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/eswitch.h>
  36
  37 #include "mlx5_core.h"
  38 #include "fs_core.h"
  39 #include "fs_cmd.h"
  40 #include "diag/fs_tracepoint.h"
  41 #include "accel/ipsec.h"
  42 #include "fpga/ipsec.h"
  43
  44 #define INIT_TREE_NODE_ARRAY_SIZE(...)  (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
  45                                          sizeof(struct init_tree_node))
  46
  47 #define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
  48                  ...) {.type = FS_TYPE_PRIO,\
  49         .min_ft_level = min_level_val,\
  50         .num_levels = num_levels_val,\
  51         .num_leaf_prios = num_prios_val,\
  52         .caps = caps_val,\
  53         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  54         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  55 }
  56
  57 #define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
  58         ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
  59                  __VA_ARGS__)\
  60
  61 #define ADD_NS(...) {.type = FS_TYPE_NAMESPACE,\
  62         .children = (struct init_tree_node[]) {__VA_ARGS__},\
  63         .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  64 }
  65
  66 #define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  67                                    sizeof(long))
  68
  69 #define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  70
  71 #define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  72                                .caps = (long[]) {__VA_ARGS__} }
  73
  74 #define FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  75                                            FS_CAP(flow_table_properties_nic_receive.modify_root), \
  76                                            FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  77                                            FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  78
  79 #define LEFTOVERS_NUM_LEVELS 1
  80 #define LEFTOVERS_NUM_PRIOS 1
  81
  82 #define BY_PASS_PRIO_NUM_LEVELS 1
  83 #define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
  84                            LEFTOVERS_NUM_PRIOS)
  85
  86 #define ETHTOOL_PRIO_NUM_LEVELS 1
  87 #define ETHTOOL_NUM_PRIOS 11
  88 #define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
  89 /* Vlan, mac, ttc, inner ttc, aRFS */
  90 #define KERNEL_NIC_PRIO_NUM_LEVELS 5
  91 #define KERNEL_NIC_NUM_PRIOS 1
  92 /* One more level for tc */
  93 #define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
  94
  95 #define KERNEL_NIC_TC_NUM_PRIOS  1
  96 #define KERNEL_NIC_TC_NUM_LEVELS 2
  97
  98 #define ANCHOR_NUM_LEVELS 1
  99 #define ANCHOR_NUM_PRIOS 1
 100 #define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 101
 102 #define OFFLOADS_MAX_FT 1
 103 #define OFFLOADS_NUM_PRIOS 1
 104 #define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + 1)
 105
 106 #define LAG_PRIO_NUM_LEVELS 1
 107 #define LAG_NUM_PRIOS 1
 108 #define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 109
 110 struct node_caps {
 111         size_t  arr_sz;
 112         long    *caps;
 113 };
 114
 115 static struct init_tree_node {
 116         enum fs_node_type       type;
 117         struct init_tree_node *children;
 118         int ar_size;
 119         struct node_caps caps;
 120         int min_ft_level;
 121         int num_leaf_prios;
 122         int prio;
 123         int num_levels;
 124 } root_fs = {
 125         .type = FS_TYPE_NAMESPACE,
 126         .ar_size = 7,
 127         .children = (struct init_tree_node[]) {
 128                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 129                          FS_CHAINING_CAPS,
 130                          ADD_NS(ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 131                                                   BY_PASS_PRIO_NUM_LEVELS))),
 132                 ADD_PRIO(0, LAG_MIN_LEVEL, 0,
 133                          FS_CHAINING_CAPS,
 134                          ADD_NS(ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 135                                                   LAG_PRIO_NUM_LEVELS))),
 136                 ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, {},
 137                          ADD_NS(ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS, OFFLOADS_MAX_FT))),
 138                 ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0,
 139                          FS_CHAINING_CAPS,
 140                          ADD_NS(ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 141                                                   ETHTOOL_PRIO_NUM_LEVELS))),
 142                 ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 143                          ADD_NS(ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS, KERNEL_NIC_TC_NUM_LEVELS),
 144                                 ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 145                                                   KERNEL_NIC_PRIO_NUM_LEVELS))),
 146                 ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0,
 147                          FS_CHAINING_CAPS,
 148                          ADD_NS(ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS, LEFTOVERS_NUM_LEVELS))),
 149                 ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 150                          ADD_NS(ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS, ANCHOR_NUM_LEVELS))),
 151         }
 152 };
 153
 154 enum fs_i_lock_class {
 155         FS_LOCK_GRANDPARENT,
 156         FS_LOCK_PARENT,
 157         FS_LOCK_CHILD
 158 };
 159
 160 static const struct rhashtable_params rhash_fte = {
 161         .key_len = FIELD_SIZEOF(struct fs_fte, val),
 162         .key_offset = offsetof(struct fs_fte, val),
 163         .head_offset = offsetof(struct fs_fte, hash),
 164         .automatic_shrinking = true,
 165         .min_size = 1,
 166 };
 167
 168 static const struct rhashtable_params rhash_fg = {
 169         .key_len = FIELD_SIZEOF(struct mlx5_flow_group, mask),
 170         .key_offset = offsetof(struct mlx5_flow_group, mask),
 171         .head_offset = offsetof(struct mlx5_flow_group, hash),
 172         .automatic_shrinking = true,
 173         .min_size = 1,
 174
 175 };
 176
 177 static void del_hw_flow_table(struct fs_node *node);
 178 static void del_hw_flow_group(struct fs_node *node);
 179 static void del_hw_fte(struct fs_node *node);
 180 static void del_sw_flow_table(struct fs_node *node);
 181 static void del_sw_flow_group(struct fs_node *node);
 182 static void del_sw_fte(struct fs_node *node);
 183 static void del_sw_prio(struct fs_node *node);
 184 static void del_sw_ns(struct fs_node *node);
 185 /* Delete rule (destination) is special case that
 186  * requires to lock the FTE for all the deletion process.
 187  */
 188 static void del_sw_hw_rule(struct fs_node *node);
 189 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 190                                 struct mlx5_flow_destination *d2);
 191 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 192 static struct mlx5_flow_rule *
 193 find_flow_rule(struct fs_fte *fte,
 194                struct mlx5_flow_destination *dest);
 195
 196 static void tree_init_node(struct fs_node *node,
 197                            void (*del_hw_func)(struct fs_node *),
 198                            void (*del_sw_func)(struct fs_node *))
 199 {
 200         refcount_set(&node->refcount, 1);
 201         INIT_LIST_HEAD(&node->list);
 202         INIT_LIST_HEAD(&node->children);
 203         init_rwsem(&node->lock);
 204         node->del_hw_func = del_hw_func;
 205         node->del_sw_func = del_sw_func;
 206         node->active = false;
 207 }
 208
 209 static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 210 {
 211         if (parent)
 212                 refcount_inc(&parent->refcount);
 213         node->parent = parent;
 214
 215         /* Parent is the root */
 216         if (!parent)
 217                 node->root = node;
 218         else
 219                 node->root = parent->root;
 220 }
 221
 222 static int tree_get_node(struct fs_node *node)
 223 {
 224         return refcount_inc_not_zero(&node->refcount);
 225 }
 226
 227 static void nested_down_read_ref_node(struct fs_node *node,
 228                                       enum fs_i_lock_class class)
 229 {
 230         if (node) {
 231                 down_read_nested(&node->lock, class);
 232                 refcount_inc(&node->refcount);
 233         }
 234 }
 235
 236 static void nested_down_write_ref_node(struct fs_node *node,
 237                                        enum fs_i_lock_class class)
 238 {
 239         if (node) {
 240                 down_write_nested(&node->lock, class);
 241                 refcount_inc(&node->refcount);
 242         }
 243 }
 244
 245 static void down_write_ref_node(struct fs_node *node)
 246 {
 247         if (node) {
 248                 down_write(&node->lock);
 249                 refcount_inc(&node->refcount);
 250         }
 251 }
 252
 253 static void up_read_ref_node(struct fs_node *node)
 254 {
 255         refcount_dec(&node->refcount);
 256         up_read(&node->lock);
 257 }
 258
 259 static void up_write_ref_node(struct fs_node *node)
 260 {
 261         refcount_dec(&node->refcount);
 262         up_write(&node->lock);
 263 }
 264
 265 static void tree_put_node(struct fs_node *node)
 266 {
 267         struct fs_node *parent_node = node->parent;
 268
 269         if (refcount_dec_and_test(&node->refcount)) {
 270                 if (node->del_hw_func)
 271                         node->del_hw_func(node);
 272                 if (parent_node) {
 273                         /* Only root namespace doesn't have parent and we just
 274                          * need to free its node.
 275                          */
 276                         down_write_ref_node(parent_node);
 277                         list_del_init(&node->list);
 278                         if (node->del_sw_func)
 279                                 node->del_sw_func(node);
 280                         up_write_ref_node(parent_node);
 281                 } else {
 282                         kfree(node);
 283                 }
 284                 node = NULL;
 285         }
 286         if (!node && parent_node)
 287                 tree_put_node(parent_node);
 288 }
 289
 290 static int tree_remove_node(struct fs_node *node)
 291 {
 292         if (refcount_read(&node->refcount) > 1) {
 293                 refcount_dec(&node->refcount);
 294                 return -EEXIST;
 295         }
 296         tree_put_node(node);
 297         return 0;
 298 }
 299
 300 static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 301                                  unsigned int prio)
 302 {
 303         struct fs_prio *iter_prio;
 304
 305         fs_for_each_prio(iter_prio, ns) {
 306                 if (iter_prio->prio == prio)
 307                         return iter_prio;
 308         }
 309
 310         return NULL;
 311 }
 312
 313 static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 314 {
 315         int i;
 316
 317         for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 318                 if (spec->match_value[i] & ~spec->match_criteria[i]) {
 319                         pr_warn("mlx5_core: match_value differs from match_criteria\n");
 320                         return false;
 321                 }
 322
 323         return true;
 324 }
 325
 326 static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 327 {
 328         struct fs_node *root;
 329         struct mlx5_flow_namespace *ns;
 330
 331         root = node->root;
 332
 333         if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 334                 pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 335                 return NULL;
 336         }
 337
 338         ns = container_of(root, struct mlx5_flow_namespace, node);
 339         return container_of(ns, struct mlx5_flow_root_namespace, ns);
 340 }
 341
 342 static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 343 {
 344         struct mlx5_flow_root_namespace *root = find_root(node);
 345
 346         if (root)
 347                 return root->dev->priv.steering;
 348         return NULL;
 349 }
 350
 351 static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 352 {
 353         struct mlx5_flow_root_namespace *root = find_root(node);
 354
 355         if (root)
 356                 return root->dev;
 357         return NULL;
 358 }
 359
 360 static void del_sw_ns(struct fs_node *node)
 361 {
 362         kfree(node);
 363 }
 364
 365 static void del_sw_prio(struct fs_node *node)
 366 {
 367         kfree(node);
 368 }
 369
 370 static void del_hw_flow_table(struct fs_node *node)
 371 {
 372         struct mlx5_flow_root_namespace *root;
 373         struct mlx5_flow_table *ft;
 374         struct mlx5_core_dev *dev;
 375         int err;
 376
 377         fs_get_obj(ft, node);
 378         dev = get_dev(&ft->node);
 379         root = find_root(&ft->node);
 380
 381         if (node->active) {
 382                 err = root->cmds->destroy_flow_table(dev, ft);
 383                 if (err)
 384                         mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 385         }
 386 }
 387
 388 static void del_sw_flow_table(struct fs_node *node)
 389 {
 390         struct mlx5_flow_table *ft;
 391         struct fs_prio *prio;
 392
 393         fs_get_obj(ft, node);
 394
 395         rhltable_destroy(&ft->fgs_hash);
 396         fs_get_obj(prio, ft->node.parent);
 397         prio->num_ft--;
 398         kfree(ft);
 399 }
 400
 401 static void del_sw_hw_rule(struct fs_node *node)
 402 {
 403         struct mlx5_flow_root_namespace *root;
 404         struct mlx5_flow_rule *rule;
 405         struct mlx5_flow_table *ft;
 406         struct mlx5_flow_group *fg;
 407         struct fs_fte *fte;
 408         int modify_mask;
 409         struct mlx5_core_dev *dev = get_dev(node);
 410         int err;
 411         bool update_fte = false;
 412
 413         fs_get_obj(rule, node);
 414         fs_get_obj(fte, rule->node.parent);
 415         fs_get_obj(fg, fte->node.parent);
 416         fs_get_obj(ft, fg->node.parent);
 417         trace_mlx5_fs_del_rule(rule);
 418         if (rule->sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
 419                 mutex_lock(&rule->dest_attr.ft->lock);
 420                 list_del(&rule->next_ft);
 421                 mutex_unlock(&rule->dest_attr.ft->lock);
 422         }
 423
 424         if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 425             --fte->dests_size) {
 426                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 427                               BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 428                 fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 429                 update_fte = true;
 430                 goto out;
 431         }
 432
 433         if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 434             --fte->dests_size) {
 435                 modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST),
 436                 update_fte = true;
 437         }
 438 out:
 439         root = find_root(&ft->node);
 440         if (update_fte && fte->dests_size) {
 441                 err = root->cmds->update_fte(dev, ft, fg->id, modify_mask, fte);
 442                 if (err)
 443                         mlx5_core_warn(dev,
 444                                        "%s can't del rule fg id=%d fte_index=%d\n",
 445                                        __func__, fg->id, fte->index);
 446         }
 447         kfree(rule);
 448 }
 449
 450 static void del_hw_fte(struct fs_node *node)
 451 {
 452         struct mlx5_flow_root_namespace *root;
 453         struct mlx5_flow_table *ft;
 454         struct mlx5_flow_group *fg;
 455         struct mlx5_core_dev *dev;
 456         struct fs_fte *fte;
 457         int err;
 458
 459         fs_get_obj(fte, node);
 460         fs_get_obj(fg, fte->node.parent);
 461         fs_get_obj(ft, fg->node.parent);
 462
 463         trace_mlx5_fs_del_fte(fte);
 464         dev = get_dev(&ft->node);
 465         root = find_root(&ft->node);
 466         if (node->active) {
 467                 err = root->cmds->delete_fte(dev, ft, fte);
 468                 if (err)
 469                         mlx5_core_warn(dev,
 470                                        "flow steering can't delete fte in index %d of flow group id %d\n",
 471                                        fte->index, fg->id);
 472         }
 473 }
 474
 475 static void del_sw_fte(struct fs_node *node)
 476 {
 477         struct mlx5_flow_steering *steering = get_steering(node);
 478         struct mlx5_flow_group *fg;
 479         struct fs_fte *fte;
 480         int err;
 481
 482         fs_get_obj(fte, node);
 483         fs_get_obj(fg, fte->node.parent);
 484
 485         err = rhashtable_remove_fast(&fg->ftes_hash,
 486                                      &fte->hash,
 487                                      rhash_fte);
 488         WARN_ON(err);
 489         ida_simple_remove(&fg->fte_allocator, fte->index - fg->start_index);
 490         kmem_cache_free(steering->ftes_cache, fte);
 491 }
 492
 493 static void del_hw_flow_group(struct fs_node *node)
 494 {
 495         struct mlx5_flow_root_namespace *root;
 496         struct mlx5_flow_group *fg;
 497         struct mlx5_flow_table *ft;
 498         struct mlx5_core_dev *dev;
 499
 500         fs_get_obj(fg, node);
 501         fs_get_obj(ft, fg->node.parent);
 502         dev = get_dev(&ft->node);
 503         trace_mlx5_fs_del_fg(fg);
 504
 505         root = find_root(&ft->node);
 506         if (fg->node.active && root->cmds->destroy_flow_group(dev, ft, fg->id))
 507                 mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 508                                fg->id, ft->id);
 509 }
 510
 511 static void del_sw_flow_group(struct fs_node *node)
 512 {
 513         struct mlx5_flow_steering *steering = get_steering(node);
 514         struct mlx5_flow_group *fg;
 515         struct mlx5_flow_table *ft;
 516         int err;
 517
 518         fs_get_obj(fg, node);
 519         fs_get_obj(ft, fg->node.parent);
 520
 521         rhashtable_destroy(&fg->ftes_hash);
 522         ida_destroy(&fg->fte_allocator);
 523         if (ft->autogroup.active)
 524                 ft->autogroup.num_groups--;
 525         err = rhltable_remove(&ft->fgs_hash,
 526                               &fg->hash,
 527                               rhash_fg);
 528         WARN_ON(err);
 529         kmem_cache_free(steering->fgs_cache, fg);
 530 }
 531
 532 static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 533 {
 534         int index;
 535         int ret;
 536
 537         index = ida_simple_get(&fg->fte_allocator, 0, fg->max_ftes, GFP_KERNEL);
 538         if (index < 0)
 539                 return index;
 540
 541         fte->index = index + fg->start_index;
 542         ret = rhashtable_insert_fast(&fg->ftes_hash,
 543                                      &fte->hash,
 544                                      rhash_fte);
 545         if (ret)
 546                 goto err_ida_remove;
 547
 548         tree_add_node(&fte->node, &fg->node);
 549         list_add_tail(&fte->node.list, &fg->node.children);
 550         return 0;
 551
 552 err_ida_remove:
 553         ida_simple_remove(&fg->fte_allocator, index);
 554         return ret;
 555 }
 556
 557 static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 558                                 u32 *match_value,
 559                                 struct mlx5_flow_act *flow_act)
 560 {
 561         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 562         struct fs_fte *fte;
 563
 564         fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 565         if (!fte)
 566                 return ERR_PTR(-ENOMEM);
 567
 568         memcpy(fte->val, match_value, sizeof(fte->val));
 569         fte->node.type =  FS_TYPE_FLOW_ENTRY;
 570         fte->action = *flow_act;
 571
 572         tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
 573
 574         return fte;
 575 }
 576
 577 static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 578                                struct mlx5_flow_group *fg)
 579 {
 580         rhashtable_destroy(&fg->ftes_hash);
 581         kmem_cache_free(steering->fgs_cache, fg);
 582 }
 583
 584 static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 585                                                 u8 match_criteria_enable,
 586                                                 void *match_criteria,
 587                                                 int start_index,
 588                                                 int end_index)
 589 {
 590         struct mlx5_flow_group *fg;
 591         int ret;
 592
 593         fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 594         if (!fg)
 595                 return ERR_PTR(-ENOMEM);
 596
 597         ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 598         if (ret) {
 599                 kmem_cache_free(steering->fgs_cache, fg);
 600                 return ERR_PTR(ret);
 601 }
 602         ida_init(&fg->fte_allocator);
 603         fg->mask.match_criteria_enable = match_criteria_enable;
 604         memcpy(&fg->mask.match_criteria, match_criteria,
 605                sizeof(fg->mask.match_criteria));
 606         fg->node.type =  FS_TYPE_FLOW_GROUP;
 607         fg->start_index = start_index;
 608         fg->max_ftes = end_index - start_index + 1;
 609
 610         return fg;
 611 }
 612
 613 static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 614                                                        u8 match_criteria_enable,
 615                                                        void *match_criteria,
 616                                                        int start_index,
 617                                                        int end_index,
 618                                                        struct list_head *prev)
 619 {
 620         struct mlx5_flow_steering *steering = get_steering(&ft->node);
 621         struct mlx5_flow_group *fg;
 622         int ret;
 623
 624         fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 625                               start_index, end_index);
 626         if (IS_ERR(fg))
 627                 return fg;
 628
 629         /* initialize refcnt, add to parent list */
 630         ret = rhltable_insert(&ft->fgs_hash,
 631                               &fg->hash,
 632                               rhash_fg);
 633         if (ret) {
 634                 dealloc_flow_group(steering, fg);
 635                 return ERR_PTR(ret);
 636         }
 637
 638         tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 639         tree_add_node(&fg->node, &ft->node);
 640         /* Add node to group list */
 641         list_add(&fg->node.list, prev);
 642         atomic_inc(&ft->node.version);
 643
 644         return fg;
 645 }
 646
 647 static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport, int max_fte,
 648                                                 enum fs_flow_table_type table_type,
 649                                                 enum fs_flow_table_op_mod op_mod,
 650                                                 u32 flags)
 651 {
 652         struct mlx5_flow_table *ft;
 653         int ret;
 654
 655         ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 656         if (!ft)
 657                 return ERR_PTR(-ENOMEM);
 658
 659         ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 660         if (ret) {
 661                 kfree(ft);
 662                 return ERR_PTR(ret);
 663         }
 664
 665         ft->level = level;
 666         ft->node.type = FS_TYPE_FLOW_TABLE;
 667         ft->op_mod = op_mod;
 668         ft->type = table_type;
 669         ft->vport = vport;
 670         ft->max_fte = max_fte;
 671         ft->flags = flags;
 672         INIT_LIST_HEAD(&ft->fwd_rules);
 673         mutex_init(&ft->lock);
 674
 675         return ft;
 676 }
 677
 678 /* If reverse is false, then we search for the first flow table in the
 679  * root sub-tree from start(closest from right), else we search for the
 680  * last flow table in the root sub-tree till start(closest from left).
 681  */
 682 static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 683                                                          struct list_head *start,
 684                                                          bool reverse)
 685 {
 686 #define list_advance_entry(pos, reverse)                \
 687         ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 688
 689 #define list_for_each_advance_continue(pos, head, reverse)      \
 690         for (pos = list_advance_entry(pos, reverse);            \
 691              &pos->list != (head);                              \
 692              pos = list_advance_entry(pos, reverse))
 693
 694         struct fs_node *iter = list_entry(start, struct fs_node, list);
 695         struct mlx5_flow_table *ft = NULL;
 696
 697         if (!root)
 698                 return NULL;
 699
 700         list_for_each_advance_continue(iter, &root->children, reverse) {
 701                 if (iter->type == FS_TYPE_FLOW_TABLE) {
 702                         fs_get_obj(ft, iter);
 703                         return ft;
 704                 }
 705                 ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 706                 if (ft)
 707                         return ft;
 708         }
 709
 710         return ft;
 711 }
 712
 713 /* If reverse if false then return the first flow table in next priority of
 714  * prio in the tree, else return the last flow table in the previous priority
 715  * of prio in the tree.
 716  */
 717 static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 718 {
 719         struct mlx5_flow_table *ft = NULL;
 720         struct fs_node *curr_node;
 721         struct fs_node *parent;
 722
 723         parent = prio->node.parent;
 724         curr_node = &prio->node;
 725         while (!ft && parent) {
 726                 ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 727                 curr_node = parent;
 728                 parent = curr_node->parent;
 729         }
 730         return ft;
 731 }
 732
 733 /* Assuming all the tree is locked by mutex chain lock */
 734 static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 735 {
 736         return find_closest_ft(prio, false);
 737 }
 738
 739 /* Assuming all the tree is locked by mutex chain lock */
 740 static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 741 {
 742         return find_closest_ft(prio, true);
 743 }
 744
 745 static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 746                                struct fs_prio *prio,
 747                                struct mlx5_flow_table *ft)
 748 {
 749         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 750         struct mlx5_flow_table *iter;
 751         int i = 0;
 752         int err;
 753
 754         fs_for_each_ft(iter, prio) {
 755                 i++;
 756                 err = root->cmds->modify_flow_table(dev, iter, ft);
 757                 if (err) {
 758                         mlx5_core_warn(dev, "Failed to modify flow table %d\n",
 759                                        iter->id);
 760                         /* The driver is out of sync with the FW */
 761                         if (i > 1)
 762                                 WARN_ON(true);
 763                         return err;
 764                 }
 765         }
 766         return 0;
 767 }
 768
 769 /* Connect flow tables from previous priority of prio to ft */
 770 static int connect_prev_fts(struct mlx5_core_dev *dev,
 771                             struct mlx5_flow_table *ft,
 772                             struct fs_prio *prio)
 773 {
 774         struct mlx5_flow_table *prev_ft;
 775
 776         prev_ft = find_prev_chained_ft(prio);
 777         if (prev_ft) {
 778                 struct fs_prio *prev_prio;
 779
 780                 fs_get_obj(prev_prio, prev_ft->node.parent);
 781                 return connect_fts_in_prio(dev, prev_prio, ft);
 782         }
 783         return 0;
 784 }
 785
 786 static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 787                                  *prio)
 788 {
 789         struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 790         struct mlx5_ft_underlay_qp *uqp;
 791         int min_level = INT_MAX;
 792         int err;
 793         u32 qpn;
 794
 795         if (root->root_ft)
 796                 min_level = root->root_ft->level;
 797
 798         if (ft->level >= min_level)
 799                 return 0;
 800
 801         if (list_empty(&root->underlay_qpns)) {
 802                 /* Don't set any QPN (zero) in case QPN list is empty */
 803                 qpn = 0;
 804                 err = root->cmds->update_root_ft(root->dev, ft, qpn, false);
 805         } else {
 806                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
 807                         qpn = uqp->qpn;
 808                         err = root->cmds->update_root_ft(root->dev, ft,
 809                                                          qpn, false);
 810                         if (err)
 811                                 break;
 812                 }
 813         }
 814
 815         if (err)
 816                 mlx5_core_warn(root->dev,
 817                                "Update root flow table of id(%u) qpn(%d) failed\n",
 818                                ft->id, qpn);
 819         else
 820                 root->root_ft = ft;
 821
 822         return err;
 823 }
 824
 825 static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 826                                          struct mlx5_flow_destination *dest)
 827 {
 828         struct mlx5_flow_root_namespace *root;
 829         struct mlx5_flow_table *ft;
 830         struct mlx5_flow_group *fg;
 831         struct fs_fte *fte;
 832         int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 833         int err = 0;
 834
 835         fs_get_obj(fte, rule->node.parent);
 836         if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 837                 return -EINVAL;
 838         down_write_ref_node(&fte->node);
 839         fs_get_obj(fg, fte->node.parent);
 840         fs_get_obj(ft, fg->node.parent);
 841
 842         memcpy(&rule->dest_attr, dest, sizeof(*dest));
 843         root = find_root(&ft->node);
 844         err = root->cmds->update_fte(get_dev(&ft->node), ft, fg->id,
 845                                      modify_mask, fte);
 846         up_write_ref_node(&fte->node);
 847
 848         return err;
 849 }
 850
 851 int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
 852                                  struct mlx5_flow_destination *new_dest,
 853                                  struct mlx5_flow_destination *old_dest)
 854 {
 855         int i;
 856
 857         if (!old_dest) {
 858                 if (handle->num_rules != 1)
 859                         return -EINVAL;
 860                 return _mlx5_modify_rule_destination(handle->rule[0],
 861                                                      new_dest);
 862         }
 863
 864         for (i = 0; i < handle->num_rules; i++) {
 865                 if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
 866                         return _mlx5_modify_rule_destination(handle->rule[i],
 867                                                              new_dest);
 868         }
 869
 870         return -EINVAL;
 871 }
 872
 873 /* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 874 static int connect_fwd_rules(struct mlx5_core_dev *dev,
 875                              struct mlx5_flow_table *new_next_ft,
 876                              struct mlx5_flow_table *old_next_ft)
 877 {
 878         struct mlx5_flow_destination dest = {};
 879         struct mlx5_flow_rule *iter;
 880         int err = 0;
 881
 882         /* new_next_ft and old_next_ft could be NULL only
 883          * when we create/destroy the anchor flow table.
 884          */
 885         if (!new_next_ft || !old_next_ft)
 886                 return 0;
 887
 888         dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
 889         dest.ft = new_next_ft;
 890
 891         mutex_lock(&old_next_ft->lock);
 892         list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
 893         mutex_unlock(&old_next_ft->lock);
 894         list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
 895                 err = _mlx5_modify_rule_destination(iter, &dest);
 896                 if (err)
 897                         pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
 898                                new_next_ft->id);
 899         }
 900         return 0;
 901 }
 902
 903 static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
 904                               struct fs_prio *prio)
 905 {
 906         struct mlx5_flow_table *next_ft;
 907         int err = 0;
 908
 909         /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
 910
 911         if (list_empty(&prio->node.children)) {
 912                 err = connect_prev_fts(dev, ft, prio);
 913                 if (err)
 914                         return err;
 915
 916                 next_ft = find_next_chained_ft(prio);
 917                 err = connect_fwd_rules(dev, ft, next_ft);
 918                 if (err)
 919                         return err;
 920         }
 921
 922         if (MLX5_CAP_FLOWTABLE(dev,
 923                                flow_table_properties_nic_receive.modify_root))
 924                 err = update_root_ft_create(ft, prio);
 925         return err;
 926 }
 927
 928 static void list_add_flow_table(struct mlx5_flow_table *ft,
 929                                 struct fs_prio *prio)
 930 {
 931         struct list_head *prev = &prio->node.children;
 932         struct mlx5_flow_table *iter;
 933
 934         fs_for_each_ft(iter, prio) {
 935                 if (iter->level > ft->level)
 936                         break;
 937                 prev = &iter->node.list;
 938         }
 939         list_add(&ft->node.list, prev);
 940 }
 941
 942 static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
 943                                                         struct mlx5_flow_table_attr *ft_attr,
 944                                                         enum fs_flow_table_op_mod op_mod,
 945                                                         u16 vport)
 946 {
 947         struct mlx5_flow_root_namespace *root = find_root(&ns->node);
 948         struct mlx5_flow_table *next_ft = NULL;
 949         struct fs_prio *fs_prio = NULL;
 950         struct mlx5_flow_table *ft;
 951         int log_table_sz;
 952         int err;
 953
 954         if (!root) {
 955                 pr_err("mlx5: flow steering failed to find root of namespace\n");
 956                 return ERR_PTR(-ENODEV);
 957         }
 958
 959         mutex_lock(&root->chain_lock);
 960         fs_prio = find_prio(ns, ft_attr->prio);
 961         if (!fs_prio) {
 962                 err = -EINVAL;
 963                 goto unlock_root;
 964         }
 965         if (ft_attr->level >= fs_prio->num_levels) {
 966                 err = -ENOSPC;
 967                 goto unlock_root;
 968         }
 969         /* The level is related to the
 970          * priority level range.
 971          */
 972         ft_attr->level += fs_prio->start_level;
 973         ft = alloc_flow_table(ft_attr->level,
 974                               vport,
 975                               ft_attr->max_fte ? roundup_pow_of_two(ft_attr->max_fte) : 0,
 976                               root->table_type,
 977                               op_mod, ft_attr->flags);
 978         if (IS_ERR(ft)) {
 979                 err = PTR_ERR(ft);
 980                 goto unlock_root;
 981         }
 982
 983         tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
 984         log_table_sz = ft->max_fte ? ilog2(ft->max_fte) : 0;
 985         next_ft = find_next_chained_ft(fs_prio);
 986         err = root->cmds->create_flow_table(root->dev, ft->vport, ft->op_mod,
 987                                             ft->type, ft->level, log_table_sz,
 988                                             next_ft, &ft->id, ft->flags);
 989         if (err)
 990                 goto free_ft;
 991
 992         err = connect_flow_table(root->dev, ft, fs_prio);
 993         if (err)
 994                 goto destroy_ft;
 995         ft->node.active = true;
 996         down_write_ref_node(&fs_prio->node);
 997         tree_add_node(&ft->node, &fs_prio->node);
 998         list_add_flow_table(ft, fs_prio);
 999         fs_prio->num_ft++;
1000         up_write_ref_node(&fs_prio->node);
1001         mutex_unlock(&root->chain_lock);
1002         return ft;
1003 destroy_ft:
1004         root->cmds->destroy_flow_table(root->dev, ft);
1005 free_ft:
1006         kfree(ft);
1007 unlock_root:
1008         mutex_unlock(&root->chain_lock);
1009         return ERR_PTR(err);
1010 }
1011
1012 struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1013                                                struct mlx5_flow_table_attr *ft_attr)
1014 {
1015         return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1016 }
1017
1018 struct mlx5_flow_table *mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1019                                                      int prio, int max_fte,
1020                                                      u32 level, u16 vport)
1021 {
1022         struct mlx5_flow_table_attr ft_attr = {};
1023
1024         ft_attr.max_fte = max_fte;
1025         ft_attr.level   = level;
1026         ft_attr.prio    = prio;
1027
1028         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1029 }
1030
1031 struct mlx5_flow_table*
1032 mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1033                                  int prio, u32 level)
1034 {
1035         struct mlx5_flow_table_attr ft_attr = {};
1036
1037         ft_attr.level = level;
1038         ft_attr.prio  = prio;
1039         return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1040 }
1041 EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1042
1043 struct mlx5_flow_table*
1044 mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1045                                     int prio,
1046                                     int num_flow_table_entries,
1047                                     int max_num_groups,
1048                                     u32 level,
1049                                     u32 flags)
1050 {
1051         struct mlx5_flow_table_attr ft_attr = {};
1052         struct mlx5_flow_table *ft;
1053
1054         if (max_num_groups > num_flow_table_entries)
1055                 return ERR_PTR(-EINVAL);
1056
1057         ft_attr.max_fte = num_flow_table_entries;
1058         ft_attr.prio    = prio;
1059         ft_attr.level   = level;
1060         ft_attr.flags   = flags;
1061
1062         ft = mlx5_create_flow_table(ns, &ft_attr);
1063         if (IS_ERR(ft))
1064                 return ft;
1065
1066         ft->autogroup.active = true;
1067         ft->autogroup.required_groups = max_num_groups;
1068
1069         return ft;
1070 }
1071 EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1072
1073 struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1074                                                u32 *fg_in)
1075 {
1076         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1077         void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1078                                             fg_in, match_criteria);
1079         u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1080                                             fg_in,
1081                                             match_criteria_enable);
1082         int start_index = MLX5_GET(create_flow_group_in, fg_in,
1083                                    start_flow_index);
1084         int end_index = MLX5_GET(create_flow_group_in, fg_in,
1085                                  end_flow_index);
1086         struct mlx5_core_dev *dev = get_dev(&ft->node);
1087         struct mlx5_flow_group *fg;
1088         int err;
1089
1090         if (ft->autogroup.active)
1091                 return ERR_PTR(-EPERM);
1092
1093         down_write_ref_node(&ft->node);
1094         fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1095                                      start_index, end_index,
1096                                      ft->node.children.prev);
1097         up_write_ref_node(&ft->node);
1098         if (IS_ERR(fg))
1099                 return fg;
1100
1101         err = root->cmds->create_flow_group(dev, ft, fg_in, &fg->id);
1102         if (err) {
1103                 tree_put_node(&fg->node);
1104                 return ERR_PTR(err);
1105         }
1106         trace_mlx5_fs_add_fg(fg);
1107         fg->node.active = true;
1108
1109         return fg;
1110 }
1111
1112 static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1113 {
1114         struct mlx5_flow_rule *rule;
1115
1116         rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1117         if (!rule)
1118                 return NULL;
1119
1120         INIT_LIST_HEAD(&rule->next_ft);
1121         rule->node.type = FS_TYPE_FLOW_DEST;
1122         if (dest)
1123                 memcpy(&rule->dest_attr, dest, sizeof(*dest));
1124
1125         return rule;
1126 }
1127
1128 static struct mlx5_flow_handle *alloc_handle(int num_rules)
1129 {
1130         struct mlx5_flow_handle *handle;
1131
1132         handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1133         if (!handle)
1134                 return NULL;
1135
1136         handle->num_rules = num_rules;
1137
1138         return handle;
1139 }
1140
1141 static void destroy_flow_handle(struct fs_fte *fte,
1142                                 struct mlx5_flow_handle *handle,
1143                                 struct mlx5_flow_destination *dest,
1144                                 int i)
1145 {
1146         for (; --i >= 0;) {
1147                 if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1148                         fte->dests_size--;
1149                         list_del(&handle->rule[i]->node.list);
1150                         kfree(handle->rule[i]);
1151                 }
1152         }
1153         kfree(handle);
1154 }
1155
1156 static struct mlx5_flow_handle *
1157 create_flow_handle(struct fs_fte *fte,
1158                    struct mlx5_flow_destination *dest,
1159                    int dest_num,
1160                    int *modify_mask,
1161                    bool *new_rule)
1162 {
1163         struct mlx5_flow_handle *handle;
1164         struct mlx5_flow_rule *rule = NULL;
1165         static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1166         static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1167         int type;
1168         int i = 0;
1169
1170         handle = alloc_handle((dest_num) ? dest_num : 1);
1171         if (!handle)
1172                 return ERR_PTR(-ENOMEM);
1173
1174         do {
1175                 if (dest) {
1176                         rule = find_flow_rule(fte, dest + i);
1177                         if (rule) {
1178                                 refcount_inc(&rule->node.refcount);
1179                                 goto rule_found;
1180                         }
1181                 }
1182
1183                 *new_rule = true;
1184                 rule = alloc_rule(dest + i);
1185                 if (!rule)
1186                         goto free_rules;
1187
1188                 /* Add dest to dests list- we need flow tables to be in the
1189                  * end of the list for forward to next prio rules.
1190                  */
1191                 tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1192                 if (dest &&
1193                     dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1194                         list_add(&rule->node.list, &fte->node.children);
1195                 else
1196                         list_add_tail(&rule->node.list, &fte->node.children);
1197                 if (dest) {
1198                         fte->dests_size++;
1199
1200                         type = dest[i].type ==
1201                                 MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1202                         *modify_mask |= type ? count : dst;
1203                 }
1204 rule_found:
1205                 handle->rule[i] = rule;
1206         } while (++i < dest_num);
1207
1208         return handle;
1209
1210 free_rules:
1211         destroy_flow_handle(fte, handle, dest, i);
1212         return ERR_PTR(-ENOMEM);
1213 }
1214
1215 /* fte should not be deleted while calling this function */
1216 static struct mlx5_flow_handle *
1217 add_rule_fte(struct fs_fte *fte,
1218              struct mlx5_flow_group *fg,
1219              struct mlx5_flow_destination *dest,
1220              int dest_num,
1221              bool update_action)
1222 {
1223         struct mlx5_flow_root_namespace *root;
1224         struct mlx5_flow_handle *handle;
1225         struct mlx5_flow_table *ft;
1226         int modify_mask = 0;
1227         int err;
1228         bool new_rule = false;
1229
1230         handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1231                                     &new_rule);
1232         if (IS_ERR(handle) || !new_rule)
1233                 goto out;
1234
1235         if (update_action)
1236                 modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1237
1238         fs_get_obj(ft, fg->node.parent);
1239         root = find_root(&fg->node);
1240         if (!(fte->status & FS_FTE_STATUS_EXISTING))
1241                 err = root->cmds->create_fte(get_dev(&ft->node),
1242                                              ft, fg, fte);
1243         else
1244                 err = root->cmds->update_fte(get_dev(&ft->node), ft, fg->id,
1245                                                      modify_mask, fte);
1246         if (err)
1247                 goto free_handle;
1248
1249         fte->node.active = true;
1250         fte->status |= FS_FTE_STATUS_EXISTING;
1251         atomic_inc(&fte->node.version);
1252
1253 out:
1254         return handle;
1255
1256 free_handle:
1257         destroy_flow_handle(fte, handle, dest, handle->num_rules);
1258         return ERR_PTR(err);
1259 }
1260
1261 static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1262                                                      struct mlx5_flow_spec *spec)
1263 {
1264         struct list_head *prev = &ft->node.children;
1265         struct mlx5_flow_group *fg;
1266         unsigned int candidate_index = 0;
1267         unsigned int group_size = 0;
1268
1269         if (!ft->autogroup.active)
1270                 return ERR_PTR(-ENOENT);
1271
1272         if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1273                 /* We save place for flow groups in addition to max types */
1274                 group_size = ft->max_fte / (ft->autogroup.required_groups + 1);
1275
1276         /*  ft->max_fte == ft->autogroup.max_types */
1277         if (group_size == 0)
1278                 group_size = 1;
1279
1280         /* sorted by start_index */
1281         fs_for_each_fg(fg, ft) {
1282                 if (candidate_index + group_size > fg->start_index)
1283                         candidate_index = fg->start_index + fg->max_ftes;
1284                 else
1285                         break;
1286                 prev = &fg->node.list;
1287         }
1288
1289         if (candidate_index + group_size > ft->max_fte)
1290                 return ERR_PTR(-ENOSPC);
1291
1292         fg = alloc_insert_flow_group(ft,
1293                                      spec->match_criteria_enable,
1294                                      spec->match_criteria,
1295                                      candidate_index,
1296                                      candidate_index + group_size - 1,
1297                                      prev);
1298         if (IS_ERR(fg))
1299                 goto out;
1300
1301         ft->autogroup.num_groups++;
1302
1303 out:
1304         return fg;
1305 }
1306
1307 static int create_auto_flow_group(struct mlx5_flow_table *ft,
1308                                   struct mlx5_flow_group *fg)
1309 {
1310         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1311         struct mlx5_core_dev *dev = get_dev(&ft->node);
1312         int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1313         void *match_criteria_addr;
1314         u8 src_esw_owner_mask_on;
1315         void *misc;
1316         int err;
1317         u32 *in;
1318
1319         in = kvzalloc(inlen, GFP_KERNEL);
1320         if (!in)
1321                 return -ENOMEM;
1322
1323         MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1324                  fg->mask.match_criteria_enable);
1325         MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1326         MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1327                  fg->max_ftes - 1);
1328
1329         misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1330                             misc_parameters);
1331         src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1332                                          source_eswitch_owner_vhca_id);
1333         MLX5_SET(create_flow_group_in, in,
1334                  source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1335
1336         match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1337                                            in, match_criteria);
1338         memcpy(match_criteria_addr, fg->mask.match_criteria,
1339                sizeof(fg->mask.match_criteria));
1340
1341         err = root->cmds->create_flow_group(dev, ft, in, &fg->id);
1342         if (!err) {
1343                 fg->node.active = true;
1344                 trace_mlx5_fs_add_fg(fg);
1345         }
1346
1347         kvfree(in);
1348         return err;
1349 }
1350
1351 static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1352                                 struct mlx5_flow_destination *d2)
1353 {
1354         if (d1->type == d2->type) {
1355                 if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1356                      d1->vport.num == d2->vport.num) ||
1357                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1358                      d1->ft == d2->ft) ||
1359                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1360                      d1->tir_num == d2->tir_num) ||
1361                     (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1362                      d1->ft_num == d2->ft_num))
1363                         return true;
1364         }
1365
1366         return false;
1367 }
1368
1369 static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1370                                              struct mlx5_flow_destination *dest)
1371 {
1372         struct mlx5_flow_rule *rule;
1373
1374         list_for_each_entry(rule, &fte->node.children, node.list) {
1375                 if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1376                         return rule;
1377         }
1378         return NULL;
1379 }
1380
1381 static bool check_conflicting_actions(u32 action1, u32 action2)
1382 {
1383         u32 xored_actions = action1 ^ action2;
1384
1385         /* if one rule only wants to count, it's ok */
1386         if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1387             action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1388                 return false;
1389
1390         if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1391                              MLX5_FLOW_CONTEXT_ACTION_ENCAP |
1392                              MLX5_FLOW_CONTEXT_ACTION_DECAP |
1393                              MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1394                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1395                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1396                              MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1397                              MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1398                 return true;
1399
1400         return false;
1401 }
1402
1403 static int check_conflicting_ftes(struct fs_fte *fte, const struct mlx5_flow_act *flow_act)
1404 {
1405         if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1406                 mlx5_core_warn(get_dev(&fte->node),
1407                                "Found two FTEs with conflicting actions\n");
1408                 return -EEXIST;
1409         }
1410
1411         if (flow_act->has_flow_tag &&
1412             fte->action.flow_tag != flow_act->flow_tag) {
1413                 mlx5_core_warn(get_dev(&fte->node),
1414                                "FTE flow tag %u already exists with different flow tag %u\n",
1415                                fte->action.flow_tag,
1416                                flow_act->flow_tag);
1417                 return -EEXIST;
1418         }
1419
1420         return 0;
1421 }
1422
1423 static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1424                                             u32 *match_value,
1425                                             struct mlx5_flow_act *flow_act,
1426                                             struct mlx5_flow_destination *dest,
1427                                             int dest_num,
1428                                             struct fs_fte *fte)
1429 {
1430         struct mlx5_flow_handle *handle;
1431         int old_action;
1432         int i;
1433         int ret;
1434
1435         ret = check_conflicting_ftes(fte, flow_act);
1436         if (ret)
1437                 return ERR_PTR(ret);
1438
1439         old_action = fte->action.action;
1440         fte->action.action |= flow_act->action;
1441         handle = add_rule_fte(fte, fg, dest, dest_num,
1442                               old_action != flow_act->action);
1443         if (IS_ERR(handle)) {
1444                 fte->action.action = old_action;
1445                 return handle;
1446         }
1447         trace_mlx5_fs_set_fte(fte, false);
1448
1449         for (i = 0; i < handle->num_rules; i++) {
1450                 if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1451                         tree_add_node(&handle->rule[i]->node, &fte->node);
1452                         trace_mlx5_fs_add_rule(handle->rule[i]);
1453                 }
1454         }
1455         return handle;
1456 }
1457
1458 struct mlx5_fc *mlx5_flow_rule_counter(struct mlx5_flow_handle *handle)
1459 {
1460         struct mlx5_flow_rule *dst;
1461         struct fs_fte *fte;
1462
1463         fs_get_obj(fte, handle->rule[0]->node.parent);
1464
1465         fs_for_each_dst(dst, fte) {
1466                 if (dst->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER)
1467                         return dst->dest_attr.counter;
1468         }
1469
1470         return NULL;
1471 }
1472
1473 static bool counter_is_valid(struct mlx5_fc *counter, u32 action)
1474 {
1475         if (!(action & MLX5_FLOW_CONTEXT_ACTION_COUNT))
1476                 return !counter;
1477
1478         if (!counter)
1479                 return false;
1480
1481         return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1482                           MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1483 }
1484
1485 static bool dest_is_valid(struct mlx5_flow_destination *dest,
1486                           u32 action,
1487                           struct mlx5_flow_table *ft)
1488 {
1489         if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1490                 return counter_is_valid(dest->counter, action);
1491
1492         if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1493                 return true;
1494
1495         if (!dest || ((dest->type ==
1496             MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1497             (dest->ft->level <= ft->level)))
1498                 return false;
1499         return true;
1500 }
1501
1502 struct match_list {
1503         struct list_head        list;
1504         struct mlx5_flow_group *g;
1505 };
1506
1507 struct match_list_head {
1508         struct list_head  list;
1509         struct match_list first;
1510 };
1511
1512 static void free_match_list(struct match_list_head *head)
1513 {
1514         if (!list_empty(&head->list)) {
1515                 struct match_list *iter, *match_tmp;
1516
1517                 list_del(&head->first.list);
1518                 tree_put_node(&head->first.g->node);
1519                 list_for_each_entry_safe(iter, match_tmp, &head->list,
1520                                          list) {
1521                         tree_put_node(&iter->g->node);
1522                         list_del(&iter->list);
1523                         kfree(iter);
1524                 }
1525         }
1526 }
1527
1528 static int build_match_list(struct match_list_head *match_head,
1529                             struct mlx5_flow_table *ft,
1530                             struct mlx5_flow_spec *spec)
1531 {
1532         struct rhlist_head *tmp, *list;
1533         struct mlx5_flow_group *g;
1534         int err = 0;
1535
1536         rcu_read_lock();
1537         INIT_LIST_HEAD(&match_head->list);
1538         /* Collect all fgs which has a matching match_criteria */
1539         list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1540         /* RCU is atomic, we can't execute FW commands here */
1541         rhl_for_each_entry_rcu(g, tmp, list, hash) {
1542                 struct match_list *curr_match;
1543
1544                 if (likely(list_empty(&match_head->list))) {
1545                         if (!tree_get_node(&g->node))
1546                                 continue;
1547                         match_head->first.g = g;
1548                         list_add_tail(&match_head->first.list,
1549                                       &match_head->list);
1550                         continue;
1551                 }
1552
1553                 curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1554                 if (!curr_match) {
1555                         free_match_list(match_head);
1556                         err = -ENOMEM;
1557                         goto out;
1558                 }
1559                 if (!tree_get_node(&g->node)) {
1560                         kfree(curr_match);
1561                         continue;
1562                 }
1563                 curr_match->g = g;
1564                 list_add_tail(&curr_match->list, &match_head->list);
1565         }
1566 out:
1567         rcu_read_unlock();
1568         return err;
1569 }
1570
1571 static u64 matched_fgs_get_version(struct list_head *match_head)
1572 {
1573         struct match_list *iter;
1574         u64 version = 0;
1575
1576         list_for_each_entry(iter, match_head, list)
1577                 version += (u64)atomic_read(&iter->g->node.version);
1578         return version;
1579 }
1580
1581 static struct fs_fte *
1582 lookup_fte_locked(struct mlx5_flow_group *g,
1583                   u32 *match_value,
1584                   bool take_write)
1585 {
1586         struct fs_fte *fte_tmp;
1587
1588         if (take_write)
1589                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1590         else
1591                 nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1592         fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1593                                          rhash_fte);
1594         if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1595                 fte_tmp = NULL;
1596                 goto out;
1597         }
1598
1599         nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1600 out:
1601         if (take_write)
1602                 up_write_ref_node(&g->node);
1603         else
1604                 up_read_ref_node(&g->node);
1605         return fte_tmp;
1606 }
1607
1608 static struct mlx5_flow_handle *
1609 try_add_to_existing_fg(struct mlx5_flow_table *ft,
1610                        struct list_head *match_head,
1611                        struct mlx5_flow_spec *spec,
1612                        struct mlx5_flow_act *flow_act,
1613                        struct mlx5_flow_destination *dest,
1614                        int dest_num,
1615                        int ft_version)
1616 {
1617         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1618         struct mlx5_flow_group *g;
1619         struct mlx5_flow_handle *rule;
1620         struct match_list *iter;
1621         bool take_write = false;
1622         struct fs_fte *fte;
1623         u64  version;
1624         int err;
1625
1626         fte = alloc_fte(ft, spec->match_value, flow_act);
1627         if (IS_ERR(fte))
1628                 return  ERR_PTR(-ENOMEM);
1629
1630 search_again_locked:
1631         version = matched_fgs_get_version(match_head);
1632         /* Try to find a fg that already contains a matching fte */
1633         list_for_each_entry(iter, match_head, list) {
1634                 struct fs_fte *fte_tmp;
1635
1636                 g = iter->g;
1637                 fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1638                 if (!fte_tmp)
1639                         continue;
1640                 rule = add_rule_fg(g, spec->match_value,
1641                                    flow_act, dest, dest_num, fte_tmp);
1642                 up_write_ref_node(&fte_tmp->node);
1643                 tree_put_node(&fte_tmp->node);
1644                 kmem_cache_free(steering->ftes_cache, fte);
1645                 return rule;
1646         }
1647
1648         /* Check the ft version, for case that new flow group
1649          * was added while the fgs weren't locked
1650          */
1651         if (atomic_read(&ft->node.version) != ft_version) {
1652                 rule = ERR_PTR(-EAGAIN);
1653                 goto out;
1654         }
1655
1656         /* Check the fgs version, for case the new FTE with the
1657          * same values was added while the fgs weren't locked
1658          */
1659         if (version != matched_fgs_get_version(match_head)) {
1660                 take_write = true;
1661                 goto search_again_locked;
1662         }
1663
1664         list_for_each_entry(iter, match_head, list) {
1665                 g = iter->g;
1666
1667                 if (!g->node.active)
1668                         continue;
1669
1670                 nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1671
1672                 err = insert_fte(g, fte);
1673                 if (err) {
1674                         up_write_ref_node(&g->node);
1675                         if (err == -ENOSPC)
1676                                 continue;
1677                         kmem_cache_free(steering->ftes_cache, fte);
1678                         return ERR_PTR(err);
1679                 }
1680
1681                 nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1682                 up_write_ref_node(&g->node);
1683                 rule = add_rule_fg(g, spec->match_value,
1684                                    flow_act, dest, dest_num, fte);
1685                 up_write_ref_node(&fte->node);
1686                 tree_put_node(&fte->node);
1687                 return rule;
1688         }
1689         rule = ERR_PTR(-ENOENT);
1690 out:
1691         kmem_cache_free(steering->ftes_cache, fte);
1692         return rule;
1693 }
1694
1695 static struct mlx5_flow_handle *
1696 _mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1697                      struct mlx5_flow_spec *spec,
1698                      struct mlx5_flow_act *flow_act,
1699                      struct mlx5_flow_destination *dest,
1700                      int dest_num)
1701
1702 {
1703         struct mlx5_flow_steering *steering = get_steering(&ft->node);
1704         struct mlx5_flow_group *g;
1705         struct mlx5_flow_handle *rule;
1706         struct match_list_head match_head;
1707         bool take_write = false;
1708         struct fs_fte *fte;
1709         int version;
1710         int err;
1711         int i;
1712
1713         if (!check_valid_spec(spec))
1714                 return ERR_PTR(-EINVAL);
1715
1716         for (i = 0; i < dest_num; i++) {
1717                 if (!dest_is_valid(&dest[i], flow_act->action, ft))
1718                         return ERR_PTR(-EINVAL);
1719         }
1720         nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1721 search_again_locked:
1722         version = atomic_read(&ft->node.version);
1723
1724         /* Collect all fgs which has a matching match_criteria */
1725         err = build_match_list(&match_head, ft, spec);
1726         if (err) {
1727                 if (take_write)
1728                         up_write_ref_node(&ft->node);
1729                 else
1730                         up_read_ref_node(&ft->node);
1731                 return ERR_PTR(err);
1732         }
1733
1734         if (!take_write)
1735                 up_read_ref_node(&ft->node);
1736
1737         rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1738                                       dest_num, version);
1739         free_match_list(&match_head);
1740         if (!IS_ERR(rule) ||
1741             (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1742                 if (take_write)
1743                         up_write_ref_node(&ft->node);
1744                 return rule;
1745         }
1746
1747         if (!take_write) {
1748                 nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1749                 take_write = true;
1750         }
1751
1752         if (PTR_ERR(rule) == -EAGAIN ||
1753             version != atomic_read(&ft->node.version))
1754                 goto search_again_locked;
1755
1756         g = alloc_auto_flow_group(ft, spec);
1757         if (IS_ERR(g)) {
1758                 rule = ERR_CAST(g);
1759                 up_write_ref_node(&ft->node);
1760                 return rule;
1761         }
1762
1763         nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1764         up_write_ref_node(&ft->node);
1765
1766         err = create_auto_flow_group(ft, g);
1767         if (err)
1768                 goto err_release_fg;
1769
1770         fte = alloc_fte(ft, spec->match_value, flow_act);
1771         if (IS_ERR(fte)) {
1772                 err = PTR_ERR(fte);
1773                 goto err_release_fg;
1774         }
1775
1776         err = insert_fte(g, fte);
1777         if (err) {
1778                 kmem_cache_free(steering->ftes_cache, fte);
1779                 goto err_release_fg;
1780         }
1781
1782         nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1783         up_write_ref_node(&g->node);
1784         rule = add_rule_fg(g, spec->match_value, flow_act, dest,
1785                            dest_num, fte);
1786         up_write_ref_node(&fte->node);
1787         tree_put_node(&fte->node);
1788         tree_put_node(&g->node);
1789         return rule;
1790
1791 err_release_fg:
1792         up_write_ref_node(&g->node);
1793         tree_put_node(&g->node);
1794         return ERR_PTR(err);
1795 }
1796
1797 static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1798 {
1799         return ((ft->type == FS_FT_NIC_RX) &&
1800                 (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1801 }
1802
1803 struct mlx5_flow_handle *
1804 mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1805                     struct mlx5_flow_spec *spec,
1806                     struct mlx5_flow_act *flow_act,
1807                     struct mlx5_flow_destination *dest,
1808                     int num_dest)
1809 {
1810         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1811         struct mlx5_flow_destination gen_dest = {};
1812         struct mlx5_flow_table *next_ft = NULL;
1813         struct mlx5_flow_handle *handle = NULL;
1814         u32 sw_action = flow_act->action;
1815         struct fs_prio *prio;
1816
1817         fs_get_obj(prio, ft->node.parent);
1818         if (flow_act->action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1819                 if (!fwd_next_prio_supported(ft))
1820                         return ERR_PTR(-EOPNOTSUPP);
1821                 if (num_dest)
1822                         return ERR_PTR(-EINVAL);
1823                 mutex_lock(&root->chain_lock);
1824                 next_ft = find_next_chained_ft(prio);
1825                 if (next_ft) {
1826                         gen_dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1827                         gen_dest.ft = next_ft;
1828                         dest = &gen_dest;
1829                         num_dest = 1;
1830                         flow_act->action = MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1831                 } else {
1832                         mutex_unlock(&root->chain_lock);
1833                         return ERR_PTR(-EOPNOTSUPP);
1834                 }
1835         }
1836
1837         handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1838
1839         if (sw_action == MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO) {
1840                 if (!IS_ERR_OR_NULL(handle) &&
1841                     (list_empty(&handle->rule[0]->next_ft))) {
1842                         mutex_lock(&next_ft->lock);
1843                         list_add(&handle->rule[0]->next_ft,
1844                                  &next_ft->fwd_rules);
1845                         mutex_unlock(&next_ft->lock);
1846                         handle->rule[0]->sw_action = MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO;
1847                 }
1848                 mutex_unlock(&root->chain_lock);
1849         }
1850         return handle;
1851 }
1852 EXPORT_SYMBOL(mlx5_add_flow_rules);
1853
1854 void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
1855 {
1856         int i;
1857
1858         for (i = handle->num_rules - 1; i >= 0; i--)
1859                 tree_remove_node(&handle->rule[i]->node);
1860         kfree(handle);
1861 }
1862 EXPORT_SYMBOL(mlx5_del_flow_rules);
1863
1864 /* Assuming prio->node.children(flow tables) is sorted by level */
1865 static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
1866 {
1867         struct fs_prio *prio;
1868
1869         fs_get_obj(prio, ft->node.parent);
1870
1871         if (!list_is_last(&ft->node.list, &prio->node.children))
1872                 return list_next_entry(ft, node.list);
1873         return find_next_chained_ft(prio);
1874 }
1875
1876 static int update_root_ft_destroy(struct mlx5_flow_table *ft)
1877 {
1878         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1879         struct mlx5_ft_underlay_qp *uqp;
1880         struct mlx5_flow_table *new_root_ft = NULL;
1881         int err = 0;
1882         u32 qpn;
1883
1884         if (root->root_ft != ft)
1885                 return 0;
1886
1887         new_root_ft = find_next_ft(ft);
1888         if (!new_root_ft) {
1889                 root->root_ft = NULL;
1890                 return 0;
1891         }
1892
1893         if (list_empty(&root->underlay_qpns)) {
1894                 /* Don't set any QPN (zero) in case QPN list is empty */
1895                 qpn = 0;
1896                 err = root->cmds->update_root_ft(root->dev, new_root_ft,
1897                                                  qpn, false);
1898         } else {
1899                 list_for_each_entry(uqp, &root->underlay_qpns, list) {
1900                         qpn = uqp->qpn;
1901                         err = root->cmds->update_root_ft(root->dev,
1902                                                          new_root_ft, qpn,
1903                                                          false);
1904                         if (err)
1905                                 break;
1906                 }
1907         }
1908
1909         if (err)
1910                 mlx5_core_warn(root->dev,
1911                                "Update root flow table of id(%u) qpn(%d) failed\n",
1912                                ft->id, qpn);
1913         else
1914                 root->root_ft = new_root_ft;
1915
1916         return 0;
1917 }
1918
1919 /* Connect flow table from previous priority to
1920  * the next flow table.
1921  */
1922 static int disconnect_flow_table(struct mlx5_flow_table *ft)
1923 {
1924         struct mlx5_core_dev *dev = get_dev(&ft->node);
1925         struct mlx5_flow_table *next_ft;
1926         struct fs_prio *prio;
1927         int err = 0;
1928
1929         err = update_root_ft_destroy(ft);
1930         if (err)
1931                 return err;
1932
1933         fs_get_obj(prio, ft->node.parent);
1934         if  (!(list_first_entry(&prio->node.children,
1935                                 struct mlx5_flow_table,
1936                                 node.list) == ft))
1937                 return 0;
1938
1939         next_ft = find_next_chained_ft(prio);
1940         err = connect_fwd_rules(dev, next_ft, ft);
1941         if (err)
1942                 return err;
1943
1944         err = connect_prev_fts(dev, next_ft, prio);
1945         if (err)
1946                 mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
1947                                ft->id);
1948         return err;
1949 }
1950
1951 int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
1952 {
1953         struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1954         int err = 0;
1955
1956         mutex_lock(&root->chain_lock);
1957         err = disconnect_flow_table(ft);
1958         if (err) {
1959                 mutex_unlock(&root->chain_lock);
1960                 return err;
1961         }
1962         if (tree_remove_node(&ft->node))
1963                 mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
1964                                ft->id);
1965         mutex_unlock(&root->chain_lock);
1966
1967         return err;
1968 }
1969 EXPORT_SYMBOL(mlx5_destroy_flow_table);
1970
1971 void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
1972 {
1973         if (tree_remove_node(&fg->node))
1974                 mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
1975                                fg->id);
1976 }
1977
1978 struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
1979                                                     enum mlx5_flow_namespace_type type)
1980 {
1981         struct mlx5_flow_steering *steering = dev->priv.steering;
1982         struct mlx5_flow_root_namespace *root_ns;
1983         int prio;
1984         struct fs_prio *fs_prio;
1985         struct mlx5_flow_namespace *ns;
1986
1987         if (!steering)
1988                 return NULL;
1989
1990         switch (type) {
1991         case MLX5_FLOW_NAMESPACE_BYPASS:
1992         case MLX5_FLOW_NAMESPACE_LAG:
1993         case MLX5_FLOW_NAMESPACE_OFFLOADS:
1994         case MLX5_FLOW_NAMESPACE_ETHTOOL:
1995         case MLX5_FLOW_NAMESPACE_KERNEL:
1996         case MLX5_FLOW_NAMESPACE_LEFTOVERS:
1997         case MLX5_FLOW_NAMESPACE_ANCHOR:
1998                 prio = type;
1999                 break;
2000         case MLX5_FLOW_NAMESPACE_FDB:
2001                 if (steering->fdb_root_ns)
2002                         return &steering->fdb_root_ns->ns;
2003                 else
2004                         return NULL;
2005         case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2006                 if (steering->sniffer_rx_root_ns)
2007                         return &steering->sniffer_rx_root_ns->ns;
2008                 else
2009                         return NULL;
2010         case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2011                 if (steering->sniffer_tx_root_ns)
2012                         return &steering->sniffer_tx_root_ns->ns;
2013                 else
2014                         return NULL;
2015         case MLX5_FLOW_NAMESPACE_EGRESS:
2016                 if (steering->egress_root_ns)
2017                         return &steering->egress_root_ns->ns;
2018                 else
2019                         return NULL;
2020         default:
2021                 return NULL;
2022         }
2023
2024         root_ns = steering->root_ns;
2025         if (!root_ns)
2026                 return NULL;
2027
2028         fs_prio = find_prio(&root_ns->ns, prio);
2029         if (!fs_prio)
2030                 return NULL;
2031
2032         ns = list_first_entry(&fs_prio->node.children,
2033                               typeof(*ns),
2034                               node.list);
2035
2036         return ns;
2037 }
2038 EXPORT_SYMBOL(mlx5_get_flow_namespace);
2039
2040 struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2041                                                               enum mlx5_flow_namespace_type type,
2042                                                               int vport)
2043 {
2044         struct mlx5_flow_steering *steering = dev->priv.steering;
2045
2046         if (!steering || vport >= MLX5_TOTAL_VPORTS(dev))
2047                 return NULL;
2048
2049         switch (type) {
2050         case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2051                 if (steering->esw_egress_root_ns &&
2052                     steering->esw_egress_root_ns[vport])
2053                         return &steering->esw_egress_root_ns[vport]->ns;
2054                 else
2055                         return NULL;
2056         case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2057                 if (steering->esw_ingress_root_ns &&
2058                     steering->esw_ingress_root_ns[vport])
2059                         return &steering->esw_ingress_root_ns[vport]->ns;
2060                 else
2061                         return NULL;
2062         default:
2063                 return NULL;
2064         }
2065 }
2066
2067 static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2068                                       unsigned int prio, int num_levels)
2069 {
2070         struct fs_prio *fs_prio;
2071
2072         fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2073         if (!fs_prio)
2074                 return ERR_PTR(-ENOMEM);
2075
2076         fs_prio->node.type = FS_TYPE_PRIO;
2077         tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2078         tree_add_node(&fs_prio->node, &ns->node);
2079         fs_prio->num_levels = num_levels;
2080         fs_prio->prio = prio;
2081         list_add_tail(&fs_prio->node.list, &ns->node.children);
2082
2083         return fs_prio;
2084 }
2085
2086 static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2087                                                      *ns)
2088 {
2089         ns->node.type = FS_TYPE_NAMESPACE;
2090
2091         return ns;
2092 }
2093
2094 static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio)
2095 {
2096         struct mlx5_flow_namespace      *ns;
2097
2098         ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2099         if (!ns)
2100                 return ERR_PTR(-ENOMEM);
2101
2102         fs_init_namespace(ns);
2103         tree_init_node(&ns->node, NULL, del_sw_ns);
2104         tree_add_node(&ns->node, &prio->node);
2105         list_add_tail(&ns->node.list, &prio->node.children);
2106
2107         return ns;
2108 }
2109
2110 static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2111                              struct init_tree_node *prio_metadata)
2112 {
2113         struct fs_prio *fs_prio;
2114         int i;
2115
2116         for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2117                 fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2118                 if (IS_ERR(fs_prio))
2119                         return PTR_ERR(fs_prio);
2120         }
2121         return 0;
2122 }
2123
2124 #define FLOW_TABLE_BIT_SZ 1
2125 #define GET_FLOW_TABLE_CAP(dev, offset) \
2126         ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2127                         offset / 32)) >>                                        \
2128           (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2129 static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2130 {
2131         int i;
2132
2133         for (i = 0; i < caps->arr_sz; i++) {
2134                 if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2135                         return false;
2136         }
2137         return true;
2138 }
2139
2140 static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2141                                     struct init_tree_node *init_node,
2142                                     struct fs_node *fs_parent_node,
2143                                     struct init_tree_node *init_parent_node,
2144                                     int prio)
2145 {
2146         int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2147                                               flow_table_properties_nic_receive.
2148                                               max_ft_level);
2149         struct mlx5_flow_namespace *fs_ns;
2150         struct fs_prio *fs_prio;
2151         struct fs_node *base;
2152         int i;
2153         int err;
2154
2155         if (init_node->type == FS_TYPE_PRIO) {
2156                 if ((init_node->min_ft_level > max_ft_level) ||
2157                     !has_required_caps(steering->dev, &init_node->caps))
2158                         return 0;
2159
2160                 fs_get_obj(fs_ns, fs_parent_node);
2161                 if (init_node->num_leaf_prios)
2162                         return create_leaf_prios(fs_ns, prio, init_node);
2163                 fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2164                 if (IS_ERR(fs_prio))
2165                         return PTR_ERR(fs_prio);
2166                 base = &fs_prio->node;
2167         } else if (init_node->type == FS_TYPE_NAMESPACE) {
2168                 fs_get_obj(fs_prio, fs_parent_node);
2169                 fs_ns = fs_create_namespace(fs_prio);
2170                 if (IS_ERR(fs_ns))
2171                         return PTR_ERR(fs_ns);
2172                 base = &fs_ns->node;
2173         } else {
2174                 return -EINVAL;
2175         }
2176         prio = 0;
2177         for (i = 0; i < init_node->ar_size; i++) {
2178                 err = init_root_tree_recursive(steering, &init_node->children[i],
2179                                                base, init_node, prio);
2180                 if (err)
2181                         return err;
2182                 if (init_node->children[i].type == FS_TYPE_PRIO &&
2183                     init_node->children[i].num_leaf_prios) {
2184                         prio += init_node->children[i].num_leaf_prios;
2185                 }
2186         }
2187
2188         return 0;
2189 }
2190
2191 static int init_root_tree(struct mlx5_flow_steering *steering,
2192                           struct init_tree_node *init_node,
2193                           struct fs_node *fs_parent_node)
2194 {
2195         int i;
2196         struct mlx5_flow_namespace *fs_ns;
2197         int err;
2198
2199         fs_get_obj(fs_ns, fs_parent_node);
2200         for (i = 0; i < init_node->ar_size; i++) {
2201                 err = init_root_tree_recursive(steering, &init_node->children[i],
2202                                                &fs_ns->node,
2203                                                init_node, i);
2204                 if (err)
2205                         return err;
2206         }
2207         return 0;
2208 }
2209
2210 static struct mlx5_flow_root_namespace
2211 *create_root_ns(struct mlx5_flow_steering *steering,
2212                 enum fs_flow_table_type table_type)
2213 {
2214         const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2215         struct mlx5_flow_root_namespace *root_ns;
2216         struct mlx5_flow_namespace *ns;
2217
2218         if (mlx5_accel_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2219             (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2220                 cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2221
2222         /* Create the root namespace */
2223         root_ns = kvzalloc(sizeof(*root_ns), GFP_KERNEL);
2224         if (!root_ns)
2225                 return NULL;
2226
2227         root_ns->dev = steering->dev;
2228         root_ns->table_type = table_type;
2229         root_ns->cmds = cmds;
2230
2231         INIT_LIST_HEAD(&root_ns->underlay_qpns);
2232
2233         ns = &root_ns->ns;
2234         fs_init_namespace(ns);
2235         mutex_init(&root_ns->chain_lock);
2236         tree_init_node(&ns->node, NULL, NULL);
2237         tree_add_node(&ns->node, NULL);
2238
2239         return root_ns;
2240 }
2241
2242 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2243
2244 static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2245 {
2246         struct fs_prio *prio;
2247
2248         fs_for_each_prio(prio, ns) {
2249                  /* This updates prio start_level and num_levels */
2250                 set_prio_attrs_in_prio(prio, acc_level);
2251                 acc_level += prio->num_levels;
2252         }
2253         return acc_level;
2254 }
2255
2256 static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2257 {
2258         struct mlx5_flow_namespace *ns;
2259         int acc_level_ns = acc_level;
2260
2261         prio->start_level = acc_level;
2262         fs_for_each_ns(ns, prio)
2263                 /* This updates start_level and num_levels of ns's priority descendants */
2264                 acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2265         if (!prio->num_levels)
2266                 prio->num_levels = acc_level_ns - prio->start_level;
2267         WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2268 }
2269
2270 static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2271 {
2272         struct mlx5_flow_namespace *ns = &root_ns->ns;
2273         struct fs_prio *prio;
2274         int start_level = 0;
2275
2276         fs_for_each_prio(prio, ns) {
2277                 set_prio_attrs_in_prio(prio, start_level);
2278                 start_level += prio->num_levels;
2279         }
2280 }
2281
2282 #define ANCHOR_PRIO 0
2283 #define ANCHOR_SIZE 1
2284 #define ANCHOR_LEVEL 0
2285 static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2286 {
2287         struct mlx5_flow_namespace *ns = NULL;
2288         struct mlx5_flow_table_attr ft_attr = {};
2289         struct mlx5_flow_table *ft;
2290
2291         ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2292         if (WARN_ON(!ns))
2293                 return -EINVAL;
2294
2295         ft_attr.max_fte = ANCHOR_SIZE;
2296         ft_attr.level   = ANCHOR_LEVEL;
2297         ft_attr.prio    = ANCHOR_PRIO;
2298
2299         ft = mlx5_create_flow_table(ns, &ft_attr);
2300         if (IS_ERR(ft)) {
2301                 mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2302                 return PTR_ERR(ft);
2303         }
2304         return 0;
2305 }
2306
2307 static int init_root_ns(struct mlx5_flow_steering *steering)
2308 {
2309         int err;
2310
2311         steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2312         if (!steering->root_ns)
2313                 return -ENOMEM;
2314
2315         err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2316         if (err)
2317                 goto out_err;
2318
2319         set_prio_attrs(steering->root_ns);
2320         err = create_anchor_flow_table(steering);
2321         if (err)
2322                 goto out_err;
2323
2324         return 0;
2325
2326 out_err:
2327         cleanup_root_ns(steering->root_ns);
2328         steering->root_ns = NULL;
2329         return err;
2330 }
2331
2332 static void clean_tree(struct fs_node *node)
2333 {
2334         if (node) {
2335                 struct fs_node *iter;
2336                 struct fs_node *temp;
2337
2338                 tree_get_node(node);
2339                 list_for_each_entry_safe(iter, temp, &node->children, list)
2340                         clean_tree(iter);
2341                 tree_put_node(node);
2342                 tree_remove_node(node);
2343         }
2344 }
2345
2346 static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2347 {
2348         if (!root_ns)
2349                 return;
2350
2351         clean_tree(&root_ns->ns.node);
2352 }
2353
2354 static void cleanup_egress_acls_root_ns(struct mlx5_core_dev *dev)
2355 {
2356         struct mlx5_flow_steering *steering = dev->priv.steering;
2357         int i;
2358
2359         if (!steering->esw_egress_root_ns)
2360                 return;
2361
2362         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2363                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2364
2365         kfree(steering->esw_egress_root_ns);
2366 }
2367
2368 static void cleanup_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2369 {
2370         struct mlx5_flow_steering *steering = dev->priv.steering;
2371         int i;
2372
2373         if (!steering->esw_ingress_root_ns)
2374                 return;
2375
2376         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++)
2377                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2378
2379         kfree(steering->esw_ingress_root_ns);
2380 }
2381
2382 void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2383 {
2384         struct mlx5_flow_steering *steering = dev->priv.steering;
2385
2386         cleanup_root_ns(steering->root_ns);
2387         cleanup_egress_acls_root_ns(dev);
2388         cleanup_ingress_acls_root_ns(dev);
2389         cleanup_root_ns(steering->fdb_root_ns);
2390         cleanup_root_ns(steering->sniffer_rx_root_ns);
2391         cleanup_root_ns(steering->sniffer_tx_root_ns);
2392         cleanup_root_ns(steering->egress_root_ns);
2393         mlx5_cleanup_fc_stats(dev);
2394         kmem_cache_destroy(steering->ftes_cache);
2395         kmem_cache_destroy(steering->fgs_cache);
2396         kfree(steering);
2397 }
2398
2399 static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2400 {
2401         struct fs_prio *prio;
2402
2403         steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2404         if (!steering->sniffer_tx_root_ns)
2405                 return -ENOMEM;
2406
2407         /* Create single prio */
2408         prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2409         if (IS_ERR(prio)) {
2410                 cleanup_root_ns(steering->sniffer_tx_root_ns);
2411                 return PTR_ERR(prio);
2412         }
2413         return 0;
2414 }
2415
2416 static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2417 {
2418         struct fs_prio *prio;
2419
2420         steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2421         if (!steering->sniffer_rx_root_ns)
2422                 return -ENOMEM;
2423
2424         /* Create single prio */
2425         prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2426         if (IS_ERR(prio)) {
2427                 cleanup_root_ns(steering->sniffer_rx_root_ns);
2428                 return PTR_ERR(prio);
2429         }
2430         return 0;
2431 }
2432
2433 static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2434 {
2435         struct fs_prio *prio;
2436
2437         steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2438         if (!steering->fdb_root_ns)
2439                 return -ENOMEM;
2440
2441         prio = fs_create_prio(&steering->fdb_root_ns->ns, 0, 2);
2442         if (IS_ERR(prio))
2443                 goto out_err;
2444
2445         prio = fs_create_prio(&steering->fdb_root_ns->ns, 1, 1);
2446         if (IS_ERR(prio))
2447                 goto out_err;
2448
2449         set_prio_attrs(steering->fdb_root_ns);
2450         return 0;
2451
2452 out_err:
2453         cleanup_root_ns(steering->fdb_root_ns);
2454         steering->fdb_root_ns = NULL;
2455         return PTR_ERR(prio);
2456 }
2457
2458 static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2459 {
2460         struct fs_prio *prio;
2461
2462         steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2463         if (!steering->esw_egress_root_ns[vport])
2464                 return -ENOMEM;
2465
2466         /* create 1 prio*/
2467         prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2468         return PTR_ERR_OR_ZERO(prio);
2469 }
2470
2471 static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2472 {
2473         struct fs_prio *prio;
2474
2475         steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2476         if (!steering->esw_ingress_root_ns[vport])
2477                 return -ENOMEM;
2478
2479         /* create 1 prio*/
2480         prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2481         return PTR_ERR_OR_ZERO(prio);
2482 }
2483
2484 static int init_egress_acls_root_ns(struct mlx5_core_dev *dev)
2485 {
2486         struct mlx5_flow_steering *steering = dev->priv.steering;
2487         int err;
2488         int i;
2489
2490         steering->esw_egress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2491                                                sizeof(*steering->esw_egress_root_ns),
2492                                                GFP_KERNEL);
2493         if (!steering->esw_egress_root_ns)
2494                 return -ENOMEM;
2495
2496         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2497                 err = init_egress_acl_root_ns(steering, i);
2498                 if (err)
2499                         goto cleanup_root_ns;
2500         }
2501
2502         return 0;
2503
2504 cleanup_root_ns:
2505         for (i--; i >= 0; i--)
2506                 cleanup_root_ns(steering->esw_egress_root_ns[i]);
2507         kfree(steering->esw_egress_root_ns);
2508         return err;
2509 }
2510
2511 static int init_ingress_acls_root_ns(struct mlx5_core_dev *dev)
2512 {
2513         struct mlx5_flow_steering *steering = dev->priv.steering;
2514         int err;
2515         int i;
2516
2517         steering->esw_ingress_root_ns = kcalloc(MLX5_TOTAL_VPORTS(dev),
2518                                                 sizeof(*steering->esw_ingress_root_ns),
2519                                                 GFP_KERNEL);
2520         if (!steering->esw_ingress_root_ns)
2521                 return -ENOMEM;
2522
2523         for (i = 0; i < MLX5_TOTAL_VPORTS(dev); i++) {
2524                 err = init_ingress_acl_root_ns(steering, i);
2525                 if (err)
2526                         goto cleanup_root_ns;
2527         }
2528
2529         return 0;
2530
2531 cleanup_root_ns:
2532         for (i--; i >= 0; i--)
2533                 cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2534         kfree(steering->esw_ingress_root_ns);
2535         return err;
2536 }
2537
2538 static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2539 {
2540         struct fs_prio *prio;
2541
2542         steering->egress_root_ns = create_root_ns(steering,
2543                                                   FS_FT_NIC_TX);
2544         if (!steering->egress_root_ns)
2545                 return -ENOMEM;
2546
2547         /* create 1 prio*/
2548         prio = fs_create_prio(&steering->egress_root_ns->ns, 0, 1);
2549         return PTR_ERR_OR_ZERO(prio);
2550 }
2551
2552 int mlx5_init_fs(struct mlx5_core_dev *dev)
2553 {
2554         struct mlx5_flow_steering *steering;
2555         int err = 0;
2556
2557         err = mlx5_init_fc_stats(dev);
2558         if (err)
2559                 return err;
2560
2561         steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2562         if (!steering)
2563                 return -ENOMEM;
2564         steering->dev = dev;
2565         dev->priv.steering = steering;
2566
2567         steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2568                                                 sizeof(struct mlx5_flow_group), 0,
2569                                                 0, NULL);
2570         steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2571                                                  0, NULL);
2572         if (!steering->ftes_cache || !steering->fgs_cache) {
2573                 err = -ENOMEM;
2574                 goto err;
2575         }
2576
2577         if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2578               (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2579              ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2580               MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2581             MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2582                 err = init_root_ns(steering);
2583                 if (err)
2584                         goto err;
2585         }
2586
2587         if (MLX5_ESWITCH_MANAGER(dev)) {
2588                 if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
2589                         err = init_fdb_root_ns(steering);
2590                         if (err)
2591                                 goto err;
2592                 }
2593                 if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
2594                         err = init_egress_acls_root_ns(dev);
2595                         if (err)
2596                                 goto err;
2597                 }
2598                 if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
2599                         err = init_ingress_acls_root_ns(dev);
2600                         if (err)
2601                                 goto err;
2602                 }
2603         }
2604
2605         if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
2606                 err = init_sniffer_rx_root_ns(steering);
2607                 if (err)
2608                         goto err;
2609         }
2610
2611         if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
2612                 err = init_sniffer_tx_root_ns(steering);
2613                 if (err)
2614                         goto err;
2615         }
2616
2617         if (MLX5_IPSEC_DEV(dev)) {
2618                 err = init_egress_root_ns(steering);
2619                 if (err)
2620                         goto err;
2621         }
2622
2623         return 0;
2624 err:
2625         mlx5_cleanup_fs(dev);
2626         return err;
2627 }
2628
2629 int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2630 {
2631         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2632         struct mlx5_ft_underlay_qp *new_uqp;
2633         int err = 0;
2634
2635         new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
2636         if (!new_uqp)
2637                 return -ENOMEM;
2638
2639         mutex_lock(&root->chain_lock);
2640
2641         if (!root->root_ft) {
2642                 err = -EINVAL;
2643                 goto update_ft_fail;
2644         }
2645
2646         err = root->cmds->update_root_ft(dev, root->root_ft, underlay_qpn,
2647                                          false);
2648         if (err) {
2649                 mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
2650                                underlay_qpn, err);
2651                 goto update_ft_fail;
2652         }
2653
2654         new_uqp->qpn = underlay_qpn;
2655         list_add_tail(&new_uqp->list, &root->underlay_qpns);
2656
2657         mutex_unlock(&root->chain_lock);
2658
2659         return 0;
2660
2661 update_ft_fail:
2662         mutex_unlock(&root->chain_lock);
2663         kfree(new_uqp);
2664         return err;
2665 }
2666 EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
2667
2668 int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
2669 {
2670         struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
2671         struct mlx5_ft_underlay_qp *uqp;
2672         bool found = false;
2673         int err = 0;
2674
2675         mutex_lock(&root->chain_lock);
2676         list_for_each_entry(uqp, &root->underlay_qpns, list) {
2677                 if (uqp->qpn == underlay_qpn) {
2678                         found = true;
2679                         break;
2680                 }
2681         }
2682
2683         if (!found) {
2684                 mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
2685                                underlay_qpn);
2686                 err = -EINVAL;
2687                 goto out;
2688         }
2689
2690         err = root->cmds->update_root_ft(dev, root->root_ft, underlay_qpn,
2691                                          true);
2692         if (err)
2693                 mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
2694                                underlay_qpn, err);
2695
2696         list_del(&uqp->list);
2697         mutex_unlock(&root->chain_lock);
2698         kfree(uqp);
2699
2700         return 0;
2701
2702 out:
2703         mutex_unlock(&root->chain_lock);
2704         return err;
2705 }
2706 EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);