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 /* Promiscuous, Vlan, mac, ttc, inner ttc, {aRFS/accel and esp/esp_err} */
 109 #define KERNEL_NIC_PRIO_NUM_LEVELS 7
 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         rhltable_destroy(&ft->fgs_hash);
1145         kfree(ft);
1146 unlock_root:
1147         mutex_unlock(&root->chain_lock);
1148         return ERR_PTR(err);
1149 }
1150
1151 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1152                                                struct mlx5_flow_table_attr *ft_attr)
1153 {
1154         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1155 }
1156 EXPORT_SYMBOL(mlx5_create_flow_table);
1157
1158 struct mlx5_flow_table *
1159 mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1160                              struct mlx5_flow_table_attr *ft_attr, u16 vport)
1161 {
1162         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1163 }
1164
1165 struct mlx5_flow_table*
1166 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1167                                  int prio, u32 level)
1168 {
1169         struct mlx5_flow_table_attr ft_attr = {};
1170
1171         ft_attr.level = level;
1172         ft_attr.prio  = prio;
1173         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1174 }
1175 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1176
1177 struct mlx5_flow_table*
1178 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1179                                     struct mlx5_flow_table_attr *ft_attr)
1180 {
1181         int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1182         int autogroups_max_fte = ft_attr->max_fte - num_reserved_entries;
1183         int max_num_groups = ft_attr->autogroup.max_num_groups;
1184         struct mlx5_flow_table *ft;
1185
1186         if (max_num_groups > autogroups_max_fte)
1187                 return ERR_PTR(-EINVAL);
1188         if (num_reserved_entries > ft_attr->max_fte)
1189                 return ERR_PTR(-EINVAL);
1190
1191         ft = mlx5_create_flow_table(ns, ft_attr);
1192         if (IS_ERR(ft))
1193                 return ft;
1194
1195         ft->autogroup.active = true;
1196         ft->autogroup.required_groups = max_num_groups;
1197         ft->autogroup.max_fte = autogroups_max_fte;
1198         /* We save place for flow groups in addition to max types */
1199         ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1200
1201         return ft;
1202 }
1203 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1204
1205 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1206                                                u32 *fg_in)
1207 {
1208         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1209         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1210                                             fg_in, match_criteria);
1211         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1212                                             fg_in,
1213                                             match_criteria_enable);
1214         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1215                                    start_flow_index);
1216         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1217                                  end_flow_index);
1218         struct mlx5_flow_group *fg;
1219         int err;
1220
1221         if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1222                 return ERR_PTR(-EPERM);
1223
1224         down_write_ref_node(&ft->node, false);
1225         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1226                                      start_index, end_index,
1227                                      ft->node.children.prev);
1228         up_write_ref_node(&ft->node, false);
1229         if (IS_ERR(fg))
1230                 return fg;
1231
1232         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1233         if (err) {
1234                 tree_put_node(&fg->node, false);
1235                 return ERR_PTR(err);
1236         }
1237         trace_mlx5_fs_add_fg(fg);
1238         fg->node.active = true;
1239
1240         return fg;
1241 }
1242 EXPORT_SYMBOL(mlx5_create_flow_group);
1243
1244 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1245 {
1246         struct mlx5_flow_rule *rule;
1247
1248         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1249         if (!rule)
1250                 return NULL;
1251
1252         INIT_LIST_HEAD(&rule->next_ft);
1253         rule->node.type = FS_TYPE_FLOW_DEST;
1254         if (dest)
1255                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1256
1257         return rule;
1258 }
1259
1260 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1261 {
1262         struct mlx5_flow_handle *handle;
1263
1264         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1265         if (!handle)
1266                 return NULL;
1267
1268         handle->num_rules = num_rules;
1269
1270         return handle;
1271 }
1272
1273 static void destroy_flow_handle(struct fs_fte *fte,
1274                                 struct mlx5_flow_handle *handle,
1275                                 struct mlx5_flow_destination *dest,
1276                                 int i)
1277 {
1278         for (; --i >= 0;) {
1279                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1280                         fte->dests_size--;
1281                         list_del(&handle->rule[i]->node.list);
1282                         kfree(handle->rule[i]);
1283                 }
1284         }
1285         kfree(handle);
1286 }
1287
1288 static struct mlx5_flow_handle *
1289 create_flow_handle(struct fs_fte *fte,
1290                    struct mlx5_flow_destination *dest,
1291                    int dest_num,
1292                    int *modify_mask,
1293                    bool *new_rule)
1294 {
1295         struct mlx5_flow_handle *handle;
1296         struct mlx5_flow_rule *rule = NULL;
1297         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1298         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1299         int type;
1300         int i = 0;
1301
1302         handle = alloc_handle((dest_num) ? dest_num : 1);
1303         if (!handle)
1304                 return ERR_PTR(-ENOMEM);
1305
1306         do {
1307                 if (dest) {
1308                         rule = find_flow_rule(fte, dest + i);
1309                         if (rule) {
1310                                 refcount_inc(&rule->node.refcount);
1311                                 goto rule_found;
1312                         }
1313                 }
1314
1315                 *new_rule = true;
1316                 rule = alloc_rule(dest + i);
1317                 if (!rule)
1318                         goto free_rules;
1319
1320                 /* Add dest to dests list- we need flow tables to be in the
1321                  * end of the list for forward to next prio rules.
1322                  */
1323                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1324                 if (dest &&
1325                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1326                         list_add(&rule->node.list, &fte->node.children);
1327                 else
1328                         list_add_tail(&rule->node.list, &fte->node.children);
1329                 if (dest) {
1330                         fte->dests_size++;
1331
1332                         type = dest[i].type ==
1333                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1334                         *modify_mask |= type ? count : dst;
1335                 }
1336 rule_found:
1337                 handle->rule[i] = rule;
1338         } while (++i < dest_num);
1339
1340         return handle;
1341
1342 free_rules:
1343         destroy_flow_handle(fte, handle, dest, i);
1344         return ERR_PTR(-ENOMEM);
1345 }
1346
1347 /* fte should not be deleted while calling this function */
1348 static struct mlx5_flow_handle *
1349 add_rule_fte(struct fs_fte *fte,
1350              struct mlx5_flow_group *fg,
1351              struct mlx5_flow_destination *dest,
1352              int dest_num,
1353              bool update_action)
1354 {
1355         struct mlx5_flow_root_namespace *root;
1356         struct mlx5_flow_handle *handle;
1357         struct mlx5_flow_table *ft;
1358         int modify_mask = 0;
1359         int err;
1360         bool new_rule = false;
1361
1362         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1363                                     &new_rule);
1364         if (IS_ERR(handle) || !new_rule)
1365                 goto out;
1366
1367         if (update_action)
1368                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1369
1370         fs_get_obj(ft, fg->node.parent);
1371         root = find_root(&fg->node);
1372         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1373                 err = root->cmds->create_fte(root, ft, fg, fte);
1374         else
1375                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1376         if (err)
1377                 goto free_handle;
1378
1379         fte->node.active = true;
1380         fte->status |= FS_FTE_STATUS_EXISTING;
1381         atomic_inc(&fg->node.version);
1382
1383 out:
1384         return handle;
1385
1386 free_handle:
1387         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1388         return ERR_PTR(err);
1389 }
1390
1391 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1392                                                      const struct mlx5_flow_spec *spec)
1393 {
1394         struct list_head *prev = &ft->node.children;
1395         u32 max_fte = ft->autogroup.max_fte;
1396         unsigned int candidate_index = 0;
1397         unsigned int group_size = 0;
1398         struct mlx5_flow_group *fg;
1399
1400         if (!ft->autogroup.active)
1401                 return ERR_PTR(-ENOENT);
1402
1403         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1404                 group_size = ft->autogroup.group_size;
1405
1406         /*  max_fte == ft->autogroup.max_types */
1407         if (group_size == 0)
1408                 group_size = 1;
1409
1410         /* sorted by start_index */
1411         fs_for_each_fg(fg, ft) {
1412                 if (candidate_index + group_size > fg->start_index)
1413                         candidate_index = fg->start_index + fg->max_ftes;
1414                 else
1415                         break;
1416                 prev = &fg->node.list;
1417         }
1418
1419         if (candidate_index + group_size > max_fte)
1420                 return ERR_PTR(-ENOSPC);
1421
1422         fg = alloc_insert_flow_group(ft,
1423                                      spec->match_criteria_enable,
1424                                      spec->match_criteria,
1425                                      candidate_index,
1426                                      candidate_index + group_size - 1,
1427                                      prev);
1428         if (IS_ERR(fg))
1429                 goto out;
1430
1431         if (group_size == ft->autogroup.group_size)
1432                 ft->autogroup.num_groups++;
1433
1434 out:
1435         return fg;
1436 }
1437
1438 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1439                                   struct mlx5_flow_group *fg)
1440 {
1441         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1442         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1443         void *match_criteria_addr;
1444         u8 src_esw_owner_mask_on;
1445         void *misc;
1446         int err;
1447         u32 *in;
1448
1449         in = kvzalloc(inlen, GFP_KERNEL);
1450         if (!in)
1451                 return -ENOMEM;
1452
1453         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1454                  fg->mask.match_criteria_enable);
1455         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1456         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1457                  fg->max_ftes - 1);
1458
1459         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1460                             misc_parameters);
1461         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1462                                          source_eswitch_owner_vhca_id);
1463         MLX5_SET(create_flow_group_in, in,
1464                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1465
1466         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1467                                            in, match_criteria);
1468         memcpy(match_criteria_addr, fg->mask.match_criteria,
1469                sizeof(fg->mask.match_criteria));
1470
1471         err = root->cmds->create_flow_group(root, ft, in, fg);
1472         if (!err) {
1473                 fg->node.active = true;
1474                 trace_mlx5_fs_add_fg(fg);
1475         }
1476
1477         kvfree(in);
1478         return err;
1479 }
1480
1481 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1482                                 struct mlx5_flow_destination *d2)
1483 {
1484         if (d1->type == d2->type) {
1485                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1486                      d1->vport.num == d2->vport.num &&
1487                      d1->vport.flags == d2->vport.flags &&
1488                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1489                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1490                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1491                       (d1->vport.pkt_reformat->id ==
1492                        d2->vport.pkt_reformat->id) : true)) ||
1493                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1494                      d1->ft == d2->ft) ||
1495                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1496                      d1->tir_num == d2->tir_num) ||
1497                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1498                      d1->ft_num == d2->ft_num))
1499                         return true;
1500         }
1501
1502         return false;
1503 }
1504
1505 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1506                                              struct mlx5_flow_destination *dest)
1507 {
1508         struct mlx5_flow_rule *rule;
1509
1510         list_for_each_entry(rule, &fte->node.children, node.list) {
1511                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1512                         return rule;
1513         }
1514         return NULL;
1515 }
1516
1517 static bool check_conflicting_actions(u32 action1, u32 action2)
1518 {
1519         u32 xored_actions = action1 ^ action2;
1520
1521         /* if one rule only wants to count, it's ok */
1522         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1523             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1524                 return false;
1525
1526         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1527                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1528                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1529                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1530                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1531                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1532                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1533                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1534                 return true;
1535
1536         return false;
1537 }
1538
1539 static int check_conflicting_ftes(struct fs_fte *fte,
1540                                   const struct mlx5_flow_context *flow_context,
1541                                   const struct mlx5_flow_act *flow_act)
1542 {
1543         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1544                 mlx5_core_warn(get_dev(&fte->node),
1545                                "Found two FTEs with conflicting actions\n");
1546                 return -EEXIST;
1547         }
1548
1549         if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1550             fte->flow_context.flow_tag != flow_context->flow_tag) {
1551                 mlx5_core_warn(get_dev(&fte->node),
1552                                "FTE flow tag %u already exists with different flow tag %u\n",
1553                                fte->flow_context.flow_tag,
1554                                flow_context->flow_tag);
1555                 return -EEXIST;
1556         }
1557
1558         return 0;
1559 }
1560
1561 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1562                                             const struct mlx5_flow_spec *spec,
1563                                             struct mlx5_flow_act *flow_act,
1564                                             struct mlx5_flow_destination *dest,
1565                                             int dest_num,
1566                                             struct fs_fte *fte)
1567 {
1568         struct mlx5_flow_handle *handle;
1569         int old_action;
1570         int i;
1571         int ret;
1572
1573         ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1574         if (ret)
1575                 return ERR_PTR(ret);
1576
1577         old_action = fte->action.action;
1578         fte->action.action |= flow_act->action;
1579         handle = add_rule_fte(fte, fg, dest, dest_num,
1580                               old_action != flow_act->action);
1581         if (IS_ERR(handle)) {
1582                 fte->action.action = old_action;
1583                 return handle;
1584         }
1585         trace_mlx5_fs_set_fte(fte, false);
1586
1587         for (i = 0; i < handle->num_rules; i++) {
1588                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1589                         tree_add_node(&handle->rule[i]->node, &fte->node);
1590                         trace_mlx5_fs_add_rule(handle->rule[i]);
1591                 }
1592         }
1593         return handle;
1594 }
1595
1596 static bool counter_is_valid(u32 action)
1597 {
1598         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1599                           MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1600                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1601 }
1602
1603 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1604                           struct mlx5_flow_act *flow_act,
1605                           struct mlx5_flow_table *ft)
1606 {
1607         bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1608         u32 action = flow_act->action;
1609
1610         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1611                 return counter_is_valid(action);
1612
1613         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1614                 return true;
1615
1616         if (ignore_level) {
1617                 if (ft->type != FS_FT_FDB &&
1618                     ft->type != FS_FT_NIC_RX)
1619                         return false;
1620
1621                 if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1622                     ft->type != dest->ft->type)
1623                         return false;
1624         }
1625
1626         if (!dest || ((dest->type ==
1627             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1628             (dest->ft->level <= ft->level && !ignore_level)))
1629                 return false;
1630         return true;
1631 }
1632
1633 struct match_list {
1634         struct list_head        list;
1635         struct mlx5_flow_group *g;
1636 };
1637
1638 static void free_match_list(struct match_list *head, bool ft_locked)
1639 {
1640         struct match_list *iter, *match_tmp;
1641
1642         list_for_each_entry_safe(iter, match_tmp, &head->list,
1643                                  list) {
1644                 tree_put_node(&iter->g->node, ft_locked);
1645                 list_del(&iter->list);
1646                 kfree(iter);
1647         }
1648 }
1649
1650 static int build_match_list(struct match_list *match_head,
1651                             struct mlx5_flow_table *ft,
1652                             const struct mlx5_flow_spec *spec,
1653                             bool ft_locked)
1654 {
1655         struct rhlist_head *tmp, *list;
1656         struct mlx5_flow_group *g;
1657         int err = 0;
1658
1659         rcu_read_lock();
1660         INIT_LIST_HEAD(&match_head->list);
1661         /* Collect all fgs which has a matching match_criteria */
1662         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1663         /* RCU is atomic, we can't execute FW commands here */
1664         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1665                 struct match_list *curr_match;
1666
1667                 if (unlikely(!tree_get_node(&g->node)))
1668                         continue;
1669
1670                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1671                 if (!curr_match) {
1672                         free_match_list(match_head, ft_locked);
1673                         err = -ENOMEM;
1674                         goto out;
1675                 }
1676                 curr_match->g = g;
1677                 list_add_tail(&curr_match->list, &match_head->list);
1678         }
1679 out:
1680         rcu_read_unlock();
1681         return err;
1682 }
1683
1684 static u64 matched_fgs_get_version(struct list_head *match_head)
1685 {
1686         struct match_list *iter;
1687         u64 version = 0;
1688
1689         list_for_each_entry(iter, match_head, list)
1690                 version += (u64)atomic_read(&iter->g->node.version);
1691         return version;
1692 }
1693
1694 static struct fs_fte *
1695 lookup_fte_locked(struct mlx5_flow_group *g,
1696                   const u32 *match_value,
1697                   bool take_write)
1698 {
1699         struct fs_fte *fte_tmp;
1700
1701         if (take_write)
1702                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1703         else
1704                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1705         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1706                                          rhash_fte);
1707         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1708                 fte_tmp = NULL;
1709                 goto out;
1710         }
1711         if (!fte_tmp->node.active) {
1712                 tree_put_node(&fte_tmp->node, false);
1713                 fte_tmp = NULL;
1714                 goto out;
1715         }
1716
1717         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1718 out:
1719         if (take_write)
1720                 up_write_ref_node(&g->node, false);
1721         else
1722                 up_read_ref_node(&g->node);
1723         return fte_tmp;
1724 }
1725
1726 static struct mlx5_flow_handle *
1727 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1728                        struct list_head *match_head,
1729                        const struct mlx5_flow_spec *spec,
1730                        struct mlx5_flow_act *flow_act,
1731                        struct mlx5_flow_destination *dest,
1732                        int dest_num,
1733                        int ft_version)
1734 {
1735         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1736         struct mlx5_flow_group *g;
1737         struct mlx5_flow_handle *rule;
1738         struct match_list *iter;
1739         bool take_write = false;
1740         struct fs_fte *fte;
1741         u64  version = 0;
1742         int err;
1743
1744         fte = alloc_fte(ft, spec, flow_act);
1745         if (IS_ERR(fte))
1746                 return  ERR_PTR(-ENOMEM);
1747
1748 search_again_locked:
1749         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1750                 goto skip_search;
1751         version = matched_fgs_get_version(match_head);
1752         /* Try to find an fte with identical match value and attempt update its
1753          * action.
1754          */
1755         list_for_each_entry(iter, match_head, list) {
1756                 struct fs_fte *fte_tmp;
1757
1758                 g = iter->g;
1759                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1760                 if (!fte_tmp)
1761                         continue;
1762                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1763                 up_write_ref_node(&fte_tmp->node, false);
1764                 tree_put_node(&fte_tmp->node, false);
1765                 kmem_cache_free(steering->ftes_cache, fte);
1766                 return rule;
1767         }
1768
1769 skip_search:
1770         /* No group with matching fte found, or we skipped the search.
1771          * Try to add a new fte to any matching fg.
1772          */
1773
1774         /* Check the ft version, for case that new flow group
1775          * was added while the fgs weren't locked
1776          */
1777         if (atomic_read(&ft->node.version) != ft_version) {
1778                 rule = ERR_PTR(-EAGAIN);
1779                 goto out;
1780         }
1781
1782         /* Check the fgs version. If version have changed it could be that an
1783          * FTE with the same match value was added while the fgs weren't
1784          * locked.
1785          */
1786         if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1787             version != matched_fgs_get_version(match_head)) {
1788                 take_write = true;
1789                 goto search_again_locked;
1790         }
1791
1792         list_for_each_entry(iter, match_head, list) {
1793                 g = iter->g;
1794
1795                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1796
1797                 if (!g->node.active) {
1798                         up_write_ref_node(&g->node, false);
1799                         continue;
1800                 }
1801
1802                 err = insert_fte(g, fte);
1803                 if (err) {
1804                         up_write_ref_node(&g->node, false);
1805                         if (err == -ENOSPC)
1806                                 continue;
1807                         kmem_cache_free(steering->ftes_cache, fte);
1808                         return ERR_PTR(err);
1809                 }
1810
1811                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1812                 up_write_ref_node(&g->node, false);
1813                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1814                 up_write_ref_node(&fte->node, false);
1815                 return rule;
1816         }
1817         rule = ERR_PTR(-ENOENT);
1818 out:
1819         kmem_cache_free(steering->ftes_cache, fte);
1820         return rule;
1821 }
1822
1823 static struct mlx5_flow_handle *
1824 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1825                      const struct mlx5_flow_spec *spec,
1826                      struct mlx5_flow_act *flow_act,
1827                      struct mlx5_flow_destination *dest,
1828                      int dest_num)
1829
1830 {
1831         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1832         struct mlx5_flow_handle *rule;
1833         struct match_list match_head;
1834         struct mlx5_flow_group *g;
1835         bool take_write = false;
1836         struct fs_fte *fte;
1837         int version;
1838         int err;
1839         int i;
1840
1841         if (!check_valid_spec(spec))
1842                 return ERR_PTR(-EINVAL);
1843
1844         for (i = 0; i < dest_num; i++) {
1845                 if (!dest_is_valid(&dest[i], flow_act, ft))
1846                         return ERR_PTR(-EINVAL);
1847         }
1848         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1849 search_again_locked:
1850         version = atomic_read(&ft->node.version);
1851
1852         /* Collect all fgs which has a matching match_criteria */
1853         err = build_match_list(&match_head, ft, spec, take_write);
1854         if (err) {
1855                 if (take_write)
1856                         up_write_ref_node(&ft->node, false);
1857                 else
1858                         up_read_ref_node(&ft->node);
1859                 return ERR_PTR(err);
1860         }
1861
1862         if (!take_write)
1863                 up_read_ref_node(&ft->node);
1864
1865         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1866                                       dest_num, version);
1867         free_match_list(&match_head, take_write);
1868         if (!IS_ERR(rule) ||
1869             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1870                 if (take_write)
1871                         up_write_ref_node(&ft->node, false);
1872                 return rule;
1873         }
1874
1875         if (!take_write) {
1876                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1877                 take_write = true;
1878         }
1879
1880         if (PTR_ERR(rule) == -EAGAIN ||
1881             version != atomic_read(&ft->node.version))
1882                 goto search_again_locked;
1883
1884         g = alloc_auto_flow_group(ft, spec);
1885         if (IS_ERR(g)) {
1886                 rule = ERR_CAST(g);
1887                 up_write_ref_node(&ft->node, false);
1888                 return rule;
1889         }
1890
1891         fte = alloc_fte(ft, spec, flow_act);
1892         if (IS_ERR(fte)) {
1893                 up_write_ref_node(&ft->node, false);
1894                 err = PTR_ERR(fte);
1895                 goto err_alloc_fte;
1896         }
1897
1898         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1899         up_write_ref_node(&ft->node, false);
1900
1901         err = create_auto_flow_group(ft, g);
1902         if (err)
1903                 goto err_release_fg;
1904
1905         err = insert_fte(g, fte);
1906         if (err)
1907                 goto err_release_fg;
1908
1909         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1910         up_write_ref_node(&g->node, false);
1911         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1912         up_write_ref_node(&fte->node, false);
1913         tree_put_node(&g->node, false);
1914         return rule;
1915
1916 err_release_fg:
1917         up_write_ref_node(&g->node, false);
1918         kmem_cache_free(steering->ftes_cache, fte);
1919 err_alloc_fte:
1920         tree_put_node(&g->node, false);
1921         return ERR_PTR(err);
1922 }
1923
1924 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1925 {
1926         return ((ft->type == FS_FT_NIC_RX) &&
1927                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1928 }
1929
1930 struct mlx5_flow_handle *
1931 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1932                     const struct mlx5_flow_spec *spec,
1933                     struct mlx5_flow_act *flow_act,
1934                     struct mlx5_flow_destination *dest,
1935                     int num_dest)
1936 {
1937         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1938         static const struct mlx5_flow_spec zero_spec = {};
1939         struct mlx5_flow_destination *gen_dest = NULL;
1940         struct mlx5_flow_table *next_ft = NULL;
1941         struct mlx5_flow_handle *handle = NULL;
1942         u32 sw_action = flow_act->action;
1943         int i;
1944
1945         if (!spec)
1946                 spec = &zero_spec;
1947
1948         if (!is_fwd_next_action(sw_action))
1949                 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1950
1951         if (!fwd_next_prio_supported(ft))
1952                 return ERR_PTR(-EOPNOTSUPP);
1953
1954         mutex_lock(&root->chain_lock);
1955         next_ft = find_next_fwd_ft(ft, flow_act);
1956         if (!next_ft) {
1957                 handle = ERR_PTR(-EOPNOTSUPP);
1958                 goto unlock;
1959         }
1960
1961         gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
1962                            GFP_KERNEL);
1963         if (!gen_dest) {
1964                 handle = ERR_PTR(-ENOMEM);
1965                 goto unlock;
1966         }
1967         for (i = 0; i < num_dest; i++)
1968                 gen_dest[i] = dest[i];
1969         gen_dest[i].type =
1970                 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1971         gen_dest[i].ft = next_ft;
1972         dest = gen_dest;
1973         num_dest++;
1974         flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
1975                               MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
1976         flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1977         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1978         if (IS_ERR(handle))
1979                 goto unlock;
1980
1981         if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
1982                 mutex_lock(&next_ft->lock);
1983                 list_add(&handle->rule[num_dest - 1]->next_ft,
1984                          &next_ft->fwd_rules);
1985                 mutex_unlock(&next_ft->lock);
1986                 handle->rule[num_dest - 1]->sw_action = sw_action;
1987                 handle->rule[num_dest - 1]->ft = ft;
1988         }
1989 unlock:
1990         mutex_unlock(&root->chain_lock);
1991         kfree(gen_dest);
1992         return handle;
1993 }
1994 EXPORT_SYMBOL(mlx5_add_flow_rules);
1995
1996 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1997 {
1998         struct fs_fte *fte;
1999         int i;
2000
2001         /* In order to consolidate the HW changes we lock the FTE for other
2002          * changes, and increase its refcount, in order not to perform the
2003          * "del" functions of the FTE. Will handle them here.
2004          * The removal of the rules is done under locked FTE.
2005          * After removing all the handle's rules, if there are remaining
2006          * rules, it means we just need to modify the FTE in FW, and
2007          * unlock/decrease the refcount we increased before.
2008          * Otherwise, it means the FTE should be deleted. First delete the
2009          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2010          * the FTE, which will handle the last decrease of the refcount, as
2011          * well as required handling of its parent.
2012          */
2013         fs_get_obj(fte, handle->rule[0]->node.parent);
2014         down_write_ref_node(&fte->node, false);
2015         for (i = handle->num_rules - 1; i >= 0; i--)
2016                 tree_remove_node(&handle->rule[i]->node, true);
2017         if (fte->dests_size) {
2018                 if (fte->modify_mask)
2019                         modify_fte(fte);
2020                 up_write_ref_node(&fte->node, false);
2021         } else if (list_empty(&fte->node.children)) {
2022                 del_hw_fte(&fte->node);
2023                 /* Avoid double call to del_hw_fte */
2024                 fte->node.del_hw_func = NULL;
2025                 up_write_ref_node(&fte->node, false);
2026                 tree_put_node(&fte->node, false);
2027         }
2028         kfree(handle);
2029 }
2030 EXPORT_SYMBOL(mlx5_del_flow_rules);
2031
2032 /* Assuming prio->node.children(flow tables) is sorted by level */
2033 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2034 {
2035         struct fs_prio *prio;
2036
2037         fs_get_obj(prio, ft->node.parent);
2038
2039         if (!list_is_last(&ft->node.list, &prio->node.children))
2040                 return list_next_entry(ft, node.list);
2041         return find_next_chained_ft(prio);
2042 }
2043
2044 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2045 {
2046         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2047         struct mlx5_ft_underlay_qp *uqp;
2048         struct mlx5_flow_table *new_root_ft = NULL;
2049         int err = 0;
2050         u32 qpn;
2051
2052         if (root->root_ft != ft)
2053                 return 0;
2054
2055         new_root_ft = find_next_ft(ft);
2056         if (!new_root_ft) {
2057                 root->root_ft = NULL;
2058                 return 0;
2059         }
2060
2061         if (list_empty(&root->underlay_qpns)) {
2062                 /* Don't set any QPN (zero) in case QPN list is empty */
2063                 qpn = 0;
2064                 err = root->cmds->update_root_ft(root, new_root_ft,
2065                                                  qpn, false);
2066         } else {
2067                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2068                         qpn = uqp->qpn;
2069                         err = root->cmds->update_root_ft(root,
2070                                                          new_root_ft, qpn,
2071                                                          false);
2072                         if (err)
2073                                 break;
2074                 }
2075         }
2076
2077         if (err)
2078                 mlx5_core_warn(root->dev,
2079                                "Update root flow table of id(%u) qpn(%d) failed\n",
2080                                ft->id, qpn);
2081         else
2082                 root->root_ft = new_root_ft;
2083
2084         return 0;
2085 }
2086
2087 /* Connect flow table from previous priority to
2088  * the next flow table.
2089  */
2090 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2091 {
2092         struct mlx5_core_dev *dev = get_dev(&ft->node);
2093         struct mlx5_flow_table *next_ft;
2094         struct fs_prio *prio;
2095         int err = 0;
2096
2097         err = update_root_ft_destroy(ft);
2098         if (err)
2099                 return err;
2100
2101         fs_get_obj(prio, ft->node.parent);
2102         if  (!(list_first_entry(&prio->node.children,
2103                                 struct mlx5_flow_table,
2104                                 node.list) == ft))
2105                 return 0;
2106
2107         next_ft = find_next_chained_ft(prio);
2108         err = connect_fwd_rules(dev, next_ft, ft);
2109         if (err)
2110                 return err;
2111
2112         err = connect_prev_fts(dev, next_ft, prio);
2113         if (err)
2114                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2115                                ft->id);
2116         return err;
2117 }
2118
2119 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2120 {
2121         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2122         int err = 0;
2123
2124         mutex_lock(&root->chain_lock);
2125         if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2126                 err = disconnect_flow_table(ft);
2127         if (err) {
2128                 mutex_unlock(&root->chain_lock);
2129                 return err;
2130         }
2131         if (tree_remove_node(&ft->node, false))
2132                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2133                                ft->id);
2134         mutex_unlock(&root->chain_lock);
2135
2136         return err;
2137 }
2138 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2139
2140 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2141 {
2142         if (tree_remove_node(&fg->node, false))
2143                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2144                                fg->id);
2145 }
2146 EXPORT_SYMBOL(mlx5_destroy_flow_group);
2147
2148 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2149                                                 int n)
2150 {
2151         struct mlx5_flow_steering *steering = dev->priv.steering;
2152
2153         if (!steering || !steering->fdb_sub_ns)
2154                 return NULL;
2155
2156         return steering->fdb_sub_ns[n];
2157 }
2158 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2159
2160 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2161                                                     enum mlx5_flow_namespace_type type)
2162 {
2163         struct mlx5_flow_steering *steering = dev->priv.steering;
2164         struct mlx5_flow_root_namespace *root_ns;
2165         int prio = 0;
2166         struct fs_prio *fs_prio;
2167         struct mlx5_flow_namespace *ns;
2168
2169         if (!steering)
2170                 return NULL;
2171
2172         switch (type) {
2173         case MLX5_FLOW_NAMESPACE_FDB:
2174                 if (steering->fdb_root_ns)
2175                         return &steering->fdb_root_ns->ns;
2176                 return NULL;
2177         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2178                 if (steering->sniffer_rx_root_ns)
2179                         return &steering->sniffer_rx_root_ns->ns;
2180                 return NULL;
2181         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2182                 if (steering->sniffer_tx_root_ns)
2183                         return &steering->sniffer_tx_root_ns->ns;
2184                 return NULL;
2185         default:
2186                 break;
2187         }
2188
2189         if (type == MLX5_FLOW_NAMESPACE_EGRESS ||
2190             type == MLX5_FLOW_NAMESPACE_EGRESS_KERNEL) {
2191                 root_ns = steering->egress_root_ns;
2192                 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2193         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX) {
2194                 root_ns = steering->rdma_rx_root_ns;
2195                 prio = RDMA_RX_BYPASS_PRIO;
2196         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL) {
2197                 root_ns = steering->rdma_rx_root_ns;
2198                 prio = RDMA_RX_KERNEL_PRIO;
2199         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_TX) {
2200                 root_ns = steering->rdma_tx_root_ns;
2201         } else { /* Must be NIC RX */
2202                 root_ns = steering->root_ns;
2203                 prio = type;
2204         }
2205
2206         if (!root_ns)
2207                 return NULL;
2208
2209         fs_prio = find_prio(&root_ns->ns, prio);
2210         if (!fs_prio)
2211                 return NULL;
2212
2213         ns = list_first_entry(&fs_prio->node.children,
2214                               typeof(*ns),
2215                               node.list);
2216
2217         return ns;
2218 }
2219 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2220
2221 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2222                                                               enum mlx5_flow_namespace_type type,
2223                                                               int vport)
2224 {
2225         struct mlx5_flow_steering *steering = dev->priv.steering;
2226
2227         if (!steering || vport >= mlx5_eswitch_get_total_vports(dev))
2228                 return NULL;
2229
2230         switch (type) {
2231         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2232                 if (steering->esw_egress_root_ns &&
2233                     steering->esw_egress_root_ns[vport])
2234                         return &steering->esw_egress_root_ns[vport]->ns;
2235                 else
2236                         return NULL;
2237         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2238                 if (steering->esw_ingress_root_ns &&
2239                     steering->esw_ingress_root_ns[vport])
2240                         return &steering->esw_ingress_root_ns[vport]->ns;
2241                 else
2242                         return NULL;
2243         default:
2244                 return NULL;
2245         }
2246 }
2247
2248 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2249                                        unsigned int prio,
2250                                        int num_levels,
2251                                        enum fs_node_type type)
2252 {
2253         struct fs_prio *fs_prio;
2254
2255         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2256         if (!fs_prio)
2257                 return ERR_PTR(-ENOMEM);
2258
2259         fs_prio->node.type = type;
2260         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2261         tree_add_node(&fs_prio->node, &ns->node);
2262         fs_prio->num_levels = num_levels;
2263         fs_prio->prio = prio;
2264         list_add_tail(&fs_prio->node.list, &ns->node.children);
2265
2266         return fs_prio;
2267 }
2268
2269 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2270                                               unsigned int prio,
2271                                               int num_levels)
2272 {
2273         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2274 }
2275
2276 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2277                                       unsigned int prio, int num_levels)
2278 {
2279         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2280 }
2281
2282 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2283                                                      *ns)
2284 {
2285         ns->node.type = FS_TYPE_NAMESPACE;
2286
2287         return ns;
2288 }
2289
2290 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2291                                                        int def_miss_act)
2292 {
2293         struct mlx5_flow_namespace      *ns;
2294
2295         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2296         if (!ns)
2297                 return ERR_PTR(-ENOMEM);
2298
2299         fs_init_namespace(ns);
2300         ns->def_miss_action = def_miss_act;
2301         tree_init_node(&ns->node, NULL, del_sw_ns);
2302         tree_add_node(&ns->node, &prio->node);
2303         list_add_tail(&ns->node.list, &prio->node.children);
2304
2305         return ns;
2306 }
2307
2308 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2309                              struct init_tree_node *prio_metadata)
2310 {
2311         struct fs_prio *fs_prio;
2312         int i;
2313
2314         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2315                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2316                 if (IS_ERR(fs_prio))
2317                         return PTR_ERR(fs_prio);
2318         }
2319         return 0;
2320 }
2321
2322 #define FLOW_TABLE_BIT_SZ 1
2323 #define GET_FLOW_TABLE_CAP(dev, offset) \
2324         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2325                         offset / 32)) >>                                        \
2326           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2327 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2328 {
2329         int i;
2330
2331         for (i = 0; i < caps->arr_sz; i++) {
2332                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2333                         return false;
2334         }
2335         return true;
2336 }
2337
2338 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2339                                     struct init_tree_node *init_node,
2340                                     struct fs_node *fs_parent_node,
2341                                     struct init_tree_node *init_parent_node,
2342                                     int prio)
2343 {
2344         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2345                                               flow_table_properties_nic_receive.
2346                                               max_ft_level);
2347         struct mlx5_flow_namespace *fs_ns;
2348         struct fs_prio *fs_prio;
2349         struct fs_node *base;
2350         int i;
2351         int err;
2352
2353         if (init_node->type == FS_TYPE_PRIO) {
2354                 if ((init_node->min_ft_level > max_ft_level) ||
2355                     !has_required_caps(steering->dev, &init_node->caps))
2356                         return 0;
2357
2358                 fs_get_obj(fs_ns, fs_parent_node);
2359                 if (init_node->num_leaf_prios)
2360                         return create_leaf_prios(fs_ns, prio, init_node);
2361                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2362                 if (IS_ERR(fs_prio))
2363                         return PTR_ERR(fs_prio);
2364                 base = &fs_prio->node;
2365         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2366                 fs_get_obj(fs_prio, fs_parent_node);
2367                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2368                 if (IS_ERR(fs_ns))
2369                         return PTR_ERR(fs_ns);
2370                 base = &fs_ns->node;
2371         } else {
2372                 return -EINVAL;
2373         }
2374         prio = 0;
2375         for (i = 0; i < init_node->ar_size; i++) {
2376                 err = init_root_tree_recursive(steering, &init_node->children[i],
2377                                                base, init_node, prio);
2378                 if (err)
2379                         return err;
2380                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2381                     init_node->children[i].num_leaf_prios) {
2382                         prio += init_node->children[i].num_leaf_prios;
2383                 }
2384         }
2385
2386         return 0;
2387 }
2388
2389 static int init_root_tree(struct mlx5_flow_steering *steering,
2390                           struct init_tree_node *init_node,
2391                           struct fs_node *fs_parent_node)
2392 {
2393         int i;
2394         struct mlx5_flow_namespace *fs_ns;
2395         int err;
2396
2397         fs_get_obj(fs_ns, fs_parent_node);
2398         for (i = 0; i < init_node->ar_size; i++) {
2399                 err = init_root_tree_recursive(steering, &init_node->children[i],
2400                                                &fs_ns->node,
2401                                                init_node, i);
2402                 if (err)
2403                         return err;
2404         }
2405         return 0;
2406 }
2407
2408 static void del_sw_root_ns(struct fs_node *node)
2409 {
2410         struct mlx5_flow_root_namespace *root_ns;
2411         struct mlx5_flow_namespace *ns;
2412
2413         fs_get_obj(ns, node);
2414         root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2415         mutex_destroy(&root_ns->chain_lock);
2416         kfree(node);
2417 }
2418
2419 static struct mlx5_flow_root_namespace
2420 *create_root_ns(struct mlx5_flow_steering *steering,
2421                 enum fs_flow_table_type table_type)
2422 {
2423         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2424         struct mlx5_flow_root_namespace *root_ns;
2425         struct mlx5_flow_namespace *ns;
2426
2427         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2428             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2429                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2430
2431         /* Create the root namespace */
2432         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2433         if (!root_ns)
2434                 return NULL;
2435
2436         root_ns->dev = steering->dev;
2437         root_ns->table_type = table_type;
2438         root_ns->cmds = cmds;
2439
2440         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2441
2442         ns = &root_ns->ns;
2443         fs_init_namespace(ns);
2444         mutex_init(&root_ns->chain_lock);
2445         tree_init_node(&ns->node, NULL, del_sw_root_ns);
2446         tree_add_node(&ns->node, NULL);
2447
2448         return root_ns;
2449 }
2450
2451 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2452
2453 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2454 {
2455         struct fs_prio *prio;
2456
2457         fs_for_each_prio(prio, ns) {
2458                  /* This updates prio start_level and num_levels */
2459                 set_prio_attrs_in_prio(prio, acc_level);
2460                 acc_level += prio->num_levels;
2461         }
2462         return acc_level;
2463 }
2464
2465 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2466 {
2467         struct mlx5_flow_namespace *ns;
2468         int acc_level_ns = acc_level;
2469
2470         prio->start_level = acc_level;
2471         fs_for_each_ns(ns, prio) {
2472                 /* This updates start_level and num_levels of ns's priority descendants */
2473                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2474
2475                 /* If this a prio with chains, and we can jump from one chain
2476                  * (namepsace) to another, so we accumulate the levels
2477                  */
2478                 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2479                         acc_level = acc_level_ns;
2480         }
2481
2482         if (!prio->num_levels)
2483                 prio->num_levels = acc_level_ns - prio->start_level;
2484         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2485 }
2486
2487 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2488 {
2489         struct mlx5_flow_namespace *ns = &root_ns->ns;
2490         struct fs_prio *prio;
2491         int start_level = 0;
2492
2493         fs_for_each_prio(prio, ns) {
2494                 set_prio_attrs_in_prio(prio, start_level);
2495                 start_level += prio->num_levels;
2496         }
2497 }
2498
2499 #define ANCHOR_PRIO 0
2500 #define ANCHOR_SIZE 1
2501 #define ANCHOR_LEVEL 0
2502 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2503 {
2504         struct mlx5_flow_namespace *ns = NULL;
2505         struct mlx5_flow_table_attr ft_attr = {};
2506         struct mlx5_flow_table *ft;
2507
2508         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2509         if (WARN_ON(!ns))
2510                 return -EINVAL;
2511
2512         ft_attr.max_fte = ANCHOR_SIZE;
2513         ft_attr.level   = ANCHOR_LEVEL;
2514         ft_attr.prio    = ANCHOR_PRIO;
2515
2516         ft = mlx5_create_flow_table(ns, &ft_attr);
2517         if (IS_ERR(ft)) {
2518                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2519                 return PTR_ERR(ft);
2520         }
2521         return 0;
2522 }
2523
2524 static int init_root_ns(struct mlx5_flow_steering *steering)
2525 {
2526         int err;
2527
2528         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2529         if (!steering->root_ns)
2530                 return -ENOMEM;
2531
2532         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2533         if (err)
2534                 goto out_err;
2535
2536         set_prio_attrs(steering->root_ns);
2537         err = create_anchor_flow_table(steering);
2538         if (err)
2539                 goto out_err;
2540
2541         return 0;
2542
2543 out_err:
2544         cleanup_root_ns(steering->root_ns);
2545         steering->root_ns = NULL;
2546         return err;
2547 }
2548
2549 static void clean_tree(struct fs_node *node)
2550 {
2551         if (node) {
2552                 struct fs_node *iter;
2553                 struct fs_node *temp;
2554
2555                 tree_get_node(node);
2556                 list_for_each_entry_safe(iter, temp, &node->children, list)
2557                         clean_tree(iter);
2558                 tree_put_node(node, false);
2559                 tree_remove_node(node, false);
2560         }
2561 }
2562
2563 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2564 {
2565         if (!root_ns)
2566                 return;
2567
2568         clean_tree(&root_ns->ns.node);
2569 }
2570
2571 static void cleanup_egress_acls_root_ns(struct mlx5_core_dev *dev)
2572 {
2573         struct mlx5_flow_steering *steering = dev->priv.steering;
2574         int i;
2575
2576         if (!steering->esw_egress_root_ns)
2577                 return;
2578
2579         for (i = 0; i < mlx5_eswitch_get_total_vports(dev); i++)
2580                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2581
2582         kfree(steering->esw_egress_root_ns);
2583         steering->esw_egress_root_ns = NULL;
2584 }
2585
2586 static void cleanup_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2587 {
2588         struct mlx5_flow_steering *steering = dev->priv.steering;
2589         int i;
2590
2591         if (!steering->esw_ingress_root_ns)
2592                 return;
2593
2594         for (i = 0; i < mlx5_eswitch_get_total_vports(dev); i++)
2595                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2596
2597         kfree(steering->esw_ingress_root_ns);
2598         steering->esw_ingress_root_ns = NULL;
2599 }
2600
2601 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2602 {
2603         struct mlx5_flow_steering *steering = dev->priv.steering;
2604
2605         cleanup_root_ns(steering->root_ns);
2606         cleanup_egress_acls_root_ns(dev);
2607         cleanup_ingress_acls_root_ns(dev);
2608         cleanup_root_ns(steering->fdb_root_ns);
2609         steering->fdb_root_ns = NULL;
2610         kfree(steering->fdb_sub_ns);
2611         steering->fdb_sub_ns = NULL;
2612         cleanup_root_ns(steering->sniffer_rx_root_ns);
2613         cleanup_root_ns(steering->sniffer_tx_root_ns);
2614         cleanup_root_ns(steering->rdma_rx_root_ns);
2615         cleanup_root_ns(steering->rdma_tx_root_ns);
2616         cleanup_root_ns(steering->egress_root_ns);
2617         mlx5_cleanup_fc_stats(dev);
2618         kmem_cache_destroy(steering->ftes_cache);
2619         kmem_cache_destroy(steering->fgs_cache);
2620         kfree(steering);
2621 }
2622
2623 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2624 {
2625         struct fs_prio *prio;
2626
2627         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2628         if (!steering->sniffer_tx_root_ns)
2629                 return -ENOMEM;
2630
2631         /* Create single prio */
2632         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2633         return PTR_ERR_OR_ZERO(prio);
2634 }
2635
2636 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2637 {
2638         struct fs_prio *prio;
2639
2640         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2641         if (!steering->sniffer_rx_root_ns)
2642                 return -ENOMEM;
2643
2644         /* Create single prio */
2645         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2646         return PTR_ERR_OR_ZERO(prio);
2647 }
2648
2649 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2650 {
2651         int err;
2652
2653         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2654         if (!steering->rdma_rx_root_ns)
2655                 return -ENOMEM;
2656
2657         err = init_root_tree(steering, &rdma_rx_root_fs,
2658                              &steering->rdma_rx_root_ns->ns.node);
2659         if (err)
2660                 goto out_err;
2661
2662         set_prio_attrs(steering->rdma_rx_root_ns);
2663
2664         return 0;
2665
2666 out_err:
2667         cleanup_root_ns(steering->rdma_rx_root_ns);
2668         steering->rdma_rx_root_ns = NULL;
2669         return err;
2670 }
2671
2672 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2673 {
2674         int err;
2675
2676         steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2677         if (!steering->rdma_tx_root_ns)
2678                 return -ENOMEM;
2679
2680         err = init_root_tree(steering, &rdma_tx_root_fs,
2681                              &steering->rdma_tx_root_ns->ns.node);
2682         if (err)
2683                 goto out_err;
2684
2685         set_prio_attrs(steering->rdma_tx_root_ns);
2686
2687         return 0;
2688
2689 out_err:
2690         cleanup_root_ns(steering->rdma_tx_root_ns);
2691         steering->rdma_tx_root_ns = NULL;
2692         return err;
2693 }
2694
2695 /* FT and tc chains are stored in the same array so we can re-use the
2696  * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2697  * When creating a new ns for each chain store it in the first available slot.
2698  * Assume tc chains are created and stored first and only then the FT chain.
2699  */
2700 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2701                                         struct mlx5_flow_namespace *ns)
2702 {
2703         int chain = 0;
2704
2705         while (steering->fdb_sub_ns[chain])
2706                 ++chain;
2707
2708         steering->fdb_sub_ns[chain] = ns;
2709 }
2710
2711 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2712                                         struct fs_prio *maj_prio)
2713 {
2714         struct mlx5_flow_namespace *ns;
2715         struct fs_prio *min_prio;
2716         int prio;
2717
2718         ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2719         if (IS_ERR(ns))
2720                 return PTR_ERR(ns);
2721
2722         for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2723                 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2724                 if (IS_ERR(min_prio))
2725                         return PTR_ERR(min_prio);
2726         }
2727
2728         store_fdb_sub_ns_prio_chain(steering, ns);
2729
2730         return 0;
2731 }
2732
2733 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2734                              int fs_prio,
2735                              int chains)
2736 {
2737         struct fs_prio *maj_prio;
2738         int levels;
2739         int chain;
2740         int err;
2741
2742         levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2743         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2744                                           fs_prio,
2745                                           levels);
2746         if (IS_ERR(maj_prio))
2747                 return PTR_ERR(maj_prio);
2748
2749         for (chain = 0; chain < chains; chain++) {
2750                 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2751                 if (err)
2752                         return err;
2753         }
2754
2755         return 0;
2756 }
2757
2758 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2759 {
2760         int err;
2761
2762         steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2763                                        sizeof(*steering->fdb_sub_ns),
2764                                        GFP_KERNEL);
2765         if (!steering->fdb_sub_ns)
2766                 return -ENOMEM;
2767
2768         err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
2769         if (err)
2770                 return err;
2771
2772         err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
2773         if (err)
2774                 return err;
2775
2776         return 0;
2777 }
2778
2779 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2780 {
2781         struct fs_prio *maj_prio;
2782         int err;
2783
2784         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2785         if (!steering->fdb_root_ns)
2786                 return -ENOMEM;
2787
2788         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2789                                   1);
2790         if (IS_ERR(maj_prio)) {
2791                 err = PTR_ERR(maj_prio);
2792                 goto out_err;
2793         }
2794         err = create_fdb_fast_path(steering);
2795         if (err)
2796                 goto out_err;
2797
2798         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2799         if (IS_ERR(maj_prio)) {
2800                 err = PTR_ERR(maj_prio);
2801                 goto out_err;
2802         }
2803
2804         /* We put this priority last, knowing that nothing will get here
2805          * unless explicitly forwarded to. This is possible because the
2806          * slow path tables have catch all rules and nothing gets passed
2807          * those tables.
2808          */
2809         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
2810         if (IS_ERR(maj_prio)) {
2811                 err = PTR_ERR(maj_prio);
2812                 goto out_err;
2813         }
2814
2815         set_prio_attrs(steering->fdb_root_ns);
2816         return 0;
2817
2818 out_err:
2819         cleanup_root_ns(steering->fdb_root_ns);
2820         kfree(steering->fdb_sub_ns);
2821         steering->fdb_sub_ns = NULL;
2822         steering->fdb_root_ns = NULL;
2823         return err;
2824 }
2825
2826 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2827 {
2828         struct fs_prio *prio;
2829
2830         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2831         if (!steering->esw_egress_root_ns[vport])
2832                 return -ENOMEM;
2833
2834         /* create 1 prio*/
2835         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2836         return PTR_ERR_OR_ZERO(prio);
2837 }
2838
2839 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2840 {
2841         struct fs_prio *prio;
2842
2843         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2844         if (!steering->esw_ingress_root_ns[vport])
2845                 return -ENOMEM;
2846
2847         /* create 1 prio*/
2848         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2849         return PTR_ERR_OR_ZERO(prio);
2850 }
2851
2852 static int init_egress_acls_root_ns(struct mlx5_core_dev *dev)
2853 {
2854         struct mlx5_flow_steering *steering = dev->priv.steering;
2855         int total_vports = mlx5_eswitch_get_total_vports(dev);
2856         int err;
2857         int i;
2858
2859         steering->esw_egress_root_ns =
2860                         kcalloc(total_vports,
2861                                 sizeof(*steering->esw_egress_root_ns),
2862                                 GFP_KERNEL);
2863         if (!steering->esw_egress_root_ns)
2864                 return -ENOMEM;
2865
2866         for (i = 0; i < total_vports; i++) {
2867                 err = init_egress_acl_root_ns(steering, i);
2868                 if (err)
2869                         goto cleanup_root_ns;
2870         }
2871
2872         return 0;
2873
2874 cleanup_root_ns:
2875         for (i--; i >= 0; i--)
2876                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2877         kfree(steering->esw_egress_root_ns);
2878         steering->esw_egress_root_ns = NULL;
2879         return err;
2880 }
2881
2882 static int init_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2883 {
2884         struct mlx5_flow_steering *steering = dev->priv.steering;
2885         int total_vports = mlx5_eswitch_get_total_vports(dev);
2886         int err;
2887         int i;
2888
2889         steering->esw_ingress_root_ns =
2890                         kcalloc(total_vports,
2891                                 sizeof(*steering->esw_ingress_root_ns),
2892                                 GFP_KERNEL);
2893         if (!steering->esw_ingress_root_ns)
2894                 return -ENOMEM;
2895
2896         for (i = 0; i < total_vports; i++) {
2897                 err = init_ingress_acl_root_ns(steering, i);
2898                 if (err)
2899                         goto cleanup_root_ns;
2900         }
2901
2902         return 0;
2903
2904 cleanup_root_ns:
2905         for (i--; i >= 0; i--)
2906                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2907         kfree(steering->esw_ingress_root_ns);
2908         steering->esw_ingress_root_ns = NULL;
2909         return err;
2910 }
2911
2912 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2913 {
2914         int err;
2915
2916         steering->egress_root_ns = create_root_ns(steering,
2917                                                   FS_FT_NIC_TX);
2918         if (!steering->egress_root_ns)
2919                 return -ENOMEM;
2920
2921         err = init_root_tree(steering, &egress_root_fs,
2922                              &steering->egress_root_ns->ns.node);
2923         if (err)
2924                 goto cleanup;
2925         set_prio_attrs(steering->egress_root_ns);
2926         return 0;
2927 cleanup:
2928         cleanup_root_ns(steering->egress_root_ns);
2929         steering->egress_root_ns = NULL;
2930         return err;
2931 }
2932
2933 int mlx5_init_fs(struct mlx5_core_dev *dev)
2934 {
2935         struct mlx5_flow_steering *steering;
2936         int err = 0;
2937
2938         err = mlx5_init_fc_stats(dev);
2939         if (err)
2940                 return err;
2941
2942         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2943         if (!steering)
2944                 return -ENOMEM;
2945         steering->dev = dev;
2946         dev->priv.steering = steering;
2947
2948         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2949                                                 sizeof(struct mlx5_flow_group), 0,
2950                                                 0, NULL);
2951         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2952                                                  0, NULL);
2953         if (!steering->ftes_cache || !steering->fgs_cache) {
2954                 err = -ENOMEM;
2955                 goto err;
2956         }
2957
2958         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2959               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2960              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2961               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2962             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2963                 err = init_root_ns(steering);
2964                 if (err)
2965                         goto err;
2966         }
2967
2968         if (MLX5_ESWITCH_MANAGER(dev)) {
2969                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2970                         err = init_fdb_root_ns(steering);
2971                         if (err)
2972                                 goto err;
2973                 }
2974                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2975                         err = init_egress_acls_root_ns(dev);
2976                         if (err)
2977                                 goto err;
2978                 }
2979                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2980                         err = init_ingress_acls_root_ns(dev);
2981                         if (err)
2982                                 goto err;
2983                 }
2984         }
2985
2986         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2987                 err = init_sniffer_rx_root_ns(steering);
2988                 if (err)
2989                         goto err;
2990         }
2991
2992         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2993                 err = init_sniffer_tx_root_ns(steering);
2994                 if (err)
2995                         goto err;
2996         }
2997
2998         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
2999             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3000                 err = init_rdma_rx_root_ns(steering);
3001                 if (err)
3002                         goto err;
3003         }
3004
3005         if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3006                 err = init_rdma_tx_root_ns(steering);
3007                 if (err)
3008                         goto err;
3009         }
3010
3011         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE ||
3012             MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3013                 err = init_egress_root_ns(steering);
3014                 if (err)
3015                         goto err;
3016         }
3017
3018         return 0;
3019 err:
3020         mlx5_cleanup_fs(dev);
3021         return err;
3022 }
3023
3024 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3025 {
3026         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3027         struct mlx5_ft_underlay_qp *new_uqp;
3028         int err = 0;
3029
3030         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3031         if (!new_uqp)
3032                 return -ENOMEM;
3033
3034         mutex_lock(&root->chain_lock);
3035
3036         if (!root->root_ft) {
3037                 err = -EINVAL;
3038                 goto update_ft_fail;
3039         }
3040
3041         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3042                                          false);
3043         if (err) {
3044                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3045                                underlay_qpn, err);
3046                 goto update_ft_fail;
3047         }
3048
3049         new_uqp->qpn = underlay_qpn;
3050         list_add_tail(&new_uqp->list, &root->underlay_qpns);
3051
3052         mutex_unlock(&root->chain_lock);
3053
3054         return 0;
3055
3056 update_ft_fail:
3057         mutex_unlock(&root->chain_lock);
3058         kfree(new_uqp);
3059         return err;
3060 }
3061 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3062
3063 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3064 {
3065         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3066         struct mlx5_ft_underlay_qp *uqp;
3067         bool found = false;
3068         int err = 0;
3069
3070         mutex_lock(&root->chain_lock);
3071         list_for_each_entry(uqp, &root->underlay_qpns, list) {
3072                 if (uqp->qpn == underlay_qpn) {
3073                         found = true;
3074                         break;
3075                 }
3076         }
3077
3078         if (!found) {
3079                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3080                                underlay_qpn);
3081                 err = -EINVAL;
3082                 goto out;
3083         }
3084
3085         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3086                                          true);
3087         if (err)
3088                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3089                                underlay_qpn, err);
3090
3091         list_del(&uqp->list);
3092         mutex_unlock(&root->chain_lock);
3093         kfree(uqp);
3094
3095         return 0;
3096
3097 out:
3098         mutex_unlock(&root->chain_lock);
3099         return err;
3100 }
3101 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3102
3103 static struct mlx5_flow_root_namespace
3104 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3105 {
3106         struct mlx5_flow_namespace *ns;
3107
3108         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3109             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3110                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3111         else
3112                 ns = mlx5_get_flow_namespace(dev, ns_type);
3113         if (!ns)
3114                 return NULL;
3115
3116         return find_root(&ns->node);
3117 }
3118
3119 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3120                                                  u8 ns_type, u8 num_actions,
3121                                                  void *modify_actions)
3122 {
3123         struct mlx5_flow_root_namespace *root;
3124         struct mlx5_modify_hdr *modify_hdr;
3125         int err;
3126
3127         root = get_root_namespace(dev, ns_type);
3128         if (!root)
3129                 return ERR_PTR(-EOPNOTSUPP);
3130
3131         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3132         if (!modify_hdr)
3133                 return ERR_PTR(-ENOMEM);
3134
3135         modify_hdr->ns_type = ns_type;
3136         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3137                                               modify_actions, modify_hdr);
3138         if (err) {
3139                 kfree(modify_hdr);
3140                 return ERR_PTR(err);
3141         }
3142
3143         return modify_hdr;
3144 }
3145 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3146
3147 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3148                                 struct mlx5_modify_hdr *modify_hdr)
3149 {
3150         struct mlx5_flow_root_namespace *root;
3151
3152         root = get_root_namespace(dev, modify_hdr->ns_type);
3153         if (WARN_ON(!root))
3154                 return;
3155         root->cmds->modify_header_dealloc(root, modify_hdr);
3156         kfree(modify_hdr);
3157 }
3158 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3159
3160 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3161                                                      int reformat_type,
3162                                                      size_t size,
3163                                                      void *reformat_data,
3164                                                      enum mlx5_flow_namespace_type ns_type)
3165 {
3166         struct mlx5_pkt_reformat *pkt_reformat;
3167         struct mlx5_flow_root_namespace *root;
3168         int err;
3169
3170         root = get_root_namespace(dev, ns_type);
3171         if (!root)
3172                 return ERR_PTR(-EOPNOTSUPP);
3173
3174         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3175         if (!pkt_reformat)
3176                 return ERR_PTR(-ENOMEM);
3177
3178         pkt_reformat->ns_type = ns_type;
3179         pkt_reformat->reformat_type = reformat_type;
3180         err = root->cmds->packet_reformat_alloc(root, reformat_type, size,
3181                                                 reformat_data, ns_type,
3182                                                 pkt_reformat);
3183         if (err) {
3184                 kfree(pkt_reformat);
3185                 return ERR_PTR(err);
3186         }
3187
3188         return pkt_reformat;
3189 }
3190 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3191
3192 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3193                                   struct mlx5_pkt_reformat *pkt_reformat)
3194 {
3195         struct mlx5_flow_root_namespace *root;
3196
3197         root = get_root_namespace(dev, pkt_reformat->ns_type);
3198         if (WARN_ON(!root))
3199                 return;
3200         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3201         kfree(pkt_reformat);
3202 }
3203 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3204
3205 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3206                                  struct mlx5_flow_root_namespace *peer_ns)
3207 {
3208         if (peer_ns && ns->mode != peer_ns->mode) {
3209                 mlx5_core_err(ns->dev,
3210                               "Can't peer namespace of different steering mode\n");
3211                 return -EINVAL;
3212         }
3213
3214         return ns->cmds->set_peer(ns, peer_ns);
3215 }
3216
3217 /* This function should be called only at init stage of the namespace.
3218  * It is not safe to call this function while steering operations
3219  * are executed in the namespace.
3220  */
3221 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3222                                  enum mlx5_flow_steering_mode mode)
3223 {
3224         struct mlx5_flow_root_namespace *root;
3225         const struct mlx5_flow_cmds *cmds;
3226         int err;
3227
3228         root = find_root(&ns->node);
3229         if (&root->ns != ns)
3230         /* Can't set cmds to non root namespace */
3231                 return -EINVAL;
3232
3233         if (root->table_type != FS_FT_FDB)
3234                 return -EOPNOTSUPP;
3235
3236         if (root->mode == mode)
3237                 return 0;
3238
3239         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3240                 cmds = mlx5_fs_cmd_get_dr_cmds();
3241         else
3242                 cmds = mlx5_fs_cmd_get_fw_cmds();
3243         if (!cmds)
3244                 return -EOPNOTSUPP;
3245
3246         err = cmds->create_ns(root);
3247         if (err) {
3248                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3249                               err);
3250                 return err;
3251         }
3252
3253         root->cmds->destroy_ns(root);
3254         root->cmds = cmds;
3255         root->mode = mode;
3256
3257         return 0;
3258 }