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(...) {.type = FS_TYPE_NAMESPACE,\
  64         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  65         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  66 }
  67
  68 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  69                                    sizeof(long))
  70
  71 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  72
  73 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  74                                .caps = (long[]) {__VA_ARGS__} }
  75
  76 #define FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  77                                            FS_CAP(flow_table_properties_nic_receive.modify_root), \
  78                                            FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  79                                            FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  80
  81 #define FS_CHAINING_CAPS_EGRESS                                                \
  82         FS_REQUIRED_CAPS(                                                      \
  83                 FS_CAP(flow_table_properties_nic_transmit.flow_modify_en),     \
  84                 FS_CAP(flow_table_properties_nic_transmit.modify_root),        \
  85                 FS_CAP(flow_table_properties_nic_transmit                      \
  86                                .identified_miss_table_mode),                   \
  87                 FS_CAP(flow_table_properties_nic_transmit.flow_table_modify))
  88
  89 #define LEFTOVERS_NUM_LEVELS 1
  90 #define LEFTOVERS_NUM_PRIOS 1
  91
  92 #define BY_PASS_PRIO_NUM_LEVELS 1
  93 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  94                            LEFTOVERS_NUM_PRIOS)
  95
  96 #define ETHTOOL_PRIO_NUM_LEVELS 1
  97 #define ETHTOOL_NUM_PRIOS 11
  98 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
  99 /* Vlan, mac, ttc, inner ttc, aRFS */
 100 #define KERNEL_NIC_PRIO_NUM_LEVELS 5
 101 #define KERNEL_NIC_NUM_PRIOS 1
 102 /* One more level for tc */
 103 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
 104
 105 #define KERNEL_NIC_TC_NUM_PRIOS  1
 106 #define KERNEL_NIC_TC_NUM_LEVELS 2
 107
 108 #define ANCHOR_NUM_LEVELS 1
 109 #define ANCHOR_NUM_PRIOS 1
 110 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 111
 112 #define OFFLOADS_MAX_FT 1
 113 #define OFFLOADS_NUM_PRIOS 1
 114 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
 115
 116 #define LAG_PRIO_NUM_LEVELS 1
 117 #define LAG_NUM_PRIOS 1
 118 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 119
 120 struct node_caps {
 121         size_t  arr_sz;
 122         long    *caps;
 123 };
 124
 125 static struct init_tree_node {
 126         enum fs_node_type       type;
 127         struct init_tree_node *children;
 128         int ar_size;
 129         struct node_caps caps;
 130         int min_ft_level;
 131         int num_leaf_prios;
 132         int prio;
 133         int num_levels;
 134 } root_fs = {
 135         .type = FS_TYPE_NAMESPACE,
 136         .ar_size = 7,
 137         .children = (struct init_tree_node[]) {
 138                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 139                          FS_CHAINING_CAPS,
 140                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 141                                                   BY_PASS_PRIO_NUM_LEVELS))),
 142                 ADD_PRIO(0, LAG_MIN_LEVEL, 0,
 143                          FS_CHAINING_CAPS,
 144                          ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 145                                                   LAG_PRIO_NUM_LEVELS))),
 146                 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
 147                          ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
 148                 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
 149                          FS_CHAINING_CAPS,
 150                          ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 151                                                   ETHTOOL_PRIO_NUM_LEVELS))),
 152                 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 153                          ADD_NS(ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS, KERNEL_NIC_TC_NUM_LEVELS),
 154                                 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 155                                                   KERNEL_NIC_PRIO_NUM_LEVELS))),
 156                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 157                          FS_CHAINING_CAPS,
 158                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
 159                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 160                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
 161         }
 162 };
 163
 164 static struct init_tree_node egress_root_fs = {
 165         .type = FS_TYPE_NAMESPACE,
 166         .ar_size = 1,
 167         .children = (struct init_tree_node[]) {
 168                 ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 169                          FS_CHAINING_CAPS_EGRESS,
 170                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 171                                                   BY_PASS_PRIO_NUM_LEVELS))),
 172         }
 173 };
 174
 175 enum fs_i_lock_class {
 176         FS_LOCK_GRANDPARENT,
 177         FS_LOCK_PARENT,
 178         FS_LOCK_CHILD
 179 };
 180
 181 static const struct rhashtable_params rhash_fte = {
 182         .key_len = FIELD_SIZEOF(struct fs_fte, val),
 183         .key_offset = offsetof(struct fs_fte, val),
 184         .head_offset = offsetof(struct fs_fte, hash),
 185         .automatic_shrinking = true,
 186         .min_size = 1,
 187 };
 188
 189 static const struct rhashtable_params rhash_fg = {
 190         .key_len = FIELD_SIZEOF(struct mlx5_flow_group, mask),
 191         .key_offset = offsetof(struct mlx5_flow_group, mask),
 192         .head_offset = offsetof(struct mlx5_flow_group, hash),
 193         .automatic_shrinking = true,
 194         .min_size = 1,
 195
 196 };
 197
 198 static void del_hw_flow_table(struct fs_node *node);
 199 static void del_hw_flow_group(struct fs_node *node);
 200 static void del_hw_fte(struct fs_node *node);
 201 static void del_sw_flow_table(struct fs_node *node);
 202 static void del_sw_flow_group(struct fs_node *node);
 203 static void del_sw_fte(struct fs_node *node);
 204 static void del_sw_prio(struct fs_node *node);
 205 static void del_sw_ns(struct fs_node *node);
 206 /* Delete rule (destination) is special case that
 207  * requires to lock the FTE for all the deletion process.
 208  */
 209 static void del_sw_hw_rule(struct fs_node *node);
 210 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 211                                 struct mlx5_flow_destination *d2);
 212 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 213 static struct mlx5_flow_rule *
 214 find_flow_rule(struct fs_fte *fte,
 215                struct mlx5_flow_destination *dest);
 216
 217 static void tree_init_node(struct fs_node *node,
 218                            void (*del_hw_func)(struct fs_node *),
 219                            void (*del_sw_func)(struct fs_node *))
 220 {
 221         refcount_set(&node->refcount, 1);
 222         INIT_LIST_HEAD(&node->list);
 223         INIT_LIST_HEAD(&node->children);
 224         init_rwsem(&node->lock);
 225         node->del_hw_func = del_hw_func;
 226         node->del_sw_func = del_sw_func;
 227         node->active = false;
 228 }
 229
 230 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 231 {
 232         if (parent)
 233                 refcount_inc(&parent->refcount);
 234         node->parent = parent;
 235
 236         /* Parent is the root */
 237         if (!parent)
 238                 node->root = node;
 239         else
 240                 node->root = parent->root;
 241 }
 242
 243 static int tree_get_node(struct fs_node *node)
 244 {
 245         return refcount_inc_not_zero(&node->refcount);
 246 }
 247
 248 static void nested_down_read_ref_node(struct fs_node *node,
 249                                       enum fs_i_lock_class class)
 250 {
 251         if (node) {
 252                 down_read_nested(&node->lock, class);
 253                 refcount_inc(&node->refcount);
 254         }
 255 }
 256
 257 static void nested_down_write_ref_node(struct fs_node *node,
 258                                        enum fs_i_lock_class class)
 259 {
 260         if (node) {
 261                 down_write_nested(&node->lock, class);
 262                 refcount_inc(&node->refcount);
 263         }
 264 }
 265
 266 static void down_write_ref_node(struct fs_node *node, bool locked)
 267 {
 268         if (node) {
 269                 if (!locked)
 270                         down_write(&node->lock);
 271                 refcount_inc(&node->refcount);
 272         }
 273 }
 274
 275 static void up_read_ref_node(struct fs_node *node)
 276 {
 277         refcount_dec(&node->refcount);
 278         up_read(&node->lock);
 279 }
 280
 281 static void up_write_ref_node(struct fs_node *node, bool locked)
 282 {
 283         refcount_dec(&node->refcount);
 284         if (!locked)
 285                 up_write(&node->lock);
 286 }
 287
 288 static void tree_put_node(struct fs_node *node, bool locked)
 289 {
 290         struct fs_node *parent_node = node->parent;
 291
 292         if (refcount_dec_and_test(&node->refcount)) {
 293                 if (node->del_hw_func)
 294                         node->del_hw_func(node);
 295                 if (parent_node) {
 296                         /* Only root namespace doesn't have parent and we just
 297                          * need to free its node.
 298                          */
 299                         down_write_ref_node(parent_node, locked);
 300                         list_del_init(&node->list);
 301                         if (node->del_sw_func)
 302                                 node->del_sw_func(node);
 303                         up_write_ref_node(parent_node, locked);
 304                 } else {
 305                         kfree(node);
 306                 }
 307                 node = NULL;
 308         }
 309         if (!node && parent_node)
 310                 tree_put_node(parent_node, locked);
 311 }
 312
 313 static int tree_remove_node(struct fs_node *node, bool locked)
 314 {
 315         if (refcount_read(&node->refcount) > 1) {
 316                 refcount_dec(&node->refcount);
 317                 return -EEXIST;
 318         }
 319         tree_put_node(node, locked);
 320         return 0;
 321 }
 322
 323 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 324                                  unsigned int prio)
 325 {
 326         struct fs_prio *iter_prio;
 327
 328         fs_for_each_prio(iter_prio, ns) {
 329                 if (iter_prio->prio == prio)
 330                         return iter_prio;
 331         }
 332
 333         return NULL;
 334 }
 335
 336 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 337 {
 338         int i;
 339
 340         for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 341                 if (spec->match_value[i] & ~spec->match_criteria[i]) {
 342                         pr_warn("mlx5_core: match_value differs from match_criteria\n");
 343                         return false;
 344                 }
 345
 346         return true;
 347 }
 348
 349 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 350 {
 351         struct fs_node *root;
 352         struct mlx5_flow_namespace *ns;
 353
 354         root = node->root;
 355
 356         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 357                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 358                 return NULL;
 359         }
 360
 361         ns = container_of(root, struct mlx5_flow_namespace, node);
 362         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 363 }
 364
 365 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 366 {
 367         struct mlx5_flow_root_namespace *root = find_root(node);
 368
 369         if (root)
 370                 return root->dev->priv.steering;
 371         return NULL;
 372 }
 373
 374 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 375 {
 376         struct mlx5_flow_root_namespace *root = find_root(node);
 377
 378         if (root)
 379                 return root->dev;
 380         return NULL;
 381 }
 382
 383 static void del_sw_ns(struct fs_node *node)
 384 {
 385         kfree(node);
 386 }
 387
 388 static void del_sw_prio(struct fs_node *node)
 389 {
 390         kfree(node);
 391 }
 392
 393 static void del_hw_flow_table(struct fs_node *node)
 394 {
 395         struct mlx5_flow_root_namespace *root;
 396         struct mlx5_flow_table *ft;
 397         struct mlx5_core_dev *dev;
 398         int err;
 399
 400         fs_get_obj(ft, node);
 401         dev = get_dev(&ft->node);
 402         root = find_root(&ft->node);
 403         trace_mlx5_fs_del_ft(ft);
 404
 405         if (node->active) {
 406                 err = root->cmds->destroy_flow_table(root, ft);
 407                 if (err)
 408                         mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 409         }
 410 }
 411
 412 static void del_sw_flow_table(struct fs_node *node)
 413 {
 414         struct mlx5_flow_table *ft;
 415         struct fs_prio *prio;
 416
 417         fs_get_obj(ft, node);
 418
 419         rhltable_destroy(&ft->fgs_hash);
 420         fs_get_obj(prio, ft->node.parent);
 421         prio->num_ft--;
 422         kfree(ft);
 423 }
 424
 425 static void modify_fte(struct fs_fte *fte)
 426 {
 427         struct mlx5_flow_root_namespace *root;
 428         struct mlx5_flow_table *ft;
 429         struct mlx5_flow_group *fg;
 430         struct mlx5_core_dev *dev;
 431         int err;
 432
 433         fs_get_obj(fg, fte->node.parent);
 434         fs_get_obj(ft, fg->node.parent);
 435         dev = get_dev(&fte->node);
 436
 437         root = find_root(&ft->node);
 438         err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte);
 439         if (err)
 440                 mlx5_core_warn(dev,
 441                                "%s can't del rule fg id=%d fte_index=%d\n",
 442                                __func__, fg->id, fte->index);
 443         fte->modify_mask = 0;
 444 }
 445
 446 static void del_sw_hw_rule(struct fs_node *node)
 447 {
 448         struct mlx5_flow_rule *rule;
 449         struct fs_fte *fte;
 450
 451         fs_get_obj(rule, node);
 452         fs_get_obj(fte, rule->node.parent);
 453         trace_mlx5_fs_del_rule(rule);
 454         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 455                 mutex_lock(&rule->dest_attr.ft->lock);
 456                 list_del(&rule->next_ft);
 457                 mutex_unlock(&rule->dest_attr.ft->lock);
 458         }
 459
 460         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 461             --fte->dests_size) {
 462                 fte->modify_mask |=
 463                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 464                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 465                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 466                 goto out;
 467         }
 468
 469         if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 470             --fte->dests_size) {
 471                 fte->modify_mask |=
 472                         BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 473         }
 474 out:
 475         kfree(rule);
 476 }
 477
 478 static void del_hw_fte(struct fs_node *node)
 479 {
 480         struct mlx5_flow_root_namespace *root;
 481         struct mlx5_flow_table *ft;
 482         struct mlx5_flow_group *fg;
 483         struct mlx5_core_dev *dev;
 484         struct fs_fte *fte;
 485         int err;
 486
 487         fs_get_obj(fte, node);
 488         fs_get_obj(fg, fte->node.parent);
 489         fs_get_obj(ft, fg->node.parent);
 490
 491         trace_mlx5_fs_del_fte(fte);
 492         dev = get_dev(&ft->node);
 493         root = find_root(&ft->node);
 494         if (node->active) {
 495                 err = root->cmds->delete_fte(root, ft, fte);
 496                 if (err)
 497                         mlx5_core_warn(dev,
 498                                        "flow steering can't delete fte in index %d of flow group id %d\n",
 499                                        fte->index, fg->id);
 500                 node->active = 0;
 501         }
 502 }
 503
 504 static void del_sw_fte(struct fs_node *node)
 505 {
 506         struct mlx5_flow_steering *steering = get_steering(node);
 507         struct mlx5_flow_group *fg;
 508         struct fs_fte *fte;
 509         int err;
 510
 511         fs_get_obj(fte, node);
 512         fs_get_obj(fg, fte->node.parent);
 513
 514         err = rhashtable_remove_fast(&fg->ftes_hash,
 515                                      &fte->hash,
 516                                      rhash_fte);
 517         WARN_ON(err);
 518         ida_simple_remove(&fg->fte_allocator, fte->index - fg->start_index);
 519         kmem_cache_free(steering->ftes_cache, fte);
 520 }
 521
 522 static void del_hw_flow_group(struct fs_node *node)
 523 {
 524         struct mlx5_flow_root_namespace *root;
 525         struct mlx5_flow_group *fg;
 526         struct mlx5_flow_table *ft;
 527         struct mlx5_core_dev *dev;
 528
 529         fs_get_obj(fg, node);
 530         fs_get_obj(ft, fg->node.parent);
 531         dev = get_dev(&ft->node);
 532         trace_mlx5_fs_del_fg(fg);
 533
 534         root = find_root(&ft->node);
 535         if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
 536                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 537                                fg->id, ft->id);
 538 }
 539
 540 static void del_sw_flow_group(struct fs_node *node)
 541 {
 542         struct mlx5_flow_steering *steering = get_steering(node);
 543         struct mlx5_flow_group *fg;
 544         struct mlx5_flow_table *ft;
 545         int err;
 546
 547         fs_get_obj(fg, node);
 548         fs_get_obj(ft, fg->node.parent);
 549
 550         rhashtable_destroy(&fg->ftes_hash);
 551         ida_destroy(&fg->fte_allocator);
 552         if (ft->autogroup.active)
 553                 ft->autogroup.num_groups--;
 554         err = rhltable_remove(&ft->fgs_hash,
 555                               &fg->hash,
 556                               rhash_fg);
 557         WARN_ON(err);
 558         kmem_cache_free(steering->fgs_cache, fg);
 559 }
 560
 561 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 562 {
 563         int index;
 564         int ret;
 565
 566         index = ida_simple_get(&fg->fte_allocator, 0, fg->max_ftes, GFP_KERNEL);
 567         if (index < 0)
 568                 return index;
 569
 570         fte->index = index + fg->start_index;
 571         ret = rhashtable_insert_fast(&fg->ftes_hash,
 572                                      &fte->hash,
 573                                      rhash_fte);
 574         if (ret)
 575                 goto err_ida_remove;
 576
 577         tree_add_node(&fte->node, &fg->node);
 578         list_add_tail(&fte->node.list, &fg->node.children);
 579         return 0;
 580
 581 err_ida_remove:
 582         ida_simple_remove(&fg->fte_allocator, index);
 583         return ret;
 584 }
 585
 586 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 587                                 u32 *match_value,
 588                                 struct mlx5_flow_act *flow_act)
 589 {
 590         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 591         struct fs_fte *fte;
 592
 593         fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 594         if (!fte)
 595                 return ERR_PTR(-ENOMEM);
 596
 597         memcpy(fte->val, match_value, sizeof(fte->val));
 598         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 599         fte->action = *flow_act;
 600
 601         tree_init_node(&fte->node, NULL, del_sw_fte);
 602
 603         return fte;
 604 }
 605
 606 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 607                                struct mlx5_flow_group *fg)
 608 {
 609         rhashtable_destroy(&fg->ftes_hash);
 610         kmem_cache_free(steering->fgs_cache, fg);
 611 }
 612
 613 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 614                                                 u8 match_criteria_enable,
 615                                                 void *match_criteria,
 616                                                 int start_index,
 617                                                 int end_index)
 618 {
 619         struct mlx5_flow_group *fg;
 620         int ret;
 621
 622         fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 623         if (!fg)
 624                 return ERR_PTR(-ENOMEM);
 625
 626         ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 627         if (ret) {
 628                 kmem_cache_free(steering->fgs_cache, fg);
 629                 return ERR_PTR(ret);
 630         }
 631
 632         ida_init(&fg->fte_allocator);
 633         fg->mask.match_criteria_enable = match_criteria_enable;
 634         memcpy(&fg->mask.match_criteria, match_criteria,
 635                sizeof(fg->mask.match_criteria));
 636         fg->node.type =  FS_TYPE_FLOW_GROUP;
 637         fg->start_index = start_index;
 638         fg->max_ftes = end_index - start_index + 1;
 639
 640         return fg;
 641 }
 642
 643 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 644                                                        u8 match_criteria_enable,
 645                                                        void *match_criteria,
 646                                                        int start_index,
 647                                                        int end_index,
 648                                                        struct list_head *prev)
 649 {
 650         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 651         struct mlx5_flow_group *fg;
 652         int ret;
 653
 654         fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 655                               start_index, end_index);
 656         if (IS_ERR(fg))
 657                 return fg;
 658
 659         /* initialize refcnt, add to parent list */
 660         ret = rhltable_insert(&ft->fgs_hash,
 661                               &fg->hash,
 662                               rhash_fg);
 663         if (ret) {
 664                 dealloc_flow_group(steering, fg);
 665                 return ERR_PTR(ret);
 666         }
 667
 668         tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 669         tree_add_node(&fg->node, &ft->node);
 670         /* Add node to group list */
 671         list_add(&fg->node.list, prev);
 672         atomic_inc(&ft->node.version);
 673
 674         return fg;
 675 }
 676
 677 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 678                                                 enum fs_flow_table_type table_type,
 679                                                 enum fs_flow_table_op_mod op_mod,
 680                                                 u32 flags)
 681 {
 682         struct mlx5_flow_table *ft;
 683         int ret;
 684
 685         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 686         if (!ft)
 687                 return ERR_PTR(-ENOMEM);
 688
 689         ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 690         if (ret) {
 691                 kfree(ft);
 692                 return ERR_PTR(ret);
 693         }
 694
 695         ft->level = level;
 696         ft->node.type = FS_TYPE_FLOW_TABLE;
 697         ft->op_mod = op_mod;
 698         ft->type = table_type;
 699         ft->vport = vport;
 700         ft->max_fte = max_fte;
 701         ft->flags = flags;
 702         INIT_LIST_HEAD(&ft->fwd_rules);
 703         mutex_init(&ft->lock);
 704
 705         return ft;
 706 }
 707
 708 /* If reverse is false, then we search for the first flow table in the
 709  * root sub-tree from start(closest from right), else we search for the
 710  * last flow table in the root sub-tree till start(closest from left).
 711  */
 712 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 713                                                          struct list_head *start,
 714                                                          bool reverse)
 715 {
 716 #define list_advance_entry(pos, reverse)                \
 717         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 718
 719 #define list_for_each_advance_continue(pos, head, reverse)      \
 720         for (pos = list_advance_entry(pos, reverse);            \
 721              &pos->list != (head);                              \
 722              pos = list_advance_entry(pos, reverse))
 723
 724         struct fs_node *iter = list_entry(start, struct fs_node, list);
 725         struct mlx5_flow_table *ft = NULL;
 726
 727         if (!root || root->type == FS_TYPE_PRIO_CHAINS)
 728                 return NULL;
 729
 730         list_for_each_advance_continue(iter, &root->children, reverse) {
 731                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 732                         fs_get_obj(ft, iter);
 733                         return ft;
 734                 }
 735                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 736                 if (ft)
 737                         return ft;
 738         }
 739
 740         return ft;
 741 }
 742
 743 /* If reverse if false then return the first flow table in next priority of
 744  * prio in the tree, else return the last flow table in the previous priority
 745  * of prio in the tree.
 746  */
 747 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 748 {
 749         struct mlx5_flow_table *ft = NULL;
 750         struct fs_node *curr_node;
 751         struct fs_node *parent;
 752
 753         parent = prio->node.parent;
 754         curr_node = &prio->node;
 755         while (!ft && parent) {
 756                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 757                 curr_node = parent;
 758                 parent = curr_node->parent;
 759         }
 760         return ft;
 761 }
 762
 763 /* Assuming all the tree is locked by mutex chain lock */
 764 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 765 {
 766         return find_closest_ft(prio, false);
 767 }
 768
 769 /* Assuming all the tree is locked by mutex chain lock */
 770 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 771 {
 772         return find_closest_ft(prio, true);
 773 }
 774
 775 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 776                                struct fs_prio *prio,
 777                                struct mlx5_flow_table *ft)
 778 {
 779         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 780         struct mlx5_flow_table *iter;
 781         int i = 0;
 782         int err;
 783
 784         fs_for_each_ft(iter, prio) {
 785                 i++;
 786                 err = root->cmds->modify_flow_table(root, iter, ft);
 787                 if (err) {
 788                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 789                                        iter->id);
 790                         /* The driver is out of sync with the FW */
 791                         if (i > 1)
 792                                 WARN_ON(true);
 793                         return err;
 794                 }
 795         }
 796         return 0;
 797 }
 798
 799 /* Connect flow tables from previous priority of prio to ft */
 800 static int connect_prev_fts(struct mlx5_core_dev *dev,
 801                             struct mlx5_flow_table *ft,
 802                             struct fs_prio *prio)
 803 {
 804         struct mlx5_flow_table *prev_ft;
 805
 806         prev_ft = find_prev_chained_ft(prio);
 807         if (prev_ft) {
 808                 struct fs_prio *prev_prio;
 809
 810                 fs_get_obj(prev_prio, prev_ft->node.parent);
 811                 return connect_fts_in_prio(dev, prev_prio, ft);
 812         }
 813         return 0;
 814 }
 815
 816 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 817                                  *prio)
 818 {
 819         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 820         struct mlx5_ft_underlay_qp *uqp;
 821         int min_level = INT_MAX;
 822         int err = 0;
 823         u32 qpn;
 824
 825         if (root->root_ft)
 826                 min_level = root->root_ft->level;
 827
 828         if (ft->level >= min_level)
 829                 return 0;
 830
 831         if (list_empty(&root->underlay_qpns)) {
 832                 /* Don't set any QPN (zero) in case QPN list is empty */
 833                 qpn = 0;
 834                 err = root->cmds->update_root_ft(root, ft, qpn, false);
 835         } else {
 836                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
 837                         qpn = uqp->qpn;
 838                         err = root->cmds->update_root_ft(root, ft,
 839                                                          qpn, false);
 840                         if (err)
 841                                 break;
 842                 }
 843         }
 844
 845         if (err)
 846                 mlx5_core_warn(root->dev,
 847                                "Update root flow table of id(%u) qpn(%d) failed\n",
 848                                ft->id, qpn);
 849         else
 850                 root->root_ft = ft;
 851
 852         return err;
 853 }
 854
 855 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 856                                          struct mlx5_flow_destination *dest)
 857 {
 858         struct mlx5_flow_root_namespace *root;
 859         struct mlx5_flow_table *ft;
 860         struct mlx5_flow_group *fg;
 861         struct fs_fte *fte;
 862         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 863         int err = 0;
 864
 865         fs_get_obj(fte, rule->node.parent);
 866         if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 867                 return -EINVAL;
 868         down_write_ref_node(&fte->node, false);
 869         fs_get_obj(fg, fte->node.parent);
 870         fs_get_obj(ft, fg->node.parent);
 871
 872         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 873         root = find_root(&ft->node);
 874         err = root->cmds->update_fte(root, ft, fg,
 875                                      modify_mask, fte);
 876         up_write_ref_node(&fte->node, false);
 877
 878         return err;
 879 }
 880
 881 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
 882                                  struct mlx5_flow_destination *new_dest,
 883                                  struct mlx5_flow_destination *old_dest)
 884 {
 885         int i;
 886
 887         if (!old_dest) {
 888                 if (handle->num_rules != 1)
 889                         return -EINVAL;
 890                 return _mlx5_modify_rule_destination(handle->rule[0],
 891                                                      new_dest);
 892         }
 893
 894         for (i = 0; i < handle->num_rules; i++) {
 895                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
 896                         return _mlx5_modify_rule_destination(handle->rule[i],
 897                                                              new_dest);
 898         }
 899
 900         return -EINVAL;
 901 }
 902
 903 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 904 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 905                              struct mlx5_flow_table *new_next_ft,
 906                              struct mlx5_flow_table *old_next_ft)
 907 {
 908         struct mlx5_flow_destination dest = {};
 909         struct mlx5_flow_rule *iter;
 910         int err = 0;
 911
 912         /* new_next_ft and old_next_ft could be NULL only
 913          * when we create/destroy the anchor flow table.
 914          */
 915         if (!new_next_ft || !old_next_ft)
 916                 return 0;
 917
 918         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 919         dest.ft = new_next_ft;
 920
 921         mutex_lock(&old_next_ft->lock);
 922         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 923         mutex_unlock(&old_next_ft->lock);
 924         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 925                 err = _mlx5_modify_rule_destination(iter, &dest);
 926                 if (err)
 927                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 928                                new_next_ft->id);
 929         }
 930         return 0;
 931 }
 932
 933 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 934                               struct fs_prio *prio)
 935 {
 936         struct mlx5_flow_table *next_ft;
 937         int err = 0;
 938
 939         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 940
 941         if (list_empty(&prio->node.children)) {
 942                 err = connect_prev_fts(dev, ft, prio);
 943                 if (err)
 944                         return err;
 945
 946                 next_ft = find_next_chained_ft(prio);
 947                 err = connect_fwd_rules(dev, ft, next_ft);
 948                 if (err)
 949                         return err;
 950         }
 951
 952         if (MLX5_CAP_FLOWTABLE(dev,
 953                                flow_table_properties_nic_receive.modify_root))
 954                 err = update_root_ft_create(ft, prio);
 955         return err;
 956 }
 957
 958 static void list_add_flow_table(struct mlx5_flow_table *ft,
 959                                 struct fs_prio *prio)
 960 {
 961         struct list_head *prev = &prio->node.children;
 962         struct mlx5_flow_table *iter;
 963
 964         fs_for_each_ft(iter, prio) {
 965                 if (iter->level > ft->level)
 966                         break;
 967                 prev = &iter->node.list;
 968         }
 969         list_add(&ft->node.list, prev);
 970 }
 971
 972 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 973                                                         struct mlx5_flow_table_attr *ft_attr,
 974                                                         enum fs_flow_table_op_mod op_mod,
 975                                                         u16 vport)
 976 {
 977         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
 978         struct mlx5_flow_table *next_ft = NULL;
 979         struct fs_prio *fs_prio = NULL;
 980         struct mlx5_flow_table *ft;
 981         int log_table_sz;
 982         int err;
 983
 984         if (!root) {
 985                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 986                 return ERR_PTR(-ENODEV);
 987         }
 988
 989         mutex_lock(&root->chain_lock);
 990         fs_prio = find_prio(ns, ft_attr->prio);
 991         if (!fs_prio) {
 992                 err = -EINVAL;
 993                 goto unlock_root;
 994         }
 995         if (ft_attr->level >= fs_prio->num_levels) {
 996                 err = -ENOSPC;
 997                 goto unlock_root;
 998         }
 999         /* The level is related to the
1000          * priority level range.
1001          */
1002         ft_attr->level += fs_prio->start_level;
1003         ft = alloc_flow_table(ft_attr->level,
1004                               vport,
1005                               ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
1006                               root->table_type,
1007                               op_mod, ft_attr->flags);
1008         if (IS_ERR(ft)) {
1009                 err = PTR_ERR(ft);
1010                 goto unlock_root;
1011         }
1012
1013         tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1014         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
1015         next_ft = find_next_chained_ft(fs_prio);
1016         err = root->cmds->create_flow_table(root, ft, log_table_sz, next_ft);
1017         if (err)
1018                 goto free_ft;
1019
1020         err = connect_flow_table(root->dev, ft, fs_prio);
1021         if (err)
1022                 goto destroy_ft;
1023         ft->node.active = true;
1024         down_write_ref_node(&fs_prio->node, false);
1025         tree_add_node(&ft->node, &fs_prio->node);
1026         list_add_flow_table(ft, fs_prio);
1027         fs_prio->num_ft++;
1028         up_write_ref_node(&fs_prio->node, false);
1029         mutex_unlock(&root->chain_lock);
1030         trace_mlx5_fs_add_ft(ft);
1031         return ft;
1032 destroy_ft:
1033         root->cmds->destroy_flow_table(root, ft);
1034 free_ft:
1035         kfree(ft);
1036 unlock_root:
1037         mutex_unlock(&root->chain_lock);
1038         return ERR_PTR(err);
1039 }
1040
1041 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1042                                                struct mlx5_flow_table_attr *ft_attr)
1043 {
1044         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1045 }
1046
1047 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1048                                                      int prio, int max_fte,
1049                                                      u32 level, u16 vport)
1050 {
1051         struct mlx5_flow_table_attr ft_attr = {};
1052
1053         ft_attr.max_fte = max_fte;
1054         ft_attr.level   = level;
1055         ft_attr.prio    = prio;
1056
1057         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1058 }
1059
1060 struct mlx5_flow_table*
1061 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1062                                  int prio, u32 level)
1063 {
1064         struct mlx5_flow_table_attr ft_attr = {};
1065
1066         ft_attr.level = level;
1067         ft_attr.prio  = prio;
1068         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1069 }
1070 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1071
1072 struct mlx5_flow_table*
1073 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1074                                     int prio,
1075                                     int num_flow_table_entries,
1076                                     int max_num_groups,
1077                                     u32 level,
1078                                     u32 flags)
1079 {
1080         struct mlx5_flow_table_attr ft_attr = {};
1081         struct mlx5_flow_table *ft;
1082
1083         if (max_num_groups > num_flow_table_entries)
1084                 return ERR_PTR(-EINVAL);
1085
1086         ft_attr.max_fte = num_flow_table_entries;
1087         ft_attr.prio    = prio;
1088         ft_attr.level   = level;
1089         ft_attr.flags   = flags;
1090
1091         ft = mlx5_create_flow_table(ns, &ft_attr);
1092         if (IS_ERR(ft))
1093                 return ft;
1094
1095         ft->autogroup.active = true;
1096         ft->autogroup.required_groups = max_num_groups;
1097
1098         return ft;
1099 }
1100 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1101
1102 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1103                                                u32 *fg_in)
1104 {
1105         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1106         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1107                                             fg_in, match_criteria);
1108         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1109                                             fg_in,
1110                                             match_criteria_enable);
1111         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1112                                    start_flow_index);
1113         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1114                                  end_flow_index);
1115         struct mlx5_flow_group *fg;
1116         int err;
1117
1118         if (ft->autogroup.active)
1119                 return ERR_PTR(-EPERM);
1120
1121         down_write_ref_node(&ft->node, false);
1122         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1123                                      start_index, end_index,
1124                                      ft->node.children.prev);
1125         up_write_ref_node(&ft->node, false);
1126         if (IS_ERR(fg))
1127                 return fg;
1128
1129         err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1130         if (err) {
1131                 tree_put_node(&fg->node, false);
1132                 return ERR_PTR(err);
1133         }
1134         trace_mlx5_fs_add_fg(fg);
1135         fg->node.active = true;
1136
1137         return fg;
1138 }
1139
1140 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1141 {
1142         struct mlx5_flow_rule *rule;
1143
1144         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1145         if (!rule)
1146                 return NULL;
1147
1148         INIT_LIST_HEAD(&rule->next_ft);
1149         rule->node.type = FS_TYPE_FLOW_DEST;
1150         if (dest)
1151                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1152
1153         return rule;
1154 }
1155
1156 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1157 {
1158         struct mlx5_flow_handle *handle;
1159
1160         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1161         if (!handle)
1162                 return NULL;
1163
1164         handle->num_rules = num_rules;
1165
1166         return handle;
1167 }
1168
1169 static void destroy_flow_handle(struct fs_fte *fte,
1170                                 struct mlx5_flow_handle *handle,
1171                                 struct mlx5_flow_destination *dest,
1172                                 int i)
1173 {
1174         for (; --i >= 0;) {
1175                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1176                         fte->dests_size--;
1177                         list_del(&handle->rule[i]->node.list);
1178                         kfree(handle->rule[i]);
1179                 }
1180         }
1181         kfree(handle);
1182 }
1183
1184 static struct mlx5_flow_handle *
1185 create_flow_handle(struct fs_fte *fte,
1186                    struct mlx5_flow_destination *dest,
1187                    int dest_num,
1188                    int *modify_mask,
1189                    bool *new_rule)
1190 {
1191         struct mlx5_flow_handle *handle;
1192         struct mlx5_flow_rule *rule = NULL;
1193         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1194         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1195         int type;
1196         int i = 0;
1197
1198         handle = alloc_handle((dest_num) ? dest_num : 1);
1199         if (!handle)
1200                 return ERR_PTR(-ENOMEM);
1201
1202         do {
1203                 if (dest) {
1204                         rule = find_flow_rule(fte, dest + i);
1205                         if (rule) {
1206                                 refcount_inc(&rule->node.refcount);
1207                                 goto rule_found;
1208                         }
1209                 }
1210
1211                 *new_rule = true;
1212                 rule = alloc_rule(dest + i);
1213                 if (!rule)
1214                         goto free_rules;
1215
1216                 /* Add dest to dests list- we need flow tables to be in the
1217                  * end of the list for forward to next prio rules.
1218                  */
1219                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1220                 if (dest &&
1221                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1222                         list_add(&rule->node.list, &fte->node.children);
1223                 else
1224                         list_add_tail(&rule->node.list, &fte->node.children);
1225                 if (dest) {
1226                         fte->dests_size++;
1227
1228                         type = dest[i].type ==
1229                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1230                         *modify_mask |= type ? count : dst;
1231                 }
1232 rule_found:
1233                 handle->rule[i] = rule;
1234         } while (++i < dest_num);
1235
1236         return handle;
1237
1238 free_rules:
1239         destroy_flow_handle(fte, handle, dest, i);
1240         return ERR_PTR(-ENOMEM);
1241 }
1242
1243 /* fte should not be deleted while calling this function */
1244 static struct mlx5_flow_handle *
1245 add_rule_fte(struct fs_fte *fte,
1246              struct mlx5_flow_group *fg,
1247              struct mlx5_flow_destination *dest,
1248              int dest_num,
1249              bool update_action)
1250 {
1251         struct mlx5_flow_root_namespace *root;
1252         struct mlx5_flow_handle *handle;
1253         struct mlx5_flow_table *ft;
1254         int modify_mask = 0;
1255         int err;
1256         bool new_rule = false;
1257
1258         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1259                                     &new_rule);
1260         if (IS_ERR(handle) || !new_rule)
1261                 goto out;
1262
1263         if (update_action)
1264                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1265
1266         fs_get_obj(ft, fg->node.parent);
1267         root = find_root(&fg->node);
1268         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1269                 err = root->cmds->create_fte(root, ft, fg, fte);
1270         else
1271                 err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1272         if (err)
1273                 goto free_handle;
1274
1275         fte->node.active = true;
1276         fte->status |= FS_FTE_STATUS_EXISTING;
1277         atomic_inc(&fte->node.version);
1278
1279 out:
1280         return handle;
1281
1282 free_handle:
1283         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1284         return ERR_PTR(err);
1285 }
1286
1287 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1288                                                      struct mlx5_flow_spec *spec)
1289 {
1290         struct list_head *prev = &ft->node.children;
1291         struct mlx5_flow_group *fg;
1292         unsigned int candidate_index = 0;
1293         unsigned int group_size = 0;
1294
1295         if (!ft->autogroup.active)
1296                 return ERR_PTR(-ENOENT);
1297
1298         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1299                 /* We save place for flow groups in addition to max types */
1300                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1301
1302         /*  ft->max_fte == ft->autogroup.max_types */
1303         if (group_size == 0)
1304                 group_size = 1;
1305
1306         /* sorted by start_index */
1307         fs_for_each_fg(fg, ft) {
1308                 if (candidate_index + group_size > fg->start_index)
1309                         candidate_index = fg->start_index + fg->max_ftes;
1310                 else
1311                         break;
1312                 prev = &fg->node.list;
1313         }
1314
1315         if (candidate_index + group_size > ft->max_fte)
1316                 return ERR_PTR(-ENOSPC);
1317
1318         fg = alloc_insert_flow_group(ft,
1319                                      spec->match_criteria_enable,
1320                                      spec->match_criteria,
1321                                      candidate_index,
1322                                      candidate_index + group_size - 1,
1323                                      prev);
1324         if (IS_ERR(fg))
1325                 goto out;
1326
1327         ft->autogroup.num_groups++;
1328
1329 out:
1330         return fg;
1331 }
1332
1333 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1334                                   struct mlx5_flow_group *fg)
1335 {
1336         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1337         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1338         void *match_criteria_addr;
1339         u8 src_esw_owner_mask_on;
1340         void *misc;
1341         int err;
1342         u32 *in;
1343
1344         in = kvzalloc(inlen, GFP_KERNEL);
1345         if (!in)
1346                 return -ENOMEM;
1347
1348         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1349                  fg->mask.match_criteria_enable);
1350         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1351         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1352                  fg->max_ftes - 1);
1353
1354         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1355                             misc_parameters);
1356         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1357                                          source_eswitch_owner_vhca_id);
1358         MLX5_SET(create_flow_group_in, in,
1359                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1360
1361         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1362                                            in, match_criteria);
1363         memcpy(match_criteria_addr, fg->mask.match_criteria,
1364                sizeof(fg->mask.match_criteria));
1365
1366         err = root->cmds->create_flow_group(root, ft, in, fg);
1367         if (!err) {
1368                 fg->node.active = true;
1369                 trace_mlx5_fs_add_fg(fg);
1370         }
1371
1372         kvfree(in);
1373         return err;
1374 }
1375
1376 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1377                                 struct mlx5_flow_destination *d2)
1378 {
1379         if (d1->type == d2->type) {
1380                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1381                      d1->vport.num == d2->vport.num &&
1382                      d1->vport.flags == d2->vport.flags &&
1383                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1384                       (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1385                      ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1386                       (d1->vport.reformat_id == d2->vport.reformat_id) : true)) ||
1387                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1388                      d1->ft == d2->ft) ||
1389                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1390                      d1->tir_num == d2->tir_num) ||
1391                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1392                      d1->ft_num == d2->ft_num))
1393                         return true;
1394         }
1395
1396         return false;
1397 }
1398
1399 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1400                                              struct mlx5_flow_destination *dest)
1401 {
1402         struct mlx5_flow_rule *rule;
1403
1404         list_for_each_entry(rule, &fte->node.children, node.list) {
1405                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1406                         return rule;
1407         }
1408         return NULL;
1409 }
1410
1411 static bool check_conflicting_actions(u32 action1, u32 action2)
1412 {
1413         u32 xored_actions = action1 ^ action2;
1414
1415         /* if one rule only wants to count, it's ok */
1416         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1417             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1418                 return false;
1419
1420         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1421                              MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1422                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1423                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1424                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1425                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1426                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1427                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1428                 return true;
1429
1430         return false;
1431 }
1432
1433 static int check_conflicting_ftes(struct fs_fte *fte, const struct mlx5_flow_act *flow_act)
1434 {
1435         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1436                 mlx5_core_warn(get_dev(&fte->node),
1437                                "Found two FTEs with conflicting actions\n");
1438                 return -EEXIST;
1439         }
1440
1441         if ((flow_act->flags & FLOW_ACT_HAS_TAG) &&
1442             fte->action.flow_tag != flow_act->flow_tag) {
1443                 mlx5_core_warn(get_dev(&fte->node),
1444                                "FTE flow tag %u already exists with different flow tag %u\n",
1445                                fte->action.flow_tag,
1446                                flow_act->flow_tag);
1447                 return -EEXIST;
1448         }
1449
1450         return 0;
1451 }
1452
1453 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1454                                             u32 *match_value,
1455                                             struct mlx5_flow_act *flow_act,
1456                                             struct mlx5_flow_destination *dest,
1457                                             int dest_num,
1458                                             struct fs_fte *fte)
1459 {
1460         struct mlx5_flow_handle *handle;
1461         int old_action;
1462         int i;
1463         int ret;
1464
1465         ret = check_conflicting_ftes(fte, flow_act);
1466         if (ret)
1467                 return ERR_PTR(ret);
1468
1469         old_action = fte->action.action;
1470         fte->action.action |= flow_act->action;
1471         handle = add_rule_fte(fte, fg, dest, dest_num,
1472                               old_action != flow_act->action);
1473         if (IS_ERR(handle)) {
1474                 fte->action.action = old_action;
1475                 return handle;
1476         }
1477         trace_mlx5_fs_set_fte(fte, false);
1478
1479         for (i = 0; i < handle->num_rules; i++) {
1480                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1481                         tree_add_node(&handle->rule[i]->node, &fte->node);
1482                         trace_mlx5_fs_add_rule(handle->rule[i]);
1483                 }
1484         }
1485         return handle;
1486 }
1487
1488 static bool counter_is_valid(u32 action)
1489 {
1490         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1491                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1492 }
1493
1494 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1495                           u32 action,
1496                           struct mlx5_flow_table *ft)
1497 {
1498         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1499                 return counter_is_valid(action);
1500
1501         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1502                 return true;
1503
1504         if (!dest || ((dest->type ==
1505             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1506             (dest->ft->level <= ft->level)))
1507                 return false;
1508         return true;
1509 }
1510
1511 struct match_list {
1512         struct list_head        list;
1513         struct mlx5_flow_group *g;
1514 };
1515
1516 struct match_list_head {
1517         struct list_head  list;
1518         struct match_list first;
1519 };
1520
1521 static void free_match_list(struct match_list_head *head)
1522 {
1523         if (!list_empty(&head->list)) {
1524                 struct match_list *iter, *match_tmp;
1525
1526                 list_del(&head->first.list);
1527                 tree_put_node(&head->first.g->node, false);
1528                 list_for_each_entry_safe(iter, match_tmp, &head->list,
1529                                          list) {
1530                         tree_put_node(&iter->g->node, false);
1531                         list_del(&iter->list);
1532                         kfree(iter);
1533                 }
1534         }
1535 }
1536
1537 static int build_match_list(struct match_list_head *match_head,
1538                             struct mlx5_flow_table *ft,
1539                             struct mlx5_flow_spec *spec)
1540 {
1541         struct rhlist_head *tmp, *list;
1542         struct mlx5_flow_group *g;
1543         int err = 0;
1544
1545         rcu_read_lock();
1546         INIT_LIST_HEAD(&match_head->list);
1547         /* Collect all fgs which has a matching match_criteria */
1548         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1549         /* RCU is atomic, we can't execute FW commands here */
1550         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1551                 struct match_list *curr_match;
1552
1553                 if (likely(list_empty(&match_head->list))) {
1554                         if (!tree_get_node(&g->node))
1555                                 continue;
1556                         match_head->first.g = g;
1557                         list_add_tail(&match_head->first.list,
1558                                       &match_head->list);
1559                         continue;
1560                 }
1561
1562                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1563                 if (!curr_match) {
1564                         free_match_list(match_head);
1565                         err = -ENOMEM;
1566                         goto out;
1567                 }
1568                 if (!tree_get_node(&g->node)) {
1569                         kfree(curr_match);
1570                         continue;
1571                 }
1572                 curr_match->g = g;
1573                 list_add_tail(&curr_match->list, &match_head->list);
1574         }
1575 out:
1576         rcu_read_unlock();
1577         return err;
1578 }
1579
1580 static u64 matched_fgs_get_version(struct list_head *match_head)
1581 {
1582         struct match_list *iter;
1583         u64 version = 0;
1584
1585         list_for_each_entry(iter, match_head, list)
1586                 version += (u64)atomic_read(&iter->g->node.version);
1587         return version;
1588 }
1589
1590 static struct fs_fte *
1591 lookup_fte_locked(struct mlx5_flow_group *g,
1592                   u32 *match_value,
1593                   bool take_write)
1594 {
1595         struct fs_fte *fte_tmp;
1596
1597         if (take_write)
1598                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1599         else
1600                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1601         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1602                                          rhash_fte);
1603         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1604                 fte_tmp = NULL;
1605                 goto out;
1606         }
1607         if (!fte_tmp->node.active) {
1608                 tree_put_node(&fte_tmp->node, false);
1609                 fte_tmp = NULL;
1610                 goto out;
1611         }
1612
1613         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1614 out:
1615         if (take_write)
1616                 up_write_ref_node(&g->node, false);
1617         else
1618                 up_read_ref_node(&g->node);
1619         return fte_tmp;
1620 }
1621
1622 static struct mlx5_flow_handle *
1623 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1624                        struct list_head *match_head,
1625                        struct mlx5_flow_spec *spec,
1626                        struct mlx5_flow_act *flow_act,
1627                        struct mlx5_flow_destination *dest,
1628                        int dest_num,
1629                        int ft_version)
1630 {
1631         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1632         struct mlx5_flow_group *g;
1633         struct mlx5_flow_handle *rule;
1634         struct match_list *iter;
1635         bool take_write = false;
1636         struct fs_fte *fte;
1637         u64  version;
1638         int err;
1639
1640         fte = alloc_fte(ft, spec->match_value, flow_act);
1641         if (IS_ERR(fte))
1642                 return  ERR_PTR(-ENOMEM);
1643
1644 search_again_locked:
1645         version = matched_fgs_get_version(match_head);
1646         if (flow_act->flags & FLOW_ACT_NO_APPEND)
1647                 goto skip_search;
1648         /* Try to find a fg that already contains a matching fte */
1649         list_for_each_entry(iter, match_head, list) {
1650                 struct fs_fte *fte_tmp;
1651
1652                 g = iter->g;
1653                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1654                 if (!fte_tmp)
1655                         continue;
1656                 rule = add_rule_fg(g, spec->match_value,
1657                                    flow_act, dest, dest_num, fte_tmp);
1658                 up_write_ref_node(&fte_tmp->node, false);
1659                 tree_put_node(&fte_tmp->node, false);
1660                 kmem_cache_free(steering->ftes_cache, fte);
1661                 return rule;
1662         }
1663
1664 skip_search:
1665         /* No group with matching fte found, or we skipped the search.
1666          * Try to add a new fte to any matching fg.
1667          */
1668
1669         /* Check the ft version, for case that new flow group
1670          * was added while the fgs weren't locked
1671          */
1672         if (atomic_read(&ft->node.version) != ft_version) {
1673                 rule = ERR_PTR(-EAGAIN);
1674                 goto out;
1675         }
1676
1677         /* Check the fgs version, for case the new FTE with the
1678          * same values was added while the fgs weren't locked
1679          */
1680         if (version != matched_fgs_get_version(match_head)) {
1681                 take_write = true;
1682                 goto search_again_locked;
1683         }
1684
1685         list_for_each_entry(iter, match_head, list) {
1686                 g = iter->g;
1687
1688                 if (!g->node.active)
1689                         continue;
1690
1691                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1692
1693                 err = insert_fte(g, fte);
1694                 if (err) {
1695                         up_write_ref_node(&g->node, false);
1696                         if (err == -ENOSPC)
1697                                 continue;
1698                         kmem_cache_free(steering->ftes_cache, fte);
1699                         return ERR_PTR(err);
1700                 }
1701
1702                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1703                 up_write_ref_node(&g->node, false);
1704                 rule = add_rule_fg(g, spec->match_value,
1705                                    flow_act, dest, dest_num, fte);
1706                 up_write_ref_node(&fte->node, false);
1707                 tree_put_node(&fte->node, false);
1708                 return rule;
1709         }
1710         rule = ERR_PTR(-ENOENT);
1711 out:
1712         kmem_cache_free(steering->ftes_cache, fte);
1713         return rule;
1714 }
1715
1716 static struct mlx5_flow_handle *
1717 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1718                      struct mlx5_flow_spec *spec,
1719                      struct mlx5_flow_act *flow_act,
1720                      struct mlx5_flow_destination *dest,
1721                      int dest_num)
1722
1723 {
1724         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1725         struct mlx5_flow_group *g;
1726         struct mlx5_flow_handle *rule;
1727         struct match_list_head match_head;
1728         bool take_write = false;
1729         struct fs_fte *fte;
1730         int version;
1731         int err;
1732         int i;
1733
1734         if (!check_valid_spec(spec))
1735                 return ERR_PTR(-EINVAL);
1736
1737         for (i = 0; i < dest_num; i++) {
1738                 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1739                         return ERR_PTR(-EINVAL);
1740         }
1741         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1742 search_again_locked:
1743         version = atomic_read(&ft->node.version);
1744
1745         /* Collect all fgs which has a matching match_criteria */
1746         err = build_match_list(&match_head, ft, spec);
1747         if (err) {
1748                 if (take_write)
1749                         up_write_ref_node(&ft->node, false);
1750                 else
1751                         up_read_ref_node(&ft->node);
1752                 return ERR_PTR(err);
1753         }
1754
1755         if (!take_write)
1756                 up_read_ref_node(&ft->node);
1757
1758         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1759                                       dest_num, version);
1760         free_match_list(&match_head);
1761         if (!IS_ERR(rule) ||
1762             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1763                 if (take_write)
1764                         up_write_ref_node(&ft->node, false);
1765                 return rule;
1766         }
1767
1768         if (!take_write) {
1769                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1770                 take_write = true;
1771         }
1772
1773         if (PTR_ERR(rule) == -EAGAIN ||
1774             version != atomic_read(&ft->node.version))
1775                 goto search_again_locked;
1776
1777         g = alloc_auto_flow_group(ft, spec);
1778         if (IS_ERR(g)) {
1779                 rule = ERR_CAST(g);
1780                 up_write_ref_node(&ft->node, false);
1781                 return rule;
1782         }
1783
1784         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1785         up_write_ref_node(&ft->node, false);
1786
1787         err = create_auto_flow_group(ft, g);
1788         if (err)
1789                 goto err_release_fg;
1790
1791         fte = alloc_fte(ft, spec->match_value, flow_act);
1792         if (IS_ERR(fte)) {
1793                 err = PTR_ERR(fte);
1794                 goto err_release_fg;
1795         }
1796
1797         err = insert_fte(g, fte);
1798         if (err) {
1799                 kmem_cache_free(steering->ftes_cache, fte);
1800                 goto err_release_fg;
1801         }
1802
1803         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1804         up_write_ref_node(&g->node, false);
1805         rule = add_rule_fg(g, spec->match_value, flow_act, dest,
1806                            dest_num, fte);
1807         up_write_ref_node(&fte->node, false);
1808         tree_put_node(&fte->node, false);
1809         tree_put_node(&g->node, false);
1810         return rule;
1811
1812 err_release_fg:
1813         up_write_ref_node(&g->node, false);
1814         tree_put_node(&g->node, false);
1815         return ERR_PTR(err);
1816 }
1817
1818 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1819 {
1820         return ((ft->type == FS_FT_NIC_RX) &&
1821                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1822 }
1823
1824 struct mlx5_flow_handle *
1825 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1826                     struct mlx5_flow_spec *spec,
1827                     struct mlx5_flow_act *flow_act,
1828                     struct mlx5_flow_destination *dest,
1829                     int num_dest)
1830 {
1831         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1832         struct mlx5_flow_destination gen_dest = {};
1833         struct mlx5_flow_table *next_ft = NULL;
1834         struct mlx5_flow_handle *handle = NULL;
1835         u32 sw_action = flow_act->action;
1836         struct fs_prio *prio;
1837
1838         fs_get_obj(prio, ft->node.parent);
1839         if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1840                 if (!fwd_next_prio_supported(ft))
1841                         return ERR_PTR(-EOPNOTSUPP);
1842                 if (num_dest)
1843                         return ERR_PTR(-EINVAL);
1844                 mutex_lock(&root->chain_lock);
1845                 next_ft = find_next_chained_ft(prio);
1846                 if (next_ft) {
1847                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1848                         gen_dest.ft = next_ft;
1849                         dest = &gen_dest;
1850                         num_dest = 1;
1851                         flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1852                 } else {
1853                         mutex_unlock(&root->chain_lock);
1854                         return ERR_PTR(-EOPNOTSUPP);
1855                 }
1856         }
1857
1858         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1859
1860         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1861                 if (!IS_ERR_OR_NULL(handle) &&
1862                     (list_empty(&handle->rule[0]->next_ft))) {
1863                         mutex_lock(&next_ft->lock);
1864                         list_add(&handle->rule[0]->next_ft,
1865                                  &next_ft->fwd_rules);
1866                         mutex_unlock(&next_ft->lock);
1867                         handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1868                 }
1869                 mutex_unlock(&root->chain_lock);
1870         }
1871         return handle;
1872 }
1873 EXPORT_SYMBOL(mlx5_add_flow_rules);
1874
1875 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1876 {
1877         struct fs_fte *fte;
1878         int i;
1879
1880         /* In order to consolidate the HW changes we lock the FTE for other
1881          * changes, and increase its refcount, in order not to perform the
1882          * "del" functions of the FTE. Will handle them here.
1883          * The removal of the rules is done under locked FTE.
1884          * After removing all the handle's rules, if there are remaining
1885          * rules, it means we just need to modify the FTE in FW, and
1886          * unlock/decrease the refcount we increased before.
1887          * Otherwise, it means the FTE should be deleted. First delete the
1888          * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
1889          * the FTE, which will handle the last decrease of the refcount, as
1890          * well as required handling of its parent.
1891          */
1892         fs_get_obj(fte, handle->rule[0]->node.parent);
1893         down_write_ref_node(&fte->node, false);
1894         for (i = handle->num_rules - 1; i >= 0; i--)
1895                 tree_remove_node(&handle->rule[i]->node, true);
1896         if (fte->modify_mask && fte->dests_size) {
1897                 modify_fte(fte);
1898                 up_write_ref_node(&fte->node, false);
1899         } else {
1900                 del_hw_fte(&fte->node);
1901                 up_write(&fte->node.lock);
1902                 tree_put_node(&fte->node, false);
1903         }
1904         kfree(handle);
1905 }
1906 EXPORT_SYMBOL(mlx5_del_flow_rules);
1907
1908 /* Assuming prio->node.children(flow tables) is sorted by level */
1909 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1910 {
1911         struct fs_prio *prio;
1912
1913         fs_get_obj(prio, ft->node.parent);
1914
1915         if (!list_is_last(&ft->node.list, &prio->node.children))
1916                 return list_next_entry(ft, node.list);
1917         return find_next_chained_ft(prio);
1918 }
1919
1920 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1921 {
1922         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1923         struct mlx5_ft_underlay_qp *uqp;
1924         struct mlx5_flow_table *new_root_ft = NULL;
1925         int err = 0;
1926         u32 qpn;
1927
1928         if (root->root_ft != ft)
1929                 return 0;
1930
1931         new_root_ft = find_next_ft(ft);
1932         if (!new_root_ft) {
1933                 root->root_ft = NULL;
1934                 return 0;
1935         }
1936
1937         if (list_empty(&root->underlay_qpns)) {
1938                 /* Don't set any QPN (zero) in case QPN list is empty */
1939                 qpn = 0;
1940                 err = root->cmds->update_root_ft(root, new_root_ft,
1941                                                  qpn, false);
1942         } else {
1943                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
1944                         qpn = uqp->qpn;
1945                         err = root->cmds->update_root_ft(root,
1946                                                          new_root_ft, qpn,
1947                                                          false);
1948                         if (err)
1949                                 break;
1950                 }
1951         }
1952
1953         if (err)
1954                 mlx5_core_warn(root->dev,
1955                                "Update root flow table of id(%u) qpn(%d) failed\n",
1956                                ft->id, qpn);
1957         else
1958                 root->root_ft = new_root_ft;
1959
1960         return 0;
1961 }
1962
1963 /* Connect flow table from previous priority to
1964  * the next flow table.
1965  */
1966 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1967 {
1968         struct mlx5_core_dev *dev = get_dev(&ft->node);
1969         struct mlx5_flow_table *next_ft;
1970         struct fs_prio *prio;
1971         int err = 0;
1972
1973         err = update_root_ft_destroy(ft);
1974         if (err)
1975                 return err;
1976
1977         fs_get_obj(prio, ft->node.parent);
1978         if  (!(list_first_entry(&prio->node.children,
1979                                 struct mlx5_flow_table,
1980                                 node.list) == ft))
1981                 return 0;
1982
1983         next_ft = find_next_chained_ft(prio);
1984         err = connect_fwd_rules(dev, next_ft, ft);
1985         if (err)
1986                 return err;
1987
1988         err = connect_prev_fts(dev, next_ft, prio);
1989         if (err)
1990                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1991                                ft->id);
1992         return err;
1993 }
1994
1995 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1996 {
1997         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1998         int err = 0;
1999
2000         mutex_lock(&root->chain_lock);
2001         err = disconnect_flow_table(ft);
2002         if (err) {
2003                 mutex_unlock(&root->chain_lock);
2004                 return err;
2005         }
2006         if (tree_remove_node(&ft->node, false))
2007                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2008                                ft->id);
2009         mutex_unlock(&root->chain_lock);
2010
2011         return err;
2012 }
2013 EXPORT_SYMBOL(mlx5_destroy_flow_table);
2014
2015 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2016 {
2017         if (tree_remove_node(&fg->node, false))
2018                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2019                                fg->id);
2020 }
2021
2022 struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2023                                                 int n)
2024 {
2025         struct mlx5_flow_steering *steering = dev->priv.steering;
2026
2027         if (!steering || !steering->fdb_sub_ns)
2028                 return NULL;
2029
2030         return steering->fdb_sub_ns[n];
2031 }
2032 EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2033
2034 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2035                                                     enum mlx5_flow_namespace_type type)
2036 {
2037         struct mlx5_flow_steering *steering = dev->priv.steering;
2038         struct mlx5_flow_root_namespace *root_ns;
2039         int prio = 0;
2040         struct fs_prio *fs_prio;
2041         struct mlx5_flow_namespace *ns;
2042
2043         if (!steering)
2044                 return NULL;
2045
2046         switch (type) {
2047         case MLX5_FLOW_NAMESPACE_FDB:
2048                 if (steering->fdb_root_ns)
2049                         return &steering->fdb_root_ns->ns;
2050                 return NULL;
2051         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2052                 if (steering->sniffer_rx_root_ns)
2053                         return &steering->sniffer_rx_root_ns->ns;
2054                 return NULL;
2055         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2056                 if (steering->sniffer_tx_root_ns)
2057                         return &steering->sniffer_tx_root_ns->ns;
2058                 return NULL;
2059         case MLX5_FLOW_NAMESPACE_RDMA_RX:
2060                 if (steering->rdma_rx_root_ns)
2061                         return &steering->rdma_rx_root_ns->ns;
2062                 return NULL;
2063         default:
2064                 break;
2065         }
2066
2067         if (type == MLX5_FLOW_NAMESPACE_EGRESS) {
2068                 root_ns = steering->egress_root_ns;
2069         } else { /* Must be NIC RX */
2070                 root_ns = steering->root_ns;
2071                 prio = type;
2072         }
2073
2074         if (!root_ns)
2075                 return NULL;
2076
2077         fs_prio = find_prio(&root_ns->ns, prio);
2078         if (!fs_prio)
2079                 return NULL;
2080
2081         ns = list_first_entry(&fs_prio->node.children,
2082                               typeof(*ns),
2083                               node.list);
2084
2085         return ns;
2086 }
2087 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2088
2089 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2090                                                               enum mlx5_flow_namespace_type type,
2091                                                               int vport)
2092 {
2093         struct mlx5_flow_steering *steering = dev->priv.steering;
2094
2095         if (!steering || vport >= MLX5_TOTAL_VPORTS(dev))
2096                 return NULL;
2097
2098         switch (type) {
2099         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2100                 if (steering->esw_egress_root_ns &&
2101                     steering->esw_egress_root_ns[vport])
2102                         return &steering->esw_egress_root_ns[vport]->ns;
2103                 else
2104                         return NULL;
2105         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2106                 if (steering->esw_ingress_root_ns &&
2107                     steering->esw_ingress_root_ns[vport])
2108                         return &steering->esw_ingress_root_ns[vport]->ns;
2109                 else
2110                         return NULL;
2111         default:
2112                 return NULL;
2113         }
2114 }
2115
2116 static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2117                                        unsigned int prio,
2118                                        int num_levels,
2119                                        enum fs_node_type type)
2120 {
2121         struct fs_prio *fs_prio;
2122
2123         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2124         if (!fs_prio)
2125                 return ERR_PTR(-ENOMEM);
2126
2127         fs_prio->node.type = type;
2128         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2129         tree_add_node(&fs_prio->node, &ns->node);
2130         fs_prio->num_levels = num_levels;
2131         fs_prio->prio = prio;
2132         list_add_tail(&fs_prio->node.list, &ns->node.children);
2133
2134         return fs_prio;
2135 }
2136
2137 static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2138                                               unsigned int prio,
2139                                               int num_levels)
2140 {
2141         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2142 }
2143
2144 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2145                                       unsigned int prio, int num_levels)
2146 {
2147         return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2148 }
2149
2150 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2151                                                      *ns)
2152 {
2153         ns->node.type = FS_TYPE_NAMESPACE;
2154
2155         return ns;
2156 }
2157
2158 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
2159 {
2160         struct mlx5_flow_namespace      *ns;
2161
2162         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2163         if (!ns)
2164                 return ERR_PTR(-ENOMEM);
2165
2166         fs_init_namespace(ns);
2167         tree_init_node(&ns->node, NULL, del_sw_ns);
2168         tree_add_node(&ns->node, &prio->node);
2169         list_add_tail(&ns->node.list, &prio->node.children);
2170
2171         return ns;
2172 }
2173
2174 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2175                              struct init_tree_node *prio_metadata)
2176 {
2177         struct fs_prio *fs_prio;
2178         int i;
2179
2180         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2181                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2182                 if (IS_ERR(fs_prio))
2183                         return PTR_ERR(fs_prio);
2184         }
2185         return 0;
2186 }
2187
2188 #define FLOW_TABLE_BIT_SZ 1
2189 #define GET_FLOW_TABLE_CAP(dev, offset) \
2190         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2191                         offset / 32)) >>                                        \
2192           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2193 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2194 {
2195         int i;
2196
2197         for (i = 0; i < caps->arr_sz; i++) {
2198                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2199                         return false;
2200         }
2201         return true;
2202 }
2203
2204 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2205                                     struct init_tree_node *init_node,
2206                                     struct fs_node *fs_parent_node,
2207                                     struct init_tree_node *init_parent_node,
2208                                     int prio)
2209 {
2210         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2211                                               flow_table_properties_nic_receive.
2212                                               max_ft_level);
2213         struct mlx5_flow_namespace *fs_ns;
2214         struct fs_prio *fs_prio;
2215         struct fs_node *base;
2216         int i;
2217         int err;
2218
2219         if (init_node->type == FS_TYPE_PRIO) {
2220                 if ((init_node->min_ft_level > max_ft_level) ||
2221                     !has_required_caps(steering->dev, &init_node->caps))
2222                         return 0;
2223
2224                 fs_get_obj(fs_ns, fs_parent_node);
2225                 if (init_node->num_leaf_prios)
2226                         return create_leaf_prios(fs_ns, prio, init_node);
2227                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2228                 if (IS_ERR(fs_prio))
2229                         return PTR_ERR(fs_prio);
2230                 base = &fs_prio->node;
2231         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2232                 fs_get_obj(fs_prio, fs_parent_node);
2233                 fs_ns = fs_create_namespace(fs_prio);
2234                 if (IS_ERR(fs_ns))
2235                         return PTR_ERR(fs_ns);
2236                 base = &fs_ns->node;
2237         } else {
2238                 return -EINVAL;
2239         }
2240         prio = 0;
2241         for (i = 0; i < init_node->ar_size; i++) {
2242                 err = init_root_tree_recursive(steering, &init_node->children[i],
2243                                                base, init_node, prio);
2244                 if (err)
2245                         return err;
2246                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2247                     init_node->children[i].num_leaf_prios) {
2248                         prio += init_node->children[i].num_leaf_prios;
2249                 }
2250         }
2251
2252         return 0;
2253 }
2254
2255 static int init_root_tree(struct mlx5_flow_steering *steering,
2256                           struct init_tree_node *init_node,
2257                           struct fs_node *fs_parent_node)
2258 {
2259         int i;
2260         struct mlx5_flow_namespace *fs_ns;
2261         int err;
2262
2263         fs_get_obj(fs_ns, fs_parent_node);
2264         for (i = 0; i < init_node->ar_size; i++) {
2265                 err = init_root_tree_recursive(steering, &init_node->children[i],
2266                                                &fs_ns->node,
2267                                                init_node, i);
2268                 if (err)
2269                         return err;
2270         }
2271         return 0;
2272 }
2273
2274 static struct mlx5_flow_root_namespace
2275 *create_root_ns(struct mlx5_flow_steering *steering,
2276                 enum fs_flow_table_type table_type)
2277 {
2278         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2279         struct mlx5_flow_root_namespace *root_ns;
2280         struct mlx5_flow_namespace *ns;
2281
2282         if (mlx5_accel_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2283             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2284                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2285
2286         /* Create the root namespace */
2287         root_ns = kvzalloc(sizeof(*root_ns), GFP_KERNEL);
2288         if (!root_ns)
2289                 return NULL;
2290
2291         root_ns->dev = steering->dev;
2292         root_ns->table_type = table_type;
2293         root_ns->cmds = cmds;
2294
2295         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2296
2297         ns = &root_ns->ns;
2298         fs_init_namespace(ns);
2299         mutex_init(&root_ns->chain_lock);
2300         tree_init_node(&ns->node, NULL, NULL);
2301         tree_add_node(&ns->node, NULL);
2302
2303         return root_ns;
2304 }
2305
2306 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2307
2308 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2309 {
2310         struct fs_prio *prio;
2311
2312         fs_for_each_prio(prio, ns) {
2313                  /* This updates prio start_level and num_levels */
2314                 set_prio_attrs_in_prio(prio, acc_level);
2315                 acc_level += prio->num_levels;
2316         }
2317         return acc_level;
2318 }
2319
2320 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2321 {
2322         struct mlx5_flow_namespace *ns;
2323         int acc_level_ns = acc_level;
2324
2325         prio->start_level = acc_level;
2326         fs_for_each_ns(ns, prio)
2327                 /* This updates start_level and num_levels of ns's priority descendants */
2328                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2329         if (!prio->num_levels)
2330                 prio->num_levels = acc_level_ns - prio->start_level;
2331         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2332 }
2333
2334 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2335 {
2336         struct mlx5_flow_namespace *ns = &root_ns->ns;
2337         struct fs_prio *prio;
2338         int start_level = 0;
2339
2340         fs_for_each_prio(prio, ns) {
2341                 set_prio_attrs_in_prio(prio, start_level);
2342                 start_level += prio->num_levels;
2343         }
2344 }
2345
2346 #define ANCHOR_PRIO 0
2347 #define ANCHOR_SIZE 1
2348 #define ANCHOR_LEVEL 0
2349 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2350 {
2351         struct mlx5_flow_namespace *ns = NULL;
2352         struct mlx5_flow_table_attr ft_attr = {};
2353         struct mlx5_flow_table *ft;
2354
2355         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2356         if (WARN_ON(!ns))
2357                 return -EINVAL;
2358
2359         ft_attr.max_fte = ANCHOR_SIZE;
2360         ft_attr.level   = ANCHOR_LEVEL;
2361         ft_attr.prio    = ANCHOR_PRIO;
2362
2363         ft = mlx5_create_flow_table(ns, &ft_attr);
2364         if (IS_ERR(ft)) {
2365                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2366                 return PTR_ERR(ft);
2367         }
2368         return 0;
2369 }
2370
2371 static int init_root_ns(struct mlx5_flow_steering *steering)
2372 {
2373         int err;
2374
2375         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2376         if (!steering->root_ns)
2377                 return -ENOMEM;
2378
2379         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2380         if (err)
2381                 goto out_err;
2382
2383         set_prio_attrs(steering->root_ns);
2384         err = create_anchor_flow_table(steering);
2385         if (err)
2386                 goto out_err;
2387
2388         return 0;
2389
2390 out_err:
2391         cleanup_root_ns(steering->root_ns);
2392         steering->root_ns = NULL;
2393         return err;
2394 }
2395
2396 static void clean_tree(struct fs_node *node)
2397 {
2398         if (node) {
2399                 struct fs_node *iter;
2400                 struct fs_node *temp;
2401
2402                 tree_get_node(node);
2403                 list_for_each_entry_safe(iter, temp, &node->children, list)
2404                         clean_tree(iter);
2405                 tree_put_node(node, false);
2406                 tree_remove_node(node, false);
2407         }
2408 }
2409
2410 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2411 {
2412         if (!root_ns)
2413                 return;
2414
2415         clean_tree(&root_ns->ns.node);
2416 }
2417
2418 static void cleanup_egress_acls_root_ns(struct mlx5_core_dev *dev)
2419 {
2420         struct mlx5_flow_steering *steering = dev->priv.steering;
2421         int i;
2422
2423         if (!steering->esw_egress_root_ns)
2424                 return;
2425
2426         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2427                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2428
2429         kfree(steering->esw_egress_root_ns);
2430 }
2431
2432 static void cleanup_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2433 {
2434         struct mlx5_flow_steering *steering = dev->priv.steering;
2435         int i;
2436
2437         if (!steering->esw_ingress_root_ns)
2438                 return;
2439
2440         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2441                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2442
2443         kfree(steering->esw_ingress_root_ns);
2444 }
2445
2446 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2447 {
2448         struct mlx5_flow_steering *steering = dev->priv.steering;
2449
2450         cleanup_root_ns(steering->root_ns);
2451         cleanup_egress_acls_root_ns(dev);
2452         cleanup_ingress_acls_root_ns(dev);
2453         cleanup_root_ns(steering->fdb_root_ns);
2454         steering->fdb_root_ns = NULL;
2455         kfree(steering->fdb_sub_ns);
2456         steering->fdb_sub_ns = NULL;
2457         cleanup_root_ns(steering->sniffer_rx_root_ns);
2458         cleanup_root_ns(steering->sniffer_tx_root_ns);
2459         cleanup_root_ns(steering->rdma_rx_root_ns);
2460         cleanup_root_ns(steering->egress_root_ns);
2461         mlx5_cleanup_fc_stats(dev);
2462         kmem_cache_destroy(steering->ftes_cache);
2463         kmem_cache_destroy(steering->fgs_cache);
2464         kfree(steering);
2465 }
2466
2467 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2468 {
2469         struct fs_prio *prio;
2470
2471         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2472         if (!steering->sniffer_tx_root_ns)
2473                 return -ENOMEM;
2474
2475         /* Create single prio */
2476         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2477         if (IS_ERR(prio)) {
2478                 cleanup_root_ns(steering->sniffer_tx_root_ns);
2479                 return PTR_ERR(prio);
2480         }
2481         return 0;
2482 }
2483
2484 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2485 {
2486         struct fs_prio *prio;
2487
2488         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2489         if (!steering->sniffer_rx_root_ns)
2490                 return -ENOMEM;
2491
2492         /* Create single prio */
2493         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2494         if (IS_ERR(prio)) {
2495                 cleanup_root_ns(steering->sniffer_rx_root_ns);
2496                 return PTR_ERR(prio);
2497         }
2498         return 0;
2499 }
2500
2501 static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2502 {
2503         struct fs_prio *prio;
2504
2505         steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2506         if (!steering->rdma_rx_root_ns)
2507                 return -ENOMEM;
2508
2509         steering->rdma_rx_root_ns->def_miss_action =
2510                 MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN;
2511
2512         /* Create single prio */
2513         prio = fs_create_prio(&steering->rdma_rx_root_ns->ns, 0, 1);
2514         if (IS_ERR(prio)) {
2515                 cleanup_root_ns(steering->rdma_rx_root_ns);
2516                 return PTR_ERR(prio);
2517         }
2518         return 0;
2519 }
2520 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2521 {
2522         struct mlx5_flow_namespace *ns;
2523         struct fs_prio *maj_prio;
2524         struct fs_prio *min_prio;
2525         int levels;
2526         int chain;
2527         int prio;
2528         int err;
2529
2530         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2531         if (!steering->fdb_root_ns)
2532                 return -ENOMEM;
2533
2534         steering->fdb_sub_ns = kzalloc(sizeof(steering->fdb_sub_ns) *
2535                                        (FDB_MAX_CHAIN + 1), GFP_KERNEL);
2536         if (!steering->fdb_sub_ns)
2537                 return -ENOMEM;
2538
2539         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2540                                   1);
2541         if (IS_ERR(maj_prio)) {
2542                 err = PTR_ERR(maj_prio);
2543                 goto out_err;
2544         }
2545
2546         levels = 2 * FDB_MAX_PRIO * (FDB_MAX_CHAIN + 1);
2547         maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2548                                           FDB_FAST_PATH,
2549                                           levels);
2550         if (IS_ERR(maj_prio)) {
2551                 err = PTR_ERR(maj_prio);
2552                 goto out_err;
2553         }
2554
2555         for (chain = 0; chain <= FDB_MAX_CHAIN; chain++) {
2556                 ns = fs_create_namespace(maj_prio);
2557                 if (IS_ERR(ns)) {
2558                         err = PTR_ERR(ns);
2559                         goto out_err;
2560                 }
2561
2562                 for (prio = 0; prio < FDB_MAX_PRIO * (chain + 1); prio++) {
2563                         min_prio = fs_create_prio(ns, prio, 2);
2564                         if (IS_ERR(min_prio)) {
2565                                 err = PTR_ERR(min_prio);
2566                                 goto out_err;
2567                         }
2568                 }
2569
2570                 steering->fdb_sub_ns[chain] = ns;
2571         }
2572
2573         maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2574         if (IS_ERR(maj_prio)) {
2575                 err = PTR_ERR(maj_prio);
2576                 goto out_err;
2577         }
2578
2579         set_prio_attrs(steering->fdb_root_ns);
2580         return 0;
2581
2582 out_err:
2583         cleanup_root_ns(steering->fdb_root_ns);
2584         kfree(steering->fdb_sub_ns);
2585         steering->fdb_sub_ns = NULL;
2586         steering->fdb_root_ns = NULL;
2587         return err;
2588 }
2589
2590 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2591 {
2592         struct fs_prio *prio;
2593
2594         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2595         if (!steering->esw_egress_root_ns[vport])
2596                 return -ENOMEM;
2597
2598         /* create 1 prio*/
2599         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2600         return PTR_ERR_OR_ZERO(prio);
2601 }
2602
2603 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2604 {
2605         struct fs_prio *prio;
2606
2607         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2608         if (!steering->esw_ingress_root_ns[vport])
2609                 return -ENOMEM;
2610
2611         /* create 1 prio*/
2612         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2613         return PTR_ERR_OR_ZERO(prio);
2614 }
2615
2616 static int init_egress_acls_root_ns(struct mlx5_core_dev *dev)
2617 {
2618         struct mlx5_flow_steering *steering = dev->priv.steering;
2619         int err;
2620         int i;
2621
2622         steering->esw_egress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2623                                                sizeof(*steering->esw_egress_root_ns),
2624                                                GFP_KERNEL);
2625         if (!steering->esw_egress_root_ns)
2626                 return -ENOMEM;
2627
2628         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2629                 err = init_egress_acl_root_ns(steering, i);
2630                 if (err)
2631                         goto cleanup_root_ns;
2632         }
2633
2634         return 0;
2635
2636 cleanup_root_ns:
2637         for (i--; i >= 0; i--)
2638                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2639         kfree(steering->esw_egress_root_ns);
2640         return err;
2641 }
2642
2643 static int init_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2644 {
2645         struct mlx5_flow_steering *steering = dev->priv.steering;
2646         int err;
2647         int i;
2648
2649         steering->esw_ingress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2650                                                 sizeof(*steering->esw_ingress_root_ns),
2651                                                 GFP_KERNEL);
2652         if (!steering->esw_ingress_root_ns)
2653                 return -ENOMEM;
2654
2655         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2656                 err = init_ingress_acl_root_ns(steering, i);
2657                 if (err)
2658                         goto cleanup_root_ns;
2659         }
2660
2661         return 0;
2662
2663 cleanup_root_ns:
2664         for (i--; i >= 0; i--)
2665                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2666         kfree(steering->esw_ingress_root_ns);
2667         return err;
2668 }
2669
2670 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2671 {
2672         int err;
2673
2674         steering->egress_root_ns = create_root_ns(steering,
2675                                                   FS_FT_NIC_TX);
2676         if (!steering->egress_root_ns)
2677                 return -ENOMEM;
2678
2679         err = init_root_tree(steering, &egress_root_fs,
2680                              &steering->egress_root_ns->ns.node);
2681         if (err)
2682                 goto cleanup;
2683         set_prio_attrs(steering->egress_root_ns);
2684         return 0;
2685 cleanup:
2686         cleanup_root_ns(steering->egress_root_ns);
2687         steering->egress_root_ns = NULL;
2688         return err;
2689 }
2690
2691 int mlx5_init_fs(struct mlx5_core_dev *dev)
2692 {
2693         struct mlx5_flow_steering *steering;
2694         int err = 0;
2695
2696         err = mlx5_init_fc_stats(dev);
2697         if (err)
2698                 return err;
2699
2700         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2701         if (!steering)
2702                 return -ENOMEM;
2703         steering->dev = dev;
2704         dev->priv.steering = steering;
2705
2706         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2707                                                 sizeof(struct mlx5_flow_group), 0,
2708                                                 0, NULL);
2709         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2710                                                  0, NULL);
2711         if (!steering->ftes_cache || !steering->fgs_cache) {
2712                 err = -ENOMEM;
2713                 goto err;
2714         }
2715
2716         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2717               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2718              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2719               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2720             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2721                 err = init_root_ns(steering);
2722                 if (err)
2723                         goto err;
2724         }
2725
2726         if (MLX5_ESWITCH_MANAGER(dev)) {
2727                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2728                         err = init_fdb_root_ns(steering);
2729                         if (err)
2730                                 goto err;
2731                 }
2732                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2733                         err = init_egress_acls_root_ns(dev);
2734                         if (err)
2735                                 goto err;
2736                 }
2737                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2738                         err = init_ingress_acls_root_ns(dev);
2739                         if (err)
2740                                 goto err;
2741                 }
2742         }
2743
2744         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2745                 err = init_sniffer_rx_root_ns(steering);
2746                 if (err)
2747                         goto err;
2748         }
2749
2750         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2751                 err = init_sniffer_tx_root_ns(steering);
2752                 if (err)
2753                         goto err;
2754         }
2755
2756         if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
2757             MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
2758                 err = init_rdma_rx_root_ns(steering);
2759                 if (err)
2760                         goto err;
2761         }
2762
2763         if (MLX5_IPSEC_DEV(dev) || MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
2764                 err = init_egress_root_ns(steering);
2765                 if (err)
2766                         goto err;
2767         }
2768
2769         return 0;
2770 err:
2771         mlx5_cleanup_fs(dev);
2772         return err;
2773 }
2774
2775 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2776 {
2777         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2778         struct mlx5_ft_underlay_qp *new_uqp;
2779         int err = 0;
2780
2781         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
2782         if (!new_uqp)
2783                 return -ENOMEM;
2784
2785         mutex_lock(&root->chain_lock);
2786
2787         if (!root->root_ft) {
2788                 err = -EINVAL;
2789                 goto update_ft_fail;
2790         }
2791
2792         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
2793                                          false);
2794         if (err) {
2795                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
2796                                underlay_qpn, err);
2797                 goto update_ft_fail;
2798         }
2799
2800         new_uqp->qpn = underlay_qpn;
2801         list_add_tail(&new_uqp->list, &root->underlay_qpns);
2802
2803         mutex_unlock(&root->chain_lock);
2804
2805         return 0;
2806
2807 update_ft_fail:
2808         mutex_unlock(&root->chain_lock);
2809         kfree(new_uqp);
2810         return err;
2811 }
2812 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
2813
2814 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2815 {
2816         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2817         struct mlx5_ft_underlay_qp *uqp;
2818         bool found = false;
2819         int err = 0;
2820
2821         mutex_lock(&root->chain_lock);
2822         list_for_each_entry(uqp, &root->underlay_qpns, list) {
2823                 if (uqp->qpn == underlay_qpn) {
2824                         found = true;
2825                         break;
2826                 }
2827         }
2828
2829         if (!found) {
2830                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
2831                                underlay_qpn);
2832                 err = -EINVAL;
2833                 goto out;
2834         }
2835
2836         err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
2837                                          true);
2838         if (err)
2839                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
2840                                underlay_qpn, err);
2841
2842         list_del(&uqp->list);
2843         mutex_unlock(&root->chain_lock);
2844         kfree(uqp);
2845
2846         return 0;
2847
2848 out:
2849         mutex_unlock(&root->chain_lock);
2850         return err;
2851 }
2852 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);