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, {UDP/ANY/aRFS/accel/{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 = false;
 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_free(&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_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, 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_free(&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                 /* No error check needed here, because insert_fte() is not called */
1764                 up_write_ref_node(&fte_tmp->node, false);
1765                 tree_put_node(&fte_tmp->node, false);
1766                 kmem_cache_free(steering->ftes_cache, fte);
1767                 return rule;
1768         }
1769
1770 skip_search:
1771         /* No group with matching fte found, or we skipped the search.
1772          * Try to add a new fte to any matching fg.
1773          */
1774
1775         /* Check the ft version, for case that new flow group
1776          * was added while the fgs weren't locked
1777          */
1778         if (atomic_read(&ft->node.version) != ft_version) {
1779                 rule = ERR_PTR(-EAGAIN);
1780                 goto out;
1781         }
1782
1783         /* Check the fgs version. If version have changed it could be that an
1784          * FTE with the same match value was added while the fgs weren't
1785          * locked.
1786          */
1787         if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1788             version != matched_fgs_get_version(match_head)) {
1789                 take_write = true;
1790                 goto search_again_locked;
1791         }
1792
1793         list_for_each_entry(iter, match_head, list) {
1794                 g = iter->g;
1795
1796                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1797
1798                 if (!g->node.active) {
1799                         up_write_ref_node(&g->node, false);
1800                         continue;
1801                 }
1802
1803                 err = insert_fte(g, fte);
1804                 if (err) {
1805                         up_write_ref_node(&g->node, false);
1806                         if (err == -ENOSPC)
1807                                 continue;
1808                         kmem_cache_free(steering->ftes_cache, fte);
1809                         return ERR_PTR(err);
1810                 }
1811
1812                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1813                 up_write_ref_node(&g->node, false);
1814                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1815                 up_write_ref_node(&fte->node, false);
1816                 if (IS_ERR(rule))
1817                         tree_put_node(&fte->node, false);
1818                 return rule;
1819         }
1820         rule = ERR_PTR(-ENOENT);
1821 out:
1822         kmem_cache_free(steering->ftes_cache, fte);
1823         return rule;
1824 }
1825
1826 static struct mlx5_flow_handle *
1827 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1828                      const struct mlx5_flow_spec *spec,
1829                      struct mlx5_flow_act *flow_act,
1830                      struct mlx5_flow_destination *dest,
1831                      int dest_num)
1832
1833 {
1834         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1835         struct mlx5_flow_handle *rule;
1836         struct match_list match_head;
1837         struct mlx5_flow_group *g;
1838         bool take_write = false;
1839         struct fs_fte *fte;
1840         int version;
1841         int err;
1842         int i;
1843
1844         if (!check_valid_spec(spec))
1845                 return ERR_PTR(-EINVAL);
1846
1847         for (i = 0; i < dest_num; i++) {
1848                 if (!dest_is_valid(&dest[i], flow_act, ft))
1849                         return ERR_PTR(-EINVAL);
1850         }
1851         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1852 search_again_locked:
1853         version = atomic_read(&ft->node.version);
1854
1855         /* Collect all fgs which has a matching match_criteria */
1856         err = build_match_list(&match_head, ft, spec, take_write);
1857         if (err) {
1858                 if (take_write)
1859                         up_write_ref_node(&ft->node, false);
1860                 else
1861                         up_read_ref_node(&ft->node);
1862                 return ERR_PTR(err);
1863         }
1864
1865         if (!take_write)
1866                 up_read_ref_node(&ft->node);
1867
1868         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1869                                       dest_num, version);
1870         free_match_list(&match_head, take_write);
1871         if (!IS_ERR(rule) ||
1872             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1873                 if (take_write)
1874                         up_write_ref_node(&ft->node, false);
1875                 return rule;
1876         }
1877
1878         if (!take_write) {
1879                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1880                 take_write = true;
1881         }
1882
1883         if (PTR_ERR(rule) == -EAGAIN ||
1884             version != atomic_read(&ft->node.version))
1885                 goto search_again_locked;
1886
1887         g = alloc_auto_flow_group(ft, spec);
1888         if (IS_ERR(g)) {
1889                 rule = ERR_CAST(g);
1890                 up_write_ref_node(&ft->node, false);
1891                 return rule;
1892         }
1893
1894         fte = alloc_fte(ft, spec, flow_act);
1895         if (IS_ERR(fte)) {
1896                 up_write_ref_node(&ft->node, false);
1897                 err = PTR_ERR(fte);
1898                 goto err_alloc_fte;
1899         }
1900
1901         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1902         up_write_ref_node(&ft->node, false);
1903
1904         err = create_auto_flow_group(ft, g);
1905         if (err)
1906                 goto err_release_fg;
1907
1908         err = insert_fte(g, fte);
1909         if (err)
1910                 goto err_release_fg;
1911
1912         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1913         up_write_ref_node(&g->node, false);
1914         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1915         up_write_ref_node(&fte->node, false);
1916         if (IS_ERR(rule))
1917                 tree_put_node(&fte->node, false);
1918         tree_put_node(&g->node, false);
1919         return rule;
1920
1921 err_release_fg:
1922         up_write_ref_node(&g->node, false);
1923         kmem_cache_free(steering->ftes_cache, fte);
1924 err_alloc_fte:
1925         tree_put_node(&g->node, false);
1926         return ERR_PTR(err);
1927 }
1928
1929 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1930 {
1931         return ((ft->type == FS_FT_NIC_RX) &&
1932                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1933 }
1934
1935 struct mlx5_flow_handle *
1936 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1937                     const struct mlx5_flow_spec *spec,
1938                     struct mlx5_flow_act *flow_act,
1939                     struct mlx5_flow_destination *dest,
1940                     int num_dest)
1941 {
1942         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1943         static const struct mlx5_flow_spec zero_spec = {};
1944         struct mlx5_flow_destination *gen_dest = NULL;
1945         struct mlx5_flow_table *next_ft = NULL;
1946         struct mlx5_flow_handle *handle = NULL;
1947         u32 sw_action = flow_act->action;
1948         int i;
1949
1950         if (!spec)
1951                 spec = &zero_spec;
1952
1953         if (!is_fwd_next_action(sw_action))
1954                 return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1955
1956         if (!fwd_next_prio_supported(ft))
1957                 return ERR_PTR(-EOPNOTSUPP);
1958
1959         mutex_lock(&root->chain_lock);
1960         next_ft = find_next_fwd_ft(ft, flow_act);
1961         if (!next_ft) {
1962                 handle = ERR_PTR(-EOPNOTSUPP);
1963                 goto unlock;
1964         }
1965
1966         gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
1967                            GFP_KERNEL);
1968         if (!gen_dest) {
1969                 handle = ERR_PTR(-ENOMEM);
1970                 goto unlock;
1971         }
1972         for (i = 0; i < num_dest; i++)
1973                 gen_dest[i] = dest[i];
1974         gen_dest[i].type =
1975                 MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1976         gen_dest[i].ft = next_ft;
1977         dest = gen_dest;
1978         num_dest++;
1979         flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
1980                               MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
1981         flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1982         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1983         if (IS_ERR(handle))
1984                 goto unlock;
1985
1986         if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
1987                 mutex_lock(&next_ft->lock);
1988                 list_add(&handle->rule[num_dest - 1]->next_ft,
1989                          &next_ft->fwd_rules);
1990                 mutex_unlock(&next_ft->lock);
1991                 handle->rule[num_dest - 1]->sw_action = sw_action;
1992                 handle->rule[num_dest - 1]->ft = ft;
1993         }
1994 unlock:
1995         mutex_unlock(&root->chain_lock);
1996         kfree(gen_dest);
1997         return handle;
1998 }
1999 EXPORT_SYMBOL(mlx5_add_flow_rules);
2000
2001 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2002 {
2003         struct fs_fte *fte;
2004         int i;
2005
2006         /* In order to consolidate the HW changes we lock the FTE for other
2007          * changes, and increase its refcount, in order not to perform the
2008          * "del" functions of the FTE. Will handle them here.
2009          * The removal of the rules is done under locked FTE.
2010          * After removing all the handle's rules, if there are remaining
2011          * rules, it means we just need to modify the FTE in FW, and
2012          * unlock/decrease the refcount we increased before.
2013          * Otherwise, it means the FTE should be deleted. First delete the
2014          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2015          * the FTE, which will handle the last decrease of the refcount, as
2016          * well as required handling of its parent.
2017          */
2018         fs_get_obj(fte, handle->rule[0]->node.parent);
2019         down_write_ref_node(&fte->node, false);
2020         for (i = handle->num_rules - 1; i >= 0; i--)
2021                 tree_remove_node(&handle->rule[i]->node, true);
2022         if (fte->dests_size) {
2023                 if (fte->modify_mask)
2024                         modify_fte(fte);
2025                 up_write_ref_node(&fte->node, false);
2026         } else if (list_empty(&fte->node.children)) {
2027                 del_hw_fte(&fte->node);
2028                 /* Avoid double call to del_hw_fte */
2029                 fte->node.del_hw_func = NULL;
2030                 up_write_ref_node(&fte->node, false);
2031                 tree_put_node(&fte->node, false);
2032         }
2033         kfree(handle);
2034 }
2035 EXPORT_SYMBOL(mlx5_del_flow_rules);
2036
2037 /* Assuming prio->node.children(flow tables) is sorted by level */
2038 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2039 {
2040         struct fs_prio *prio;
2041
2042         fs_get_obj(prio, ft->node.parent);
2043
2044         if (!list_is_last(&ft->node.list, &prio->node.children))
2045                 return list_next_entry(ft, node.list);
2046         return find_next_chained_ft(prio);
2047 }
2048
2049 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2050 {
2051         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2052         struct mlx5_ft_underlay_qp *uqp;
2053         struct mlx5_flow_table *new_root_ft = NULL;
2054         int err = 0;
2055         u32 qpn;
2056
2057         if (root->root_ft != ft)
2058                 return 0;
2059
2060         new_root_ft = find_next_ft(ft);
2061         if (!new_root_ft) {
2062                 root->root_ft = NULL;
2063                 return 0;
2064         }
2065
2066         if (list_empty(&root->underlay_qpns)) {
2067                 /* Don't set any QPN (zero) in case QPN list is empty */
2068                 qpn = 0;
2069                 err = root->cmds->update_root_ft(root, new_root_ft,
2070                                                  qpn, false);
2071         } else {
2072                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
2073                         qpn = uqp->qpn;
2074                         err = root->cmds->update_root_ft(root,
2075                                                          new_root_ft, qpn,
2076                                                          false);
2077                         if (err)
2078                                 break;
2079                 }
2080         }
2081
2082         if (err)
2083                 mlx5_core_warn(root->dev,
2084                                "Update root flow table of id(%u) qpn(%d) failed\n",
2085                                ft->id, qpn);
2086         else
2087                 root->root_ft = new_root_ft;
2088
2089         return 0;
2090 }
2091
2092 /* Connect flow table from previous priority to
2093  * the next flow table.
2094  */
2095 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2096 {
2097         struct mlx5_core_dev *dev = get_dev(&ft->node);
2098         struct mlx5_flow_table *next_ft;
2099         struct fs_prio *prio;
2100         int err = 0;
2101
2102         err = update_root_ft_destroy(ft);
2103         if (err)
2104                 return err;
2105
2106         fs_get_obj(prio, ft->node.parent);
2107         if  (!(list_first_entry(&prio->node.children,
2108                                 struct mlx5_flow_table,
2109                                 node.list) == ft))
2110                 return 0;
2111
2112         next_ft = find_next_chained_ft(prio);
2113         err = connect_fwd_rules(dev, next_ft, ft);
2114         if (err)
2115                 return err;
2116
2117         err = connect_prev_fts(dev, next_ft, prio);
2118         if (err)
2119                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2120                                ft->id);
2121         return err;
2122 }
2123
2124 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2125 {
2126         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2127         int err = 0;
2128
2129         mutex_lock(&root->chain_lock);
2130         if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2131                 err = disconnect_flow_table(ft);
2132         if (err) {
2133                 mutex_unlock(&root->chain_lock);
2134                 return err;
2135         }
2136         if (tree_remove_node(&ft->node, false))
2137                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2138                                ft->id);
2139         mutex_unlock(&root->chain_lock);
2140
2141         return err;
2142 }
2143 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2144
2145 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2146 {
2147         if (tree_remove_node(&fg->node, false))
2148                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2149                                fg->id);
2150 }
2151 EXPORT_SYMBOL(mlx5_destroy_flow_group);
2152
2153 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2154                                                 int n)
2155 {
2156         struct mlx5_flow_steering *steering = dev->priv.steering;
2157
2158         if (!steering || !steering->fdb_sub_ns)
2159                 return NULL;
2160
2161         return steering->fdb_sub_ns[n];
2162 }
2163 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2164
2165 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2166                                                     enum mlx5_flow_namespace_type type)
2167 {
2168         struct mlx5_flow_steering *steering = dev->priv.steering;
2169         struct mlx5_flow_root_namespace *root_ns;
2170         int prio = 0;
2171         struct fs_prio *fs_prio;
2172         struct mlx5_flow_namespace *ns;
2173
2174         if (!steering)
2175                 return NULL;
2176
2177         switch (type) {
2178         case MLX5_FLOW_NAMESPACE_FDB:
2179                 if (steering->fdb_root_ns)
2180                         return &steering->fdb_root_ns->ns;
2181                 return NULL;
2182         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2183                 if (steering->sniffer_rx_root_ns)
2184                         return &steering->sniffer_rx_root_ns->ns;
2185                 return NULL;
2186         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2187                 if (steering->sniffer_tx_root_ns)
2188                         return &steering->sniffer_tx_root_ns->ns;
2189                 return NULL;
2190         default:
2191                 break;
2192         }
2193
2194         if (type == MLX5_FLOW_NAMESPACE_EGRESS ||
2195             type == MLX5_FLOW_NAMESPACE_EGRESS_KERNEL) {
2196                 root_ns = steering->egress_root_ns;
2197                 prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2198         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX) {
2199                 root_ns = steering->rdma_rx_root_ns;
2200                 prio = RDMA_RX_BYPASS_PRIO;
2201         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL) {
2202                 root_ns = steering->rdma_rx_root_ns;
2203                 prio = RDMA_RX_KERNEL_PRIO;
2204         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_TX) {
2205                 root_ns = steering->rdma_tx_root_ns;
2206         } else { /* Must be NIC RX */
2207                 root_ns = steering->root_ns;
2208                 prio = type;
2209         }
2210
2211         if (!root_ns)
2212                 return NULL;
2213
2214         fs_prio = find_prio(&root_ns->ns, prio);
2215         if (!fs_prio)
2216                 return NULL;
2217
2218         ns = list_first_entry(&fs_prio->node.children,
2219                               typeof(*ns),
2220                               node.list);
2221
2222         return ns;
2223 }
2224 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2225
2226 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2227                                                               enum mlx5_flow_namespace_type type,
2228                                                               int vport)
2229 {
2230         struct mlx5_flow_steering *steering = dev->priv.steering;
2231
2232         if (!steering)
2233                 return NULL;
2234
2235         switch (type) {
2236         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2237                 if (vport >= steering->esw_egress_acl_vports)
2238                         return NULL;
2239                 if (steering->esw_egress_root_ns &&
2240                     steering->esw_egress_root_ns[vport])
2241                         return &steering->esw_egress_root_ns[vport]->ns;
2242                 else
2243                         return NULL;
2244         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2245                 if (vport >= steering->esw_ingress_acl_vports)
2246                         return NULL;
2247                 if (steering->esw_ingress_root_ns &&
2248                     steering->esw_ingress_root_ns[vport])
2249                         return &steering->esw_ingress_root_ns[vport]->ns;
2250                 else
2251                         return NULL;
2252         default:
2253                 return NULL;
2254         }
2255 }
2256
2257 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2258                                        unsigned int prio,
2259                                        int num_levels,
2260                                        enum fs_node_type type)
2261 {
2262         struct fs_prio *fs_prio;
2263
2264         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2265         if (!fs_prio)
2266                 return ERR_PTR(-ENOMEM);
2267
2268         fs_prio->node.type = type;
2269         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2270         tree_add_node(&fs_prio->node, &ns->node);
2271         fs_prio->num_levels = num_levels;
2272         fs_prio->prio = prio;
2273         list_add_tail(&fs_prio->node.list, &ns->node.children);
2274
2275         return fs_prio;
2276 }
2277
2278 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2279                                               unsigned int prio,
2280                                               int num_levels)
2281 {
2282         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2283 }
2284
2285 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2286                                       unsigned int prio, int num_levels)
2287 {
2288         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2289 }
2290
2291 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2292                                                      *ns)
2293 {
2294         ns->node.type = FS_TYPE_NAMESPACE;
2295
2296         return ns;
2297 }
2298
2299 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2300                                                        int def_miss_act)
2301 {
2302         struct mlx5_flow_namespace      *ns;
2303
2304         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2305         if (!ns)
2306                 return ERR_PTR(-ENOMEM);
2307
2308         fs_init_namespace(ns);
2309         ns->def_miss_action = def_miss_act;
2310         tree_init_node(&ns->node, NULL, del_sw_ns);
2311         tree_add_node(&ns->node, &prio->node);
2312         list_add_tail(&ns->node.list, &prio->node.children);
2313
2314         return ns;
2315 }
2316
2317 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2318                              struct init_tree_node *prio_metadata)
2319 {
2320         struct fs_prio *fs_prio;
2321         int i;
2322
2323         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2324                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2325                 if (IS_ERR(fs_prio))
2326                         return PTR_ERR(fs_prio);
2327         }
2328         return 0;
2329 }
2330
2331 #define FLOW_TABLE_BIT_SZ 1
2332 #define GET_FLOW_TABLE_CAP(dev, offset) \
2333         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2334                         offset / 32)) >>                                        \
2335           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2336 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2337 {
2338         int i;
2339
2340         for (i = 0; i < caps->arr_sz; i++) {
2341                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2342                         return false;
2343         }
2344         return true;
2345 }
2346
2347 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2348                                     struct init_tree_node *init_node,
2349                                     struct fs_node *fs_parent_node,
2350                                     struct init_tree_node *init_parent_node,
2351                                     int prio)
2352 {
2353         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2354                                               flow_table_properties_nic_receive.
2355                                               max_ft_level);
2356         struct mlx5_flow_namespace *fs_ns;
2357         struct fs_prio *fs_prio;
2358         struct fs_node *base;
2359         int i;
2360         int err;
2361
2362         if (init_node->type == FS_TYPE_PRIO) {
2363                 if ((init_node->min_ft_level > max_ft_level) ||
2364                     !has_required_caps(steering->dev, &init_node->caps))
2365                         return 0;
2366
2367                 fs_get_obj(fs_ns, fs_parent_node);
2368                 if (init_node->num_leaf_prios)
2369                         return create_leaf_prios(fs_ns, prio, init_node);
2370                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2371                 if (IS_ERR(fs_prio))
2372                         return PTR_ERR(fs_prio);
2373                 base = &fs_prio->node;
2374         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2375                 fs_get_obj(fs_prio, fs_parent_node);
2376                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2377                 if (IS_ERR(fs_ns))
2378                         return PTR_ERR(fs_ns);
2379                 base = &fs_ns->node;
2380         } else {
2381                 return -EINVAL;
2382         }
2383         prio = 0;
2384         for (i = 0; i < init_node->ar_size; i++) {
2385                 err = init_root_tree_recursive(steering, &init_node->children[i],
2386                                                base, init_node, prio);
2387                 if (err)
2388                         return err;
2389                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2390                     init_node->children[i].num_leaf_prios) {
2391                         prio += init_node->children[i].num_leaf_prios;
2392                 }
2393         }
2394
2395         return 0;
2396 }
2397
2398 static int init_root_tree(struct mlx5_flow_steering *steering,
2399                           struct init_tree_node *init_node,
2400                           struct fs_node *fs_parent_node)
2401 {
2402         int err;
2403         int i;
2404
2405         for (i = 0; i < init_node->ar_size; i++) {
2406                 err = init_root_tree_recursive(steering, &init_node->children[i],
2407                                                fs_parent_node,
2408                                                init_node, i);
2409                 if (err)
2410                         return err;
2411         }
2412         return 0;
2413 }
2414
2415 static void del_sw_root_ns(struct fs_node *node)
2416 {
2417         struct mlx5_flow_root_namespace *root_ns;
2418         struct mlx5_flow_namespace *ns;
2419
2420         fs_get_obj(ns, node);
2421         root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2422         mutex_destroy(&root_ns->chain_lock);
2423         kfree(node);
2424 }
2425
2426 static struct mlx5_flow_root_namespace
2427 *create_root_ns(struct mlx5_flow_steering *steering,
2428                 enum fs_flow_table_type table_type)
2429 {
2430         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2431         struct mlx5_flow_root_namespace *root_ns;
2432         struct mlx5_flow_namespace *ns;
2433
2434         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2435             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2436                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2437
2438         /* Create the root namespace */
2439         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2440         if (!root_ns)
2441                 return NULL;
2442
2443         root_ns->dev = steering->dev;
2444         root_ns->table_type = table_type;
2445         root_ns->cmds = cmds;
2446
2447         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2448
2449         ns = &root_ns->ns;
2450         fs_init_namespace(ns);
2451         mutex_init(&root_ns->chain_lock);
2452         tree_init_node(&ns->node, NULL, del_sw_root_ns);
2453         tree_add_node(&ns->node, NULL);
2454
2455         return root_ns;
2456 }
2457
2458 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2459
2460 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2461 {
2462         struct fs_prio *prio;
2463
2464         fs_for_each_prio(prio, ns) {
2465                  /* This updates prio start_level and num_levels */
2466                 set_prio_attrs_in_prio(prio, acc_level);
2467                 acc_level += prio->num_levels;
2468         }
2469         return acc_level;
2470 }
2471
2472 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2473 {
2474         struct mlx5_flow_namespace *ns;
2475         int acc_level_ns = acc_level;
2476
2477         prio->start_level = acc_level;
2478         fs_for_each_ns(ns, prio) {
2479                 /* This updates start_level and num_levels of ns's priority descendants */
2480                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2481
2482                 /* If this a prio with chains, and we can jump from one chain
2483                  * (namepsace) to another, so we accumulate the levels
2484                  */
2485                 if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2486                         acc_level = acc_level_ns;
2487         }
2488
2489         if (!prio->num_levels)
2490                 prio->num_levels = acc_level_ns - prio->start_level;
2491         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2492 }
2493
2494 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2495 {
2496         struct mlx5_flow_namespace *ns = &root_ns->ns;
2497         struct fs_prio *prio;
2498         int start_level = 0;
2499
2500         fs_for_each_prio(prio, ns) {
2501                 set_prio_attrs_in_prio(prio, start_level);
2502                 start_level += prio->num_levels;
2503         }
2504 }
2505
2506 #define ANCHOR_PRIO 0
2507 #define ANCHOR_SIZE 1
2508 #define ANCHOR_LEVEL 0
2509 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2510 {
2511         struct mlx5_flow_namespace *ns = NULL;
2512         struct mlx5_flow_table_attr ft_attr = {};
2513         struct mlx5_flow_table *ft;
2514
2515         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2516         if (WARN_ON(!ns))
2517                 return -EINVAL;
2518
2519         ft_attr.max_fte = ANCHOR_SIZE;
2520         ft_attr.level   = ANCHOR_LEVEL;
2521         ft_attr.prio    = ANCHOR_PRIO;
2522
2523         ft = mlx5_create_flow_table(ns, &ft_attr);
2524         if (IS_ERR(ft)) {
2525                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2526                 return PTR_ERR(ft);
2527         }
2528         return 0;
2529 }
2530
2531 static int init_root_ns(struct mlx5_flow_steering *steering)
2532 {
2533         int err;
2534
2535         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2536         if (!steering->root_ns)
2537                 return -ENOMEM;
2538
2539         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2540         if (err)
2541                 goto out_err;
2542
2543         set_prio_attrs(steering->root_ns);
2544         err = create_anchor_flow_table(steering);
2545         if (err)
2546                 goto out_err;
2547
2548         return 0;
2549
2550 out_err:
2551         cleanup_root_ns(steering->root_ns);
2552         steering->root_ns = NULL;
2553         return err;
2554 }
2555
2556 static void clean_tree(struct fs_node *node)
2557 {
2558         if (node) {
2559                 struct fs_node *iter;
2560                 struct fs_node *temp;
2561
2562                 tree_get_node(node);
2563                 list_for_each_entry_safe(iter, temp, &node->children, list)
2564                         clean_tree(iter);
2565                 tree_put_node(node, false);
2566                 tree_remove_node(node, false);
2567         }
2568 }
2569
2570 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2571 {
2572         if (!root_ns)
2573                 return;
2574
2575         clean_tree(&root_ns->ns.node);
2576 }
2577
2578 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2579 {
2580         struct mlx5_flow_steering *steering = dev->priv.steering;
2581
2582         cleanup_root_ns(steering->root_ns);
2583         cleanup_root_ns(steering->fdb_root_ns);
2584         steering->fdb_root_ns = NULL;
2585         kfree(steering->fdb_sub_ns);
2586         steering->fdb_sub_ns = NULL;
2587         cleanup_root_ns(steering->sniffer_rx_root_ns);
2588         cleanup_root_ns(steering->sniffer_tx_root_ns);
2589         cleanup_root_ns(steering->rdma_rx_root_ns);
2590         cleanup_root_ns(steering->rdma_tx_root_ns);
2591         cleanup_root_ns(steering->egress_root_ns);
2592         mlx5_cleanup_fc_stats(dev);
2593         kmem_cache_destroy(steering->ftes_cache);
2594         kmem_cache_destroy(steering->fgs_cache);
2595         kfree(steering);
2596 }
2597
2598 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2599 {
2600         struct fs_prio *prio;
2601
2602         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2603         if (!steering->sniffer_tx_root_ns)
2604                 return -ENOMEM;
2605
2606         /* Create single prio */
2607         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2608         return PTR_ERR_OR_ZERO(prio);
2609 }
2610
2611 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2612 {
2613         struct fs_prio *prio;
2614
2615         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2616         if (!steering->sniffer_rx_root_ns)
2617                 return -ENOMEM;
2618
2619         /* Create single prio */
2620         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2621         return PTR_ERR_OR_ZERO(prio);
2622 }
2623
2624 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2625 {
2626         int err;
2627
2628         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2629         if (!steering->rdma_rx_root_ns)
2630                 return -ENOMEM;
2631
2632         err = init_root_tree(steering, &rdma_rx_root_fs,
2633                              &steering->rdma_rx_root_ns->ns.node);
2634         if (err)
2635                 goto out_err;
2636
2637         set_prio_attrs(steering->rdma_rx_root_ns);
2638
2639         return 0;
2640
2641 out_err:
2642         cleanup_root_ns(steering->rdma_rx_root_ns);
2643         steering->rdma_rx_root_ns = NULL;
2644         return err;
2645 }
2646
2647 static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2648 {
2649         int err;
2650
2651         steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2652         if (!steering->rdma_tx_root_ns)
2653                 return -ENOMEM;
2654
2655         err = init_root_tree(steering, &rdma_tx_root_fs,
2656                              &steering->rdma_tx_root_ns->ns.node);
2657         if (err)
2658                 goto out_err;
2659
2660         set_prio_attrs(steering->rdma_tx_root_ns);
2661
2662         return 0;
2663
2664 out_err:
2665         cleanup_root_ns(steering->rdma_tx_root_ns);
2666         steering->rdma_tx_root_ns = NULL;
2667         return err;
2668 }
2669
2670 /* FT and tc chains are stored in the same array so we can re-use the
2671  * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2672  * When creating a new ns for each chain store it in the first available slot.
2673  * Assume tc chains are created and stored first and only then the FT chain.
2674  */
2675 static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2676                                         struct mlx5_flow_namespace *ns)
2677 {
2678         int chain = 0;
2679
2680         while (steering->fdb_sub_ns[chain])
2681                 ++chain;
2682
2683         steering->fdb_sub_ns[chain] = ns;
2684 }
2685
2686 static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2687                                         struct fs_prio *maj_prio)
2688 {
2689         struct mlx5_flow_namespace *ns;
2690         struct fs_prio *min_prio;
2691         int prio;
2692
2693         ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2694         if (IS_ERR(ns))
2695                 return PTR_ERR(ns);
2696
2697         for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2698                 min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2699                 if (IS_ERR(min_prio))
2700                         return PTR_ERR(min_prio);
2701         }
2702
2703         store_fdb_sub_ns_prio_chain(steering, ns);
2704
2705         return 0;
2706 }
2707
2708 static int create_fdb_chains(struct mlx5_flow_steering *steering,
2709                              int fs_prio,
2710                              int chains)
2711 {
2712         struct fs_prio *maj_prio;
2713         int levels;
2714         int chain;
2715         int err;
2716
2717         levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2718         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2719                                           fs_prio,
2720                                           levels);
2721         if (IS_ERR(maj_prio))
2722                 return PTR_ERR(maj_prio);
2723
2724         for (chain = 0; chain < chains; chain++) {
2725                 err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2726                 if (err)
2727                         return err;
2728         }
2729
2730         return 0;
2731 }
2732
2733 static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2734 {
2735         int err;
2736
2737         steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2738                                        sizeof(*steering->fdb_sub_ns),
2739                                        GFP_KERNEL);
2740         if (!steering->fdb_sub_ns)
2741                 return -ENOMEM;
2742
2743         err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
2744         if (err)
2745                 return err;
2746
2747         err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
2748         if (err)
2749                 return err;
2750
2751         return 0;
2752 }
2753
2754 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2755 {
2756         struct fs_prio *maj_prio;
2757         int err;
2758
2759         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2760         if (!steering->fdb_root_ns)
2761                 return -ENOMEM;
2762
2763         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2764                                   1);
2765         if (IS_ERR(maj_prio)) {
2766                 err = PTR_ERR(maj_prio);
2767                 goto out_err;
2768         }
2769         err = create_fdb_fast_path(steering);
2770         if (err)
2771                 goto out_err;
2772
2773         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2774         if (IS_ERR(maj_prio)) {
2775                 err = PTR_ERR(maj_prio);
2776                 goto out_err;
2777         }
2778
2779         /* We put this priority last, knowing that nothing will get here
2780          * unless explicitly forwarded to. This is possible because the
2781          * slow path tables have catch all rules and nothing gets passed
2782          * those tables.
2783          */
2784         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
2785         if (IS_ERR(maj_prio)) {
2786                 err = PTR_ERR(maj_prio);
2787                 goto out_err;
2788         }
2789
2790         set_prio_attrs(steering->fdb_root_ns);
2791         return 0;
2792
2793 out_err:
2794         cleanup_root_ns(steering->fdb_root_ns);
2795         kfree(steering->fdb_sub_ns);
2796         steering->fdb_sub_ns = NULL;
2797         steering->fdb_root_ns = NULL;
2798         return err;
2799 }
2800
2801 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2802 {
2803         struct fs_prio *prio;
2804
2805         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2806         if (!steering->esw_egress_root_ns[vport])
2807                 return -ENOMEM;
2808
2809         /* create 1 prio*/
2810         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2811         return PTR_ERR_OR_ZERO(prio);
2812 }
2813
2814 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2815 {
2816         struct fs_prio *prio;
2817
2818         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2819         if (!steering->esw_ingress_root_ns[vport])
2820                 return -ENOMEM;
2821
2822         /* create 1 prio*/
2823         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2824         return PTR_ERR_OR_ZERO(prio);
2825 }
2826
2827 int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2828 {
2829         struct mlx5_flow_steering *steering = dev->priv.steering;
2830         int err;
2831         int i;
2832
2833         steering->esw_egress_root_ns =
2834                         kcalloc(total_vports,
2835                                 sizeof(*steering->esw_egress_root_ns),
2836                                 GFP_KERNEL);
2837         if (!steering->esw_egress_root_ns)
2838                 return -ENOMEM;
2839
2840         for (i = 0; i < total_vports; i++) {
2841                 err = init_egress_acl_root_ns(steering, i);
2842                 if (err)
2843                         goto cleanup_root_ns;
2844         }
2845         steering->esw_egress_acl_vports = total_vports;
2846         return 0;
2847
2848 cleanup_root_ns:
2849         for (i--; i >= 0; i--)
2850                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2851         kfree(steering->esw_egress_root_ns);
2852         steering->esw_egress_root_ns = NULL;
2853         return err;
2854 }
2855
2856 void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev)
2857 {
2858         struct mlx5_flow_steering *steering = dev->priv.steering;
2859         int i;
2860
2861         if (!steering->esw_egress_root_ns)
2862                 return;
2863
2864         for (i = 0; i < steering->esw_egress_acl_vports; i++)
2865                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2866
2867         kfree(steering->esw_egress_root_ns);
2868         steering->esw_egress_root_ns = NULL;
2869 }
2870
2871 int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2872 {
2873         struct mlx5_flow_steering *steering = dev->priv.steering;
2874         int err;
2875         int i;
2876
2877         steering->esw_ingress_root_ns =
2878                         kcalloc(total_vports,
2879                                 sizeof(*steering->esw_ingress_root_ns),
2880                                 GFP_KERNEL);
2881         if (!steering->esw_ingress_root_ns)
2882                 return -ENOMEM;
2883
2884         for (i = 0; i < total_vports; i++) {
2885                 err = init_ingress_acl_root_ns(steering, i);
2886                 if (err)
2887                         goto cleanup_root_ns;
2888         }
2889         steering->esw_ingress_acl_vports = total_vports;
2890         return 0;
2891
2892 cleanup_root_ns:
2893         for (i--; i >= 0; i--)
2894                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2895         kfree(steering->esw_ingress_root_ns);
2896         steering->esw_ingress_root_ns = NULL;
2897         return err;
2898 }
2899
2900 void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev)
2901 {
2902         struct mlx5_flow_steering *steering = dev->priv.steering;
2903         int i;
2904
2905         if (!steering->esw_ingress_root_ns)
2906                 return;
2907
2908         for (i = 0; i < steering->esw_ingress_acl_vports; i++)
2909                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2910
2911         kfree(steering->esw_ingress_root_ns);
2912         steering->esw_ingress_root_ns = NULL;
2913 }
2914
2915 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2916 {
2917         int err;
2918
2919         steering->egress_root_ns = create_root_ns(steering,
2920                                                   FS_FT_NIC_TX);
2921         if (!steering->egress_root_ns)
2922                 return -ENOMEM;
2923
2924         err = init_root_tree(steering, &egress_root_fs,
2925                              &steering->egress_root_ns->ns.node);
2926         if (err)
2927                 goto cleanup;
2928         set_prio_attrs(steering->egress_root_ns);
2929         return 0;
2930 cleanup:
2931         cleanup_root_ns(steering->egress_root_ns);
2932         steering->egress_root_ns = NULL;
2933         return err;
2934 }
2935
2936 int mlx5_init_fs(struct mlx5_core_dev *dev)
2937 {
2938         struct mlx5_flow_steering *steering;
2939         int err = 0;
2940
2941         err = mlx5_init_fc_stats(dev);
2942         if (err)
2943                 return err;
2944
2945         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2946         if (!steering)
2947                 return -ENOMEM;
2948         steering->dev = dev;
2949         dev->priv.steering = steering;
2950
2951         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2952                                                 sizeof(struct mlx5_flow_group), 0,
2953                                                 0, NULL);
2954         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2955                                                  0, NULL);
2956         if (!steering->ftes_cache || !steering->fgs_cache) {
2957                 err = -ENOMEM;
2958                 goto err;
2959         }
2960
2961         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2962               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2963              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2964               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2965             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2966                 err = init_root_ns(steering);
2967                 if (err)
2968                         goto err;
2969         }
2970
2971         if (MLX5_ESWITCH_MANAGER(dev)) {
2972                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2973                         err = init_fdb_root_ns(steering);
2974                         if (err)
2975                                 goto err;
2976                 }
2977         }
2978
2979         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2980                 err = init_sniffer_rx_root_ns(steering);
2981                 if (err)
2982                         goto err;
2983         }
2984
2985         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2986                 err = init_sniffer_tx_root_ns(steering);
2987                 if (err)
2988                         goto err;
2989         }
2990
2991         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
2992             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
2993                 err = init_rdma_rx_root_ns(steering);
2994                 if (err)
2995                         goto err;
2996         }
2997
2998         if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
2999                 err = init_rdma_tx_root_ns(steering);
3000                 if (err)
3001                         goto err;
3002         }
3003
3004         if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE ||
3005             MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3006                 err = init_egress_root_ns(steering);
3007                 if (err)
3008                         goto err;
3009         }
3010
3011         return 0;
3012 err:
3013         mlx5_cleanup_fs(dev);
3014         return err;
3015 }
3016
3017 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3018 {
3019         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3020         struct mlx5_ft_underlay_qp *new_uqp;
3021         int err = 0;
3022
3023         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3024         if (!new_uqp)
3025                 return -ENOMEM;
3026
3027         mutex_lock(&root->chain_lock);
3028
3029         if (!root->root_ft) {
3030                 err = -EINVAL;
3031                 goto update_ft_fail;
3032         }
3033
3034         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3035                                          false);
3036         if (err) {
3037                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3038                                underlay_qpn, err);
3039                 goto update_ft_fail;
3040         }
3041
3042         new_uqp->qpn = underlay_qpn;
3043         list_add_tail(&new_uqp->list, &root->underlay_qpns);
3044
3045         mutex_unlock(&root->chain_lock);
3046
3047         return 0;
3048
3049 update_ft_fail:
3050         mutex_unlock(&root->chain_lock);
3051         kfree(new_uqp);
3052         return err;
3053 }
3054 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3055
3056 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3057 {
3058         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3059         struct mlx5_ft_underlay_qp *uqp;
3060         bool found = false;
3061         int err = 0;
3062
3063         mutex_lock(&root->chain_lock);
3064         list_for_each_entry(uqp, &root->underlay_qpns, list) {
3065                 if (uqp->qpn == underlay_qpn) {
3066                         found = true;
3067                         break;
3068                 }
3069         }
3070
3071         if (!found) {
3072                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3073                                underlay_qpn);
3074                 err = -EINVAL;
3075                 goto out;
3076         }
3077
3078         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3079                                          true);
3080         if (err)
3081                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3082                                underlay_qpn, err);
3083
3084         list_del(&uqp->list);
3085         mutex_unlock(&root->chain_lock);
3086         kfree(uqp);
3087
3088         return 0;
3089
3090 out:
3091         mutex_unlock(&root->chain_lock);
3092         return err;
3093 }
3094 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3095
3096 static struct mlx5_flow_root_namespace
3097 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3098 {
3099         struct mlx5_flow_namespace *ns;
3100
3101         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3102             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3103                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3104         else
3105                 ns = mlx5_get_flow_namespace(dev, ns_type);
3106         if (!ns)
3107                 return NULL;
3108
3109         return find_root(&ns->node);
3110 }
3111
3112 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3113                                                  u8 ns_type, u8 num_actions,
3114                                                  void *modify_actions)
3115 {
3116         struct mlx5_flow_root_namespace *root;
3117         struct mlx5_modify_hdr *modify_hdr;
3118         int err;
3119
3120         root = get_root_namespace(dev, ns_type);
3121         if (!root)
3122                 return ERR_PTR(-EOPNOTSUPP);
3123
3124         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3125         if (!modify_hdr)
3126                 return ERR_PTR(-ENOMEM);
3127
3128         modify_hdr->ns_type = ns_type;
3129         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3130                                               modify_actions, modify_hdr);
3131         if (err) {
3132                 kfree(modify_hdr);
3133                 return ERR_PTR(err);
3134         }
3135
3136         return modify_hdr;
3137 }
3138 EXPORT_SYMBOL(mlx5_modify_header_alloc);
3139
3140 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3141                                 struct mlx5_modify_hdr *modify_hdr)
3142 {
3143         struct mlx5_flow_root_namespace *root;
3144
3145         root = get_root_namespace(dev, modify_hdr->ns_type);
3146         if (WARN_ON(!root))
3147                 return;
3148         root->cmds->modify_header_dealloc(root, modify_hdr);
3149         kfree(modify_hdr);
3150 }
3151 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3152
3153 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3154                                                      int reformat_type,
3155                                                      size_t size,
3156                                                      void *reformat_data,
3157                                                      enum mlx5_flow_namespace_type ns_type)
3158 {
3159         struct mlx5_pkt_reformat *pkt_reformat;
3160         struct mlx5_flow_root_namespace *root;
3161         int err;
3162
3163         root = get_root_namespace(dev, ns_type);
3164         if (!root)
3165                 return ERR_PTR(-EOPNOTSUPP);
3166
3167         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3168         if (!pkt_reformat)
3169                 return ERR_PTR(-ENOMEM);
3170
3171         pkt_reformat->ns_type = ns_type;
3172         pkt_reformat->reformat_type = reformat_type;
3173         err = root->cmds->packet_reformat_alloc(root, reformat_type, size,
3174                                                 reformat_data, ns_type,
3175                                                 pkt_reformat);
3176         if (err) {
3177                 kfree(pkt_reformat);
3178                 return ERR_PTR(err);
3179         }
3180
3181         return pkt_reformat;
3182 }
3183 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3184
3185 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3186                                   struct mlx5_pkt_reformat *pkt_reformat)
3187 {
3188         struct mlx5_flow_root_namespace *root;
3189
3190         root = get_root_namespace(dev, pkt_reformat->ns_type);
3191         if (WARN_ON(!root))
3192                 return;
3193         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3194         kfree(pkt_reformat);
3195 }
3196 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3197
3198 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3199                                  struct mlx5_flow_root_namespace *peer_ns)
3200 {
3201         if (peer_ns && ns->mode != peer_ns->mode) {
3202                 mlx5_core_err(ns->dev,
3203                               "Can't peer namespace of different steering mode\n");
3204                 return -EINVAL;
3205         }
3206
3207         return ns->cmds->set_peer(ns, peer_ns);
3208 }
3209
3210 /* This function should be called only at init stage of the namespace.
3211  * It is not safe to call this function while steering operations
3212  * are executed in the namespace.
3213  */
3214 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3215                                  enum mlx5_flow_steering_mode mode)
3216 {
3217         struct mlx5_flow_root_namespace *root;
3218         const struct mlx5_flow_cmds *cmds;
3219         int err;
3220
3221         root = find_root(&ns->node);
3222         if (&root->ns != ns)
3223         /* Can't set cmds to non root namespace */
3224                 return -EINVAL;
3225
3226         if (root->table_type != FS_FT_FDB)
3227                 return -EOPNOTSUPP;
3228
3229         if (root->mode == mode)
3230                 return 0;
3231
3232         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3233                 cmds = mlx5_fs_cmd_get_dr_cmds();
3234         else
3235                 cmds = mlx5_fs_cmd_get_fw_cmds();
3236         if (!cmds)
3237                 return -EOPNOTSUPP;
3238
3239         err = cmds->create_ns(root);
3240         if (err) {
3241                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3242                               err);
3243                 return err;
3244         }
3245
3246         root->cmds->destroy_ns(root);
3247         root->cmds = cmds;
3248         root->mode = mode;
3249
3250         return 0;
3251 }