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 #include "eswitch.h"
  45
  46 #define INIT_TREE_NODE_ARRAY_SIZE(...)  (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
  47                                          sizeof(struct init_tree_node))
  48
  49 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
  50                  ...) {.type = FS_TYPE_PRIO,\
  51         .min_ft_level = min_level_val,\
  52         .num_levels = num_levels_val,\
  53         .num_leaf_prios = num_prios_val,\
  54         .caps = caps_val,\
  55         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  56         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  57 }
  58
  59 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
  60         ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
  61                  __VA_ARGS__)\
  62
  63 #define ADD_NS(def_miss_act, ...) {.type = FS_TYPE_NAMESPACE,   \
  64         .def_miss_action = def_miss_act,\
  65         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  66         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  67 }
  68
  69 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  70                                    sizeof(long))
  71
  72 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  73
  74 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  75                                .caps = (long[]) {__VA_ARGS__} }
  76
  77 #define FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  78                                            FS_CAP(flow_table_properties_nic_receive.modify_root), \
  79                                            FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  80                                            FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  81
  82 #define FS_CHAINING_CAPS_EGRESS                                                \
  83         FS_REQUIRED_CAPS(                                                      \
  84                 FS_CAP(flow_table_properties_nic_transmit.flow_modify_en),     \
  85                 FS_CAP(flow_table_properties_nic_transmit.modify_root),        \
  86                 FS_CAP(flow_table_properties_nic_transmit                      \
  87                                .identified_miss_table_mode),                   \
  88                 FS_CAP(flow_table_properties_nic_transmit.flow_table_modify))
  89
  90 #define LEFTOVERS_NUM_LEVELS 1
  91 #define LEFTOVERS_NUM_PRIOS 1
  92
  93 #define BY_PASS_PRIO_NUM_LEVELS 1
  94 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  95                            LEFTOVERS_NUM_PRIOS)
  96
  97 #define ETHTOOL_PRIO_NUM_LEVELS 1
  98 #define ETHTOOL_NUM_PRIOS 11
  99 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
 100 /* Vlan, mac, ttc, inner ttc, aRFS */
 101 #define KERNEL_NIC_PRIO_NUM_LEVELS 5
 102 #define KERNEL_NIC_NUM_PRIOS 1
 103 /* One more level for tc */
 104 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
 105
 106 #define KERNEL_NIC_TC_NUM_PRIOS  1
 107 #define KERNEL_NIC_TC_NUM_LEVELS 2
 108
 109 #define ANCHOR_NUM_LEVELS 1
 110 #define ANCHOR_NUM_PRIOS 1
 111 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 112
 113 #define OFFLOADS_MAX_FT 1
 114 #define OFFLOADS_NUM_PRIOS 1
 115 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
 116
 117 #define LAG_PRIO_NUM_LEVELS 1
 118 #define LAG_NUM_PRIOS 1
 119 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 120
 121 struct node_caps {
 122         size_t  arr_sz;
 123         long    *caps;
 124 };
 125
 126 static struct init_tree_node {
 127         enum fs_node_type       type;
 128         struct init_tree_node *children;
 129         int ar_size;
 130         struct node_caps caps;
 131         int min_ft_level;
 132         int num_leaf_prios;
 133         int prio;
 134         int num_levels;
 135         enum mlx5_flow_table_miss_action def_miss_action;
 136 } root_fs = {
 137         .type = FS_TYPE_NAMESPACE,
 138         .ar_size = 7,
 139           .children = (struct init_tree_node[]){
 140                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 141                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 142                                   ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 143                                                     BY_PASS_PRIO_NUM_LEVELS))),
 144                   ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 145                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 146                                   ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 147                                                     LAG_PRIO_NUM_LEVELS))),
 148                   ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
 149                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 150                                   ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS,
 151                                                     OFFLOADS_MAX_FT))),
 152                   ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 153                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 154                                   ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 155                                                     ETHTOOL_PRIO_NUM_LEVELS))),
 156                   ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 157                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 158                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS,
 159                                                     KERNEL_NIC_TC_NUM_LEVELS),
 160                                   ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 161                                                     KERNEL_NIC_PRIO_NUM_LEVELS))),
 162                   ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 163                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 164                                   ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS,
 165                                                     LEFTOVERS_NUM_LEVELS))),
 166                   ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 167                            ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 168                                   ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS,
 169                                                     ANCHOR_NUM_LEVELS))),
 170         }
 171 };
 172
 173 static struct init_tree_node egress_root_fs = {
 174         .type = FS_TYPE_NAMESPACE,
 175         .ar_size = 1,
 176         .children = (struct init_tree_node[]) {
 177                 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 178                          FS_CHAINING_CAPS_EGRESS,
 179                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 180                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 181                                                   BY_PASS_PRIO_NUM_LEVELS))),
 182         }
 183 };
 184
 185 #define RDMA_RX_BYPASS_PRIO 0
 186 #define RDMA_RX_KERNEL_PRIO 1
 187 static struct init_tree_node rdma_rx_root_fs = {
 188         .type = FS_TYPE_NAMESPACE,
 189         .ar_size = 2,
 190         .children = (struct init_tree_node[]) {
 191                 [RDMA_RX_BYPASS_PRIO] =
 192                 ADD_PRIO(0, MLX5_BY_PASS_NUM_REGULAR_PRIOS, 0,
 193                          FS_CHAINING_CAPS,
 194                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 195                                 ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS,
 196                                                   BY_PASS_PRIO_NUM_LEVELS))),
 197                 [RDMA_RX_KERNEL_PRIO] =
 198                 ADD_PRIO(0, MLX5_BY_PASS_NUM_REGULAR_PRIOS + 1, 0,
 199                          FS_CHAINING_CAPS,
 200                          ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN,
 201                                 ADD_MULTIPLE_PRIO(1, 1))),
 202         }
 203 };
 204
 205 enum fs_i_lock_class {
 206         FS_LOCK_GRANDPARENT,
 207         FS_LOCK_PARENT,
 208         FS_LOCK_CHILD
 209 };
 210
 211 static const struct rhashtable_params rhash_fte = {
 212         .key_len = FIELD_SIZEOF(struct fs_fte, val),
 213         .key_offset = offsetof(struct fs_fte, val),
 214         .head_offset = offsetof(struct fs_fte, hash),
 215         .automatic_shrinking = true,
 216         .min_size = 1,
 217 };
 218
 219 static const struct rhashtable_params rhash_fg = {
 220         .key_len = FIELD_SIZEOF(struct mlx5_flow_group, mask),
 221         .key_offset = offsetof(struct mlx5_flow_group, mask),
 222         .head_offset = offsetof(struct mlx5_flow_group, hash),
 223         .automatic_shrinking = true,
 224         .min_size = 1,
 225
 226 };
 227
 228 static void del_hw_flow_table(struct fs_node *node);
 229 static void del_hw_flow_group(struct fs_node *node);
 230 static void del_hw_fte(struct fs_node *node);
 231 static void del_sw_flow_table(struct fs_node *node);
 232 static void del_sw_flow_group(struct fs_node *node);
 233 static void del_sw_fte(struct fs_node *node);
 234 static void del_sw_prio(struct fs_node *node);
 235 static void del_sw_ns(struct fs_node *node);
 236 /* Delete rule (destination) is special case that
 237  * requires to lock the FTE for all the deletion process.
 238  */
 239 static void del_sw_hw_rule(struct fs_node *node);
 240 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 241                                 struct mlx5_flow_destination *d2);
 242 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 243 static struct mlx5_flow_rule *
 244 find_flow_rule(struct fs_fte *fte,
 245                struct mlx5_flow_destination *dest);
 246
 247 static void tree_init_node(struct fs_node *node,
 248                            void (*del_hw_func)(struct fs_node *),
 249                            void (*del_sw_func)(struct fs_node *))
 250 {
 251         refcount_set(&node->refcount, 1);
 252         INIT_LIST_HEAD(&node->list);
 253         INIT_LIST_HEAD(&node->children);
 254         init_rwsem(&node->lock);
 255         node->del_hw_func = del_hw_func;
 256         node->del_sw_func = del_sw_func;
 257         node->active = false;
 258 }
 259
 260 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 261 {
 262         if (parent)
 263                 refcount_inc(&parent->refcount);
 264         node->parent = parent;
 265
 266         /* Parent is the root */
 267         if (!parent)
 268                 node->root = node;
 269         else
 270                 node->root = parent->root;
 271 }
 272
 273 static int tree_get_node(struct fs_node *node)
 274 {
 275         return refcount_inc_not_zero(&node->refcount);
 276 }
 277
 278 static void nested_down_read_ref_node(struct fs_node *node,
 279                                       enum fs_i_lock_class class)
 280 {
 281         if (node) {
 282                 down_read_nested(&node->lock, class);
 283                 refcount_inc(&node->refcount);
 284         }
 285 }
 286
 287 static void nested_down_write_ref_node(struct fs_node *node,
 288                                        enum fs_i_lock_class class)
 289 {
 290         if (node) {
 291                 down_write_nested(&node->lock, class);
 292                 refcount_inc(&node->refcount);
 293         }
 294 }
 295
 296 static void down_write_ref_node(struct fs_node *node, bool locked)
 297 {
 298         if (node) {
 299                 if (!locked)
 300                         down_write(&node->lock);
 301                 refcount_inc(&node->refcount);
 302         }
 303 }
 304
 305 static void up_read_ref_node(struct fs_node *node)
 306 {
 307         refcount_dec(&node->refcount);
 308         up_read(&node->lock);
 309 }
 310
 311 static void up_write_ref_node(struct fs_node *node, bool locked)
 312 {
 313         refcount_dec(&node->refcount);
 314         if (!locked)
 315                 up_write(&node->lock);
 316 }
 317
 318 static void tree_put_node(struct fs_node *node, bool locked)
 319 {
 320         struct fs_node *parent_node = node->parent;
 321
 322         if (refcount_dec_and_test(&node->refcount)) {
 323                 if (node->del_hw_func)
 324                         node->del_hw_func(node);
 325                 if (parent_node) {
 326                         /* Only root namespace doesn't have parent and we just
 327                          * need to free its node.
 328                          */
 329                         down_write_ref_node(parent_node, locked);
 330                         list_del_init(&node->list);
 331                         if (node->del_sw_func)
 332                                 node->del_sw_func(node);
 333                         up_write_ref_node(parent_node, locked);
 334                 } else {
 335                         kfree(node);
 336                 }
 337                 node = NULL;
 338         }
 339         if (!node && parent_node)
 340                 tree_put_node(parent_node, locked);
 341 }
 342
 343 static int tree_remove_node(struct fs_node *node, bool locked)
 344 {
 345         if (refcount_read(&node->refcount) > 1) {
 346                 refcount_dec(&node->refcount);
 347                 return -EEXIST;
 348         }
 349         tree_put_node(node, locked);
 350         return 0;
 351 }
 352
 353 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 354                                  unsigned int prio)
 355 {
 356         struct fs_prio *iter_prio;
 357
 358         fs_for_each_prio(iter_prio, ns) {
 359                 if (iter_prio->prio == prio)
 360                         return iter_prio;
 361         }
 362
 363         return NULL;
 364 }
 365
 366 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 367 {
 368         int i;
 369
 370         for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 371                 if (spec->match_value[i] & ~spec->match_criteria[i]) {
 372                         pr_warn("mlx5_core: match_value differs from match_criteria\n");
 373                         return false;
 374                 }
 375
 376         return true;
 377 }
 378
 379 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 380 {
 381         struct fs_node *root;
 382         struct mlx5_flow_namespace *ns;
 383
 384         root = node->root;
 385
 386         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 387                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 388                 return NULL;
 389         }
 390
 391         ns = container_of(root, struct mlx5_flow_namespace, node);
 392         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 393 }
 394
 395 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 396 {
 397         struct mlx5_flow_root_namespace *root = find_root(node);
 398
 399         if (root)
 400                 return root->dev->priv.steering;
 401         return NULL;
 402 }
 403
 404 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 405 {
 406         struct mlx5_flow_root_namespace *root = find_root(node);
 407
 408         if (root)
 409                 return root->dev;
 410         return NULL;
 411 }
 412
 413 static void del_sw_ns(struct fs_node *node)
 414 {
 415         kfree(node);
 416 }
 417
 418 static void del_sw_prio(struct fs_node *node)
 419 {
 420         kfree(node);
 421 }
 422
 423 static void del_hw_flow_table(struct fs_node *node)
 424 {
 425         struct mlx5_flow_root_namespace *root;
 426         struct mlx5_flow_table *ft;
 427         struct mlx5_core_dev *dev;
 428         int err;
 429
 430         fs_get_obj(ft, node);
 431         dev = get_dev(&ft->node);
 432         root = find_root(&ft->node);
 433         trace_mlx5_fs_del_ft(ft);
 434
 435         if (node->active) {
 436                 err = root->cmds->destroy_flow_table(root, ft);
 437                 if (err)
 438                         mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 439         }
 440 }
 441
 442 static void del_sw_flow_table(struct fs_node *node)
 443 {
 444         struct mlx5_flow_table *ft;
 445         struct fs_prio *prio;
 446
 447         fs_get_obj(ft, node);
 448
 449         rhltable_destroy(&ft->fgs_hash);
 450         fs_get_obj(prio, ft->node.parent);
 451         prio->num_ft--;
 452         kfree(ft);
 453 }
 454
 455 static void modify_fte(struct fs_fte *fte)
 456 {
 457         struct mlx5_flow_root_namespace *root;
 458         struct mlx5_flow_table *ft;
 459         struct mlx5_flow_group *fg;
 460         struct mlx5_core_dev *dev;
 461         int err;
 462
 463         fs_get_obj(fg, fte->node.parent);
 464         fs_get_obj(ft, fg->node.parent);
 465         dev = get_dev(&fte->node);
 466
 467         root = find_root(&ft->node);
 468         err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte);
 469         if (err)
 470                 mlx5_core_warn(dev,
 471                                "%s can't del rule fg id=%d fte_index=%d\n",
 472                                __func__, fg->id, fte->index);
 473         fte->modify_mask = 0;
 474 }
 475
 476 static void del_sw_hw_rule(struct fs_node *node)
 477 {
 478         struct mlx5_flow_rule *rule;
 479         struct fs_fte *fte;
 480
 481         fs_get_obj(rule, node);
 482         fs_get_obj(fte, rule->node.parent);
 483         trace_mlx5_fs_del_rule(rule);
 484         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 485                 mutex_lock(&rule->dest_attr.ft->lock);
 486                 list_del(&rule->next_ft);
 487                 mutex_unlock(&rule->dest_attr.ft->lock);
 488         }
 489
 490         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 491             --fte->dests_size) {
 492                 fte->modify_mask |=
 493                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 494                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 495                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 496                 goto out;
 497         }
 498
 499         if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 500             --fte->dests_size) {
 501                 fte->modify_mask |=
 502                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 503         }
 504 out:
 505         kfree(rule);
 506 }
 507
 508 static void del_hw_fte(struct fs_node *node)
 509 {
 510         struct mlx5_flow_root_namespace *root;
 511         struct mlx5_flow_table *ft;
 512         struct mlx5_flow_group *fg;
 513         struct mlx5_core_dev *dev;
 514         struct fs_fte *fte;
 515         int err;
 516
 517         fs_get_obj(fte, node);
 518         fs_get_obj(fg, fte->node.parent);
 519         fs_get_obj(ft, fg->node.parent);
 520
 521         trace_mlx5_fs_del_fte(fte);
 522         dev = get_dev(&ft->node);
 523         root = find_root(&ft->node);
 524         if (node->active) {
 525                 err = root->cmds->delete_fte(root, ft, fte);
 526                 if (err)
 527                         mlx5_core_warn(dev,
 528                                        "flow steering can't delete fte in index %d of flow group id %d\n",
 529                                        fte->index, fg->id);
 530                 node->active = 0;
 531         }
 532 }
 533
 534 static void del_sw_fte(struct fs_node *node)
 535 {
 536         struct mlx5_flow_steering *steering = get_steering(node);
 537         struct mlx5_flow_group *fg;
 538         struct fs_fte *fte;
 539         int err;
 540
 541         fs_get_obj(fte, node);
 542         fs_get_obj(fg, fte->node.parent);
 543
 544         err = rhashtable_remove_fast(&fg->ftes_hash,
 545                                      &fte->hash,
 546                                      rhash_fte);
 547         WARN_ON(err);
 548         ida_simple_remove(&fg->fte_allocator, fte->index - fg->start_index);
 549         kmem_cache_free(steering->ftes_cache, fte);
 550 }
 551
 552 static void del_hw_flow_group(struct fs_node *node)
 553 {
 554         struct mlx5_flow_root_namespace *root;
 555         struct mlx5_flow_group *fg;
 556         struct mlx5_flow_table *ft;
 557         struct mlx5_core_dev *dev;
 558
 559         fs_get_obj(fg, node);
 560         fs_get_obj(ft, fg->node.parent);
 561         dev = get_dev(&ft->node);
 562         trace_mlx5_fs_del_fg(fg);
 563
 564         root = find_root(&ft->node);
 565         if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
 566                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 567                                fg->id, ft->id);
 568 }
 569
 570 static void del_sw_flow_group(struct fs_node *node)
 571 {
 572         struct mlx5_flow_steering *steering = get_steering(node);
 573         struct mlx5_flow_group *fg;
 574         struct mlx5_flow_table *ft;
 575         int err;
 576
 577         fs_get_obj(fg, node);
 578         fs_get_obj(ft, fg->node.parent);
 579
 580         rhashtable_destroy(&fg->ftes_hash);
 581         ida_destroy(&fg->fte_allocator);
 582         if (ft->autogroup.active && fg->max_ftes == ft->autogroup.group_size)
 583                 ft->autogroup.num_groups--;
 584         err = rhltable_remove(&ft->fgs_hash,
 585                               &fg->hash,
 586                               rhash_fg);
 587         WARN_ON(err);
 588         kmem_cache_free(steering->fgs_cache, fg);
 589 }
 590
 591 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 592 {
 593         int index;
 594         int ret;
 595
 596         index = ida_simple_get(&fg->fte_allocator, 0, fg->max_ftes, GFP_KERNEL);
 597         if (index < 0)
 598                 return index;
 599
 600         fte->index = index + fg->start_index;
 601         ret = rhashtable_insert_fast(&fg->ftes_hash,
 602                                      &fte->hash,
 603                                      rhash_fte);
 604         if (ret)
 605                 goto err_ida_remove;
 606
 607         tree_add_node(&fte->node, &fg->node);
 608         list_add_tail(&fte->node.list, &fg->node.children);
 609         return 0;
 610
 611 err_ida_remove:
 612         ida_simple_remove(&fg->fte_allocator, index);
 613         return ret;
 614 }
 615
 616 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 617                                 const struct mlx5_flow_spec *spec,
 618                                 struct mlx5_flow_act *flow_act)
 619 {
 620         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 621         struct fs_fte *fte;
 622
 623         fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 624         if (!fte)
 625                 return ERR_PTR(-ENOMEM);
 626
 627         memcpy(fte->val, &spec->match_value, sizeof(fte->val));
 628         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 629         fte->action = *flow_act;
 630         fte->flow_context = spec->flow_context;
 631
 632         tree_init_node(&fte->node, NULL, del_sw_fte);
 633
 634         return fte;
 635 }
 636
 637 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 638                                struct mlx5_flow_group *fg)
 639 {
 640         rhashtable_destroy(&fg->ftes_hash);
 641         kmem_cache_free(steering->fgs_cache, fg);
 642 }
 643
 644 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 645                                                 u8 match_criteria_enable,
 646                                                 const void *match_criteria,
 647                                                 int start_index,
 648                                                 int end_index)
 649 {
 650         struct mlx5_flow_group *fg;
 651         int ret;
 652
 653         fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 654         if (!fg)
 655                 return ERR_PTR(-ENOMEM);
 656
 657         ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 658         if (ret) {
 659                 kmem_cache_free(steering->fgs_cache, fg);
 660                 return ERR_PTR(ret);
 661         }
 662
 663         ida_init(&fg->fte_allocator);
 664         fg->mask.match_criteria_enable = match_criteria_enable;
 665         memcpy(&fg->mask.match_criteria, match_criteria,
 666                sizeof(fg->mask.match_criteria));
 667         fg->node.type =  FS_TYPE_FLOW_GROUP;
 668         fg->start_index = start_index;
 669         fg->max_ftes = end_index - start_index + 1;
 670
 671         return fg;
 672 }
 673
 674 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 675                                                        u8 match_criteria_enable,
 676                                                        const void *match_criteria,
 677                                                        int start_index,
 678                                                        int end_index,
 679                                                        struct list_head *prev)
 680 {
 681         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 682         struct mlx5_flow_group *fg;
 683         int ret;
 684
 685         fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 686                               start_index, end_index);
 687         if (IS_ERR(fg))
 688                 return fg;
 689
 690         /* initialize refcnt, add to parent list */
 691         ret = rhltable_insert(&ft->fgs_hash,
 692                               &fg->hash,
 693                               rhash_fg);
 694         if (ret) {
 695                 dealloc_flow_group(steering, fg);
 696                 return ERR_PTR(ret);
 697         }
 698
 699         tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 700         tree_add_node(&fg->node, &ft->node);
 701         /* Add node to group list */
 702         list_add(&fg->node.list, prev);
 703         atomic_inc(&ft->node.version);
 704
 705         return fg;
 706 }
 707
 708 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 709                                                 enum fs_flow_table_type table_type,
 710                                                 enum fs_flow_table_op_mod op_mod,
 711                                                 u32 flags)
 712 {
 713         struct mlx5_flow_table *ft;
 714         int ret;
 715
 716         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 717         if (!ft)
 718                 return ERR_PTR(-ENOMEM);
 719
 720         ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 721         if (ret) {
 722                 kfree(ft);
 723                 return ERR_PTR(ret);
 724         }
 725
 726         ft->level = level;
 727         ft->node.type = FS_TYPE_FLOW_TABLE;
 728         ft->op_mod = op_mod;
 729         ft->type = table_type;
 730         ft->vport = vport;
 731         ft->max_fte = max_fte;
 732         ft->flags = flags;
 733         INIT_LIST_HEAD(&ft->fwd_rules);
 734         mutex_init(&ft->lock);
 735
 736         return ft;
 737 }
 738
 739 /* If reverse is false, then we search for the first flow table in the
 740  * root sub-tree from start(closest from right), else we search for the
 741  * last flow table in the root sub-tree till start(closest from left).
 742  */
 743 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 744                                                          struct list_head *start,
 745                                                          bool reverse)
 746 {
 747 #define list_advance_entry(pos, reverse)                \
 748         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 749
 750 #define list_for_each_advance_continue(pos, head, reverse)      \
 751         for (pos = list_advance_entry(pos, reverse);            \
 752              &pos->list != (head);                              \
 753              pos = list_advance_entry(pos, reverse))
 754
 755         struct fs_node *iter = list_entry(start, struct fs_node, list);
 756         struct mlx5_flow_table *ft = NULL;
 757
 758         if (!root || root->type == FS_TYPE_PRIO_CHAINS)
 759                 return NULL;
 760
 761         list_for_each_advance_continue(iter, &root->children, reverse) {
 762                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 763                         fs_get_obj(ft, iter);
 764                         return ft;
 765                 }
 766                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 767                 if (ft)
 768                         return ft;
 769         }
 770
 771         return ft;
 772 }
 773
 774 /* If reverse if false then return the first flow table in next priority of
 775  * prio in the tree, else return the last flow table in the previous priority
 776  * of prio in the tree.
 777  */
 778 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 779 {
 780         struct mlx5_flow_table *ft = NULL;
 781         struct fs_node *curr_node;
 782         struct fs_node *parent;
 783
 784         parent = prio->node.parent;
 785         curr_node = &prio->node;
 786         while (!ft && parent) {
 787                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 788                 curr_node = parent;
 789                 parent = curr_node->parent;
 790         }
 791         return ft;
 792 }
 793
 794 /* Assuming all the tree is locked by mutex chain lock */
 795 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 796 {
 797         return find_closest_ft(prio, false);
 798 }
 799
 800 /* Assuming all the tree is locked by mutex chain lock */
 801 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 802 {
 803         return find_closest_ft(prio, true);
 804 }
 805
 806 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 807                                struct fs_prio *prio,
 808                                struct mlx5_flow_table *ft)
 809 {
 810         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 811         struct mlx5_flow_table *iter;
 812         int i = 0;
 813         int err;
 814
 815         fs_for_each_ft(iter, prio) {
 816                 i++;
 817                 err = root->cmds->modify_flow_table(root, iter, ft);
 818                 if (err) {
 819                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 820                                        iter->id);
 821                         /* The driver is out of sync with the FW */
 822                         if (i > 1)
 823                                 WARN_ON(true);
 824                         return err;
 825                 }
 826         }
 827         return 0;
 828 }
 829
 830 /* Connect flow tables from previous priority of prio to ft */
 831 static int connect_prev_fts(struct mlx5_core_dev *dev,
 832                             struct mlx5_flow_table *ft,
 833                             struct fs_prio *prio)
 834 {
 835         struct mlx5_flow_table *prev_ft;
 836
 837         prev_ft = find_prev_chained_ft(prio);
 838         if (prev_ft) {
 839                 struct fs_prio *prev_prio;
 840
 841                 fs_get_obj(prev_prio, prev_ft->node.parent);
 842                 return connect_fts_in_prio(dev, prev_prio, ft);
 843         }
 844         return 0;
 845 }
 846
 847 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 848                                  *prio)
 849 {
 850         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 851         struct mlx5_ft_underlay_qp *uqp;
 852         int min_level = INT_MAX;
 853         int err = 0;
 854         u32 qpn;
 855
 856         if (root->root_ft)
 857                 min_level = root->root_ft->level;
 858
 859         if (ft->level >= min_level)
 860                 return 0;
 861
 862         if (list_empty(&root->underlay_qpns)) {
 863                 /* Don't set any QPN (zero) in case QPN list is empty */
 864                 qpn = 0;
 865                 err = root->cmds->update_root_ft(root, ft, qpn, false);
 866         } else {
 867                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
 868                         qpn = uqp->qpn;
 869                         err = root->cmds->update_root_ft(root, ft,
 870                                                          qpn, false);
 871                         if (err)
 872                                 break;
 873                 }
 874         }
 875
 876         if (err)
 877                 mlx5_core_warn(root->dev,
 878                                "Update root flow table of id(%u) qpn(%d) failed\n",
 879                                ft->id, qpn);
 880         else
 881                 root->root_ft = ft;
 882
 883         return err;
 884 }
 885
 886 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 887                                          struct mlx5_flow_destination *dest)
 888 {
 889         struct mlx5_flow_root_namespace *root;
 890         struct mlx5_flow_table *ft;
 891         struct mlx5_flow_group *fg;
 892         struct fs_fte *fte;
 893         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 894         int err = 0;
 895
 896         fs_get_obj(fte, rule->node.parent);
 897         if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 898                 return -EINVAL;
 899         down_write_ref_node(&fte->node, false);
 900         fs_get_obj(fg, fte->node.parent);
 901         fs_get_obj(ft, fg->node.parent);
 902
 903         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 904         root = find_root(&ft->node);
 905         err = root->cmds->update_fte(root, ft, fg,
 906                                      modify_mask, fte);
 907         up_write_ref_node(&fte->node, false);
 908
 909         return err;
 910 }
 911
 912 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
 913                                  struct mlx5_flow_destination *new_dest,
 914                                  struct mlx5_flow_destination *old_dest)
 915 {
 916         int i;
 917
 918         if (!old_dest) {
 919                 if (handle->num_rules != 1)
 920                         return -EINVAL;
 921                 return _mlx5_modify_rule_destination(handle->rule[0],
 922                                                      new_dest);
 923         }
 924
 925         for (i = 0; i < handle->num_rules; i++) {
 926                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
 927                         return _mlx5_modify_rule_destination(handle->rule[i],
 928                                                              new_dest);
 929         }
 930
 931         return -EINVAL;
 932 }
 933
 934 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 935 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 936                              struct mlx5_flow_table *new_next_ft,
 937                              struct mlx5_flow_table *old_next_ft)
 938 {
 939         struct mlx5_flow_destination dest = {};
 940         struct mlx5_flow_rule *iter;
 941         int err = 0;
 942
 943         /* new_next_ft and old_next_ft could be NULL only
 944          * when we create/destroy the anchor flow table.
 945          */
 946         if (!new_next_ft || !old_next_ft)
 947                 return 0;
 948
 949         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 950         dest.ft = new_next_ft;
 951
 952         mutex_lock(&old_next_ft->lock);
 953         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 954         mutex_unlock(&old_next_ft->lock);
 955         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 956                 err = _mlx5_modify_rule_destination(iter, &dest);
 957                 if (err)
 958                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 959                                new_next_ft->id);
 960         }
 961         return 0;
 962 }
 963
 964 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 965                               struct fs_prio *prio)
 966 {
 967         struct mlx5_flow_table *next_ft;
 968         int err = 0;
 969
 970         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 971
 972         if (list_empty(&prio->node.children)) {
 973                 err = connect_prev_fts(dev, ft, prio);
 974                 if (err)
 975                         return err;
 976
 977                 next_ft = find_next_chained_ft(prio);
 978                 err = connect_fwd_rules(dev, ft, next_ft);
 979                 if (err)
 980                         return err;
 981         }
 982
 983         if (MLX5_CAP_FLOWTABLE(dev,
 984                                flow_table_properties_nic_receive.modify_root))
 985                 err = update_root_ft_create(ft, prio);
 986         return err;
 987 }
 988
 989 static void list_add_flow_table(struct mlx5_flow_table *ft,
 990                                 struct fs_prio *prio)
 991 {
 992         struct list_head *prev = &prio->node.children;
 993         struct mlx5_flow_table *iter;
 994
 995         fs_for_each_ft(iter, prio) {
 996                 if (iter->level > ft->level)
 997                         break;
 998                 prev = &iter->node.list;
 999         }
1000         list_add(&ft->node.list, prev);
1001 }
1002
1003 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1004                                                         struct mlx5_flow_table_attr *ft_attr,
1005                                                         enum fs_flow_table_op_mod op_mod,
1006                                                         u16 vport)
1007 {
1008         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1009         struct mlx5_flow_table *next_ft = NULL;
1010         struct fs_prio *fs_prio = NULL;
1011         struct mlx5_flow_table *ft;
1012         int log_table_sz;
1013         int err;
1014
1015         if (!root) {
1016                 pr_err("mlx5: flow steering failed to find root of namespace\n");
1017                 return ERR_PTR(-ENODEV);
1018         }
1019
1020         mutex_lock(&root->chain_lock);
1021         fs_prio = find_prio(ns, ft_attr->prio);
1022         if (!fs_prio) {
1023                 err = -EINVAL;
1024                 goto unlock_root;
1025         }
1026         if (ft_attr->level >= fs_prio->num_levels) {
1027                 err = -ENOSPC;
1028                 goto unlock_root;
1029         }
1030         /* The level is related to the
1031          * priority level range.
1032          */
1033         ft_attr->level += fs_prio->start_level;
1034         ft = alloc_flow_table(ft_attr->level,
1035                               vport,
1036                               ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
1037                               root->table_type,
1038                               op_mod, ft_attr->flags);
1039         if (IS_ERR(ft)) {
1040                 err = PTR_ERR(ft);
1041                 goto unlock_root;
1042         }
1043
1044         tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1045         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
1046         next_ft = find_next_chained_ft(fs_prio);
1047         ft->def_miss_action = ns->def_miss_action;
1048         err = root->cmds->create_flow_table(root, ft, log_table_sz, next_ft);
1049         if (err)
1050                 goto free_ft;
1051
1052         err = connect_flow_table(root->dev, ft, fs_prio);
1053         if (err)
1054                 goto destroy_ft;
1055         ft->node.active = true;
1056         down_write_ref_node(&fs_prio->node, false);
1057         tree_add_node(&ft->node, &fs_prio->node);
1058         list_add_flow_table(ft, fs_prio);
1059         fs_prio->num_ft++;
1060         up_write_ref_node(&fs_prio->node, false);
1061         mutex_unlock(&root->chain_lock);
1062         trace_mlx5_fs_add_ft(ft);
1063         return ft;
1064 destroy_ft:
1065         root->cmds->destroy_flow_table(root, ft);
1066 free_ft:
1067         kfree(ft);
1068 unlock_root:
1069         mutex_unlock(&root->chain_lock);
1070         return ERR_PTR(err);
1071 }
1072
1073 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1074                                                struct mlx5_flow_table_attr *ft_attr)
1075 {
1076         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1077 }
1078
1079 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1080                                                      int prio, int max_fte,
1081                                                      u32 level, u16 vport)
1082 {
1083         struct mlx5_flow_table_attr ft_attr = {};
1084
1085         ft_attr.max_fte = max_fte;
1086         ft_attr.level   = level;
1087         ft_attr.prio    = prio;
1088
1089         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1090 }
1091
1092 struct mlx5_flow_table*
1093 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1094                                  int prio, u32 level)
1095 {
1096         struct mlx5_flow_table_attr ft_attr = {};
1097
1098         ft_attr.level = level;
1099         ft_attr.prio  = prio;
1100         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1101 }
1102 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1103
1104 struct mlx5_flow_table*
1105 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1106                                     int prio,
1107                                     int num_flow_table_entries,
1108                                     int max_num_groups,
1109                                     u32 level,
1110                                     u32 flags)
1111 {
1112         struct mlx5_flow_table_attr ft_attr = {};
1113         struct mlx5_flow_table *ft;
1114
1115         if (max_num_groups > num_flow_table_entries)
1116                 return ERR_PTR(-EINVAL);
1117
1118         ft_attr.max_fte = num_flow_table_entries;
1119         ft_attr.prio    = prio;
1120         ft_attr.level   = level;
1121         ft_attr.flags   = flags;
1122
1123         ft = mlx5_create_flow_table(ns, &ft_attr);
1124         if (IS_ERR(ft))
1125                 return ft;
1126
1127         ft->autogroup.active = true;
1128         ft->autogroup.required_groups = max_num_groups;
1129         /* We save place for flow groups in addition to max types */
1130         ft->autogroup.group_size = ft->max_fte / (max_num_groups + 1);
1131
1132         return ft;
1133 }
1134 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1135
1136 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1137                                                u32 *fg_in)
1138 {
1139         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1140         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1141                                             fg_in, match_criteria);
1142         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1143                                             fg_in,
1144                                             match_criteria_enable);
1145         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1146                                    start_flow_index);
1147         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1148                                  end_flow_index);
1149         struct mlx5_flow_group *fg;
1150         int err;
1151
1152         if (ft->autogroup.active)
1153                 return ERR_PTR(-EPERM);
1154
1155         down_write_ref_node(&ft->node, false);
1156         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1157                                      start_index, end_index,
1158                                      ft->node.children.prev);
1159         up_write_ref_node(&ft->node, false);
1160         if (IS_ERR(fg))
1161                 return fg;
1162
1163         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1164         if (err) {
1165                 tree_put_node(&fg->node, false);
1166                 return ERR_PTR(err);
1167         }
1168         trace_mlx5_fs_add_fg(fg);
1169         fg->node.active = true;
1170
1171         return fg;
1172 }
1173
1174 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1175 {
1176         struct mlx5_flow_rule *rule;
1177
1178         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1179         if (!rule)
1180                 return NULL;
1181
1182         INIT_LIST_HEAD(&rule->next_ft);
1183         rule->node.type = FS_TYPE_FLOW_DEST;
1184         if (dest)
1185                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1186
1187         return rule;
1188 }
1189
1190 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1191 {
1192         struct mlx5_flow_handle *handle;
1193
1194         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1195         if (!handle)
1196                 return NULL;
1197
1198         handle->num_rules = num_rules;
1199
1200         return handle;
1201 }
1202
1203 static void destroy_flow_handle(struct fs_fte *fte,
1204                                 struct mlx5_flow_handle *handle,
1205                                 struct mlx5_flow_destination *dest,
1206                                 int i)
1207 {
1208         for (; --i >= 0;) {
1209                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1210                         fte->dests_size--;
1211                         list_del(&handle->rule[i]->node.list);
1212                         kfree(handle->rule[i]);
1213                 }
1214         }
1215         kfree(handle);
1216 }
1217
1218 static struct mlx5_flow_handle *
1219 create_flow_handle(struct fs_fte *fte,
1220                    struct mlx5_flow_destination *dest,
1221                    int dest_num,
1222                    int *modify_mask,
1223                    bool *new_rule)
1224 {
1225         struct mlx5_flow_handle *handle;
1226         struct mlx5_flow_rule *rule = NULL;
1227         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1228         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1229         int type;
1230         int i = 0;
1231
1232         handle = alloc_handle((dest_num) ? dest_num : 1);
1233         if (!handle)
1234                 return ERR_PTR(-ENOMEM);
1235
1236         do {
1237                 if (dest) {
1238                         rule = find_flow_rule(fte, dest + i);
1239                         if (rule) {
1240                                 refcount_inc(&rule->node.refcount);
1241                                 goto rule_found;
1242                         }
1243                 }
1244
1245                 *new_rule = true;
1246                 rule = alloc_rule(dest + i);
1247                 if (!rule)
1248                         goto free_rules;
1249
1250                 /* Add dest to dests list- we need flow tables to be in the
1251                  * end of the list for forward to next prio rules.
1252                  */
1253                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1254                 if (dest &&
1255                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1256                         list_add(&rule->node.list, &fte->node.children);
1257                 else
1258                         list_add_tail(&rule->node.list, &fte->node.children);
1259                 if (dest) {
1260                         fte->dests_size++;
1261
1262                         type = dest[i].type ==
1263                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1264                         *modify_mask |= type ? count : dst;
1265                 }
1266 rule_found:
1267                 handle->rule[i] = rule;
1268         } while (++i < dest_num);
1269
1270         return handle;
1271
1272 free_rules:
1273         destroy_flow_handle(fte, handle, dest, i);
1274         return ERR_PTR(-ENOMEM);
1275 }
1276
1277 /* fte should not be deleted while calling this function */
1278 static struct mlx5_flow_handle *
1279 add_rule_fte(struct fs_fte *fte,
1280              struct mlx5_flow_group *fg,
1281              struct mlx5_flow_destination *dest,
1282              int dest_num,
1283              bool update_action)
1284 {
1285         struct mlx5_flow_root_namespace *root;
1286         struct mlx5_flow_handle *handle;
1287         struct mlx5_flow_table *ft;
1288         int modify_mask = 0;
1289         int err;
1290         bool new_rule = false;
1291
1292         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1293                                     &new_rule);
1294         if (IS_ERR(handle) || !new_rule)
1295                 goto out;
1296
1297         if (update_action)
1298                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1299
1300         fs_get_obj(ft, fg->node.parent);
1301         root = find_root(&fg->node);
1302         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1303                 err = root->cmds->create_fte(root, ft, fg, fte);
1304         else
1305                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1306         if (err)
1307                 goto free_handle;
1308
1309         fte->node.active = true;
1310         fte->status |= FS_FTE_STATUS_EXISTING;
1311         atomic_inc(&fte->node.version);
1312
1313 out:
1314         return handle;
1315
1316 free_handle:
1317         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1318         return ERR_PTR(err);
1319 }
1320
1321 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1322                                                      const struct mlx5_flow_spec *spec)
1323 {
1324         struct list_head *prev = &ft->node.children;
1325         struct mlx5_flow_group *fg;
1326         unsigned int candidate_index = 0;
1327         unsigned int group_size = 0;
1328
1329         if (!ft->autogroup.active)
1330                 return ERR_PTR(-ENOENT);
1331
1332         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1333                 group_size = ft->autogroup.group_size;
1334
1335         /*  ft->max_fte == ft->autogroup.max_types */
1336         if (group_size == 0)
1337                 group_size = 1;
1338
1339         /* sorted by start_index */
1340         fs_for_each_fg(fg, ft) {
1341                 if (candidate_index + group_size > fg->start_index)
1342                         candidate_index = fg->start_index + fg->max_ftes;
1343                 else
1344                         break;
1345                 prev = &fg->node.list;
1346         }
1347
1348         if (candidate_index + group_size > ft->max_fte)
1349                 return ERR_PTR(-ENOSPC);
1350
1351         fg = alloc_insert_flow_group(ft,
1352                                      spec->match_criteria_enable,
1353                                      spec->match_criteria,
1354                                      candidate_index,
1355                                      candidate_index + group_size - 1,
1356                                      prev);
1357         if (IS_ERR(fg))
1358                 goto out;
1359
1360         if (group_size == ft->autogroup.group_size)
1361                 ft->autogroup.num_groups++;
1362
1363 out:
1364         return fg;
1365 }
1366
1367 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1368                                   struct mlx5_flow_group *fg)
1369 {
1370         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1371         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1372         void *match_criteria_addr;
1373         u8 src_esw_owner_mask_on;
1374         void *misc;
1375         int err;
1376         u32 *in;
1377
1378         in = kvzalloc(inlen, GFP_KERNEL);
1379         if (!in)
1380                 return -ENOMEM;
1381
1382         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1383                  fg->mask.match_criteria_enable);
1384         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1385         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1386                  fg->max_ftes - 1);
1387
1388         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1389                             misc_parameters);
1390         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1391                                          source_eswitch_owner_vhca_id);
1392         MLX5_SET(create_flow_group_in, in,
1393                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1394
1395         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1396                                            in, match_criteria);
1397         memcpy(match_criteria_addr, fg->mask.match_criteria,
1398                sizeof(fg->mask.match_criteria));
1399
1400         err = root->cmds->create_flow_group(root, ft, in, fg);
1401         if (!err) {
1402                 fg->node.active = true;
1403                 trace_mlx5_fs_add_fg(fg);
1404         }
1405
1406         kvfree(in);
1407         return err;
1408 }
1409
1410 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1411                                 struct mlx5_flow_destination *d2)
1412 {
1413         if (d1->type == d2->type) {
1414                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1415                      d1->vport.num == d2->vport.num &&
1416                      d1->vport.flags == d2->vport.flags &&
1417                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1418                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1419                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1420                       (d1->vport.pkt_reformat->id ==
1421                        d2->vport.pkt_reformat->id) : true)) ||
1422                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1423                      d1->ft == d2->ft) ||
1424                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1425                      d1->tir_num == d2->tir_num) ||
1426                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1427                      d1->ft_num == d2->ft_num))
1428                         return true;
1429         }
1430
1431         return false;
1432 }
1433
1434 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1435                                              struct mlx5_flow_destination *dest)
1436 {
1437         struct mlx5_flow_rule *rule;
1438
1439         list_for_each_entry(rule, &fte->node.children, node.list) {
1440                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1441                         return rule;
1442         }
1443         return NULL;
1444 }
1445
1446 static bool check_conflicting_actions(u32 action1, u32 action2)
1447 {
1448         u32 xored_actions = action1 ^ action2;
1449
1450         /* if one rule only wants to count, it's ok */
1451         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1452             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1453                 return false;
1454
1455         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1456                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1457                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1458                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1459                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1460                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1461                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1462                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1463                 return true;
1464
1465         return false;
1466 }
1467
1468 static int check_conflicting_ftes(struct fs_fte *fte,
1469                                   const struct mlx5_flow_context *flow_context,
1470                                   const struct mlx5_flow_act *flow_act)
1471 {
1472         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1473                 mlx5_core_warn(get_dev(&fte->node),
1474                                "Found two FTEs with conflicting actions\n");
1475                 return -EEXIST;
1476         }
1477
1478         if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1479             fte->flow_context.flow_tag != flow_context->flow_tag) {
1480                 mlx5_core_warn(get_dev(&fte->node),
1481                                "FTE flow tag %u already exists with different flow tag %u\n",
1482                                fte->flow_context.flow_tag,
1483                                flow_context->flow_tag);
1484                 return -EEXIST;
1485         }
1486
1487         return 0;
1488 }
1489
1490 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1491                                             const struct mlx5_flow_spec *spec,
1492                                             struct mlx5_flow_act *flow_act,
1493                                             struct mlx5_flow_destination *dest,
1494                                             int dest_num,
1495                                             struct fs_fte *fte)
1496 {
1497         struct mlx5_flow_handle *handle;
1498         int old_action;
1499         int i;
1500         int ret;
1501
1502         ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1503         if (ret)
1504                 return ERR_PTR(ret);
1505
1506         old_action = fte->action.action;
1507         fte->action.action |= flow_act->action;
1508         handle = add_rule_fte(fte, fg, dest, dest_num,
1509                               old_action != flow_act->action);
1510         if (IS_ERR(handle)) {
1511                 fte->action.action = old_action;
1512                 return handle;
1513         }
1514         trace_mlx5_fs_set_fte(fte, false);
1515
1516         for (i = 0; i < handle->num_rules; i++) {
1517                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1518                         tree_add_node(&handle->rule[i]->node, &fte->node);
1519                         trace_mlx5_fs_add_rule(handle->rule[i]);
1520                 }
1521         }
1522         return handle;
1523 }
1524
1525 static bool counter_is_valid(u32 action)
1526 {
1527         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1528                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1529 }
1530
1531 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1532                           u32 action,
1533                           struct mlx5_flow_table *ft)
1534 {
1535         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1536                 return counter_is_valid(action);
1537
1538         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1539                 return true;
1540
1541         if (!dest || ((dest->type ==
1542             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1543             (dest->ft->level <= ft->level)))
1544                 return false;
1545         return true;
1546 }
1547
1548 struct match_list {
1549         struct list_head        list;
1550         struct mlx5_flow_group *g;
1551 };
1552
1553 struct match_list_head {
1554         struct list_head  list;
1555         struct match_list first;
1556 };
1557
1558 static void free_match_list(struct match_list_head *head, bool ft_locked)
1559 {
1560         if (!list_empty(&head->list)) {
1561                 struct match_list *iter, *match_tmp;
1562
1563                 list_del(&head->first.list);
1564                 tree_put_node(&head->first.g->node, ft_locked);
1565                 list_for_each_entry_safe(iter, match_tmp, &head->list,
1566                                          list) {
1567                         tree_put_node(&iter->g->node, ft_locked);
1568                         list_del(&iter->list);
1569                         kfree(iter);
1570                 }
1571         }
1572 }
1573
1574 static int build_match_list(struct match_list_head *match_head,
1575                             struct mlx5_flow_table *ft,
1576                             const struct mlx5_flow_spec *spec,
1577                             bool ft_locked)
1578 {
1579         struct rhlist_head *tmp, *list;
1580         struct mlx5_flow_group *g;
1581         int err = 0;
1582
1583         rcu_read_lock();
1584         INIT_LIST_HEAD(&match_head->list);
1585         /* Collect all fgs which has a matching match_criteria */
1586         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1587         /* RCU is atomic, we can't execute FW commands here */
1588         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1589                 struct match_list *curr_match;
1590
1591                 if (likely(list_empty(&match_head->list))) {
1592                         if (!tree_get_node(&g->node))
1593                                 continue;
1594                         match_head->first.g = g;
1595                         list_add_tail(&match_head->first.list,
1596                                       &match_head->list);
1597                         continue;
1598                 }
1599
1600                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1601                 if (!curr_match) {
1602                         free_match_list(match_head, ft_locked);
1603                         err = -ENOMEM;
1604                         goto out;
1605                 }
1606                 if (!tree_get_node(&g->node)) {
1607                         kfree(curr_match);
1608                         continue;
1609                 }
1610                 curr_match->g = g;
1611                 list_add_tail(&curr_match->list, &match_head->list);
1612         }
1613 out:
1614         rcu_read_unlock();
1615         return err;
1616 }
1617
1618 static u64 matched_fgs_get_version(struct list_head *match_head)
1619 {
1620         struct match_list *iter;
1621         u64 version = 0;
1622
1623         list_for_each_entry(iter, match_head, list)
1624                 version += (u64)atomic_read(&iter->g->node.version);
1625         return version;
1626 }
1627
1628 static struct fs_fte *
1629 lookup_fte_locked(struct mlx5_flow_group *g,
1630                   const u32 *match_value,
1631                   bool take_write)
1632 {
1633         struct fs_fte *fte_tmp;
1634
1635         if (take_write)
1636                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1637         else
1638                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1639         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1640                                          rhash_fte);
1641         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1642                 fte_tmp = NULL;
1643                 goto out;
1644         }
1645         if (!fte_tmp->node.active) {
1646                 tree_put_node(&fte_tmp->node, false);
1647                 fte_tmp = NULL;
1648                 goto out;
1649         }
1650
1651         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1652 out:
1653         if (take_write)
1654                 up_write_ref_node(&g->node, false);
1655         else
1656                 up_read_ref_node(&g->node);
1657         return fte_tmp;
1658 }
1659
1660 static struct mlx5_flow_handle *
1661 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1662                        struct list_head *match_head,
1663                        const struct mlx5_flow_spec *spec,
1664                        struct mlx5_flow_act *flow_act,
1665                        struct mlx5_flow_destination *dest,
1666                        int dest_num,
1667                        int ft_version)
1668 {
1669         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1670         struct mlx5_flow_group *g;
1671         struct mlx5_flow_handle *rule;
1672         struct match_list *iter;
1673         bool take_write = false;
1674         struct fs_fte *fte;
1675         u64  version;
1676         int err;
1677
1678         fte = alloc_fte(ft, spec, flow_act);
1679         if (IS_ERR(fte))
1680                 return  ERR_PTR(-ENOMEM);
1681
1682 search_again_locked:
1683         version = matched_fgs_get_version(match_head);
1684         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1685                 goto skip_search;
1686         /* Try to find a fg that already contains a matching fte */
1687         list_for_each_entry(iter, match_head, list) {
1688                 struct fs_fte *fte_tmp;
1689
1690                 g = iter->g;
1691                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1692                 if (!fte_tmp)
1693                         continue;
1694                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1695                 up_write_ref_node(&fte_tmp->node, false);
1696                 tree_put_node(&fte_tmp->node, false);
1697                 kmem_cache_free(steering->ftes_cache, fte);
1698                 return rule;
1699         }
1700
1701 skip_search:
1702         /* No group with matching fte found, or we skipped the search.
1703          * Try to add a new fte to any matching fg.
1704          */
1705
1706         /* Check the ft version, for case that new flow group
1707          * was added while the fgs weren't locked
1708          */
1709         if (atomic_read(&ft->node.version) != ft_version) {
1710                 rule = ERR_PTR(-EAGAIN);
1711                 goto out;
1712         }
1713
1714         /* Check the fgs version, for case the new FTE with the
1715          * same values was added while the fgs weren't locked
1716          */
1717         if (version != matched_fgs_get_version(match_head)) {
1718                 take_write = true;
1719                 goto search_again_locked;
1720         }
1721
1722         list_for_each_entry(iter, match_head, list) {
1723                 g = iter->g;
1724
1725                 if (!g->node.active)
1726                         continue;
1727
1728                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1729
1730                 err = insert_fte(g, fte);
1731                 if (err) {
1732                         up_write_ref_node(&g->node, false);
1733                         if (err == -ENOSPC)
1734                                 continue;
1735                         kmem_cache_free(steering->ftes_cache, fte);
1736                         return ERR_PTR(err);
1737                 }
1738
1739                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1740                 up_write_ref_node(&g->node, false);
1741                 rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1742                 up_write_ref_node(&fte->node, false);
1743                 tree_put_node(&fte->node, false);
1744                 return rule;
1745         }
1746         rule = ERR_PTR(-ENOENT);
1747 out:
1748         kmem_cache_free(steering->ftes_cache, fte);
1749         return rule;
1750 }
1751
1752 static struct mlx5_flow_handle *
1753 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1754                      const struct mlx5_flow_spec *spec,
1755                      struct mlx5_flow_act *flow_act,
1756                      struct mlx5_flow_destination *dest,
1757                      int dest_num)
1758
1759 {
1760         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1761         struct mlx5_flow_group *g;
1762         struct mlx5_flow_handle *rule;
1763         struct match_list_head match_head;
1764         bool take_write = false;
1765         struct fs_fte *fte;
1766         int version;
1767         int err;
1768         int i;
1769
1770         if (!check_valid_spec(spec))
1771                 return ERR_PTR(-EINVAL);
1772
1773         for (i = 0; i < dest_num; i++) {
1774                 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1775                         return ERR_PTR(-EINVAL);
1776         }
1777         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1778 search_again_locked:
1779         version = atomic_read(&ft->node.version);
1780
1781         /* Collect all fgs which has a matching match_criteria */
1782         err = build_match_list(&match_head, ft, spec, take_write);
1783         if (err) {
1784                 if (take_write)
1785                         up_write_ref_node(&ft->node, false);
1786                 else
1787                         up_read_ref_node(&ft->node);
1788                 return ERR_PTR(err);
1789         }
1790
1791         if (!take_write)
1792                 up_read_ref_node(&ft->node);
1793
1794         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1795                                       dest_num, version);
1796         free_match_list(&match_head, take_write);
1797         if (!IS_ERR(rule) ||
1798             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1799                 if (take_write)
1800                         up_write_ref_node(&ft->node, false);
1801                 return rule;
1802         }
1803
1804         if (!take_write) {
1805                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1806                 take_write = true;
1807         }
1808
1809         if (PTR_ERR(rule) == -EAGAIN ||
1810             version != atomic_read(&ft->node.version))
1811                 goto search_again_locked;
1812
1813         g = alloc_auto_flow_group(ft, spec);
1814         if (IS_ERR(g)) {
1815                 rule = ERR_CAST(g);
1816                 up_write_ref_node(&ft->node, false);
1817                 return rule;
1818         }
1819
1820         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1821         up_write_ref_node(&ft->node, false);
1822
1823         err = create_auto_flow_group(ft, g);
1824         if (err)
1825                 goto err_release_fg;
1826
1827         fte = alloc_fte(ft, spec, flow_act);
1828         if (IS_ERR(fte)) {
1829                 err = PTR_ERR(fte);
1830                 goto err_release_fg;
1831         }
1832
1833         err = insert_fte(g, fte);
1834         if (err) {
1835                 kmem_cache_free(steering->ftes_cache, fte);
1836                 goto err_release_fg;
1837         }
1838
1839         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1840         up_write_ref_node(&g->node, false);
1841         rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1842         up_write_ref_node(&fte->node, false);
1843         tree_put_node(&fte->node, false);
1844         tree_put_node(&g->node, false);
1845         return rule;
1846
1847 err_release_fg:
1848         up_write_ref_node(&g->node, false);
1849         tree_put_node(&g->node, false);
1850         return ERR_PTR(err);
1851 }
1852
1853 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1854 {
1855         return ((ft->type == FS_FT_NIC_RX) &&
1856                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1857 }
1858
1859 struct mlx5_flow_handle *
1860 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1861                     const struct mlx5_flow_spec *spec,
1862                     struct mlx5_flow_act *flow_act,
1863                     struct mlx5_flow_destination *dest,
1864                     int num_dest)
1865 {
1866         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1867         struct mlx5_flow_destination gen_dest = {};
1868         struct mlx5_flow_table *next_ft = NULL;
1869         struct mlx5_flow_handle *handle = NULL;
1870         u32 sw_action = flow_act->action;
1871         struct fs_prio *prio;
1872
1873         fs_get_obj(prio, ft->node.parent);
1874         if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1875                 if (!fwd_next_prio_supported(ft))
1876                         return ERR_PTR(-EOPNOTSUPP);
1877                 if (num_dest)
1878                         return ERR_PTR(-EINVAL);
1879                 mutex_lock(&root->chain_lock);
1880                 next_ft = find_next_chained_ft(prio);
1881                 if (next_ft) {
1882                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1883                         gen_dest.ft = next_ft;
1884                         dest = &gen_dest;
1885                         num_dest = 1;
1886                         flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1887                 } else {
1888                         mutex_unlock(&root->chain_lock);
1889                         return ERR_PTR(-EOPNOTSUPP);
1890                 }
1891         }
1892
1893         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1894
1895         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1896                 if (!IS_ERR_OR_NULL(handle) &&
1897                     (list_empty(&handle->rule[0]->next_ft))) {
1898                         mutex_lock(&next_ft->lock);
1899                         list_add(&handle->rule[0]->next_ft,
1900                                  &next_ft->fwd_rules);
1901                         mutex_unlock(&next_ft->lock);
1902                         handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1903                 }
1904                 mutex_unlock(&root->chain_lock);
1905         }
1906         return handle;
1907 }
1908 EXPORT_SYMBOL(mlx5_add_flow_rules);
1909
1910 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1911 {
1912         struct fs_fte *fte;
1913         int i;
1914
1915         /* In order to consolidate the HW changes we lock the FTE for other
1916          * changes, and increase its refcount, in order not to perform the
1917          * "del" functions of the FTE. Will handle them here.
1918          * The removal of the rules is done under locked FTE.
1919          * After removing all the handle's rules, if there are remaining
1920          * rules, it means we just need to modify the FTE in FW, and
1921          * unlock/decrease the refcount we increased before.
1922          * Otherwise, it means the FTE should be deleted. First delete the
1923          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
1924          * the FTE, which will handle the last decrease of the refcount, as
1925          * well as required handling of its parent.
1926          */
1927         fs_get_obj(fte, handle->rule[0]->node.parent);
1928         down_write_ref_node(&fte->node, false);
1929         for (i = handle->num_rules - 1; i >= 0; i--)
1930                 tree_remove_node(&handle->rule[i]->node, true);
1931         if (fte->modify_mask && fte->dests_size) {
1932                 modify_fte(fte);
1933                 up_write_ref_node(&fte->node, false);
1934         } else {
1935                 del_hw_fte(&fte->node);
1936                 up_write(&fte->node.lock);
1937                 tree_put_node(&fte->node, false);
1938         }
1939         kfree(handle);
1940 }
1941 EXPORT_SYMBOL(mlx5_del_flow_rules);
1942
1943 /* Assuming prio->node.children(flow tables) is sorted by level */
1944 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1945 {
1946         struct fs_prio *prio;
1947
1948         fs_get_obj(prio, ft->node.parent);
1949
1950         if (!list_is_last(&ft->node.list, &prio->node.children))
1951                 return list_next_entry(ft, node.list);
1952         return find_next_chained_ft(prio);
1953 }
1954
1955 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1956 {
1957         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1958         struct mlx5_ft_underlay_qp *uqp;
1959         struct mlx5_flow_table *new_root_ft = NULL;
1960         int err = 0;
1961         u32 qpn;
1962
1963         if (root->root_ft != ft)
1964                 return 0;
1965
1966         new_root_ft = find_next_ft(ft);
1967         if (!new_root_ft) {
1968                 root->root_ft = NULL;
1969                 return 0;
1970         }
1971
1972         if (list_empty(&root->underlay_qpns)) {
1973                 /* Don't set any QPN (zero) in case QPN list is empty */
1974                 qpn = 0;
1975                 err = root->cmds->update_root_ft(root, new_root_ft,
1976                                                  qpn, false);
1977         } else {
1978                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
1979                         qpn = uqp->qpn;
1980                         err = root->cmds->update_root_ft(root,
1981                                                          new_root_ft, qpn,
1982                                                          false);
1983                         if (err)
1984                                 break;
1985                 }
1986         }
1987
1988         if (err)
1989                 mlx5_core_warn(root->dev,
1990                                "Update root flow table of id(%u) qpn(%d) failed\n",
1991                                ft->id, qpn);
1992         else
1993                 root->root_ft = new_root_ft;
1994
1995         return 0;
1996 }
1997
1998 /* Connect flow table from previous priority to
1999  * the next flow table.
2000  */
2001 static int disconnect_flow_table(struct mlx5_flow_table *ft)
2002 {
2003         struct mlx5_core_dev *dev = get_dev(&ft->node);
2004         struct mlx5_flow_table *next_ft;
2005         struct fs_prio *prio;
2006         int err = 0;
2007
2008         err = update_root_ft_destroy(ft);
2009         if (err)
2010                 return err;
2011
2012         fs_get_obj(prio, ft->node.parent);
2013         if  (!(list_first_entry(&prio->node.children,
2014                                 struct mlx5_flow_table,
2015                                 node.list) == ft))
2016                 return 0;
2017
2018         next_ft = find_next_chained_ft(prio);
2019         err = connect_fwd_rules(dev, next_ft, ft);
2020         if (err)
2021                 return err;
2022
2023         err = connect_prev_fts(dev, next_ft, prio);
2024         if (err)
2025                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2026                                ft->id);
2027         return err;
2028 }
2029
2030 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2031 {
2032         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2033         int err = 0;
2034
2035         mutex_lock(&root->chain_lock);
2036         err = disconnect_flow_table(ft);
2037         if (err) {
2038                 mutex_unlock(&root->chain_lock);
2039                 return err;
2040         }
2041         if (tree_remove_node(&ft->node, false))
2042                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2043                                ft->id);
2044         mutex_unlock(&root->chain_lock);
2045
2046         return err;
2047 }
2048 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2049
2050 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2051 {
2052         if (tree_remove_node(&fg->node, false))
2053                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2054                                fg->id);
2055 }
2056
2057 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2058                                                 int n)
2059 {
2060         struct mlx5_flow_steering *steering = dev->priv.steering;
2061
2062         if (!steering || !steering->fdb_sub_ns)
2063                 return NULL;
2064
2065         return steering->fdb_sub_ns[n];
2066 }
2067 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2068
2069 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2070                                                     enum mlx5_flow_namespace_type type)
2071 {
2072         struct mlx5_flow_steering *steering = dev->priv.steering;
2073         struct mlx5_flow_root_namespace *root_ns;
2074         int prio = 0;
2075         struct fs_prio *fs_prio;
2076         struct mlx5_flow_namespace *ns;
2077
2078         if (!steering)
2079                 return NULL;
2080
2081         switch (type) {
2082         case MLX5_FLOW_NAMESPACE_FDB:
2083                 if (steering->fdb_root_ns)
2084                         return &steering->fdb_root_ns->ns;
2085                 return NULL;
2086         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2087                 if (steering->sniffer_rx_root_ns)
2088                         return &steering->sniffer_rx_root_ns->ns;
2089                 return NULL;
2090         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2091                 if (steering->sniffer_tx_root_ns)
2092                         return &steering->sniffer_tx_root_ns->ns;
2093                 return NULL;
2094         default:
2095                 break;
2096         }
2097
2098         if (type == MLX5_FLOW_NAMESPACE_EGRESS) {
2099                 root_ns = steering->egress_root_ns;
2100         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX) {
2101                 root_ns = steering->rdma_rx_root_ns;
2102                 prio = RDMA_RX_BYPASS_PRIO;
2103         } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL) {
2104                 root_ns = steering->rdma_rx_root_ns;
2105                 prio = RDMA_RX_KERNEL_PRIO;
2106         } else { /* Must be NIC RX */
2107                 root_ns = steering->root_ns;
2108                 prio = type;
2109         }
2110
2111         if (!root_ns)
2112                 return NULL;
2113
2114         fs_prio = find_prio(&root_ns->ns, prio);
2115         if (!fs_prio)
2116                 return NULL;
2117
2118         ns = list_first_entry(&fs_prio->node.children,
2119                               typeof(*ns),
2120                               node.list);
2121
2122         return ns;
2123 }
2124 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2125
2126 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2127                                                               enum mlx5_flow_namespace_type type,
2128                                                               int vport)
2129 {
2130         struct mlx5_flow_steering *steering = dev->priv.steering;
2131
2132         if (!steering || vport >= mlx5_eswitch_get_total_vports(dev))
2133                 return NULL;
2134
2135         switch (type) {
2136         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2137                 if (steering->esw_egress_root_ns &&
2138                     steering->esw_egress_root_ns[vport])
2139                         return &steering->esw_egress_root_ns[vport]->ns;
2140                 else
2141                         return NULL;
2142         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2143                 if (steering->esw_ingress_root_ns &&
2144                     steering->esw_ingress_root_ns[vport])
2145                         return &steering->esw_ingress_root_ns[vport]->ns;
2146                 else
2147                         return NULL;
2148         default:
2149                 return NULL;
2150         }
2151 }
2152
2153 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2154                                        unsigned int prio,
2155                                        int num_levels,
2156                                        enum fs_node_type type)
2157 {
2158         struct fs_prio *fs_prio;
2159
2160         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2161         if (!fs_prio)
2162                 return ERR_PTR(-ENOMEM);
2163
2164         fs_prio->node.type = type;
2165         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2166         tree_add_node(&fs_prio->node, &ns->node);
2167         fs_prio->num_levels = num_levels;
2168         fs_prio->prio = prio;
2169         list_add_tail(&fs_prio->node.list, &ns->node.children);
2170
2171         return fs_prio;
2172 }
2173
2174 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2175                                               unsigned int prio,
2176                                               int num_levels)
2177 {
2178         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2179 }
2180
2181 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2182                                       unsigned int prio, int num_levels)
2183 {
2184         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2185 }
2186
2187 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2188                                                      *ns)
2189 {
2190         ns->node.type = FS_TYPE_NAMESPACE;
2191
2192         return ns;
2193 }
2194
2195 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2196                                                        int def_miss_act)
2197 {
2198         struct mlx5_flow_namespace      *ns;
2199
2200         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2201         if (!ns)
2202                 return ERR_PTR(-ENOMEM);
2203
2204         fs_init_namespace(ns);
2205         ns->def_miss_action = def_miss_act;
2206         tree_init_node(&ns->node, NULL, del_sw_ns);
2207         tree_add_node(&ns->node, &prio->node);
2208         list_add_tail(&ns->node.list, &prio->node.children);
2209
2210         return ns;
2211 }
2212
2213 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2214                              struct init_tree_node *prio_metadata)
2215 {
2216         struct fs_prio *fs_prio;
2217         int i;
2218
2219         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2220                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2221                 if (IS_ERR(fs_prio))
2222                         return PTR_ERR(fs_prio);
2223         }
2224         return 0;
2225 }
2226
2227 #define FLOW_TABLE_BIT_SZ 1
2228 #define GET_FLOW_TABLE_CAP(dev, offset) \
2229         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2230                         offset / 32)) >>                                        \
2231           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2232 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2233 {
2234         int i;
2235
2236         for (i = 0; i < caps->arr_sz; i++) {
2237                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2238                         return false;
2239         }
2240         return true;
2241 }
2242
2243 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2244                                     struct init_tree_node *init_node,
2245                                     struct fs_node *fs_parent_node,
2246                                     struct init_tree_node *init_parent_node,
2247                                     int prio)
2248 {
2249         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2250                                               flow_table_properties_nic_receive.
2251                                               max_ft_level);
2252         struct mlx5_flow_namespace *fs_ns;
2253         struct fs_prio *fs_prio;
2254         struct fs_node *base;
2255         int i;
2256         int err;
2257
2258         if (init_node->type == FS_TYPE_PRIO) {
2259                 if ((init_node->min_ft_level > max_ft_level) ||
2260                     !has_required_caps(steering->dev, &init_node->caps))
2261                         return 0;
2262
2263                 fs_get_obj(fs_ns, fs_parent_node);
2264                 if (init_node->num_leaf_prios)
2265                         return create_leaf_prios(fs_ns, prio, init_node);
2266                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2267                 if (IS_ERR(fs_prio))
2268                         return PTR_ERR(fs_prio);
2269                 base = &fs_prio->node;
2270         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2271                 fs_get_obj(fs_prio, fs_parent_node);
2272                 fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2273                 if (IS_ERR(fs_ns))
2274                         return PTR_ERR(fs_ns);
2275                 base = &fs_ns->node;
2276         } else {
2277                 return -EINVAL;
2278         }
2279         prio = 0;
2280         for (i = 0; i < init_node->ar_size; i++) {
2281                 err = init_root_tree_recursive(steering, &init_node->children[i],
2282                                                base, init_node, prio);
2283                 if (err)
2284                         return err;
2285                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2286                     init_node->children[i].num_leaf_prios) {
2287                         prio += init_node->children[i].num_leaf_prios;
2288                 }
2289         }
2290
2291         return 0;
2292 }
2293
2294 static int init_root_tree(struct mlx5_flow_steering *steering,
2295                           struct init_tree_node *init_node,
2296                           struct fs_node *fs_parent_node)
2297 {
2298         int i;
2299         struct mlx5_flow_namespace *fs_ns;
2300         int err;
2301
2302         fs_get_obj(fs_ns, fs_parent_node);
2303         for (i = 0; i < init_node->ar_size; i++) {
2304                 err = init_root_tree_recursive(steering, &init_node->children[i],
2305                                                &fs_ns->node,
2306                                                init_node, i);
2307                 if (err)
2308                         return err;
2309         }
2310         return 0;
2311 }
2312
2313 static struct mlx5_flow_root_namespace
2314 *create_root_ns(struct mlx5_flow_steering *steering,
2315                 enum fs_flow_table_type table_type)
2316 {
2317         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2318         struct mlx5_flow_root_namespace *root_ns;
2319         struct mlx5_flow_namespace *ns;
2320
2321         if (mlx5_accel_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2322             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2323                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2324
2325         /* Create the root namespace */
2326         root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2327         if (!root_ns)
2328                 return NULL;
2329
2330         root_ns->dev = steering->dev;
2331         root_ns->table_type = table_type;
2332         root_ns->cmds = cmds;
2333
2334         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2335
2336         ns = &root_ns->ns;
2337         fs_init_namespace(ns);
2338         mutex_init(&root_ns->chain_lock);
2339         tree_init_node(&ns->node, NULL, NULL);
2340         tree_add_node(&ns->node, NULL);
2341
2342         return root_ns;
2343 }
2344
2345 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2346
2347 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2348 {
2349         struct fs_prio *prio;
2350
2351         fs_for_each_prio(prio, ns) {
2352                  /* This updates prio start_level and num_levels */
2353                 set_prio_attrs_in_prio(prio, acc_level);
2354                 acc_level += prio->num_levels;
2355         }
2356         return acc_level;
2357 }
2358
2359 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2360 {
2361         struct mlx5_flow_namespace *ns;
2362         int acc_level_ns = acc_level;
2363
2364         prio->start_level = acc_level;
2365         fs_for_each_ns(ns, prio)
2366                 /* This updates start_level and num_levels of ns's priority descendants */
2367                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2368         if (!prio->num_levels)
2369                 prio->num_levels = acc_level_ns - prio->start_level;
2370         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2371 }
2372
2373 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2374 {
2375         struct mlx5_flow_namespace *ns = &root_ns->ns;
2376         struct fs_prio *prio;
2377         int start_level = 0;
2378
2379         fs_for_each_prio(prio, ns) {
2380                 set_prio_attrs_in_prio(prio, start_level);
2381                 start_level += prio->num_levels;
2382         }
2383 }
2384
2385 #define ANCHOR_PRIO 0
2386 #define ANCHOR_SIZE 1
2387 #define ANCHOR_LEVEL 0
2388 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2389 {
2390         struct mlx5_flow_namespace *ns = NULL;
2391         struct mlx5_flow_table_attr ft_attr = {};
2392         struct mlx5_flow_table *ft;
2393
2394         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2395         if (WARN_ON(!ns))
2396                 return -EINVAL;
2397
2398         ft_attr.max_fte = ANCHOR_SIZE;
2399         ft_attr.level   = ANCHOR_LEVEL;
2400         ft_attr.prio    = ANCHOR_PRIO;
2401
2402         ft = mlx5_create_flow_table(ns, &ft_attr);
2403         if (IS_ERR(ft)) {
2404                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2405                 return PTR_ERR(ft);
2406         }
2407         return 0;
2408 }
2409
2410 static int init_root_ns(struct mlx5_flow_steering *steering)
2411 {
2412         int err;
2413
2414         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2415         if (!steering->root_ns)
2416                 return -ENOMEM;
2417
2418         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2419         if (err)
2420                 goto out_err;
2421
2422         set_prio_attrs(steering->root_ns);
2423         err = create_anchor_flow_table(steering);
2424         if (err)
2425                 goto out_err;
2426
2427         return 0;
2428
2429 out_err:
2430         cleanup_root_ns(steering->root_ns);
2431         steering->root_ns = NULL;
2432         return err;
2433 }
2434
2435 static void clean_tree(struct fs_node *node)
2436 {
2437         if (node) {
2438                 struct fs_node *iter;
2439                 struct fs_node *temp;
2440
2441                 tree_get_node(node);
2442                 list_for_each_entry_safe(iter, temp, &node->children, list)
2443                         clean_tree(iter);
2444                 tree_put_node(node, false);
2445                 tree_remove_node(node, false);
2446         }
2447 }
2448
2449 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2450 {
2451         if (!root_ns)
2452                 return;
2453
2454         clean_tree(&root_ns->ns.node);
2455 }
2456
2457 static void cleanup_egress_acls_root_ns(struct mlx5_core_dev *dev)
2458 {
2459         struct mlx5_flow_steering *steering = dev->priv.steering;
2460         int i;
2461
2462         if (!steering->esw_egress_root_ns)
2463                 return;
2464
2465         for (i = 0; i < mlx5_eswitch_get_total_vports(dev); i++)
2466                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2467
2468         kfree(steering->esw_egress_root_ns);
2469         steering->esw_egress_root_ns = NULL;
2470 }
2471
2472 static void cleanup_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2473 {
2474         struct mlx5_flow_steering *steering = dev->priv.steering;
2475         int i;
2476
2477         if (!steering->esw_ingress_root_ns)
2478                 return;
2479
2480         for (i = 0; i < mlx5_eswitch_get_total_vports(dev); i++)
2481                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2482
2483         kfree(steering->esw_ingress_root_ns);
2484         steering->esw_ingress_root_ns = NULL;
2485 }
2486
2487 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2488 {
2489         struct mlx5_flow_steering *steering = dev->priv.steering;
2490
2491         cleanup_root_ns(steering->root_ns);
2492         cleanup_egress_acls_root_ns(dev);
2493         cleanup_ingress_acls_root_ns(dev);
2494         cleanup_root_ns(steering->fdb_root_ns);
2495         steering->fdb_root_ns = NULL;
2496         kfree(steering->fdb_sub_ns);
2497         steering->fdb_sub_ns = NULL;
2498         cleanup_root_ns(steering->sniffer_rx_root_ns);
2499         cleanup_root_ns(steering->sniffer_tx_root_ns);
2500         cleanup_root_ns(steering->rdma_rx_root_ns);
2501         cleanup_root_ns(steering->egress_root_ns);
2502         mlx5_cleanup_fc_stats(dev);
2503         kmem_cache_destroy(steering->ftes_cache);
2504         kmem_cache_destroy(steering->fgs_cache);
2505         kfree(steering);
2506 }
2507
2508 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2509 {
2510         struct fs_prio *prio;
2511
2512         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2513         if (!steering->sniffer_tx_root_ns)
2514                 return -ENOMEM;
2515
2516         /* Create single prio */
2517         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2518         return PTR_ERR_OR_ZERO(prio);
2519 }
2520
2521 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2522 {
2523         struct fs_prio *prio;
2524
2525         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2526         if (!steering->sniffer_rx_root_ns)
2527                 return -ENOMEM;
2528
2529         /* Create single prio */
2530         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2531         return PTR_ERR_OR_ZERO(prio);
2532 }
2533
2534 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2535 {
2536         int err;
2537
2538         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2539         if (!steering->rdma_rx_root_ns)
2540                 return -ENOMEM;
2541
2542         err = init_root_tree(steering, &rdma_rx_root_fs,
2543                              &steering->rdma_rx_root_ns->ns.node);
2544         if (err)
2545                 goto out_err;
2546
2547         set_prio_attrs(steering->rdma_rx_root_ns);
2548
2549         return 0;
2550
2551 out_err:
2552         cleanup_root_ns(steering->rdma_rx_root_ns);
2553         steering->rdma_rx_root_ns = NULL;
2554         return err;
2555 }
2556 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2557 {
2558         struct mlx5_flow_namespace *ns;
2559         struct fs_prio *maj_prio;
2560         struct fs_prio *min_prio;
2561         int levels;
2562         int chain;
2563         int prio;
2564         int err;
2565
2566         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2567         if (!steering->fdb_root_ns)
2568                 return -ENOMEM;
2569
2570         steering->fdb_sub_ns = kzalloc(sizeof(steering->fdb_sub_ns) *
2571                                        (FDB_MAX_CHAIN + 1), GFP_KERNEL);
2572         if (!steering->fdb_sub_ns)
2573                 return -ENOMEM;
2574
2575         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2576                                   1);
2577         if (IS_ERR(maj_prio)) {
2578                 err = PTR_ERR(maj_prio);
2579                 goto out_err;
2580         }
2581
2582         levels = 2 * FDB_MAX_PRIO * (FDB_MAX_CHAIN + 1);
2583         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2584                                           FDB_FAST_PATH,
2585                                           levels);
2586         if (IS_ERR(maj_prio)) {
2587                 err = PTR_ERR(maj_prio);
2588                 goto out_err;
2589         }
2590
2591         for (chain = 0; chain <= FDB_MAX_CHAIN; chain++) {
2592                 ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2593                 if (IS_ERR(ns)) {
2594                         err = PTR_ERR(ns);
2595                         goto out_err;
2596                 }
2597
2598                 for (prio = 0; prio < FDB_MAX_PRIO * (chain + 1); prio++) {
2599                         min_prio = fs_create_prio(ns, prio, 2);
2600                         if (IS_ERR(min_prio)) {
2601                                 err = PTR_ERR(min_prio);
2602                                 goto out_err;
2603                         }
2604                 }
2605
2606                 steering->fdb_sub_ns[chain] = ns;
2607         }
2608
2609         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2610         if (IS_ERR(maj_prio)) {
2611                 err = PTR_ERR(maj_prio);
2612                 goto out_err;
2613         }
2614
2615         set_prio_attrs(steering->fdb_root_ns);
2616         return 0;
2617
2618 out_err:
2619         cleanup_root_ns(steering->fdb_root_ns);
2620         kfree(steering->fdb_sub_ns);
2621         steering->fdb_sub_ns = NULL;
2622         steering->fdb_root_ns = NULL;
2623         return err;
2624 }
2625
2626 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2627 {
2628         struct fs_prio *prio;
2629
2630         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2631         if (!steering->esw_egress_root_ns[vport])
2632                 return -ENOMEM;
2633
2634         /* create 1 prio*/
2635         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2636         return PTR_ERR_OR_ZERO(prio);
2637 }
2638
2639 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2640 {
2641         struct fs_prio *prio;
2642
2643         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2644         if (!steering->esw_ingress_root_ns[vport])
2645                 return -ENOMEM;
2646
2647         /* create 1 prio*/
2648         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2649         return PTR_ERR_OR_ZERO(prio);
2650 }
2651
2652 static int init_egress_acls_root_ns(struct mlx5_core_dev *dev)
2653 {
2654         struct mlx5_flow_steering *steering = dev->priv.steering;
2655         int total_vports = mlx5_eswitch_get_total_vports(dev);
2656         int err;
2657         int i;
2658
2659         steering->esw_egress_root_ns =
2660                         kcalloc(total_vports,
2661                                 sizeof(*steering->esw_egress_root_ns),
2662                                 GFP_KERNEL);
2663         if (!steering->esw_egress_root_ns)
2664                 return -ENOMEM;
2665
2666         for (i = 0; i < total_vports; i++) {
2667                 err = init_egress_acl_root_ns(steering, i);
2668                 if (err)
2669                         goto cleanup_root_ns;
2670         }
2671
2672         return 0;
2673
2674 cleanup_root_ns:
2675         for (i--; i >= 0; i--)
2676                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2677         kfree(steering->esw_egress_root_ns);
2678         steering->esw_egress_root_ns = NULL;
2679         return err;
2680 }
2681
2682 static int init_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2683 {
2684         struct mlx5_flow_steering *steering = dev->priv.steering;
2685         int total_vports = mlx5_eswitch_get_total_vports(dev);
2686         int err;
2687         int i;
2688
2689         steering->esw_ingress_root_ns =
2690                         kcalloc(total_vports,
2691                                 sizeof(*steering->esw_ingress_root_ns),
2692                                 GFP_KERNEL);
2693         if (!steering->esw_ingress_root_ns)
2694                 return -ENOMEM;
2695
2696         for (i = 0; i < total_vports; i++) {
2697                 err = init_ingress_acl_root_ns(steering, i);
2698                 if (err)
2699                         goto cleanup_root_ns;
2700         }
2701
2702         return 0;
2703
2704 cleanup_root_ns:
2705         for (i--; i >= 0; i--)
2706                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2707         kfree(steering->esw_ingress_root_ns);
2708         steering->esw_ingress_root_ns = NULL;
2709         return err;
2710 }
2711
2712 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2713 {
2714         int err;
2715
2716         steering->egress_root_ns = create_root_ns(steering,
2717                                                   FS_FT_NIC_TX);
2718         if (!steering->egress_root_ns)
2719                 return -ENOMEM;
2720
2721         err = init_root_tree(steering, &egress_root_fs,
2722                              &steering->egress_root_ns->ns.node);
2723         if (err)
2724                 goto cleanup;
2725         set_prio_attrs(steering->egress_root_ns);
2726         return 0;
2727 cleanup:
2728         cleanup_root_ns(steering->egress_root_ns);
2729         steering->egress_root_ns = NULL;
2730         return err;
2731 }
2732
2733 int mlx5_init_fs(struct mlx5_core_dev *dev)
2734 {
2735         struct mlx5_flow_steering *steering;
2736         int err = 0;
2737
2738         err = mlx5_init_fc_stats(dev);
2739         if (err)
2740                 return err;
2741
2742         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2743         if (!steering)
2744                 return -ENOMEM;
2745         steering->dev = dev;
2746         dev->priv.steering = steering;
2747
2748         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2749                                                 sizeof(struct mlx5_flow_group), 0,
2750                                                 0, NULL);
2751         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2752                                                  0, NULL);
2753         if (!steering->ftes_cache || !steering->fgs_cache) {
2754                 err = -ENOMEM;
2755                 goto err;
2756         }
2757
2758         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2759               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2760              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2761               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2762             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2763                 err = init_root_ns(steering);
2764                 if (err)
2765                         goto err;
2766         }
2767
2768         if (MLX5_ESWITCH_MANAGER(dev)) {
2769                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2770                         err = init_fdb_root_ns(steering);
2771                         if (err)
2772                                 goto err;
2773                 }
2774                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2775                         err = init_egress_acls_root_ns(dev);
2776                         if (err)
2777                                 goto err;
2778                 }
2779                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2780                         err = init_ingress_acls_root_ns(dev);
2781                         if (err)
2782                                 goto err;
2783                 }
2784         }
2785
2786         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2787                 err = init_sniffer_rx_root_ns(steering);
2788                 if (err)
2789                         goto err;
2790         }
2791
2792         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2793                 err = init_sniffer_tx_root_ns(steering);
2794                 if (err)
2795                         goto err;
2796         }
2797
2798         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
2799             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
2800                 err = init_rdma_rx_root_ns(steering);
2801                 if (err)
2802                         goto err;
2803         }
2804
2805         if (MLX5_IPSEC_DEV(dev) || MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
2806                 err = init_egress_root_ns(steering);
2807                 if (err)
2808                         goto err;
2809         }
2810
2811         return 0;
2812 err:
2813         mlx5_cleanup_fs(dev);
2814         return err;
2815 }
2816
2817 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2818 {
2819         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2820         struct mlx5_ft_underlay_qp *new_uqp;
2821         int err = 0;
2822
2823         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
2824         if (!new_uqp)
2825                 return -ENOMEM;
2826
2827         mutex_lock(&root->chain_lock);
2828
2829         if (!root->root_ft) {
2830                 err = -EINVAL;
2831                 goto update_ft_fail;
2832         }
2833
2834         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
2835                                          false);
2836         if (err) {
2837                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
2838                                underlay_qpn, err);
2839                 goto update_ft_fail;
2840         }
2841
2842         new_uqp->qpn = underlay_qpn;
2843         list_add_tail(&new_uqp->list, &root->underlay_qpns);
2844
2845         mutex_unlock(&root->chain_lock);
2846
2847         return 0;
2848
2849 update_ft_fail:
2850         mutex_unlock(&root->chain_lock);
2851         kfree(new_uqp);
2852         return err;
2853 }
2854 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
2855
2856 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2857 {
2858         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2859         struct mlx5_ft_underlay_qp *uqp;
2860         bool found = false;
2861         int err = 0;
2862
2863         mutex_lock(&root->chain_lock);
2864         list_for_each_entry(uqp, &root->underlay_qpns, list) {
2865                 if (uqp->qpn == underlay_qpn) {
2866                         found = true;
2867                         break;
2868                 }
2869         }
2870
2871         if (!found) {
2872                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
2873                                underlay_qpn);
2874                 err = -EINVAL;
2875                 goto out;
2876         }
2877
2878         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
2879                                          true);
2880         if (err)
2881                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
2882                                underlay_qpn, err);
2883
2884         list_del(&uqp->list);
2885         mutex_unlock(&root->chain_lock);
2886         kfree(uqp);
2887
2888         return 0;
2889
2890 out:
2891         mutex_unlock(&root->chain_lock);
2892         return err;
2893 }
2894 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
2895
2896 static struct mlx5_flow_root_namespace
2897 *get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
2898 {
2899         struct mlx5_flow_namespace *ns;
2900
2901         if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
2902             ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
2903                 ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
2904         else
2905                 ns = mlx5_get_flow_namespace(dev, ns_type);
2906         if (!ns)
2907                 return NULL;
2908
2909         return find_root(&ns->node);
2910 }
2911
2912 struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
2913                                                  u8 ns_type, u8 num_actions,
2914                                                  void *modify_actions)
2915 {
2916         struct mlx5_flow_root_namespace *root;
2917         struct mlx5_modify_hdr *modify_hdr;
2918         int err;
2919
2920         root = get_root_namespace(dev, ns_type);
2921         if (!root)
2922                 return ERR_PTR(-EOPNOTSUPP);
2923
2924         modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
2925         if (!modify_hdr)
2926                 return ERR_PTR(-ENOMEM);
2927
2928         modify_hdr->ns_type = ns_type;
2929         err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
2930                                               modify_actions, modify_hdr);
2931         if (err) {
2932                 kfree(modify_hdr);
2933                 return ERR_PTR(err);
2934         }
2935
2936         return modify_hdr;
2937 }
2938 EXPORT_SYMBOL(mlx5_modify_header_alloc);
2939
2940 void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
2941                                 struct mlx5_modify_hdr *modify_hdr)
2942 {
2943         struct mlx5_flow_root_namespace *root;
2944
2945         root = get_root_namespace(dev, modify_hdr->ns_type);
2946         if (WARN_ON(!root))
2947                 return;
2948         root->cmds->modify_header_dealloc(root, modify_hdr);
2949         kfree(modify_hdr);
2950 }
2951 EXPORT_SYMBOL(mlx5_modify_header_dealloc);
2952
2953 struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
2954                                                      int reformat_type,
2955                                                      size_t size,
2956                                                      void *reformat_data,
2957                                                      enum mlx5_flow_namespace_type ns_type)
2958 {
2959         struct mlx5_pkt_reformat *pkt_reformat;
2960         struct mlx5_flow_root_namespace *root;
2961         int err;
2962
2963         root = get_root_namespace(dev, ns_type);
2964         if (!root)
2965                 return ERR_PTR(-EOPNOTSUPP);
2966
2967         pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
2968         if (!pkt_reformat)
2969                 return ERR_PTR(-ENOMEM);
2970
2971         pkt_reformat->ns_type = ns_type;
2972         pkt_reformat->reformat_type = reformat_type;
2973         err = root->cmds->packet_reformat_alloc(root, reformat_type, size,
2974                                                 reformat_data, ns_type,
2975                                                 pkt_reformat);
2976         if (err) {
2977                 kfree(pkt_reformat);
2978                 return ERR_PTR(err);
2979         }
2980
2981         return pkt_reformat;
2982 }
2983 EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
2984
2985 void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
2986                                   struct mlx5_pkt_reformat *pkt_reformat)
2987 {
2988         struct mlx5_flow_root_namespace *root;
2989
2990         root = get_root_namespace(dev, pkt_reformat->ns_type);
2991         if (WARN_ON(!root))
2992                 return;
2993         root->cmds->packet_reformat_dealloc(root, pkt_reformat);
2994         kfree(pkt_reformat);
2995 }
2996 EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
2997
2998 int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
2999                                  struct mlx5_flow_root_namespace *peer_ns)
3000 {
3001         if (peer_ns && ns->mode != peer_ns->mode) {
3002                 mlx5_core_err(ns->dev,
3003                               "Can't peer namespace of different steering mode\n");
3004                 return -EINVAL;
3005         }
3006
3007         return ns->cmds->set_peer(ns, peer_ns);
3008 }
3009
3010 /* This function should be called only at init stage of the namespace.
3011  * It is not safe to call this function while steering operations
3012  * are executed in the namespace.
3013  */
3014 int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3015                                  enum mlx5_flow_steering_mode mode)
3016 {
3017         struct mlx5_flow_root_namespace *root;
3018         const struct mlx5_flow_cmds *cmds;
3019         int err;
3020
3021         root = find_root(&ns->node);
3022         if (&root->ns != ns)
3023         /* Can't set cmds to non root namespace */
3024                 return -EINVAL;
3025
3026         if (root->table_type != FS_FT_FDB)
3027                 return -EOPNOTSUPP;
3028
3029         if (root->mode == mode)
3030                 return 0;
3031
3032         if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3033                 cmds = mlx5_fs_cmd_get_dr_cmds();
3034         else
3035                 cmds = mlx5_fs_cmd_get_fw_cmds();
3036         if (!cmds)
3037                 return -EOPNOTSUPP;
3038
3039         err = cmds->create_ns(root);
3040         if (err) {
3041                 mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3042                               err);
3043                 return err;
3044         }
3045
3046         root->cmds->destroy_ns(root);
3047         root->cmds = cmds;
3048         root->mode = mode;
3049
3050         return 0;
3051 }