fs_del_ft(ft);
 470
 471         if (node->active) {
 472                 err = root->cmds->destroy_flow_table(root, ft);
 473                 if (err)
 474                         mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 475         }
 476 }
 477
 478 static void del_sw_flow_table(struct fs_node *node)
 479 {
 480         struct mlx5_flow_table *ft;
 481         struct fs_prio *prio;
 482
 483         fs_get_obj(ft, node);
 484
 485         rhltable_destroy(&ft->fgs_hash);
 486         if (ft->node.parent) {
 487                 fs_get_obj(prio, ft->node.parent);
 488                 prio->num_ft--;
 489         }
 490         kfree(ft);
 491 }
 492
 493 static void modify_fte(struct fs_fte *fte)
 494 {
 495         struct mlx5_flow_root_namespace *root;
 496         struct mlx5_flow_table *ft;
 497         struct mlx5_flow_group *fg;
 498         struct mlx5_core_dev *dev;
 499         int err;
 500
 501         fs_get_obj(fg, fte->node.parent);
 502         fs_get_obj(ft, fg->node.parent);
 503         dev = get_dev(&fte->node);
 504
 505         root = find_root(&ft->node);
 506         err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte);
 507         if (err)
 508                 mlx5_core_warn(dev,
 509                                "%s can't del rule fg id=%d fte_index=%d\n",
 510                                __func__, fg->id, fte->index);
 511         fte->modify_mask = 0;
 512 }
 513
 514 static void del_sw_hw_rule(struct fs_node *node)
 515 {
 516         struct mlx5_flow_rule *rule;
 517         struct fs_fte *fte;
 518
 519         fs_get_obj(rule, node);
 520         fs_get_obj(fte, rule->node.parent);
 521         trace_mlx5_fs_del_rule(rule);
 522         if (is_fwd_next_action(rule->sw_action)) {
 523                 mutex_lock(&rule->dest_attr.ft->lock);
 524                 list_del(&rule->next_ft);
 525                 mutex_unlock(&rule->dest_attr.ft->lock);
 526         }
 527
 528         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 529             --fte->dests_size) {
 530                 fte->modify_mask |=
 531                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 532                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 533                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 534                 goto out;
 535         }
 536
 537         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT &&
 538             --fte->dests_size) {
 539                 fte->modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
 540                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW;
 541                 goto out;
 542         }
 543
 544         if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 545             --fte->dests_size) {
 546                 fte->modify_mask |=
 547                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 548         }
 549 out:
 550         kfree(rule);
 551 }
 552
 553 static void del_hw_fte(struct fs_node *node)
 554 {
 555         struct mlx5_flow_root_namespace *root;
 556         struct mlx5_flow_table *ft;
 557         struct mlx5_flow_group *fg;
 558         struct mlx5_core_dev *dev;
 559         struct fs_fte *fte;
 560         int err;
 561
 562         fs_get_obj(fte, node);
 563         fs_get_obj(fg, fte->node.parent);
 564         fs_get_obj(ft, fg->node.parent);
 565
 566         trace_mlx5_fs_del_fte(fte);
 567         dev = get_dev(&ft->node);
 568         root = find_root(&ft->node);
 569         if (node->active) {
 570                 err = root->cmds->delete_fte(root, ft, fte);
 571                 if (err)
 572                         mlx5_core_warn(dev,
 573                                        "flow steering can't delete fte in index %d of flow group id %d\n",
 574                                        fte->index, fg->id);
 575                 node->active = 0;
 576         }
 577 }
 578
 579 static void del_sw_fte(struct fs_node *node)
 580 {
 581         struct mlx5_flow_steering *steering = get_steering(node);
 582         struct mlx5_flow_group *fg;
 583         struct fs_fte *fte;
 584         int err;
 585
 586         fs_get_obj(fte, node);
 587         fs_get_obj(fg, fte->node.parent);
 588
 589         err = rhashtable_remove_fast(&fg->ftes_hash,
 590                                      &fte->hash,
 591                                      rhash_fte);
 592         WARN_ON(err);
 593         ida_simple_remove(&fg->fte_allocator, fte->index - fg->start_index);
 594         kmem_cache_free(steering->ftes_cache, fte);
 595 }
 596
 597 static void del_hw_flow_group(struct fs_node *node)
 598 {
 599         struct mlx5_flow_root_namespace *root;
 600         struct mlx5_flow_group *fg;
 601         struct mlx5_flow_table *ft;
 602         struct mlx5_core_dev *dev;
 603
 604         fs_get_obj(fg, node);
 605         fs_get_obj(ft, fg->node.parent);
 606         dev = get_dev(&ft->node);
 607         trace_mlx5_fs_del_fg(fg);
 608
 609         root = find_root(&ft->node);
 610         if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
 611                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 612                                fg->id, ft->id);
 613 }
 614
 615 static void del_sw_flow_group(struct fs_node *node)
 616 {
 617         struct mlx5_flow_steering *steering = get_steering(node);
 618         struct mlx5_flow_group *fg;
 619         struct mlx5_flow_table *ft;
 620         int err;
 621
 622         fs_get_obj(fg, node);
 623         fs_get_obj(ft, fg->node.parent);
 624
 625         rhashtable_destroy(&fg->ftes_hash);
 626         ida_destroy(&fg->fte_allocator);
 627         if (ft->autogroup.active &&
 628             fg->max_ftes == ft->autogroup.group_size &&
 629             fg->start_index < ft->autogroup.max_fte)
 630                 ft->autogroup.num_groups--;
 631         err = rhltable_remove(&ft->fgs_hash,
 632                               &fg->hash,
 633                               rhash_fg);
 634         WARN_ON(err);
 635         kmem_cache_free(steering->fgs_cache, fg);
 636 }
 637
 638 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 639 {
 640         int index;
 641         int ret;
 642
 643         index = ida_simple_get(&fg->fte_allocator, 0, fg->max_ftes, GFP_KERNEL);
 644         if (index < 0)
 645                 return index;
 646
 647         fte->index = index + fg->start_index;
 648         ret = rhashtable_insert_fast(&fg->ftes_hash,
 649                                      &fte->hash,
 650                                      rhash_fte);
 651         if (ret)
 652                 goto err_ida_remove;
 653
 654         tree_add_node(&fte->node, &fg->node);
 655         list_add_tail(&fte->node.list, &fg->node.children);
 656         return 0;
 657
 658 err_ida_remove:
 659         ida_simple_remove(&fg->fte_allocator, index);
 660         return ret;
 661 }
 662
 663 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 664                                 const struct mlx5_flow_spec *spec,
 665                                 struct mlx5_flow_act *flow_act)
 666 {
 667         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 668         struct fs_fte *fte;
 669
 670         fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 671         if (!fte)
 672                 return ERR_PTR(-ENOMEM);
 673
 674         memcpy(fte->val, &spec->match_value, sizeof(fte->val));
 675         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 676         fte->action = *flow_act;
 677         fte->flow_context = spec->flow_context;
 678
 679         tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
 680
 681         return fte;
 682 }
 683
 684 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 685                                struct mlx5_flow_group *fg)
 686 {
 687         rhashtable_destroy(&fg->ftes_hash);
 688         kmem_cache_free(steering->fgs_cache, fg);
 689 }
 690
 691 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 692                                                 u8 match_criteria_enable,
 693                                                 const void *match_criteria,
 694                                                 int start_index,
 695                                                 int end_index)
 696 {
 697         struct mlx5_flow_group *fg;
 698         int ret;
 699
 700         fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 701         if (!fg)
 702                 return ERR_PTR(-ENOMEM);
 703
 704         ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 705         if (ret) {
 706                 kmem_cache_free(steering->fgs_cache, fg);
 707                 return ERR_PTR(ret);
 708         }
 709
 710         ida_init(&fg->fte_allocator);
 711         fg->mask.match_criteria_enable = match_criteria_enable;
 712         memcpy(&fg->mask.match_criteria, match_criteria,
 713                sizeof(fg->mask.match_criteria));
 714         fg->node.type =  FS_TYPE_FLOW_GROUP;
 715         fg->start_index = start_index;
 716         fg->max_ftes = end_index - start_index + 1;
 717
 718         return fg;
 719 }
 720
 721 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 722                                                        u8 match_criteria_enable,
 723                                                        const void *match_criteria,
 724                                                        int start_index,
 725                                                        int end_index,
 726                                                        struct list_head *prev)
 727 {
 728         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 729         struct mlx5_flow_group *fg;
 730         int ret;
 731
 732         fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 733                               start_index, end_index);
 734         if (IS_ERR(fg))
 735                 return fg;
 736
 737         /* initialize refcnt, add to parent list */
 738         ret = rhltable_insert(&ft->fgs_hash,
 739                               &fg->hash,
 740                               rhash_fg);
 741         if (ret) {
 742                 dealloc_flow_group(steering, fg);
 743                 return ERR_PTR(ret);
 744         }
 745
 746         tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 747         tree_add_node(&fg->node, &ft->node);
 748         /* Add node to group list */
 749         list_add(&fg->node.list, prev);
 750         atomic_inc(&ft->node.version);
 751
 752         return fg;
 753 }
 754
 755 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 756                                                 enum fs_flow_table_type table_type,
 757                                                 enum fs_flow_table_op_mod op_mod,
 758                                                 u32 flags)
 759 {
 760         struct mlx5_flow_table *ft;
 761         int ret;
 762
 763         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 764         if (!ft)
 765                 return ERR_PTR(-ENOMEM);
 766
 767         ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 768         if (ret) {
 769                 kfree(ft);
 770                 return ERR_PTR(ret);
 771         }
 772
 773         ft->level = level;
 774         ft->node.type = FS_TYPE_FLOW_TABLE;
 775         ft->op_mod = op_mod;
 776         ft->type = table_type;
 777         ft->vport = vport;
 778         ft->max_fte = max_fte;
 779         ft->flags = flags;
 780         INIT_LIST_HEAD(&ft->fwd_rules);
 781         mutex_init(&ft->lock);
 782
 783         return ft;
 784 }
 785
 786 /* If reverse is false, then we search for the first flow table in the
 787  * root sub-tree from start(closest from right), else we search for the
 788  * last flow table in the root sub-tree till start(closest from left).
 789  */
 790 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 791                                                          struct list_head *start,
 792                                                          bool reverse)
 793 {
 794 #define list_advance_entry(pos, reverse)                \
 795         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 796
 797 #define list_for_each_advance_continue(pos, head, reverse)      \
 798         for (pos = list_advance_entry(pos, reverse);            \
 799              &pos->list != (head);                              \
 800              pos = list_advance_entry(pos, reverse))
 801
 802         struct fs_node *iter = list_entry(start, struct fs_node, list);
 803         struct mlx5_flow_table *ft = NULL;
 804
 805         if (!root || root->type == FS_TYPE_PRIO_CHAINS)
 806                 return NULL;
 807
 808         list_for_each_advance_continue(iter, &root->children, reverse) {
 809                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 810                         fs_get_obj(ft, iter);
 811                         return ft;
 812                 }
 813                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 814                 if (ft)
 815                         return ft;
 816         }
 817
 818         return ft;
 819 }
 820
 821 /* If reverse is false then return the first flow table in next priority of
 822  * prio in the tree, else return the last flow table in the previous priority
 823  * of prio in the tree.
 824  */
 825 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 826 {
 827         struct mlx5_flow_table *ft = NULL;
 828         struct fs_node *curr_node;
 829         struct fs_node *parent;
 830
 831         parent = prio->node.parent;
 832         curr_node = &prio->node;
 833         while (!ft && parent) {
 834                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 835                 curr_node = parent;
 836                 parent = curr_node->parent;
 837         }
 838         return ft;
 839 }
 840
 841 /* Assuming all the tree is locked by mutex chain lock */
 842 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 843 {
 844         return find_closest_ft(prio, false);
 845 }
 846
 847 /* Assuming all the tree is locked by mutex chain lock */
 848 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 849 {
 850         return find_closest_ft(prio, true);
 851 }
 852
 853 static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft,
 854                                                 struct mlx5_flow_act *flow_act)
 855 {
 856         struct fs_prio *prio;
 857         bool next_ns;
 858
 859         next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
 860         fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent);
 861
 862         return find_next_chained_ft(prio);
 863 }
 864
 865 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 866                                struct fs_prio *prio,
 867                                struct mlx5_flow_table *ft)
 868 {
 869         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 870         struct mlx5_flow_table *iter;
 871         int err;
 872
 873         fs_for_each_ft(iter, prio) {
 874                 err = root->cmds->modify_flow_table(root, iter, ft);
 875                 if (err) {
 876                         mlx5_core_err(dev,
 877                                       "Failed to modify flow table id %d, type %d, err %d\n",
 878                                       iter->id, iter->type, err);
 879                         /* The driver is out of sync with the FW */
 880                         return err;
 881                 }
 882         }
 883         return 0;
 884 }
 885
 886 /* Connect flow tables from previous priority of prio to ft */
 887 static int connect_prev_fts(struct mlx5_core_dev *dev,
 888                             struct mlx5_flow_table *ft,
 889                             struct fs_prio *prio)
 890 {
 891         struct mlx5_flow_table *prev_ft;
 892
 893         prev_ft = find_prev_chained_ft(prio);
 894         if (prev_ft) {
 895                 struct fs_prio *prev_prio;
 896
 897                 fs_get_obj(prev_prio, prev_ft->node.parent);
 898                 return connect_fts_in_prio(dev, prev_prio, ft);
 899         }
 900         return 0;
 901 }
 902
 903 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 904                                  *prio)
 905 {
 906         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 907         struct mlx5_ft_underlay_qp *uqp;
 908         int min_level = INT_MAX;
 909         int err = 0;
 910         u32 qpn;
 911
 912         if (root->root_ft)
 913                 min_level = root->root_ft->level;
 914
 915         if (ft->level >= min_level)
 916                 return 0;
 917
 918         if (list_empty(&root->underlay_qpns)) {
 919                 /* Don't set any QPN (zero) in case QPN list is empty */
 920                 qpn = 0;
 921                 err = root->cmds->update_root_ft(root, ft, qpn, false);
 922         } else {
 923                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
 924                         qpn = uqp->qpn;
 925                         err = root->cmds->update_root_ft(root, ft,
 926                                                          qpn, false);
 927                         if (err)
 928                                 break;
 929                 }
 930         }
 931
 932         if (err)
 933                 mlx5_core_warn(root->dev,
 934                                "Update root flow table of id(%u) qpn(%d) failed\n",
 935                                ft->id, qpn);
 936         else
 937                 root->root_ft = ft;
 938
 939         return err;
 940 }
 941
 942 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 943                                          struct mlx5_flow_destination *dest)
 944 {
 945         struct mlx5_flow_root_namespace *root;
 946         struct mlx5_flow_table *ft;
 947         struct mlx5_flow_group *fg;
 948         struct fs_fte *fte;
 949         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 950         int err = 0;
 951
 952         fs_get_obj(fte, rule->node.parent);
 953         if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 954                 return -EINVAL;
 955         down_write_ref_node(&fte->node, false);
 956         fs_get_obj(fg, fte->node.parent);
 957         fs_get_obj(ft, fg->node.parent);
 958
 959         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 960         root = find_root(&ft->node);
 961         err = root->cmds->update_fte(root, ft, fg,
 962                                      modify_mask, fte);
 963         up_write_ref_node(&fte->node, false);
 964
 965         return err;
 966 }
 967
 968 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
 969                                  struct mlx5_flow_destination *new_dest,
 970                                  struct mlx5_flow_destination *old_dest)
 971 {
 972         int i;
 973
 974         if (!old_dest) {
 975                 if (handle->num_rules != 1)
 976                         return -EINVAL;
 977                 return _mlx5_modify_rule_destination(handle->rule[0],
 978                                                      new_dest);
 979         }
 980
 981         for (i = 0; i < handle->num_rules; i++) {
 982                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
 983                         return _mlx5_modify_rule_destination(handle->rule[i],
 984                                                              new_dest);
 985         }
 986
 987         return -EINVAL;
 988 }
 989
 990 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 991 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 992                              struct mlx5_flow_table *new_next_ft,
 993                              struct mlx5_flow_table *old_next_ft)
 994 {
 995         struct mlx5_flow_destination dest = {};
 996         struct mlx5_flow_rule *iter;
 997         int err = 0;
 998
 999         /* new_next_ft and old_next_ft could be NULL only
1000          * when we create/destroy the anchor flow table.
1001          */
1002         if (!new_next_ft || !old_next_ft)
1003                 return 0;
1004
1005         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1006         dest.ft = new_next_ft;
1007
1008         mutex_lock(&old_next_ft->lock);
1009         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
1010         mutex_unlock(&old_next_ft->lock);
1011         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
1012                 if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) &&
1013                     iter->ft->ns == new_next_ft->ns)
1014                         continue;
1015
1016                 err = _mlx5_modify_rule_destination(iter, &dest);
1017                 if (err)
1018                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
1019                                new_next_ft->id);
1020         }
1021         return 0;
1022 }
1023
1024 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
1025                               struct fs_prio *prio)
1026 {
1027         struct mlx5_flow_table *next_ft;
1028         int err = 0;
1029
1030         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
1031
1032         if (list_empty(&prio->node.children)) {
1033                 err = connect_prev_fts(dev, ft, prio);
1034                 if (err)
1035                         return err;
1036
1037                 next_ft = find_next_chained_ft(prio);
1038                 err = connect_fwd_rules(dev, ft, next_ft);
1039                 if (err)
1040                         return err;
1041         }
1042
1043         if (MLX5_CAP_FLOWTABLE(dev,
1044                                flow_table_properties_nic_receive.modify_root))
1045                 err = update_root_ft_create(ft, prio);
1046         return err;
1047 }
1048
1049 static void list_add_flow_table(struct mlx5_flow_table *ft,
1050                                 struct fs_prio *prio)
1051 {
1052         struct list_head *prev = &prio->node.children;
1053         struct mlx5_flow_table *iter;
1054
1055         fs_for_each_ft(iter, prio) {
1056                 if (iter->level > ft->level)
1057                         break;
1058                 prev = &iter->node.list;
1059         }
1060         list_add(&ft->node.list, prev);
1061 }
1062
1063 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1064                                                         struct mlx5_flow_table_attr *ft_attr,
1065                                                         enum fs_flow_table_op_mod op_mod,
1066                                                         u16 vport)
1067 {
1068         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1069         bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED;
1070         struct mlx5_flow_table *next_ft;
1071         struct fs_prio *fs_prio = NULL;
1072         struct mlx5_flow_table *ft;
1073         int log_table_sz;
1074         int err;
1075
1076         if (!root) {
1077                 pr_err("mlx5: flow steering failed to find root of namespace\n");
1078                 return ERR_PTR(-ENODEV);
1079         }
1080
1081         mutex_lock(&root->chain_lock);
1082         fs_prio = find_prio(ns, ft_attr->prio);
1083         if (!fs_prio) {
1084                 err = -EINVAL;
1085                 goto unlock_root;
1086         }
1087         if (!unmanaged) {
1088                 /* The level is related to the
1089                  * priority level range.
1090                  */
1091                 if (ft_attr->level >= fs_prio->num_levels) {
1092                         err = -ENOSPC;
1093                         goto unlock_root;
1094                 }
1095
1096                 ft_attr->level += fs_prio->start_level;
1097         }
1098
1099         /* The level is related to the
1100          * priority level range.
1101          */
1102         ft = alloc_flow_table(ft_attr->level,
1103                               vport,
1104                               ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
1105                               root->table_type,
1106                               op_mod, ft_attr->flags);
1107         if (IS_ERR(ft)) {
1108                 err = PTR_ERR(ft);
1109                 goto unlock_root;
1110         }
1111
1112         tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1113         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
1114         next_ft = unmanaged ? ft_attr->next_ft :
1115                               find_next_chained_ft(fs_prio);
1116         ft->def_miss_action = ns->def_miss_action;
1117         ft->ns = ns;
1118         err = root->cmds->create_flow_table(root, ft, log_table_sz, next_ft);
1119         if (err)
1120                 goto free_ft;
1121
1122         if (!unmanaged) {
1123                 err = connect_flow_table(root->dev, ft, fs_prio);
1124                 if (err)
1125                         goto destroy_ft;
1126         }
1127
1128         ft->node.active = true;
1129         down_write_ref_node(&fs_prio->node, false);
1130         if (!unmanaged) {
1131                 tree_add_node(&ft->node, &fs_prio->node);
1132                 list_add_flow_table(ft, fs_prio);
1133         } else {
1134                 ft->node.root = fs_prio->node.root;
1135         }
1136         fs_prio->num_ft++;
1137         up_write_ref_node(&fs_prio->node, false);
1138         mutex_unlock(&root->chain_lock);
1139         trace_mlx5_fs_add_ft(ft);
1140         return ft;
1141 destroy_ft:
1142         root->cmds->destroy_flow_table(root, ft);
1143 free_ft:
1144         kfree(ft);
1145 unlock_root:
1146         mutex_unlock(&root->chain_lock);
1147         return ERR_PTR(err);
1148 }
1149
1150 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1151                                                struct mlx5_flow_table_attr *ft_attr)
1152 {
1153         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1154 }
1155
1156 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1157                                                      int prio, int max_fte,
1158                                                      u32 level, u16 vport)
1159 {
1160         struct mlx5_flow_table_attr ft_attr = {};
1161
1162         ft_attr.max_fte = max_fte;
1163         ft_attr.level   = level;
1164         ft_attr.prio    = prio;
1165
1166         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1167 }
1168
1169 struct mlx5_flow_table*
1170 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1171                                  int prio, u32 level)
1172 {
1173         struct mlx5_flow_table_attr ft_attr = {};
1174
1175         ft_attr.level = level;
1176         ft_attr.prio  = prio;
1177         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1178 }
1179 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1180
1181 struct mlx5_flow_table*
1182 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1183                                     struct mlx5_flow_table_attr *ft_attr)
1184 {
1185         int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1186         int autogroups_max_fte = ft_attr->max_fte - num_reserved_entries;
1187         int max_num_groups = ft_attr->autogroup.max_num_groups;
1188         struct mlx5_flow_table *ft;
1189
1190         if (max_num_groups > autogroups_max_fte)
1191                 return ERR_PTR(-EINVAL);
1192         if (num_reserved_entries > ft_attr->max_fte)
1193                 return ERR_PTR(-EINVAL);
1194
1195         ft = mlx5_create_flow_table(ns, ft_attr);
1196         if (IS_ERR(ft))
1197                 return ft;
1198
1199         ft->autogroup.active = true;
1200         ft->autogroup.required_groups = max_num_groups;
1201         ft->autogroup.max_fte = autogroups_max_fte;
1202         /* We save place for flow groups in addition to max types */
1203         ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1204
1205         return ft;
1206 }
1207 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1208
1209 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1210                                                u32 *fg_in)
1211 {
1212         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1213         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1214                                             fg_in, match_criteria);
1215         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1216                                             fg_in,
1217                                             match_criteria_enable);
1218         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1219                                    start_flow_index);
1220         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1221                                  end_flow_index);
1222         struct mlx5_flow_group *fg;
1223         int err;
1224
1225         if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1226                 return ERR_PTR(-EPERM);
1227
1228         down_write_ref_node(&ft->node, false);
1229         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1230                                      start_index, end_index,
1231                                      ft->node.children.prev);
1232         up_write_ref_node(&ft->node, false);
1233         if (IS_ERR(fg))
1234                 return fg;
1235
1236         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1237         if (err) {
1238                 tree_put_node(&fg->node, false);
1239                 return ERR_PTR(err);
1240         }
1241         trace_mlx5_fs_add_fg(fg);
1242         fg->node.active = true;
1243
1244         return fg;
1245 }
1246
1247 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1248 {
1249         struct mlx5_flow_rule *rule;
1250
1251         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1252         if (!rule)
1253                 return NULL;
1254
1255         INIT_LIST_HEAD(&rule->next_ft);
1256         rule->node.type = FS_TYPE_FLOW_DEST;
1257         if (dest)
1258                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1259
1260         return rule;
1261 }
1262
1263 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1264 {
1265         struct mlx5_flow_handle *handle;
1266
1267         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1268         if (!handle)
1269                 return NULL;
1270
1271         handle->num_rules = num_rules;
1272
1273         return handle;
1274 }
1275
1276 static void destroy_flow_handle(struct fs_fte *fte,
1277                                 struct mlx5_flow_handle *handle,
1278                                 struct mlx5_flow_destination *dest,
1279                                 int i)
1280 {
1281         for (; --i >= 0;) {
1282                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1283                         fte->dests_size--;
1284                         list_del(&handle->rule[i]->node.list);
1285                         kfree(handle->rule[i]);
1286                 }
1287         }
1288         kfree(handle);
1289 }
1290
1291 static struct mlx5_flow_handle *
1292 create_flow_handle(struct fs_fte *fte,
1293                    struct mlx5_flow_destination *dest,
1294                    int dest_num,
1295                    int *modify_mask,
1296                    bool *new_rule)
1297 {
1298         struct mlx5_flow_handle *handle;
1299         struct mlx5_flow_rule *rule = NULL;
1300         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1301         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1302         int type;
1303         int i = 0;
1304
1305         handle = alloc_handle((dest_num) ? dest_num : 1);
1306         if (!handle)
1307                 return ERR_PTR(-ENOMEM);
1308
1309         do {
1310                 if (dest) {
1311                         rule = find_flow_rule(fte, dest + i);
1312                         if (rule) {
1313                                 refcount_inc(&rule->node.refcount);
1314                                 goto rule_found;
1315                         }
1316                 }
1317
1318                 *new_rule = true;
1319                 rule = alloc_rule(dest + i);
1320                 if (!rule)
1321                         goto free_rules;
1322
1323                 /* Add dest to dests list- we need flow tables to be in the
1324                  * end of the list for forward to next prio rules.
1325                  */
1326                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1327                 if (dest &&
1328                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1329                         list_add(&rule->node.list, &fte->node.children);
1330                 else
1331                         list_add_tail(&rule->node.list, &fte->node.children);
1332                 if (dest) {
1333                         fte->dests_size++;
1334
1335                         type = dest[i].type ==
1336                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1337                         *modify_mask |= type ? count : dst;
1338                 }
1339 rule_found:
1340                 handle->rule[i] = rule;
1341         } while (++i < dest_num);
1342
1343         return handle;
1344
1345 free_rules:
1346         destroy_flow_handle(fte, handle, dest, i);
1347         return ERR_PTR(-ENOMEM);
1348 }
1349
1350 /* fte should not be deleted while calling this function */
1351 static struct mlx5_flow_handle *
1352 add_rule_fte(struct fs_fte *fte,
1353              struct mlx5_flow_group *fg,
1354              struct mlx5_flow_destination *dest,
1355              int dest_num,
1356              bool update_action)
1357 {
1358         struct mlx5_flow_root_namespace *root;
1359         struct mlx5_flow_handle *handle;
1360         struct mlx5_flow_table *ft;
1361         int modify_mask = 0;
1362         int err;
1363         bool new_rule = false;
1364
1365         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1366                                     &new_rule);
1367         if (IS_ERR(handle) || !new_rule)
1368                 goto out;
1369
1370         if (update_action)
1371                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1372
1373         fs_get_obj(ft, fg->node.parent);
1374         root = find_root(&fg->node);
1375         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1376                 err = root->cmds->create_fte(root, ft, fg, fte);
1377         else
1378                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1379         if (err)
1380                 goto free_handle;
1381
1382         fte->node.active = true;
1383         fte->status |= FS_FTE_STATUS_EXISTING;
1384         atomic_inc(&fg->node.version);
1385
1386 out:
1387         return handle;
1388
1389 free_handle:
1390         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1391         return ERR_PTR(err);
1392 }
1393
1394 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1395                                                      const struct mlx5_flow_spec *spec)
1396 {
1397         struct list_head *prev = &ft->node.children;
1398         u32 max_fte = ft->autogroup.max_fte;
1399         unsigned int candidate_index = 0;
1400         unsigned int group_size = 0;
1401         struct mlx5_flow_group *fg;
1402
1403         if (!ft->autogroup.active)
1404                 return ERR_PTR(-ENOENT);
1405
1406         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1407                 group_size = ft->autogroup.group_size;
1408
1409         /*  max_fte == ft->autogroup.max_types */
1410         if (group_size == 0)
1411                 group_size = 1;
1412
1413         /* sorted by start_index */
1414         fs_for_each_fg(fg, ft) {
1415                 if (candidate_index + group_size > fg->start_index)
1416                         candidate_index = fg->start_index + fg->max_ftes;
1417                 else
1418                         break;
1419                 prev = &fg->node.list;
1420         }
1421
1422         if (candidate_index + group_size > max_fte)
1423                 return ERR_PTR(-ENOSPC);
1424
1425         fg = alloc_insert_flow_group(ft,
1426                                      spec->match_criteria_enable,
1427                                      spec->match_criteria,
1428                                      candidate_index,
1429                                      candidate_index + group_size - 1,
1430                                      prev);
1431         if (IS_ERR(fg))
1432                 goto out;
1433
1434         if (group_size == ft->autogroup.group_size)
1435                 ft->autogroup.num_groups++;
1436
1437 out:
1438         return fg;
1439 }
1440
1441 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1442                                   struct mlx5_flow_group *fg)
1443 {
1444         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1445         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1446         void *match_criteria_addr;
1447         u8 src_esw_owner_mask_on;
1448         void *misc;
1449         int err;
1450         u32 *in;
1451
1452         in = kvzalloc(inlen, GFP_KERNEL);
1453         if (!in)
1454                 return -ENOMEM;
1455
1456         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1457                  fg->mask.match_criteria_enable);
1458         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1459         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1460                  fg->max_ftes - 1);
1461
1462         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1463                             misc_parameters);
1464         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1465                                          source_eswitch_owner_vhca_id);
1466         MLX5_SET(create_flow_group_in, in,
1467                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1468
1469         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1470                                            in, match_criteria);
1471         memcpy(match_criteria_addr, fg->mask.match_criteria,
1472                sizeof(fg->mask.match_criteria));
1473
1474         err = root->cmds->create_flow_group(root, ft, in, fg);
1475         if (!err) {
1476                 fg->node.active = true;
1477                 trace_mlx5_fs_add_fg(fg);
1478         }
1479
1480         kvfree(in);
1481         return err;
1482 }
1483
1484 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1485                                 struct mlx5_flow_destination *d2)
1486 {
1487         if (d1->type == d2->type) {
1488                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1489                      d1->vport.num == d2->vport.num &&
1490                      d1->vport.flags == d2->vport.flags &&
1491                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1492                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1493                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1494                       (d1->vport.pkt_reformat->id ==
1495                        d2->vport.pkt_reformat->id) : true)) ||
1496                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1497                      d1->ft == d2->ft) ||
1498                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1499                      d1->tir_num == d2->tir_num) ||
1500                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1501                      d1->ft_num == d2->ft_num))
1502                         return true;
1503         }
1504
1505         return false;
1506 }
1507
1508 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1509                                              struct mlx5_flow_destination *dest)
1510 {
1511         struct mlx5_flow_rule *rule;
1512
1513         list_for_each_entry(rule, &fte->node.children, node.list) {
1514                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1515                         return rule;
1516         }
1517         return NULL;
1518 }
1519
1520 static bool check_conflicting_actions(u32 action1, u32 action2)
1521 {
1522         u32 xored_actions = action1 ^ action2;
1523
1524         /* if one rule only wants to count, it's ok */
1525         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1526             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1527                 return false;
1528
1529         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1530                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1531                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1532                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1533                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1534                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1535                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1536                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1537                 return true;
1538
1539         return false;
1540 }
1541
1542 static int check_conflicting_ftes(struct fs_fte *fte,
1543                                   const struct mlx5_flow_context *flow_context,
1544                                   const struct mlx5_flow_act *flow_act)
1545 {
1546         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1547                 mlx5_core_warn(get_dev(&fte->node),
1548                                "Found two FTEs with conflicting actions\n");
1549                 return -EEXIST;
1550         }
1551
1552         if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1553             fte->flow_context.flow_tag != flow_context->flow_tag) {
1554                 mlx5_core_warn(get_dev(&fte->node),
1555                                "FTE flow tag %u already exists with different flow tag %u\n",
1556                                fte->flow_context.flow_tag,
1557                                flow_context->flow_tag);
1558                 return -EEXIST;
1559         }
1560
1561         return 0;
1562 }
1563
1564 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1565                                             const struct mlx5_flow_spec *spec,
1566                                             struct mlx5_flow_act *flow_act,
1567                                             struct mlx5_flow_destination *dest,
1568                                             int dest_num,
1569                                             struct fs_fte *fte)
1570 {
1571         struct mlx5_flow_handle *handle;
1572         int old_action;
1573         int i;
1574         int ret;
1575
1576         ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1577         if (ret)
1578                 return ERR_PTR(ret);
1579
1580         old_action = fte->action.action;
1581         fte->action.action |= flow_act->action;
1582         handle = add_rule_fte(fte, fg, dest, dest_num,
1583                               old_action != flow_act->action);
1584         if (IS_ERR(handle)) {
1585                 fte->action.action = old_action;
1586                 return handle;
1587         }
1588         trace_mlx5_fs_set_fte(fte, false);
1589
1590         for (i = 0; i < handle->num_rules; i++) {
1591                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1592                         tree_add_node(&handle->rule[i]->node, &fte->node);
1593                         trace_mlx5_fs_add_rule(handle->rule[i]);
1594                 }
1595         }
1596         return handle;
1597 }
1598
1599 static bool counter_is_valid(u32 action)
1600 {
1601         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1602                           MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1603                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1604 }
1605
1606 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1607                           struct mlx5_flow_act *flow_act,
1608                           struct mlx5_flow_table *ft)
1609 {
1610         bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1611         u32 action = flow_act->action;
1612
1613         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1614                 return counter_is_valid(action);
1615
1616         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1617                 return true;
1618
1619         if (ignore_level) {
1620                 if (ft->type != FS_FT_FDB &&
1621                     ft->type != FS_FT_NIC_RX)
1622                         return false;
1623
1624                 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1625                     ft->type != dest->ft->type)
1626                         return false;
1627         }
1628
1629         if (!dest || ((dest->type ==
1630             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1631             (dest->ft->level <= ft->level && !ignore_level)))
1632                 return false;
1633         return true;
1634 }
1635
1636 struct match_list {
1637         struct list_head        list;
1638         struct mlx5_flow_group *g;
1639 };
1640
1641 static void free_match_list(struct match_list *head, bool ft_locked)
1642 {
1643         struct match_list *iter, *match_tmp;
1644
1645         list_for_each_entry_safe(iter, match_tmp, &head->list,
1646                                  list) {
1647                 tree_put_node(&iter->g->node, ft_locked);
1648                 list_del(&iter->list);
1649                 kfree(iter);
1650         }
1651 }
1652
1653 static int build_match_list(struct match_list *match_head,
1654                             struct mlx5_flow_table *ft,
1655                             const struct mlx5_flow_spec *spec,
1656                             bool ft_locked)
1657 {
1658         struct rhlist_head *tmp, *list;
1659         struct mlx5_flow_group *g;
1660         int err = 0;
1661
1662         rcu_read_lock();
1663         INIT_LIST_HEAD(&match_head->list);
1664         /* Collect all fgs which has a matching match_criteria */
1665         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1666         /* RCU is atomic, we can't execute FW commands here */
1667         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1668                 struct match_list *curr_match;
1669
1670                 if (unlikely(!tree_get_node(&g->node)))
1671                         continue;
1672
1673                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1674                 if (!curr_match) {
1675                         free_match_list(match_head, ft_locked);
1676                         err = -ENOMEM;
1677                         goto out;
1678                 }
1679                 curr_match->g = g;
1680                 list_add_tail(&curr_match->list, &match_head->list);
1681         }
1682 out:
1683         rcu_read_unlock();
1684         return err;
1685 }
1686
1687 static u64 matched_fgs_get_version(struct list_head *match_head)
1688 {
1689         struct match_list *iter;
1690         u64 version = 0;
1691
1692         list_for_each_entry(iter, match_head, list)
1693                 version += (u64)atomic_read(&iter->g->node.version);
1694         return version;
1695 }
1696
1697 static struct fs_fte *
1698 lookup_fte_locked(struct mlx5_flow_group *g,
1699                   const u32 *match_value,
1700                   bool take_write)
1701 {
1702         struct fs_fte *fte_tmp;
1703
1704         if (take_write)
1705                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1706         else
1707                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1708         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1709                                          rhash_fte);
1710         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1711                 fte_tmp = NULL;
1712                 goto out;
1713         }
1714         if (!fte_tmp->node.active) {
1715                 tree_put_node(&fte_tmp->node, false);
1716                 fte_tmp = NULL;
1717                 goto out;
1718         }
1719
1720         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1721 out:
1722         if (take_write)
1723                 up_write_ref_node(&g->node, false);
1724         else
1725                 up_read_ref_node(&g->node);
1726         return fte_tmp;
1727 }
1728
1729 static struct mlx5_flow_handle *
1730 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1731                        struct list_head *match_head,
1732                        const struct mlx5_flow_spec *spec,
1733                        struct mlx5_flow_act *flow_act,
1734                        struct mlx5_flow_destination *dest,
1735                        int dest_num,
1736                        int ft_version)
1737 {
1738         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1739         struct mlx5_flow_group *g;
1740         struct mlx5_flow_handle *rule;
1741         struct match_list *iter;
1742         bool take_write = false;
1743         struct fs_fte *fte;
1744         u64  version = 0;
1745         int err;
1746
1747         fte = alloc_fte(ft, spec, flow_act);
1748         if (IS_ERR(fte))
1749                 return  ERR_PTR(-ENOMEM);
1750
1751 search_again_locked:
1752         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1753                 goto skip_search;
1754         version = matched_fgs_get_version(match_head);
1755         /* Try to find an fte with identical match value and attempt update its
1756          * action.
1757          */
1758         list_for_each_entry(iter, match_head, list) {
1759                 struct fs_fte *fte_tmp;
1760
1761                 g = iter->g;
1762                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1763                 if (!fte_tmp)
1764                         continue;
1765                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1766                 up_write_ref_node(&fte_tmp->node, false);
1767                 tree_put_node(&fte_tmp->node, false);
1768                 kmem_cache_free(steering->ftes_cache, fte);
1769                 return rule;
1770         }
1771
1772 skip_search:
1773         /* No group with matching fte found, or we skipped the search.
1774          * Try to add a new fte to any matching fg.
1775          */
1776
1777         /* Check the ft version, for case that new flow group
1778          * was added while the fgs weren't locked
1779          */
1780         if (atomic_read(&ft->node.version) != ft_version) {
1781                 rule = ERR_PTR(-EAGAIN);
1782                 goto out;
1783         }
1784
1785         /* Check the fgs version. If version have changed it could be that an
1786          * FTE with the same match value was added while the fgs weren't
1787          * locked.
1788          */
1789         if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1790             version != matched_fgs_get_version(match_head)) {
1791                 take_write = true;
1792                 goto search_again_locked;
1793         }
1794
1795         list_for_each_entry(iter, match_head, list) {
1796                 g = iter->g;
1797
1798                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1799
1800                 if (!g->node.active) {
1801                         up_write_ref_node(&g->node, false);
1802                         continue;
1803                 }
1804
1805                 err = insert_fte(g, fte);
1806                 if (err) {
1807                         up_write_ref_node(&g->node, false);
1808                         if (err == -ENOSPC)
1809                                 continue;
1810                         kmem_cache_free(steering->ftes_cache, fte);
1811                         return ERR_PTR(err);
1812                 }
1813
1814                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1815                 up_write_ref_node(&g->node, false);
1816                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1817                 up_write_ref_node(&fte->node, false);
1818                 return rule;
1819         }
1820         rule = ERR_PTR(-ENOENT);
1821 out:
1822         kmem_cache_free(steering->ftes_cache, fte);
1823         return rule;
1824 }
1825
1826 static struct mlx5_flow_handle *
1827 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1828                      const struct mlx5_flow_spec *spec,
1829                      struct mlx5_flow_act *flow_act,
1830                      struct mlx5_flow_destination *dest,
1831                      int dest_num)
1832
1833 {
1834         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1835         struct mlx5_flow_handle *rule;
1836         struct match_list match_head;
1837         struct mlx5_flow_group *g;
1838         bool take_write = false;
1839         struct fs_fte *fte;
1840         int version;
1841         int err;
1842         int i;
1843
1844         if (!check_valid_spec(spec))
1845                 return ERR_PTR(-EINVAL);
1846
1847         for (i = 0; i < dest_num; i++) {
1848                 if (!dest_is_valid(&dest[i], flow_act, ft))
1849                         return ERR_PTR(-EINVAL);
1850         }
1851         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1852 search_again_locked:
1853         version = atomic_read(&ft->node.version);
1854
1855         /* Collect all fgs which has a matching match_criteria */
1856         err = build_match_list(&match_head, ft, spec, take_write);
1857         if (err) {
1858                 if (take_write)
1859                         up_write_ref_node(&ft->node, false);
1860                 else
1861                         up_read_ref_node(&ft->node);
1862                 return ERR_PTR(err);
1863         }
1864
1865         if (!take_write)
1866                 up_read_ref_node(&ft->node);
1867
1868         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1869                                       dest_num, version);
1870         free_match_list(&match_head, take_write);
1871         if (!IS_ERR(rule) ||
1872             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1873                 if (take_write)
1874                         up_write_ref_node(&ft->node, false);
1875                 return rule;
1876         }
1877
1878         if (!take_write) {
1879                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1880                 take_write = true;
1881         }
1882
1883         if (PTR_ERR(rule) == -EAGAIN ||
1884             version != atomic_read(&ft->node.version))
1885                 goto search_again_locked;
1886
1887         g = alloc_auto_flow_group(ft, spec);
1888         if (IS_ERR(g)) {
1889                 rule = ERR_CAST(g);
1890                 up_write_ref_node(&ft->node, false);
1891                 return rule;
1892         }
1893
1894         fte = alloc_fte(ft, spec, flow_act);
1895         if (IS_ERR(fte)) {
1896                 up_write_ref_node(&ft->node, false);
1897                 err = PTR_ERR(fte);
1898                 goto err_alloc_fte;
1899         }
1900
1901         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1902         up_write_ref_node(&ft->node, false);
1903
1904         err = create_auto_flow_group(ft, g);
1905         if (err)
1906                 goto err_release_fg;
1907
1908         err = insert_fte(g, fte);
1909         if (err)
1910                 goto err_release_fg;
1911
1912         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1913         up_write_ref_node(&g->node, false);
1914         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1915         up_write_ref_node(&fte->node, false);
1916         tree_put_node(&g->node, false);
1917         return rule;
1918
1919 err_release_fg:
1920         up_write_ref_node(&g->node, false);
1921         kmem_cache_free(steering->ftes_cache, fte);
1922 err_alloc_fte:
1923         tree_put_node(&g->node, false);
1924         return ERR_PTR(err);
1925 }
1926
1927 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1928 {
1929         return ((ft->type == FS_FT_NIC_RX) &&
1930                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1931 }
1932
1933 struct mlx5_flow_handle *
1934 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1935                     const struct mlx5_flow_spec *spec,
1936                     struct mlx5_flow_act *flow_act,
1937                     struct mlx5_flow_destination *dest,
1938                     int num_dest)
1939 {
1940         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1941         static const struct mlx5_flow_spec zero_spec = {};
1942         struct mlx5_flow_destination *gen_dest = NULL;
1943         struct mlx5_flow_table *next_ft = NULL;
1944         struct mlx5_flow_handle *handle = NULL;
1945         u32 sw_action = flow_act->action;
1946         int i;
1947
1948         if (!spec)
1949                 spec = &zero_spec;
1950
1951         if (!is_fwd_next_action(sw_action))
1952                 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1953
1954         if (!fwd_next_prio_supported(ft))
1955                 return ERR_PTR(-EOPNOTSUPP);
1956
1957         mutex_lock(&root->chain_lock);
1958         next_ft = find_next_fwd_ft(ft, flow_act);
1959         if (!next_ft) {
1960                 handle = ERR_PTR(-EOPNOTSUPP);
1961                 goto unlock;
1962         }
1963
1964         gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
1965                            GFP_KERNEL);
1966         if (!gen_dest) {
1967                 handle = ERR_PTR(-ENOMEM);
1968                 goto unlock;
1969         }
1970         for (i = 0; i < num_dest; i++)
1971                 gen_dest[i] = dest[i];
1972         gen_dest[i].type =
1973                 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1974         gen_dest[i].ft = next_ft;
1975         dest = gen_dest;
1976         num_dest++;
1977         flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
1978                               MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
1979         flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1980         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1981         if (IS_ERR(handle))
1982                 goto unlock;
1983
1984         if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
1985                 mutex_lock(&next_ft->lock);
1986                 list_add(&handle->rule[num_dest - 1]->next_ft,
1987                          &next_ft->fwd_rules);
1988                 mutex_unlock(&next_ft->lock);
1989                 handle->rule[num_dest - 1]->sw_action = sw_action;
1990                 handle->rule[num_dest - 1]->ft = ft;
1991         }
1992 unlock:
1993         mutex_unlock(&root->chain_lock);
1994         kfree(gen_dest);
1995         return handle;
1996 }
1997 EXPORT_SYMBOL(mlx5_add_flow_rules);
1998
1999 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2000 {
2001         struct fs_fte *fte;
2002         int i;
2003
2004         /* In order to consolidate the HW changes we lock the FTE for other
2005          * changes, and increase its refcount, in order not to perform the
2006          * "del" functions of the FTE. Will handle them here.
2007          * The removal of the rules is done under locked FTE.
2008          * After removing all the handle's rules, if there are remaining
2009          * rules, it means we just need to modify the FTE in FW, and
2010          * unlock/decrease the refcount we increased before.
2011          * Otherwise, it means the FTE should be deleted. First delete the
2012          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2013          * the FTE, which will handle the last decrease of the refcount, as
2014          * well as required handling of its parent.
2015          */
2016         fs_get_obj(fte, handle->rule[0]->node.parent);
2017         down_write_ref_node(&fte->node, false);
2018         for (i = handle->num_rules - 1; i >= 0; i--)
2019                 tree_remove_node(&handle->rule[i]->node, true);
2020         if (fte->dests_size) {
2021                 if (fte->modify_mask)
2022                         modify_fte(fte);
2023                 up_write_ref_node(&fte->node, false);
2024         } else if (list_empty(&fte->node.children)) {
2025                 del_hw_fte(&fte->node);
2026                 /* Avoid double call to del_hw_fte */
2027                 fte->node.del_hw_func = NULL;
2028                 up_write_ref_node(&fte->node, false);
2029                 tree_put_node(&fte->node, false);
2030         }
2031         kfree(handle);
2032 }
2033 EXPORT_SYMBOL(mlx5_del_flow_rules);
2034
2035 /* Assuming prio->node.children(flow tables) is sorted by level */
2036 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2037 {
2038         struct fs_prio *prio;
2039
2040         fs_get_obj(prio, ft->node.parent);
2041
2042         if (!list_is_last(&ft->node.list, &prio->node.children))
2043                 return list_next_entry(ft, node.list);
2044         return find_next_chained_ft(prio);
2045 }
2046
2047 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2048 {
2049         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2050         struct mlx5_ft_underlay_qp *uqp;
2051         struct mlx5_flow_table *new_root_ft = NULL;
2052         int err = 0;
2053         u32 qpn;
2054
2055         if (root->root_ft != ft)
2056                 return 0;
2057
2058         new_root_ft = find_next_ft(ft);
2059         if (!new_root_ft) {
2060                 root->root_ft = NULL;
2061                 return 0;
2062         }
2063
2064         if (list_empty(&root->underlay_qpns)) {
2065                 /* Don't set any QPN (zero) in case QPN list is empty */
2066                 qpn = 0;
2067                 err = root->cmds->update_root_ft(root, new_root_ft,
2068                                                  qpn, false);
2069         } else {
2070                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2071                         qpn = uqp->qpn;
2072                         err = root->cmds->update_root_ft(root,
2073                                                          new_root_ft, qpn,
2074                                                          false);
2075                         if (err)
2076                                 break;
2077                 }
2078         }
2079
2080         if (err)
2081                 mlx5_core_warn(root->dev,
2082                                "Update root flow table of id(%u) qpn(%d) failed\n",
2083                                ft->id, qpn);
2084         else
2085                 root->root_ft = new_root_ft;
2086
2087         return 0;
2088 }
2089
2090 /* Connect flow table from previous priority to
2091  * the next flow table.
2092  */
2093 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2094 {
2095         struct mlx5_core_dev *dev = get_dev(&ft->node);
2096         struct mlx5_flow_table *next_ft;
2097         struct fs_prio *prio;
2098         int err = 0;
2099
2100         err = update_root_ft_destroy(ft);
2101         if (err)
2102                 return err;
2103
2104         fs_get_obj(prio, ft->node.parent);
2105         if  (!(list_first_entry(&prio->node.children,
2106                                 struct mlx5_flow_table,
2107                                 node.list) == ft))
2108                 return 0;
2109
2110         next_ft = find_next_chained_ft(prio);
2111         err = connect_fwd_rules(dev, next_ft, ft);
2112         if (err)
2113                 return err;
2114
2115         err = connect_prev_fts(dev, next_ft, prio);
2116         if (err)
2117                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2118                                ft->id);
2119         return err;
2120 }
2121
2122 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2123 {
2124         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2125         int err = 0;
2126
2127         mutex_lock(&root->chain_lock);
2128         if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2129                 err = disconnect_flow_table(ft);
2130         if (err) {
2131                 mutex_unlock(&root->chain_lock);
2132                 return err;
2133         }
2134         if (tree_remove_node(&ft->node, false))
2135                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2136                                ft->id);
2137         mutex_unlock(&root->chain_lock);
2138
2139         return err;
2140 }
2141 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2142
2143 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2144 {
2145         if (tree_remove_node(&fg->node, false))
2146                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2147                                fg->id);
2148 }
2149
2150 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2151                                                 int n)
2152 {
2153         struct mlx5_flow_steering *steering = dev->priv.steering;
2154
2155         if (!steering || !steering->fdb_sub_ns)
2156                 return NULL;
2157
2158         return steering->fdb_sub_ns[n];
2159 }
2160 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2161
2162 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2163                                                     enum mlx5_flow_namespace_type type)
2164 {
2165         struct mlx5_flow_steering *steering = dev->priv.steering;
2166         struct mlx5_flow_root_namespace *root_ns;
2167         int prio = 0;
2168         struct fs_prio *fs_prio;
2169         struct mlx5_flow_namespace *ns;
2170
2171         if (!steering)
2172                 return NULL;
2173
2174         switch (type) {
2175         case MLX5_FLOW_NAMESPACE_FDB:
2176                 if (steering->fdb_root_ns)
2177                         return &steering->fdb_root_ns->ns;
2178                 return NULL;
2179         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2180                 if (steering->sniffer_rx_root_ns)
2181                         return &steering->sniffer_rx_root_ns->ns;
2182                 return NULL;
2183         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2184                 if (steering->sniffer_tx_root_ns)
2185                         return &steering->sniffer_tx_root_ns->ns;
2186                 return NULL;
2187         default:
2188                 break;
2189         }
2190
2191         if (type == MLX5_FLOW_NAMESPACE_EGRESS ||
2192             type == MLX5_FLOW_NAMESPACE_EGRESS_KERNEL) {
2193                 root_ns = steering->egress_root_ns;
2194                 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2195         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX) {
2196                 root_ns = steering->rdma_rx_root_ns;
2197                 prio = RDMA_RX_BYPASS_PRIO;
2198         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL) {
2199                 root_ns = steering->rdma_rx_root_ns;
2200                 prio = RDMA_RX_KERNEL_PRIO;
2201         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_TX) {
2202                 root_ns = steering->rdma_tx_root_ns;
2203         } else { /* Must be NIC RX */
2204                 root_ns = steering->root_ns;
2205                 prio = type;
2206         }
2207
2208         if (!root_ns)
2209                 return NULL;
2210
2211         fs_prio = find_prio(&root_ns->ns, prio);
2212         if (!fs_prio)
2213                 return NULL;
2214
2215         ns = list_first_entry(&fs_prio->node.children,
2216                               typeof(*ns),
2217                               node.list);
2218
2219         return ns;
2220 }
2221 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2222
2223 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2224                                                               enum mlx5_flow_namespace_type type,
2225                                                               int vport)
2226 {
2227         struct mlx5_flow_steering *steering = dev->priv.steering;
2228
2229         if (!steering || vport >= mlx5_eswitch_get_total_vports(dev))
2230                 return NULL;
2231
2232         switch (type) {
2233         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2234                 if (steering->esw_egress_root_ns &&
2235                     steering->esw_egress_root_ns[vport])
2236                         return &steering->esw_egress_root_ns[vport]->ns;
2237                 else
2238                         return NULL;
2239         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2240                 if (steering->esw_ingress_root_ns &&
2241                     steering->esw_ingress_root_ns[vport])
2242                         return &steering->esw_ingress_root_ns[vport]->ns;
2243                 else
2244                         return NULL;
2245         default:
2246                 return NULL;
2247         }
2248 }
2249
2250 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2251                                        unsigned int prio,
2252                                        int num_levels,
2253                                        enum fs_node_type type)
2254 {
2255         struct fs_prio *fs_prio;
2256
2257         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2258         if (!fs_prio)
2259                 return ERR_PTR(-ENOMEM);
2260
2261         fs_prio->node.type = type;
2262         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2263         tree_add_node(&fs_prio->node, &ns->node);
2264         fs_prio->num_levels = num_levels;
2265         fs_prio->prio = prio;
2266         list_add_tail(&fs_prio->node.list, &ns->node.children);
2267
2268         return fs_prio;
2269 }
2270
2271 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2272                                               unsigned int prio,
2273                                               int num_levels)
2274 {
2275         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2276 }
2277
2278 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2279                                       unsigned int prio, int num_levels)
2280 {
2281         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2282 }
2283
2284 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2285                                                      *ns)
2286 {
2287         ns->node.type = FS_TYPE_NAMESPACE;
2288
2289         return ns;
2290 }
2291
2292 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2293                                                        int def_miss_act)
2294 {
2295         struct mlx5_flow_namespace      *ns;
2296
2297         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2298         if (!ns)
2299                 return ERR_PTR(-ENOMEM);
2300
2301         fs_init_namespace(ns);
2302         ns->def_miss_action = def_miss_act;
2303         tree_init_node(&ns->node, NULL, del_sw_ns);
2304         tree_add_node(&ns->node, &prio->node);
2305         list_add_tail(&ns->node.list, &prio->node.children);
2306
2307         return ns;
2308 }
2309
2310 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2311                              struct init_tree_node *prio_metadata)
2312 {
2313         struct fs_prio *fs_prio;
2314         int i;
2315
2316         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2317                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2318                 if (IS_ERR(fs_prio))
2319                         return PTR_ERR(fs_prio);
2320         }
2321         return 0;
2322 }
2323
2324 #define FLOW_TABLE_BIT_SZ 1
2325 #define GET_FLOW_TABLE_CAP(dev, offset) \
2326         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2327                         offset / 32)) >>                                        \
2328           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2329 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2330 {
2331         int i;
2332
2333         for (i = 0; i < caps->arr_sz; i++) {
2334                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2335                         return false;
2336         }
2337         return true;
2338 }
2339
2340 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2341                                     struct init_tree_node *init_node,
2342                                     struct fs_node *fs_parent_node,
2343                                     struct init_tree_node *init_parent_node,
2344                                     int prio)
2345 {
2346         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2347                                               flow_table_properties_nic_receive.
2348                                               max_ft_level);
2349         struct mlx5_flow_namespace *fs_ns;
2350         struct fs_prio *fs_prio;
2351         struct fs_node *base;
2352         int i;
2353         int err;
2354
2355         if (init_node->type == FS_TYPE_PRIO) {
2356                 if ((init_node->min_ft_level > max_ft_level) ||
2357                     !has_required_caps(steering->dev, &init_node->caps))
2358                         return 0;
2359
2360                 fs_get_obj(fs_ns, fs_parent_node);
2361                 if (init_node->num_leaf_prios)
2362                         return create_leaf_prios(fs_ns, prio, init_node);
2363                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2364                 if (IS_ERR(fs_prio))
2365                         return PTR_ERR(fs_prio);
2366                 base = &fs_prio->node;
2367         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2368                 fs_get_obj(fs_prio, fs_parent_node);
2369                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2370                 if (IS_ERR(fs_ns))
2371                         return PTR_ERR(fs_ns);
2372                 base = &fs_ns->node;
2373         } else {
2374                 return -EINVAL;
2375         }
2376         prio = 0;
2377         for (i = 0; i < init_node->ar_size; i++) {
2378                 err = init_root_tree_recursive(steering, &init_node->children[i],
2379                                                base, init_node, prio);
2380                 if (err)
2381                         return err;
2382                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2383                     init_node->children[i].num_leaf_prios) {
2384                         prio += init_node->children[i].num_leaf_prios;
2385                 }
2386         }
2387
2388         return 0;
2389 }
2390
2391 static int init_root_tree(struct mlx5_flow_steering *steering,
2392                           struct init_tree_node *init_node,
2393                           struct fs_node *fs_parent_node)
2394 {
2395         int i;
2396         struct mlx5_flow_namespace *fs_ns;
2397         int err;
2398
2399         fs_get_obj(fs_ns, fs_parent_node);
2400         for (i = 0; i < init_node->ar_size; i++) {
2401                 err = init_root_tree_recursive(steering, &init_node->children[i],
2402                                                &fs_ns->node,
2403                                                init_node, i);
2404                 if (err)
2405                         return err;
2406         }
2407         return 0;
2408 }
2409
2410 static void del_sw_root_ns(struct fs_node *node)
2411 {
2412         struct mlx5_flow_root_namespace *root_ns;
2413         struct mlx5_flow_namespace *ns;
2414
2415         fs_get_obj(ns, node);
2416         root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2417         mutex_destroy(&root_ns->chain_lock);
2418         kfree(node);
2419 }
2420
2421 static struct mlx5_flow_root_namespace
2422 *create_root_ns(struct mlx5_flow_steering *steering,
2423                 enum fs_flow_table_type table_type)
2424 {
2425         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2426         struct mlx5_flow_root_namespace *root_ns;
2427         struct mlx5_flow_namespace *ns;
2428
2429         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2430             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2431                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2432
2433         /* Create the root namespace */
2434         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2435         if (!root_ns)
2436                 return NULL;
2437
2438         root_ns->dev = steering->dev;
2439         root_ns->table_type = table_type;
2440         root_ns->cmds = cmds;
2441
2442         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2443
2444         ns = &root_ns->ns;
2445         fs_init_namespace(ns);
2446         mutex_init(&root_ns->chain_lock);
2447         tree_init_node(&ns->node, NULL, del_sw_root_ns);
2448         tree_add_node(&ns->node, NULL);
2449
2450         return root_ns;
2451 }
2452
2453 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2454
2455 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2456 {
2457         struct fs_prio *prio;
2458
2459         fs_for_each_prio(prio, ns) {
2460                  /* This updates prio start_level and num_levels */
2461                 set_prio_attrs_in_prio(prio, acc_level);
2462                 acc_level += prio->num_levels;
2463         }
2464         return acc_level;
2465 }
2466
2467 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2468 {
2469         struct mlx5_flow_namespace *ns;
2470         int acc_level_ns = acc_level;
2471
2472         prio->start_level = acc_level;
2473         fs_for_each_ns(ns, prio) {
2474                 /* This updates start_level and num_levels of ns's priority descendants */
2475                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2476
2477                 /* If this a prio with chains, and we can jump from one chain
2478                  * (namepsace) to another, so we accumulate the levels
2479                  */
2480                 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2481                         acc_level = acc_level_ns;
2482         }
2483
2484         if (!prio->num_levels)
2485                 prio->num_levels = acc_level_ns - prio->start_level;
2486         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2487 }
2488
2489 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2490 {
2491         struct mlx5_flow_namespace *ns = &root_ns->ns;
2492         struct fs_prio *prio;
2493         int start_level = 0;
2494
2495         fs_for_each_prio(prio, ns) {
2496                 set_prio_attrs_in_prio(prio, start_level);
2497                 start_level += prio->num_levels;
2498         }
2499 }
2500
2501 #define ANCHOR_PRIO 0
2502 #define ANCHOR_SIZE 1
2503 #define ANCHOR_LEVEL 0
2504 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2505 {
2506         struct mlx5_flow_namespace *ns = NULL;
2507         struct mlx5_flow_table_attr ft_attr = {};
2508         struct mlx5_flow_table *ft;
2509
2510         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2511         if (WARN_ON(!ns))
2512                 return -EINVAL;
2513
2514         ft_attr.max_fte = ANCHOR_SIZE;
2515         ft_attr.level   = ANCHOR_LEVEL;
2516         ft_attr.prio    = ANCHOR_PRIO;
2517
2518         ft = mlx5_create_flow_table(ns, &ft_attr);
2519         if (IS_ERR(ft)) {
2520                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2521                 return PTR_ERR(ft);
2522         }
2523         return 0;
2524 }
2525
2526 static int init_root_ns(struct mlx5_flow_steering *steering)
2527 {
2528         int err;
2529
2530         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2531         if (!steering->root_ns)
2532                 return -ENOMEM;
2533
2534         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2535         if (err)
2536                 goto out_err;
2537
2538         set_prio_attrs(steering->root_ns);
2539         err = create_anchor_flow_table(steering);
2540         if (err)
2541                 goto out_err;
2542
2543         return 0;
2544
2545 out_err:
2546         cleanup_root_ns(steering->root_ns);
2547         steering->root_ns = NULL;
2548         return err;
2549 }
2550
2551 static void clean_tree(struct fs_node *node)
2552 {
2553         if (node) {
2554                 struct fs_node *iter;
2555                 struct fs_node *temp;
2556
2557                 tree_get_node(node);
2558                 list_for_each_entry_safe(iter, temp, &node->children, list)
2559                         clean_tree(iter);
2560                 tree_put_node(node, false);
2561                 tree_remove_node(node, false);
2562         }
2563 }
2564
2565 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2566 {
2567         if (!root_ns)
2568                 return;
2569
2570         clean_tree(&root_ns->ns.node);
2571 }
2572
2573 static void cleanup_egress_acls_root_ns(struct mlx5_core_dev *dev)
2574 {
2575         struct mlx5_flow_steering *steering = dev->priv.steering;
2576         int i;
2577
2578         if (!steering->esw_egress_root_ns)
2579                 return;
2580
2581         for (i = 0; i < mlx5_eswitch_get_total_vports(dev); i++)
2582                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2583
2584         kfree(steering->esw_egress_root_ns);
2585         steering->esw_egress_root_ns = NULL;
2586 }
2587
2588 static void cleanup_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2589 {
2590         struct mlx5_flow_steering *steering = dev->priv.steering;
2591         int i;
2592
2593         if (!steering->esw_ingress_root_ns)
2594                 return;
2595
2596         for (i = 0; i < mlx5_eswitch_get_total_vports(dev); i++)
2597                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2598
2599         kfree(steering->esw_ingress_root_ns);
2600         steering->esw_ingress_root_ns = NULL;
2601 }
2602
2603 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2604 {
2605         struct mlx5_flow_steering *steering = dev->priv.steering;
2606
2607         cleanup_root_ns(steering->root_ns);
2608         cleanup_egress_acls_root_ns(dev);
2609         cleanup_ingress_acls_root_ns(dev);
2610         cleanup_root_ns(steering->fdb_root_ns);
2611         steering->fdb_root_ns = NULL;
2612         kfree(steering->fdb_sub_ns);
2613         steering->fdb_sub_ns = NULL;
2614         cleanup_root_ns(steering->sniffer_rx_root_ns);
2615         cleanup_root_ns(steering->sniffer_tx_root_ns);
2616         cleanup_root_ns(steering->rdma_rx_root_ns);
2617         cleanup_root_ns(steering->rdma_tx_root_ns);
2618         cleanup_root_ns(steering->egress_root_ns);
2619         mlx5_cleanup_fc_stats(dev);
2620         kmem_cache_destroy(steering->ftes_cache);
2621         kmem_cache_destroy(steering->fgs_cache);
2622         kfree(steering);
2623 }
2624
2625 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2626 {
2627         struct fs_prio *prio;
2628
2629         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2630         if (!steering->sniffer_tx_root_ns)
2631                 return -ENOMEM;
2632
2633         /* Create single prio */
2634         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2635         return PTR_ERR_OR_ZERO(prio);
2636 }
2637
2638 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2639 {
2640         struct fs_prio *prio;
2641
2642         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2643         if (!steering->sniffer_rx_root_ns)
2644                 return -ENOMEM;
2645
2646         /* Create single prio */
2647         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2648         return PTR_ERR_OR_ZERO(prio);
2649 }
2650
2651 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2652 {
2653         int err;
2654
2655         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2656         if (!steering->rdma_rx_root_ns)
2657                 return -ENOMEM;
2658
2659         err = init_root_tree(steering, &rdma_rx_root_fs,
2660                              &steering->rdma_rx_root_ns->ns.node);
2661         if (err)
2662                 goto out_err;
2663
2664         set_prio_attrs(steering->rdma_rx_root_ns);
2665
2666         return 0;
2667
2668 out_err:
2669         cleanup_root_ns(steering->rdma_rx_root_ns);
2670         steering->rdma_rx_root_ns = NULL;
2671         return err;
2672 }
2673
2674 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2675 {
2676         int err;
2677
2678         steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2679         if (!steering->rdma_tx_root_ns)
2680                 return -ENOMEM;
2681
2682         err = init_root_tree(steering, &rdma_tx_root_fs,
2683                              &steering->rdma_tx_root_ns->ns.node);
2684         if (err)
2685                 goto out_err;
2686
2687         set_prio_attrs(steering->rdma_tx_root_ns);
2688
2689         return 0;
2690
2691 out_err:
2692         cleanup_root_ns(steering->rdma_tx_root_ns);
2693         steering->rdma_tx_root_ns = NULL;
2694         return err;
2695 }
2696
2697 /* FT and tc chains are stored in the same array so we can re-use the
2698  * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2699  * When creating a new ns for each chain store it in the first available slot.
2700  * Assume tc chains are created and stored first and only then the FT chain.
2701  */
2702 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2703                                         struct mlx5_flow_namespace *ns)
2704 {
2705         int chain = 0;
2706
2707         while (steering->fdb_sub_ns[chain])
2708                 ++chain;
2709
2710         steering->fdb_sub_ns[chain] = ns;
2711 }
2712
2713 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2714                                         struct fs_prio *maj_prio)
2715 {
2716         struct mlx5_flow_namespace *ns;
2717         struct fs_prio *min_prio;
2718         int prio;
2719
2720         ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2721         if (IS_ERR(ns))
2722                 return PTR_ERR(ns);
2723
2724         for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2725                 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2726                 if (IS_ERR(min_prio))
2727                         return PTR_ERR(min_prio);
2728         }
2729
2730         store_fdb_sub_ns_prio_chain(steering, ns);
2731
2732         return 0;
2733 }
2734
2735 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2736                              int fs_prio,
2737                              int chains)
2738 {
2739         struct fs_prio *maj_prio;
2740         int levels;
2741         int chain;
2742         int err;
2743
2744         levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2745         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2746                                           fs_prio,
2747                                           levels);
2748         if (IS_ERR(maj_prio))
2749                 return PTR_ERR(maj_prio);
2750
2751         for (chain = 0; chain < chains; chain++) {
2752                 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2753                 if (err)
2754                         return err;
2755         }
2756
2757         return 0;
2758 }
2759
2760 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2761 {
2762         int err;
2763
2764         steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2765                                        sizeof(*steering->fdb_sub_ns),
2766                                        GFP_KERNEL);
2767         if (!steering->fdb_sub_ns)
2768                 return -ENOMEM;
2769
2770         err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
2771         if (err)
2772                 return err;
2773
2774         err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
2775         if (err)
2776                 return err;
2777
2778         return 0;
2779 }
2780
2781 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2782 {
2783         struct fs_prio *maj_prio;
2784         int err;
2785
2786         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2787         if (!steering->fdb_root_ns)
2788                 return -ENOMEM;
2789
2790         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2791                                   1);
2792         if (IS_ERR(maj_prio)) {
2793                 err = PTR_ERR(maj_prio);
2794                 goto out_err;
2795         }
2796         err = create_fdb_fast_path(steering);
2797         if (err)
2798                 goto out_err;
2799
2800         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2801         if (IS_ERR(maj_prio)) {
2802                 err = PTR_ERR(maj_prio);
2803                 goto out_err;
2804         }
2805
2806         /* We put this priority last, knowing that nothing will get here
2807          * unless explicitly forwarded to. This is possible because the
2808          * slow path tables have catch all rules and nothing gets passed
2809          * those tables.
2810          */
2811         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
2812         if (IS_ERR(maj_prio)) {
2813                 err = PTR_ERR(maj_prio);
2814                 goto out_err;
2815         }
2816
2817         set_prio_attrs(steering->fdb_root_ns);
2818         return 0;
2819
2820 out_err:
2821         cleanup_root_ns(steering->fdb_root_ns);
2822         kfree(steering->fdb_sub_ns);
2823         steering->fdb_sub_ns = NULL;
2824         steering->fdb_root_ns = NULL;
2825         return err;
2826 }
2827
2828 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2829 {
2830         struct fs_prio *prio;
2831
2832         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2833         if (!steering->esw_egress_root_ns[vport])
2834                 return -ENOMEM;
2835
2836         /* create 1 prio*/
2837         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2838         return PTR_ERR_OR_ZERO(prio);
2839 }
2840
2841 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2842 {
2843         struct fs_prio *prio;
2844
2845         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2846         if (!steering->esw_ingress_root_ns[vport])
2847                 return -ENOMEM;
2848
2849         /* create 1 prio*/
2850         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2851         return PTR_ERR_OR_ZERO(prio);
2852 }
2853
2854 static int init_egress_acls_root_ns(struct mlx5_core_dev *dev)
2855 {
2856         struct mlx5_flow_steering *steering = dev->priv.steering;
2857         int total_vports = mlx5_eswitch_get_total_vports(dev);
2858         int err;
2859         int i;
2860
2861         steering->esw_egress_root_ns =
2862                         kcalloc(total_vports,
2863                                 sizeof(*steering->esw_egress_root_ns),
2864                                 GFP_KERNEL);
2865         if (!steering->esw_egress_root_ns)
2866                 return -ENOMEM;
2867
2868         for (i = 0; i < total_vports; i++) {
2869                 err = init_egress_acl_root_ns(steering, i);
2870                 if (err)
2871                         goto cleanup_root_ns;
2872         }
2873
2874         return 0;
2875
2876 cleanup_root_ns:
2877         for (i--; i >= 0; i--)
2878                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2879         kfree(steering->esw_egress_root_ns);
2880         steering->esw_egress_root_ns = NULL;
2881         return err;
2882 }
2883
2884 static int init_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2885 {
2886         struct mlx5_flow_steering *steering = dev->priv.steering;
2887         int total_vports = mlx5_eswitch_get_total_vports(dev);
2888         int err;
2889         int i;
2890
2891         steering->esw_ingress_root_ns =
2892                         kcalloc(total_vports,
2893                                 sizeof(*steering->esw_ingress_root_ns),
2894                                 GFP_KERNEL);
2895         if (!steering->esw_ingress_root_ns)
2896                 return -ENOMEM;
2897
2898         for (i = 0; i < total_vports; i++) {
2899                 err = init_ingress_acl_root_ns(steering, i);
2900                 if (err)
2901                         goto cleanup_root_ns;
2902         }
2903
2904         return 0;
2905
2906 cleanup_root_ns:
2907         for (i--; i >= 0; i--)
2908                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2909         kfree(steering->esw_ingress_root_ns);
2910         steering->esw_ingress_root_ns = NULL;
2911         return err;
2912 }
2913
2914 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2915 {
2916         int err;
2917
2918         steering->egress_root_ns = create_root_ns(steering,
2919                                                   FS_FT_NIC_TX);
2920         if (!steering->egress_root_ns)
2921                 return -ENOMEM;
2922
2923         err = init_root_tree(steering, &egress_root_fs,
2924                              &steering->egress_root_ns->ns.node);
2925         if (err)
2926                 goto cleanup;
2927         set_prio_attrs(steering->egress_root_ns);
2928         return 0;
2929 cleanup:
2930         cleanup_root_ns(steering->egress_root_ns);
2931         steering->egress_root_ns = NULL;
2932         return err;
2933 }
2934
2935 int mlx5_init_fs(struct mlx5_core_dev *dev)
2936 {
2937         struct mlx5_flow_steering *steering;
2938         int err = 0;
2939
2940         err = mlx5_init_fc_stats(dev);
2941         if (err)
2942                 return err;
2943
2944         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2945         if (!steering)
2946                 return -ENOMEM;
2947         steering->dev = dev;
2948         dev->priv.steering = steering;
2949
2950         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2951                                                 sizeof(struct mlx5_flow_group), 0,
2952                                                 0, NULL);
2953         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2954                                                  0, NULL);
2955         if (!steering->ftes_cache || !steering->fgs_cache) {
2956                 err = -ENOMEM;
2957                 goto err;
2958         }
2959
2960         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2961               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2962              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2963               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2964             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2965                 err = init_root_ns(steering);
2966                 if (err)
2967                         goto err;
2968         }
2969
2970         if (MLX5_ESWITCH_MANAGER(dev)) {
2971                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2972                         err = init_fdb_root_ns(steering);
2973                         if (err)
2974                                 goto err;
2975                 }
2976                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2977                         err = init_egress_acls_root_ns(dev);
2978                         if (err)
2979                                 goto err;
2980                 }
2981                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2982                         err = init_ingress_acls_root_ns(dev);
2983                         if (err)
2984                                 goto err;
2985                 }
2986         }
2987
2988         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2989                 err = init_sniffer_rx_root_ns(steering);
2990                 if (err)
2991                         goto err;
2992         }
2993
2994         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2995                 err = init_sniffer_tx_root_ns(steering);
2996                 if (err)
2997                         goto err;
2998         }
2999
3000         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
3001             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3002                 err = init_rdma_rx_root_ns(steering);
3003                 if (err)
3004                         goto err;
3005         }
3006
3007         if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3008                 err = init_rdma_tx_root_ns(steering);
3009                 if (err)
3010                         goto err;
3011         }
3012
3013         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE ||
3014             MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3015                 err = init_egress_root_ns(steering);
3016                 if (err)
3017                         goto err;
3018         }
3019
3020         return 0;
3021 err:
3022         mlx5_cleanup_fs(dev);
3023         return err;
3024 }
3025
3026 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3027 {
3028         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3029         struct mlx5_ft_underlay_qp *new_uqp;
3030         int err = 0;
3031
3032         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3033         if (!new_uqp)
3034                 return -ENOMEM;
3035
3036         mutex_lock(&root->chain_lock);
3037
3038         if (!root->root_ft) {
3039                 err = -EINVAL;
3040                 goto update_ft_fail;
3041         }
3042
3043         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3044                                          false);
3045         if (err) {
3046                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3047                                underlay_qpn, err);
3048                 goto update_ft_fail;
3049         }
3050
3051         new_uqp->qpn = underlay_qpn;
3052         list_add_tail(&new_uqp->list, &root->underlay_qpns);
3053
3054         mutex_unlock(&root->chain_lock);
3055
3056         return 0;
3057
3058 update_ft_fail:
3059         mutex_unlock(&root->chain_lock);
3060         kfree(new_uqp);
3061         return err;
3062 }
3063 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3064
3065 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3066 {
3067         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3068         struct mlx5_ft_underlay_qp *uqp;
3069         bool found = false;
3070         int err = 0;
3071
3072         mutex_lock(&root->chain_lock);
3073         list_for_each_entry(uqp, &root->underlay_qpns, list) {
3074                 if (uqp->qpn == underlay_qpn) {
3075                         found = true;
3076                         break;
3077                 }
3078         }
3079
3080         if (!found) {
3081                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3082                                underlay_qpn);
3083                 err = -EINVAL;
3084                 goto out;
3085         }
3086
3087         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3088                                          true);
3089         if (err)
3090                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3091                                underlay_qpn, err);
3092
3093         list_del(&uqp->list);
3094         mutex_unlock(&root->chain_lock);
3095         kfree(uqp);
3096
3097         return 0;
3098
3099 out:
3100         mutex_unlock(&root->chain_lock);
3101         return err;
3102 }
3103 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3104
3105 static struct mlx5_flow_root_namespace
3106 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3107 {
3108         struct mlx5_flow_namespace *ns;
3109
3110         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3111             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3112                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3113         else
3114                 ns = mlx5_get_flow_namespace(dev, ns_type);
3115         if (!ns)
3116                 return NULL;
3117
3118         return find_root(&ns->node);
3119 }
3120
3121 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3122                                                  u8 ns_type, u8 num_actions,
3123                                                  void *modify_actions)
3124 {
3125         struct mlx5_flow_root_namespace *root;
3126         struct mlx5_modify_hdr *modify_hdr;
3127         int err;
3128
3129         root = get_root_namespace(dev, ns_type);
3130         if (!root)
3131                 return ERR_PTR(-EOPNOTSUPP);
3132
3133         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3134         if (!modify_hdr)
3135                 return ERR_PTR(-ENOMEM);
3136
3137         modify_hdr->ns_type = ns_type;
3138         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3139                                               modify_actions, modify_hdr);
3140         if (err) {
3141                 kfree(modify_hdr);
3142                 return ERR_PTR(err);
3143         }
3144
3145         return modify_hdr;
3146 }
3147 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3148
3149 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3150                                 struct mlx5_modify_hdr *modify_hdr)
3151 {
3152         struct mlx5_flow_root_namespace *root;
3153
3154         root = get_root_namespace(dev, modify_hdr->ns_type);
3155         if (WARN_ON(!root))
3156                 return;
3157         root->cmds->modify_header_dealloc(root, modify_hdr);
3158         kfree(modify_hdr);
3159 }
3160 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3161
3162 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3163                                                      int reformat_type,
3164                                                      size_t size,
3165                                                      void *reformat_data,
3166                                                      enum mlx5_flow_namespace_type ns_type)
3167 {
3168         struct mlx5_pkt_reformat *pkt_reformat;
3169         struct mlx5_flow_root_namespace *root;
3170         int err;
3171
3172         root = get_root_namespace(dev, ns_type);
3173         if (!root)
3174                 return ERR_PTR(-EOPNOTSUPP);
3175
3176         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3177         if (!pkt_reformat)
3178                 return ERR_PTR(-ENOMEM);
3179
3180         pkt_reformat->ns_type = ns_type;
3181         pkt_reformat->reformat_type = reformat_type;
3182         err = root->cmds->packet_reformat_alloc(root, reformat_type, size,
3183                                                 reformat_data, ns_type,
3184                                                 pkt_reformat);
3185         if (err) {
3186                 kfree(pkt_reformat);
3187                 return ERR_PTR(err);
3188         }
3189
3190         return pkt_reformat;
3191 }
3192 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3193
3194 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3195                                   struct mlx5_pkt_reformat *pkt_reformat)
3196 {
3197         struct mlx5_flow_root_namespace *root;
3198
3199         root = get_root_namespace(dev, pkt_reformat->ns_type);
3200         if (WARN_ON(!root))
3201                 return;
3202         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3203         kfree(pkt_reformat);
3204 }
3205 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3206
3207 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3208                                  struct mlx5_flow_root_namespace *peer_ns)
3209 {
3210         if (peer_ns && ns->mode != peer_ns->mode) {
3211                 mlx5_core_err(ns->dev,
3212                               "Can't peer namespace of different steering mode\n");
3213                 return -EINVAL;
3214         }
3215
3216         return ns->cmds->set_peer(ns, peer_ns);
3217 }
3218
3219 /* This function should be called only at init stage of the namespace.
3220  * It is not safe to call this function while steering operations
3221  * are executed in the namespace.
3222  */
3223 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3224                                  enum mlx5_flow_steering_mode mode)
3225 {
3226         struct mlx5_flow_root_namespace *root;
3227         const struct mlx5_flow_cmds *cmds;
3228         int err;
3229
3230         root = find_root(&ns->node);
3231         if (&root->ns != ns)
3232         /* Can't set cmds to non root namespace */
3233                 return -EINVAL;
3234
3235         if (root->table_type != FS_FT_FDB)
3236                 return -EOPNOTSUPP;
3237
3238         if (root->mode == mode)
3239                 return 0;
3240
3241         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3242                 cmds = mlx5_fs_cmd_get_dr_cmds();
3243         else
3244                 cmds = mlx5_fs_cmd_get_fw_cmds();
3245         if (!cmds)
3246                 return -EOPNOTSUPP;
3247
3248         err = cmds->create_ns(root);
3249         if (err) {
3250                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3251                               err);
3252                 return err;
3253         }
3254
3255         root->cmds->destroy_ns(root);
3256         root->cmds = cmds;
3257         root->mode = mode;
3258
3259         return 0;
3260 }