linux/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2015, Mellanox Technologies. All rights reserved.
   3 *
   4 * This software is available to you under a choice of one of two
   5 * licenses.  You may choose to be licensed under the terms of the GNU
   6 * General Public License (GPL) Version 2, available from the file
   7 * COPYING in the main directory of this source tree, or the
   8 * OpenIB.org BSD license below:
   9 *
  10 *     Redistribution and use in source and binary forms, with or
  11 *     without modification, are permitted provided that the following
  12 *     conditions are met:
  13 *
  14 *      - Redistributions of source code must retain the above
  15 *        copyright notice, this list of conditions and the following
  16 *        disclaimer.
  17 *
  18 *      - Redistributions in binary form must reproduce the above
  19 *        copyright notice, this list of conditions and the following
  20 *        disclaimer in the documentation and/or other materials
  21 *        provided with the distribution.
  22 *
  23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30 * SOFTWARE.
  31 */
  32
  33#include <linux/mutex.h>
  34#include <linux/mlx5/driver.h>
  35#include <linux/mlx5/vport.h>
  36#include <linux/mlx5/eswitch.h>
  37
  38#include "mlx5_core.h"
  39#include "fs_core.h"
  40#include "fs_cmd.h"
  41#include "fs_ft_pool.h"
  42#include "diag/fs_tracepoint.h"
  43#include "accel/ipsec.h"
  44#include "fpga/ipsec.h"
  45
  46#define INIT_TREE_NODE_ARRAY_SIZE(...)  (sizeof((struct init_tree_node[]){__VA_ARGS__}) /\
  47                                         sizeof(struct init_tree_node))
  48
  49#define ADD_PRIO(num_prios_val, min_level_val, num_levels_val, caps_val,\
  50                 ...) {.type = FS_TYPE_PRIO,\
  51        .min_ft_level = min_level_val,\
  52        .num_levels = num_levels_val,\
  53        .num_leaf_prios = num_prios_val,\
  54        .caps = caps_val,\
  55        .children = (struct init_tree_node[]) {__VA_ARGS__},\
  56        .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  57}
  58
  59#define ADD_MULTIPLE_PRIO(num_prios_val, num_levels_val, ...)\
  60        ADD_PRIO(num_prios_val, 0, num_levels_val, {},\
  61                 __VA_ARGS__)\
  62
  63#define ADD_NS(def_miss_act, ...) {.type = FS_TYPE_NAMESPACE,   \
  64        .def_miss_action = def_miss_act,\
  65        .children = (struct init_tree_node[]) {__VA_ARGS__},\
  66        .ar_size = INIT_TREE_NODE_ARRAY_SIZE(__VA_ARGS__) \
  67}
  68
  69#define INIT_CAPS_ARRAY_SIZE(...) (sizeof((long[]){__VA_ARGS__}) /\
  70                                   sizeof(long))
  71
  72#define FS_CAP(cap) (__mlx5_bit_off(flow_table_nic_cap, cap))
  73
  74#define FS_REQUIRED_CAPS(...) {.arr_sz = INIT_CAPS_ARRAY_SIZE(__VA_ARGS__), \
  75                               .caps = (long[]) {__VA_ARGS__} }
  76
  77#define FS_CHAINING_CAPS  FS_REQUIRED_CAPS(FS_CAP(flow_table_properties_nic_receive.flow_modify_en), \
  78                                           FS_CAP(flow_table_properties_nic_receive.modify_root), \
  79                                           FS_CAP(flow_table_properties_nic_receive.identified_miss_table_mode), \
  80                                           FS_CAP(flow_table_properties_nic_receive.flow_table_modify))
  81
  82#define FS_CHAINING_CAPS_EGRESS                                                \
  83        FS_REQUIRED_CAPS(                                                      \
  84                FS_CAP(flow_table_properties_nic_transmit.flow_modify_en),     \
  85                FS_CAP(flow_table_properties_nic_transmit.modify_root),        \
  86                FS_CAP(flow_table_properties_nic_transmit                      \
  87                               .identified_miss_table_mode),                   \
  88                FS_CAP(flow_table_properties_nic_transmit.flow_table_modify))
  89
  90#define FS_CHAINING_CAPS_RDMA_TX                                                \
  91        FS_REQUIRED_CAPS(                                                       \
  92                FS_CAP(flow_table_properties_nic_transmit_rdma.flow_modify_en), \
  93                FS_CAP(flow_table_properties_nic_transmit_rdma.modify_root),    \
  94                FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  95                               .identified_miss_table_mode),                    \
  96                FS_CAP(flow_table_properties_nic_transmit_rdma                  \
  97                               .flow_table_modify))
  98
  99#define LEFTOVERS_NUM_LEVELS 1
 100#define LEFTOVERS_NUM_PRIOS 1
 101
 102#define BY_PASS_PRIO_NUM_LEVELS 1
 103#define BY_PASS_MIN_LEVEL (ETHTOOL_MIN_LEVEL + MLX5_BY_PASS_NUM_PRIOS +\
 104                           LEFTOVERS_NUM_PRIOS)
 105
 106#define ETHTOOL_PRIO_NUM_LEVELS 1
 107#define ETHTOOL_NUM_PRIOS 11
 108#define ETHTOOL_MIN_LEVEL (KERNEL_MIN_LEVEL + ETHTOOL_NUM_PRIOS)
 109/* Promiscuous, Vlan, mac, ttc, inner ttc, {UDP/ANY/aRFS/accel/{esp, esp_err}} */
 110#define KERNEL_NIC_PRIO_NUM_LEVELS 7
 111#define KERNEL_NIC_NUM_PRIOS 1
 112/* One more level for tc */
 113#define KERNEL_MIN_LEVEL (KERNEL_NIC_PRIO_NUM_LEVELS + 1)
 114
 115#define KERNEL_NIC_TC_NUM_PRIOS  1
 116#define KERNEL_NIC_TC_NUM_LEVELS 2
 117
 118#define ANCHOR_NUM_LEVELS 1
 119#define ANCHOR_NUM_PRIOS 1
 120#define ANCHOR_MIN_LEVEL (BY_PASS_MIN_LEVEL + 1)
 121
 122#define OFFLOADS_MAX_FT 2
 123#define OFFLOADS_NUM_PRIOS 2
 124#define OFFLOADS_MIN_LEVEL (ANCHOR_MIN_LEVEL + OFFLOADS_NUM_PRIOS)
 125
 126#define LAG_PRIO_NUM_LEVELS 1
 127#define LAG_NUM_PRIOS 1
 128#define LAG_MIN_LEVEL (OFFLOADS_MIN_LEVEL + 1)
 129
 130#define KERNEL_TX_IPSEC_NUM_PRIOS  1
 131#define KERNEL_TX_IPSEC_NUM_LEVELS 1
 132#define KERNEL_TX_MIN_LEVEL        (KERNEL_TX_IPSEC_NUM_LEVELS)
 133
 134struct node_caps {
 135        size_t  arr_sz;
 136        long    *caps;
 137};
 138
 139static struct init_tree_node {
 140        enum fs_node_type       type;
 141        struct init_tree_node *children;
 142        int ar_size;
 143        struct node_caps caps;
 144        int min_ft_level;
 145        int num_leaf_prios;
 146        int prio;
 147        int num_levels;
 148        enum mlx5_flow_table_miss_action def_miss_action;
 149} root_fs = {
 150        .type = FS_TYPE_NAMESPACE,
 151        .ar_size = 7,
 152          .children = (struct init_tree_node[]){
 153                  ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 154                           ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 155                                  ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 156                                                    BY_PASS_PRIO_NUM_LEVELS))),
 157                  ADD_PRIO(0, LAG_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 158                           ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 159                                  ADD_MULTIPLE_PRIO(LAG_NUM_PRIOS,
 160                                                    LAG_PRIO_NUM_LEVELS))),
 161                  ADD_PRIO(0, OFFLOADS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 162                           ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 163                                  ADD_MULTIPLE_PRIO(OFFLOADS_NUM_PRIOS,
 164                                                    OFFLOADS_MAX_FT))),
 165                  ADD_PRIO(0, ETHTOOL_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 166                           ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 167                                  ADD_MULTIPLE_PRIO(ETHTOOL_NUM_PRIOS,
 168                                                    ETHTOOL_PRIO_NUM_LEVELS))),
 169                  ADD_PRIO(0, KERNEL_MIN_LEVEL, 0, {},
 170                           ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 171                                  ADD_MULTIPLE_PRIO(KERNEL_NIC_TC_NUM_PRIOS,
 172                                                    KERNEL_NIC_TC_NUM_LEVELS),
 173                                  ADD_MULTIPLE_PRIO(KERNEL_NIC_NUM_PRIOS,
 174                                                    KERNEL_NIC_PRIO_NUM_LEVELS))),
 175                  ADD_PRIO(0, BY_PASS_MIN_LEVEL, 0, FS_CHAINING_CAPS,
 176                           ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 177                                  ADD_MULTIPLE_PRIO(LEFTOVERS_NUM_PRIOS,
 178                                                    LEFTOVERS_NUM_LEVELS))),
 179                  ADD_PRIO(0, ANCHOR_MIN_LEVEL, 0, {},
 180                           ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 181                                  ADD_MULTIPLE_PRIO(ANCHOR_NUM_PRIOS,
 182                                                    ANCHOR_NUM_LEVELS))),
 183        }
 184};
 185
 186static struct init_tree_node egress_root_fs = {
 187        .type = FS_TYPE_NAMESPACE,
 188#ifdef CONFIG_MLX5_IPSEC
 189        .ar_size = 2,
 190#else
 191        .ar_size = 1,
 192#endif
 193        .children = (struct init_tree_node[]) {
 194                ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 195                         FS_CHAINING_CAPS_EGRESS,
 196                         ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 197                                ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 198                                                  BY_PASS_PRIO_NUM_LEVELS))),
 199#ifdef CONFIG_MLX5_IPSEC
 200                ADD_PRIO(0, KERNEL_TX_MIN_LEVEL, 0,
 201                         FS_CHAINING_CAPS_EGRESS,
 202                         ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 203                                ADD_MULTIPLE_PRIO(KERNEL_TX_IPSEC_NUM_PRIOS,
 204                                                  KERNEL_TX_IPSEC_NUM_LEVELS))),
 205#endif
 206        }
 207};
 208
 209#define RDMA_RX_BYPASS_PRIO 0
 210#define RDMA_RX_KERNEL_PRIO 1
 211static struct init_tree_node rdma_rx_root_fs = {
 212        .type = FS_TYPE_NAMESPACE,
 213        .ar_size = 2,
 214        .children = (struct init_tree_node[]) {
 215                [RDMA_RX_BYPASS_PRIO] =
 216                ADD_PRIO(0, MLX5_BY_PASS_NUM_REGULAR_PRIOS, 0,
 217                         FS_CHAINING_CAPS,
 218                         ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 219                                ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_REGULAR_PRIOS,
 220                                                  BY_PASS_PRIO_NUM_LEVELS))),
 221                [RDMA_RX_KERNEL_PRIO] =
 222                ADD_PRIO(0, MLX5_BY_PASS_NUM_REGULAR_PRIOS + 1, 0,
 223                         FS_CHAINING_CAPS,
 224                         ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_SWITCH_DOMAIN,
 225                                ADD_MULTIPLE_PRIO(1, 1))),
 226        }
 227};
 228
 229static struct init_tree_node rdma_tx_root_fs = {
 230        .type = FS_TYPE_NAMESPACE,
 231        .ar_size = 1,
 232        .children = (struct init_tree_node[]) {
 233                ADD_PRIO(0, MLX5_BY_PASS_NUM_PRIOS, 0,
 234                         FS_CHAINING_CAPS_RDMA_TX,
 235                         ADD_NS(MLX5_FLOW_TABLE_MISS_ACTION_DEF,
 236                                ADD_MULTIPLE_PRIO(MLX5_BY_PASS_NUM_PRIOS,
 237                                                  BY_PASS_PRIO_NUM_LEVELS))),
 238        }
 239};
 240
 241enum fs_i_lock_class {
 242        FS_LOCK_GRANDPARENT,
 243        FS_LOCK_PARENT,
 244        FS_LOCK_CHILD
 245};
 246
 247static const struct rhashtable_params rhash_fte = {
 248        .key_len = sizeof_field(struct fs_fte, val),
 249        .key_offset = offsetof(struct fs_fte, val),
 250        .head_offset = offsetof(struct fs_fte, hash),
 251        .automatic_shrinking = true,
 252        .min_size = 1,
 253};
 254
 255static const struct rhashtable_params rhash_fg = {
 256        .key_len = sizeof_field(struct mlx5_flow_group, mask),
 257        .key_offset = offsetof(struct mlx5_flow_group, mask),
 258        .head_offset = offsetof(struct mlx5_flow_group, hash),
 259        .automatic_shrinking = true,
 260        .min_size = 1,
 261
 262};
 263
 264static void del_hw_flow_table(struct fs_node *node);
 265static void del_hw_flow_group(struct fs_node *node);
 266static void del_hw_fte(struct fs_node *node);
 267static void del_sw_flow_table(struct fs_node *node);
 268static void del_sw_flow_group(struct fs_node *node);
 269static void del_sw_fte(struct fs_node *node);
 270static void del_sw_prio(struct fs_node *node);
 271static void del_sw_ns(struct fs_node *node);
 272/* Delete rule (destination) is special case that
 273 * requires to lock the FTE for all the deletion process.
 274 */
 275static void del_sw_hw_rule(struct fs_node *node);
 276static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
 277                                struct mlx5_flow_destination *d2);
 278static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns);
 279static struct mlx5_flow_rule *
 280find_flow_rule(struct fs_fte *fte,
 281               struct mlx5_flow_destination *dest);
 282
 283static void tree_init_node(struct fs_node *node,
 284                           void (*del_hw_func)(struct fs_node *),
 285                           void (*del_sw_func)(struct fs_node *))
 286{
 287        refcount_set(&node->refcount, 1);
 288        INIT_LIST_HEAD(&node->list);
 289        INIT_LIST_HEAD(&node->children);
 290        init_rwsem(&node->lock);
 291        node->del_hw_func = del_hw_func;
 292        node->del_sw_func = del_sw_func;
 293        node->active = false;
 294}
 295
 296static void tree_add_node(struct fs_node *node, struct fs_node *parent)
 297{
 298        if (parent)
 299                refcount_inc(&parent->refcount);
 300        node->parent = parent;
 301
 302        /* Parent is the root */
 303        if (!parent)
 304                node->root = node;
 305        else
 306                node->root = parent->root;
 307}
 308
 309static int tree_get_node(struct fs_node *node)
 310{
 311        return refcount_inc_not_zero(&node->refcount);
 312}
 313
 314static void nested_down_read_ref_node(struct fs_node *node,
 315                                      enum fs_i_lock_class class)
 316{
 317        if (node) {
 318                down_read_nested(&node->lock, class);
 319                refcount_inc(&node->refcount);
 320        }
 321}
 322
 323static void nested_down_write_ref_node(struct fs_node *node,
 324                                       enum fs_i_lock_class class)
 325{
 326        if (node) {
 327                down_write_nested(&node->lock, class);
 328                refcount_inc(&node->refcount);
 329        }
 330}
 331
 332static void down_write_ref_node(struct fs_node *node, bool locked)
 333{
 334        if (node) {
 335                if (!locked)
 336                        down_write(&node->lock);
 337                refcount_inc(&node->refcount);
 338        }
 339}
 340
 341static void up_read_ref_node(struct fs_node *node)
 342{
 343        refcount_dec(&node->refcount);
 344        up_read(&node->lock);
 345}
 346
 347static void up_write_ref_node(struct fs_node *node, bool locked)
 348{
 349        refcount_dec(&node->refcount);
 350        if (!locked)
 351                up_write(&node->lock);
 352}
 353
 354static void tree_put_node(struct fs_node *node, bool locked)
 355{
 356        struct fs_node *parent_node = node->parent;
 357
 358        if (refcount_dec_and_test(&node->refcount)) {
 359                if (node->del_hw_func)
 360                        node->del_hw_func(node);
 361                if (parent_node) {
 362                        down_write_ref_node(parent_node, locked);
 363                        list_del_init(&node->list);
 364                }
 365                node->del_sw_func(node);
 366                if (parent_node)
 367                        up_write_ref_node(parent_node, locked);
 368                node = NULL;
 369        }
 370        if (!node && parent_node)
 371                tree_put_node(parent_node, locked);
 372}
 373
 374static int tree_remove_node(struct fs_node *node, bool locked)
 375{
 376        if (refcount_read(&node->refcount) > 1) {
 377                refcount_dec(&node->refcount);
 378                return -EEXIST;
 379        }
 380        tree_put_node(node, locked);
 381        return 0;
 382}
 383
 384static struct fs_prio *find_prio(struct mlx5_flow_namespace *ns,
 385                                 unsigned int prio)
 386{
 387        struct fs_prio *iter_prio;
 388
 389        fs_for_each_prio(iter_prio, ns) {
 390                if (iter_prio->prio == prio)
 391                        return iter_prio;
 392        }
 393
 394        return NULL;
 395}
 396
 397static bool is_fwd_next_action(u32 action)
 398{
 399        return action & (MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
 400                         MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
 401}
 402
 403static bool check_valid_spec(const struct mlx5_flow_spec *spec)
 404{
 405        int i;
 406
 407        for (i = 0; i < MLX5_ST_SZ_DW_MATCH_PARAM; i++)
 408                if (spec->match_value[i] & ~spec->match_criteria[i]) {
 409                        pr_warn("mlx5_core: match_value differs from match_criteria\n");
 410                        return false;
 411                }
 412
 413        return true;
 414}
 415
 416static struct mlx5_flow_root_namespace *find_root(struct fs_node *node)
 417{
 418        struct fs_node *root;
 419        struct mlx5_flow_namespace *ns;
 420
 421        root = node->root;
 422
 423        if (WARN_ON(root->type != FS_TYPE_NAMESPACE)) {
 424                pr_warn("mlx5: flow steering node is not in tree or garbaged\n");
 425                return NULL;
 426        }
 427
 428        ns = container_of(root, struct mlx5_flow_namespace, node);
 429        return container_of(ns, struct mlx5_flow_root_namespace, ns);
 430}
 431
 432static inline struct mlx5_flow_steering *get_steering(struct fs_node *node)
 433{
 434        struct mlx5_flow_root_namespace *root = find_root(node);
 435
 436        if (root)
 437                return root->dev->priv.steering;
 438        return NULL;
 439}
 440
 441static inline struct mlx5_core_dev *get_dev(struct fs_node *node)
 442{
 443        struct mlx5_flow_root_namespace *root = find_root(node);
 444
 445        if (root)
 446                return root->dev;
 447        return NULL;
 448}
 449
 450static void del_sw_ns(struct fs_node *node)
 451{
 452        kfree(node);
 453}
 454
 455static void del_sw_prio(struct fs_node *node)
 456{
 457        kfree(node);
 458}
 459
 460static void del_hw_flow_table(struct fs_node *node)
 461{
 462        struct mlx5_flow_root_namespace *root;
 463        struct mlx5_flow_table *ft;
 464        struct mlx5_core_dev *dev;
 465        int err;
 466
 467        fs_get_obj(ft, node);
 468        dev = get_dev(&ft->node);
 469        root = find_root(&ft->node);
 470        trace_mlx5_fs_del_ft(ft);
 471
 472        if (node->active) {
 473                err = root->cmds->destroy_flow_table(root, ft);
 474                if (err)
 475                        mlx5_core_warn(dev, "flow steering can't destroy ft\n");
 476        }
 477}
 478
 479static void del_sw_flow_table(struct fs_node *node)
 480{
 481        struct mlx5_flow_table *ft;
 482        struct fs_prio *prio;
 483
 484        fs_get_obj(ft, node);
 485
 486        rhltable_destroy(&ft->fgs_hash);
 487        if (ft->node.parent) {
 488                fs_get_obj(prio, ft->node.parent);
 489                prio->num_ft--;
 490        }
 491        kfree(ft);
 492}
 493
 494static void modify_fte(struct fs_fte *fte)
 495{
 496        struct mlx5_flow_root_namespace *root;
 497        struct mlx5_flow_table *ft;
 498        struct mlx5_flow_group *fg;
 499        struct mlx5_core_dev *dev;
 500        int err;
 501
 502        fs_get_obj(fg, fte->node.parent);
 503        fs_get_obj(ft, fg->node.parent);
 504        dev = get_dev(&fte->node);
 505
 506        root = find_root(&ft->node);
 507        err = root->cmds->update_fte(root, ft, fg, fte->modify_mask, fte);
 508        if (err)
 509                mlx5_core_warn(dev,
 510                               "%s can't del rule fg id=%d fte_index=%d\n",
 511                               __func__, fg->id, fte->index);
 512        fte->modify_mask = 0;
 513}
 514
 515static void del_sw_hw_rule(struct fs_node *node)
 516{
 517        struct mlx5_flow_rule *rule;
 518        struct fs_fte *fte;
 519
 520        fs_get_obj(rule, node);
 521        fs_get_obj(fte, rule->node.parent);
 522        trace_mlx5_fs_del_rule(rule);
 523        if (is_fwd_next_action(rule->sw_action)) {
 524                mutex_lock(&rule->dest_attr.ft->lock);
 525                list_del(&rule->next_ft);
 526                mutex_unlock(&rule->dest_attr.ft->lock);
 527        }
 528
 529        if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_COUNTER  &&
 530            --fte->dests_size) {
 531                fte->modify_mask |=
 532                        BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION) |
 533                        BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
 534                fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_COUNT;
 535                goto out;
 536        }
 537
 538        if (rule->dest_attr.type == MLX5_FLOW_DESTINATION_TYPE_PORT &&
 539            --fte->dests_size) {
 540                fte->modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
 541                fte->action.action &= ~MLX5_FLOW_CONTEXT_ACTION_ALLOW;
 542                goto out;
 543        }
 544
 545        if ((fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) &&
 546            --fte->dests_size) {
 547                fte->modify_mask |=
 548                        BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 549        }
 550out:
 551        kfree(rule);
 552}
 553
 554static void del_hw_fte(struct fs_node *node)
 555{
 556        struct mlx5_flow_root_namespace *root;
 557        struct mlx5_flow_table *ft;
 558        struct mlx5_flow_group *fg;
 559        struct mlx5_core_dev *dev;
 560        struct fs_fte *fte;
 561        int err;
 562
 563        fs_get_obj(fte, node);
 564        fs_get_obj(fg, fte->node.parent);
 565        fs_get_obj(ft, fg->node.parent);
 566
 567        trace_mlx5_fs_del_fte(fte);
 568        dev = get_dev(&ft->node);
 569        root = find_root(&ft->node);
 570        if (node->active) {
 571                err = root->cmds->delete_fte(root, ft, fte);
 572                if (err)
 573                        mlx5_core_warn(dev,
 574                                       "flow steering can't delete fte in index %d of flow group id %d\n",
 575                                       fte->index, fg->id);
 576                node->active = false;
 577        }
 578}
 579
 580static void del_sw_fte(struct fs_node *node)
 581{
 582        struct mlx5_flow_steering *steering = get_steering(node);
 583        struct mlx5_flow_group *fg;
 584        struct fs_fte *fte;
 585        int err;
 586
 587        fs_get_obj(fte, node);
 588        fs_get_obj(fg, fte->node.parent);
 589
 590        err = rhashtable_remove_fast(&fg->ftes_hash,
 591                                     &fte->hash,
 592                                     rhash_fte);
 593        WARN_ON(err);
 594        ida_free(&fg->fte_allocator, fte->index - fg->start_index);
 595        kmem_cache_free(steering->ftes_cache, fte);
 596}
 597
 598static void del_hw_flow_group(struct fs_node *node)
 599{
 600        struct mlx5_flow_root_namespace *root;
 601        struct mlx5_flow_group *fg;
 602        struct mlx5_flow_table *ft;
 603        struct mlx5_core_dev *dev;
 604
 605        fs_get_obj(fg, node);
 606        fs_get_obj(ft, fg->node.parent);
 607        dev = get_dev(&ft->node);
 608        trace_mlx5_fs_del_fg(fg);
 609
 610        root = find_root(&ft->node);
 611        if (fg->node.active && root->cmds->destroy_flow_group(root, ft, fg))
 612                mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n",
 613                               fg->id, ft->id);
 614}
 615
 616static void del_sw_flow_group(struct fs_node *node)
 617{
 618        struct mlx5_flow_steering *steering = get_steering(node);
 619        struct mlx5_flow_group *fg;
 620        struct mlx5_flow_table *ft;
 621        int err;
 622
 623        fs_get_obj(fg, node);
 624        fs_get_obj(ft, fg->node.parent);
 625
 626        rhashtable_destroy(&fg->ftes_hash);
 627        ida_destroy(&fg->fte_allocator);
 628        if (ft->autogroup.active &&
 629            fg->max_ftes == ft->autogroup.group_size &&
 630            fg->start_index < ft->autogroup.max_fte)
 631                ft->autogroup.num_groups--;
 632        err = rhltable_remove(&ft->fgs_hash,
 633                              &fg->hash,
 634                              rhash_fg);
 635        WARN_ON(err);
 636        kmem_cache_free(steering->fgs_cache, fg);
 637}
 638
 639static int insert_fte(struct mlx5_flow_group *fg, struct fs_fte *fte)
 640{
 641        int index;
 642        int ret;
 643
 644        index = ida_alloc_max(&fg->fte_allocator, fg->max_ftes - 1, GFP_KERNEL);
 645        if (index < 0)
 646                return index;
 647
 648        fte->index = index + fg->start_index;
 649        ret = rhashtable_insert_fast(&fg->ftes_hash,
 650                                     &fte->hash,
 651                                     rhash_fte);
 652        if (ret)
 653                goto err_ida_remove;
 654
 655        tree_add_node(&fte->node, &fg->node);
 656        list_add_tail(&fte->node.list, &fg->node.children);
 657        return 0;
 658
 659err_ida_remove:
 660        ida_free(&fg->fte_allocator, index);
 661        return ret;
 662}
 663
 664static struct fs_fte *alloc_fte(struct mlx5_flow_table *ft,
 665                                const struct mlx5_flow_spec *spec,
 666                                struct mlx5_flow_act *flow_act)
 667{
 668        struct mlx5_flow_steering *steering = get_steering(&ft->node);
 669        struct fs_fte *fte;
 670
 671        fte = kmem_cache_zalloc(steering->ftes_cache, GFP_KERNEL);
 672        if (!fte)
 673                return ERR_PTR(-ENOMEM);
 674
 675        memcpy(fte->val, &spec->match_value, sizeof(fte->val));
 676        fte->node.type =  FS_TYPE_FLOW_ENTRY;
 677        fte->action = *flow_act;
 678        fte->flow_context = spec->flow_context;
 679
 680        tree_init_node(&fte->node, del_hw_fte, del_sw_fte);
 681
 682        return fte;
 683}
 684
 685static void dealloc_flow_group(struct mlx5_flow_steering *steering,
 686                               struct mlx5_flow_group *fg)
 687{
 688        rhashtable_destroy(&fg->ftes_hash);
 689        kmem_cache_free(steering->fgs_cache, fg);
 690}
 691
 692static struct mlx5_flow_group *alloc_flow_group(struct mlx5_flow_steering *steering,
 693                                                u8 match_criteria_enable,
 694                                                const void *match_criteria,
 695                                                int start_index,
 696                                                int end_index)
 697{
 698        struct mlx5_flow_group *fg;
 699        int ret;
 700
 701        fg = kmem_cache_zalloc(steering->fgs_cache, GFP_KERNEL);
 702        if (!fg)
 703                return ERR_PTR(-ENOMEM);
 704
 705        ret = rhashtable_init(&fg->ftes_hash, &rhash_fte);
 706        if (ret) {
 707                kmem_cache_free(steering->fgs_cache, fg);
 708                return ERR_PTR(ret);
 709        }
 710
 711        ida_init(&fg->fte_allocator);
 712        fg->mask.match_criteria_enable = match_criteria_enable;
 713        memcpy(&fg->mask.match_criteria, match_criteria,
 714               sizeof(fg->mask.match_criteria));
 715        fg->node.type =  FS_TYPE_FLOW_GROUP;
 716        fg->start_index = start_index;
 717        fg->max_ftes = end_index - start_index + 1;
 718
 719        return fg;
 720}
 721
 722static struct mlx5_flow_group *alloc_insert_flow_group(struct mlx5_flow_table *ft,
 723                                                       u8 match_criteria_enable,
 724                                                       const void *match_criteria,
 725                                                       int start_index,
 726                                                       int end_index,
 727                                                       struct list_head *prev)
 728{
 729        struct mlx5_flow_steering *steering = get_steering(&ft->node);
 730        struct mlx5_flow_group *fg;
 731        int ret;
 732
 733        fg = alloc_flow_group(steering, match_criteria_enable, match_criteria,
 734                              start_index, end_index);
 735        if (IS_ERR(fg))
 736                return fg;
 737
 738        /* initialize refcnt, add to parent list */
 739        ret = rhltable_insert(&ft->fgs_hash,
 740                              &fg->hash,
 741                              rhash_fg);
 742        if (ret) {
 743                dealloc_flow_group(steering, fg);
 744                return ERR_PTR(ret);
 745        }
 746
 747        tree_init_node(&fg->node, del_hw_flow_group, del_sw_flow_group);
 748        tree_add_node(&fg->node, &ft->node);
 749        /* Add node to group list */
 750        list_add(&fg->node.list, prev);
 751        atomic_inc(&ft->node.version);
 752
 753        return fg;
 754}
 755
 756static struct mlx5_flow_table *alloc_flow_table(int level, u16 vport,
 757                                                enum fs_flow_table_type table_type,
 758                                                enum fs_flow_table_op_mod op_mod,
 759                                                u32 flags)
 760{
 761        struct mlx5_flow_table *ft;
 762        int ret;
 763
 764        ft  = kzalloc(sizeof(*ft), GFP_KERNEL);
 765        if (!ft)
 766                return ERR_PTR(-ENOMEM);
 767
 768        ret = rhltable_init(&ft->fgs_hash, &rhash_fg);
 769        if (ret) {
 770                kfree(ft);
 771                return ERR_PTR(ret);
 772        }
 773
 774        ft->level = level;
 775        ft->node.type = FS_TYPE_FLOW_TABLE;
 776        ft->op_mod = op_mod;
 777        ft->type = table_type;
 778        ft->vport = vport;
 779        ft->flags = flags;
 780        INIT_LIST_HEAD(&ft->fwd_rules);
 781        mutex_init(&ft->lock);
 782
 783        return ft;
 784}
 785
 786/* If reverse is false, then we search for the first flow table in the
 787 * root sub-tree from start(closest from right), else we search for the
 788 * last flow table in the root sub-tree till start(closest from left).
 789 */
 790static struct mlx5_flow_table *find_closest_ft_recursive(struct fs_node  *root,
 791                                                         struct list_head *start,
 792                                                         bool reverse)
 793{
 794#define list_advance_entry(pos, reverse)                \
 795        ((reverse) ? list_prev_entry(pos, list) : list_next_entry(pos, list))
 796
 797#define list_for_each_advance_continue(pos, head, reverse)      \
 798        for (pos = list_advance_entry(pos, reverse);            \
 799             &pos->list != (head);                              \
 800             pos = list_advance_entry(pos, reverse))
 801
 802        struct fs_node *iter = list_entry(start, struct fs_node, list);
 803        struct mlx5_flow_table *ft = NULL;
 804
 805        if (!root || root->type == FS_TYPE_PRIO_CHAINS)
 806                return NULL;
 807
 808        list_for_each_advance_continue(iter, &root->children, reverse) {
 809                if (iter->type == FS_TYPE_FLOW_TABLE) {
 810                        fs_get_obj(ft, iter);
 811                        return ft;
 812                }
 813                ft = find_closest_ft_recursive(iter, &iter->children, reverse);
 814                if (ft)
 815                        return ft;
 816        }
 817
 818        return ft;
 819}
 820
 821/* If reverse is false then return the first flow table in next priority of
 822 * prio in the tree, else return the last flow table in the previous priority
 823 * of prio in the tree.
 824 */
 825static struct mlx5_flow_table *find_closest_ft(struct fs_prio *prio, bool reverse)
 826{
 827        struct mlx5_flow_table *ft = NULL;
 828        struct fs_node *curr_node;
 829        struct fs_node *parent;
 830
 831        parent = prio->node.parent;
 832        curr_node = &prio->node;
 833        while (!ft && parent) {
 834                ft = find_closest_ft_recursive(parent, &curr_node->list, reverse);
 835                curr_node = parent;
 836                parent = curr_node->parent;
 837        }
 838        return ft;
 839}
 840
 841/* Assuming all the tree is locked by mutex chain lock */
 842static struct mlx5_flow_table *find_next_chained_ft(struct fs_prio *prio)
 843{
 844        return find_closest_ft(prio, false);
 845}
 846
 847/* Assuming all the tree is locked by mutex chain lock */
 848static struct mlx5_flow_table *find_prev_chained_ft(struct fs_prio *prio)
 849{
 850        return find_closest_ft(prio, true);
 851}
 852
 853static struct mlx5_flow_table *find_next_fwd_ft(struct mlx5_flow_table *ft,
 854                                                struct mlx5_flow_act *flow_act)
 855{
 856        struct fs_prio *prio;
 857        bool next_ns;
 858
 859        next_ns = flow_act->action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS;
 860        fs_get_obj(prio, next_ns ? ft->ns->node.parent : ft->node.parent);
 861
 862        return find_next_chained_ft(prio);
 863}
 864
 865static int connect_fts_in_prio(struct mlx5_core_dev *dev,
 866                               struct fs_prio *prio,
 867                               struct mlx5_flow_table *ft)
 868{
 869        struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 870        struct mlx5_flow_table *iter;
 871        int err;
 872
 873        fs_for_each_ft(iter, prio) {
 874                err = root->cmds->modify_flow_table(root, iter, ft);
 875                if (err) {
 876                        mlx5_core_err(dev,
 877                                      "Failed to modify flow table id %d, type %d, err %d\n",
 878                                      iter->id, iter->type, err);
 879                        /* The driver is out of sync with the FW */
 880                        return err;
 881                }
 882        }
 883        return 0;
 884}
 885
 886/* Connect flow tables from previous priority of prio to ft */
 887static int connect_prev_fts(struct mlx5_core_dev *dev,
 888                            struct mlx5_flow_table *ft,
 889                            struct fs_prio *prio)
 890{
 891        struct mlx5_flow_table *prev_ft;
 892
 893        prev_ft = find_prev_chained_ft(prio);
 894        if (prev_ft) {
 895                struct fs_prio *prev_prio;
 896
 897                fs_get_obj(prev_prio, prev_ft->node.parent);
 898                return connect_fts_in_prio(dev, prev_prio, ft);
 899        }
 900        return 0;
 901}
 902
 903static int update_root_ft_create(struct mlx5_flow_table *ft, struct fs_prio
 904                                 *prio)
 905{
 906        struct mlx5_flow_root_namespace *root = find_root(&prio->node);
 907        struct mlx5_ft_underlay_qp *uqp;
 908        int min_level = INT_MAX;
 909        int err = 0;
 910        u32 qpn;
 911
 912        if (root->root_ft)
 913                min_level = root->root_ft->level;
 914
 915        if (ft->level >= min_level)
 916                return 0;
 917
 918        if (list_empty(&root->underlay_qpns)) {
 919                /* Don't set any QPN (zero) in case QPN list is empty */
 920                qpn = 0;
 921                err = root->cmds->update_root_ft(root, ft, qpn, false);
 922        } else {
 923                list_for_each_entry(uqp, &root->underlay_qpns, list) {
 924                        qpn = uqp->qpn;
 925                        err = root->cmds->update_root_ft(root, ft,
 926                                                         qpn, false);
 927                        if (err)
 928                                break;
 929                }
 930        }
 931
 932        if (err)
 933                mlx5_core_warn(root->dev,
 934                               "Update root flow table of id(%u) qpn(%d) failed\n",
 935                               ft->id, qpn);
 936        else
 937                root->root_ft = ft;
 938
 939        return err;
 940}
 941
 942static int _mlx5_modify_rule_destination(struct mlx5_flow_rule *rule,
 943                                         struct mlx5_flow_destination *dest)
 944{
 945        struct mlx5_flow_root_namespace *root;
 946        struct mlx5_flow_table *ft;
 947        struct mlx5_flow_group *fg;
 948        struct fs_fte *fte;
 949        int modify_mask = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
 950        int err = 0;
 951
 952        fs_get_obj(fte, rule->node.parent);
 953        if (!(fte->action.action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
 954                return -EINVAL;
 955        down_write_ref_node(&fte->node, false);
 956        fs_get_obj(fg, fte->node.parent);
 957        fs_get_obj(ft, fg->node.parent);
 958
 959        memcpy(&rule->dest_attr, dest, sizeof(*dest));
 960        root = find_root(&ft->node);
 961        err = root->cmds->update_fte(root, ft, fg,
 962                                     modify_mask, fte);
 963        up_write_ref_node(&fte->node, false);
 964
 965        return err;
 966}
 967
 968int mlx5_modify_rule_destination(struct mlx5_flow_handle *handle,
 969                                 struct mlx5_flow_destination *new_dest,
 970                                 struct mlx5_flow_destination *old_dest)
 971{
 972        int i;
 973
 974        if (!old_dest) {
 975                if (handle->num_rules != 1)
 976                        return -EINVAL;
 977                return _mlx5_modify_rule_destination(handle->rule[0],
 978                                                     new_dest);
 979        }
 980
 981        for (i = 0; i < handle->num_rules; i++) {
 982                if (mlx5_flow_dests_cmp(new_dest, &handle->rule[i]->dest_attr))
 983                        return _mlx5_modify_rule_destination(handle->rule[i],
 984                                                             new_dest);
 985        }
 986
 987        return -EINVAL;
 988}
 989
 990/* Modify/set FWD rules that point on old_next_ft to point on new_next_ft  */
 991static int connect_fwd_rules(struct mlx5_core_dev *dev,
 992                             struct mlx5_flow_table *new_next_ft,
 993                             struct mlx5_flow_table *old_next_ft)
 994{
 995        struct mlx5_flow_destination dest = {};
 996        struct mlx5_flow_rule *iter;
 997        int err = 0;
 998
 999        /* new_next_ft and old_next_ft could be NULL only
1000         * when we create/destroy the anchor flow table.
1001         */
1002        if (!new_next_ft || !old_next_ft)
1003                return 0;
1004
1005        dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1006        dest.ft = new_next_ft;
1007
1008        mutex_lock(&old_next_ft->lock);
1009        list_splice_init(&old_next_ft->fwd_rules, &new_next_ft->fwd_rules);
1010        mutex_unlock(&old_next_ft->lock);
1011        list_for_each_entry(iter, &new_next_ft->fwd_rules, next_ft) {
1012                if ((iter->sw_action & MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS) &&
1013                    iter->ft->ns == new_next_ft->ns)
1014                        continue;
1015
1016                err = _mlx5_modify_rule_destination(iter, &dest);
1017                if (err)
1018                        pr_err("mlx5_core: failed to modify rule to point on flow table %d\n",
1019                               new_next_ft->id);
1020        }
1021        return 0;
1022}
1023
1024static int connect_flow_table(struct mlx5_core_dev *dev, struct mlx5_flow_table *ft,
1025                              struct fs_prio *prio)
1026{
1027        struct mlx5_flow_table *next_ft, *first_ft;
1028        int err = 0;
1029
1030        /* Connect_prev_fts and update_root_ft_create are mutually exclusive */
1031
1032        first_ft = list_first_entry_or_null(&prio->node.children,
1033                                            struct mlx5_flow_table, node.list);
1034        if (!first_ft || first_ft->level > ft->level) {
1035                err = connect_prev_fts(dev, ft, prio);
1036                if (err)
1037                        return err;
1038
1039                next_ft = first_ft ? first_ft : find_next_chained_ft(prio);
1040                err = connect_fwd_rules(dev, ft, next_ft);
1041                if (err)
1042                        return err;
1043        }
1044
1045        if (MLX5_CAP_FLOWTABLE(dev,
1046                               flow_table_properties_nic_receive.modify_root))
1047                err = update_root_ft_create(ft, prio);
1048        return err;
1049}
1050
1051static void list_add_flow_table(struct mlx5_flow_table *ft,
1052                                struct fs_prio *prio)
1053{
1054        struct list_head *prev = &prio->node.children;
1055        struct mlx5_flow_table *iter;
1056
1057        fs_for_each_ft(iter, prio) {
1058                if (iter->level > ft->level)
1059                        break;
1060                prev = &iter->node.list;
1061        }
1062        list_add(&ft->node.list, prev);
1063}
1064
1065static struct mlx5_flow_table *__mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1066                                                        struct mlx5_flow_table_attr *ft_attr,
1067                                                        enum fs_flow_table_op_mod op_mod,
1068                                                        u16 vport)
1069{
1070        struct mlx5_flow_root_namespace *root = find_root(&ns->node);
1071        bool unmanaged = ft_attr->flags & MLX5_FLOW_TABLE_UNMANAGED;
1072        struct mlx5_flow_table *next_ft;
1073        struct fs_prio *fs_prio = NULL;
1074        struct mlx5_flow_table *ft;
1075        int err;
1076
1077        if (!root) {
1078                pr_err("mlx5: flow steering failed to find root of namespace\n");
1079                return ERR_PTR(-ENODEV);
1080        }
1081
1082        mutex_lock(&root->chain_lock);
1083        fs_prio = find_prio(ns, ft_attr->prio);
1084        if (!fs_prio) {
1085                err = -EINVAL;
1086                goto unlock_root;
1087        }
1088        if (!unmanaged) {
1089                /* The level is related to the
1090                 * priority level range.
1091                 */
1092                if (ft_attr->level >= fs_prio->num_levels) {
1093                        err = -ENOSPC;
1094                        goto unlock_root;
1095                }
1096
1097                ft_attr->level += fs_prio->start_level;
1098        }
1099
1100        /* The level is related to the
1101         * priority level range.
1102         */
1103        ft = alloc_flow_table(ft_attr->level,
1104                              vport,
1105                              root->table_type,
1106                              op_mod, ft_attr->flags);
1107        if (IS_ERR(ft)) {
1108                err = PTR_ERR(ft);
1109                goto unlock_root;
1110        }
1111
1112        tree_init_node(&ft->node, del_hw_flow_table, del_sw_flow_table);
1113        next_ft = unmanaged ? ft_attr->next_ft :
1114                              find_next_chained_ft(fs_prio);
1115        ft->def_miss_action = ns->def_miss_action;
1116        ft->ns = ns;
1117        err = root->cmds->create_flow_table(root, ft, ft_attr->max_fte, next_ft);
1118        if (err)
1119                goto free_ft;
1120
1121        if (!unmanaged) {
1122                err = connect_flow_table(root->dev, ft, fs_prio);
1123                if (err)
1124                        goto destroy_ft;
1125        }
1126
1127        ft->node.active = true;
1128        down_write_ref_node(&fs_prio->node, false);
1129        if (!unmanaged) {
1130                tree_add_node(&ft->node, &fs_prio->node);
1131                list_add_flow_table(ft, fs_prio);
1132        } else {
1133                ft->node.root = fs_prio->node.root;
1134        }
1135        fs_prio->num_ft++;
1136        up_write_ref_node(&fs_prio->node, false);
1137        mutex_unlock(&root->chain_lock);
1138        trace_mlx5_fs_add_ft(ft);
1139        return ft;
1140destroy_ft:
1141        root->cmds->destroy_flow_table(root, ft);
1142free_ft:
1143        rhltable_destroy(&ft->fgs_hash);
1144        kfree(ft);
1145unlock_root:
1146        mutex_unlock(&root->chain_lock);
1147        return ERR_PTR(err);
1148}
1149
1150struct mlx5_flow_table *mlx5_create_flow_table(struct mlx5_flow_namespace *ns,
1151                                               struct mlx5_flow_table_attr *ft_attr)
1152{
1153        return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, 0);
1154}
1155EXPORT_SYMBOL(mlx5_create_flow_table);
1156
1157struct mlx5_flow_table *
1158mlx5_create_vport_flow_table(struct mlx5_flow_namespace *ns,
1159                             struct mlx5_flow_table_attr *ft_attr, u16 vport)
1160{
1161        return __mlx5_create_flow_table(ns, ft_attr, FS_FT_OP_MOD_NORMAL, vport);
1162}
1163
1164struct mlx5_flow_table*
1165mlx5_create_lag_demux_flow_table(struct mlx5_flow_namespace *ns,
1166                                 int prio, u32 level)
1167{
1168        struct mlx5_flow_table_attr ft_attr = {};
1169
1170        ft_attr.level = level;
1171        ft_attr.prio  = prio;
1172        ft_attr.max_fte = 1;
1173
1174        return __mlx5_create_flow_table(ns, &ft_attr, FS_FT_OP_MOD_LAG_DEMUX, 0);
1175}
1176EXPORT_SYMBOL(mlx5_create_lag_demux_flow_table);
1177
1178#define MAX_FLOW_GROUP_SIZE BIT(24)
1179struct mlx5_flow_table*
1180mlx5_create_auto_grouped_flow_table(struct mlx5_flow_namespace *ns,
1181                                    struct mlx5_flow_table_attr *ft_attr)
1182{
1183        int num_reserved_entries = ft_attr->autogroup.num_reserved_entries;
1184        int max_num_groups = ft_attr->autogroup.max_num_groups;
1185        struct mlx5_flow_table *ft;
1186        int autogroups_max_fte;
1187
1188        ft = mlx5_create_flow_table(ns, ft_attr);
1189        if (IS_ERR(ft))
1190                return ft;
1191
1192        autogroups_max_fte = ft->max_fte - num_reserved_entries;
1193        if (max_num_groups > autogroups_max_fte)
1194                goto err_validate;
1195        if (num_reserved_entries > ft->max_fte)
1196                goto err_validate;
1197
1198        /* Align the number of groups according to the largest group size */
1199        if (autogroups_max_fte / (max_num_groups + 1) > MAX_FLOW_GROUP_SIZE)
1200                max_num_groups = (autogroups_max_fte / MAX_FLOW_GROUP_SIZE) - 1;
1201
1202        ft->autogroup.active = true;
1203        ft->autogroup.required_groups = max_num_groups;
1204        ft->autogroup.max_fte = autogroups_max_fte;
1205        /* We save place for flow groups in addition to max types */
1206        ft->autogroup.group_size = autogroups_max_fte / (max_num_groups + 1);
1207
1208        return ft;
1209
1210err_validate:
1211        mlx5_destroy_flow_table(ft);
1212        return ERR_PTR(-ENOSPC);
1213}
1214EXPORT_SYMBOL(mlx5_create_auto_grouped_flow_table);
1215
1216struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft,
1217                                               u32 *fg_in)
1218{
1219        struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1220        void *match_criteria = MLX5_ADDR_OF(create_flow_group_in,
1221                                            fg_in, match_criteria);
1222        u8 match_criteria_enable = MLX5_GET(create_flow_group_in,
1223                                            fg_in,
1224                                            match_criteria_enable);
1225        int start_index = MLX5_GET(create_flow_group_in, fg_in,
1226                                   start_flow_index);
1227        int end_index = MLX5_GET(create_flow_group_in, fg_in,
1228                                 end_flow_index);
1229        struct mlx5_flow_group *fg;
1230        int err;
1231
1232        if (ft->autogroup.active && start_index < ft->autogroup.max_fte)
1233                return ERR_PTR(-EPERM);
1234
1235        down_write_ref_node(&ft->node, false);
1236        fg = alloc_insert_flow_group(ft, match_criteria_enable, match_criteria,
1237                                     start_index, end_index,
1238                                     ft->node.children.prev);
1239        up_write_ref_node(&ft->node, false);
1240        if (IS_ERR(fg))
1241                return fg;
1242
1243        err = root->cmds->create_flow_group(root, ft, fg_in, fg);
1244        if (err) {
1245                tree_put_node(&fg->node, false);
1246                return ERR_PTR(err);
1247        }
1248        trace_mlx5_fs_add_fg(fg);
1249        fg->node.active = true;
1250
1251        return fg;
1252}
1253EXPORT_SYMBOL(mlx5_create_flow_group);
1254
1255static struct mlx5_flow_rule *alloc_rule(struct mlx5_flow_destination *dest)
1256{
1257        struct mlx5_flow_rule *rule;
1258
1259        rule = kzalloc(sizeof(*rule), GFP_KERNEL);
1260        if (!rule)
1261                return NULL;
1262
1263        INIT_LIST_HEAD(&rule->next_ft);
1264        rule->node.type = FS_TYPE_FLOW_DEST;
1265        if (dest)
1266                memcpy(&rule->dest_attr, dest, sizeof(*dest));
1267
1268        return rule;
1269}
1270
1271static struct mlx5_flow_handle *alloc_handle(int num_rules)
1272{
1273        struct mlx5_flow_handle *handle;
1274
1275        handle = kzalloc(struct_size(handle, rule, num_rules), GFP_KERNEL);
1276        if (!handle)
1277                return NULL;
1278
1279        handle->num_rules = num_rules;
1280
1281        return handle;
1282}
1283
1284static void destroy_flow_handle(struct fs_fte *fte,
1285                                struct mlx5_flow_handle *handle,
1286                                struct mlx5_flow_destination *dest,
1287                                int i)
1288{
1289        for (; --i >= 0;) {
1290                if (refcount_dec_and_test(&handle->rule[i]->node.refcount)) {
1291                        fte->dests_size--;
1292                        list_del(&handle->rule[i]->node.list);
1293                        kfree(handle->rule[i]);
1294                }
1295        }
1296        kfree(handle);
1297}
1298
1299static struct mlx5_flow_handle *
1300create_flow_handle(struct fs_fte *fte,
1301                   struct mlx5_flow_destination *dest,
1302                   int dest_num,
1303                   int *modify_mask,
1304                   bool *new_rule)
1305{
1306        struct mlx5_flow_handle *handle;
1307        struct mlx5_flow_rule *rule = NULL;
1308        static int count = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_FLOW_COUNTERS);
1309        static int dst = BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_DESTINATION_LIST);
1310        int type;
1311        int i = 0;
1312
1313        handle = alloc_handle((dest_num) ? dest_num : 1);
1314        if (!handle)
1315                return ERR_PTR(-ENOMEM);
1316
1317        do {
1318                if (dest) {
1319                        rule = find_flow_rule(fte, dest + i);
1320                        if (rule) {
1321                                refcount_inc(&rule->node.refcount);
1322                                goto rule_found;
1323                        }
1324                }
1325
1326                *new_rule = true;
1327                rule = alloc_rule(dest + i);
1328                if (!rule)
1329                        goto free_rules;
1330
1331                /* Add dest to dests list- we need flow tables to be in the
1332                 * end of the list for forward to next prio rules.
1333                 */
1334                tree_init_node(&rule->node, NULL, del_sw_hw_rule);
1335                if (dest &&
1336                    dest[i].type != MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE)
1337                        list_add(&rule->node.list, &fte->node.children);
1338                else
1339                        list_add_tail(&rule->node.list, &fte->node.children);
1340                if (dest) {
1341                        fte->dests_size++;
1342
1343                        type = dest[i].type ==
1344                                MLX5_FLOW_DESTINATION_TYPE_COUNTER;
1345                        *modify_mask |= type ? count : dst;
1346                }
1347rule_found:
1348                handle->rule[i] = rule;
1349        } while (++i < dest_num);
1350
1351        return handle;
1352
1353free_rules:
1354        destroy_flow_handle(fte, handle, dest, i);
1355        return ERR_PTR(-ENOMEM);
1356}
1357
1358/* fte should not be deleted while calling this function */
1359static struct mlx5_flow_handle *
1360add_rule_fte(struct fs_fte *fte,
1361             struct mlx5_flow_group *fg,
1362             struct mlx5_flow_destination *dest,
1363             int dest_num,
1364             bool update_action)
1365{
1366        struct mlx5_flow_root_namespace *root;
1367        struct mlx5_flow_handle *handle;
1368        struct mlx5_flow_table *ft;
1369        int modify_mask = 0;
1370        int err;
1371        bool new_rule = false;
1372
1373        handle = create_flow_handle(fte, dest, dest_num, &modify_mask,
1374                                    &new_rule);
1375        if (IS_ERR(handle) || !new_rule)
1376                goto out;
1377
1378        if (update_action)
1379                modify_mask |= BIT(MLX5_SET_FTE_MODIFY_ENABLE_MASK_ACTION);
1380
1381        fs_get_obj(ft, fg->node.parent);
1382        root = find_root(&fg->node);
1383        if (!(fte->status & FS_FTE_STATUS_EXISTING))
1384                err = root->cmds->create_fte(root, ft, fg, fte);
1385        else
1386                err = root->cmds->update_fte(root, ft, fg, modify_mask, fte);
1387        if (err)
1388                goto free_handle;
1389
1390        fte->node.active = true;
1391        fte->status |= FS_FTE_STATUS_EXISTING;
1392        atomic_inc(&fg->node.version);
1393
1394out:
1395        return handle;
1396
1397free_handle:
1398        destroy_flow_handle(fte, handle, dest, handle->num_rules);
1399        return ERR_PTR(err);
1400}
1401
1402static struct mlx5_flow_group *alloc_auto_flow_group(struct mlx5_flow_table  *ft,
1403                                                     const struct mlx5_flow_spec *spec)
1404{
1405        struct list_head *prev = &ft->node.children;
1406        u32 max_fte = ft->autogroup.max_fte;
1407        unsigned int candidate_index = 0;
1408        unsigned int group_size = 0;
1409        struct mlx5_flow_group *fg;
1410
1411        if (!ft->autogroup.active)
1412                return ERR_PTR(-ENOENT);
1413
1414        if (ft->autogroup.num_groups < ft->autogroup.required_groups)
1415                group_size = ft->autogroup.group_size;
1416
1417        /*  max_fte == ft->autogroup.max_types */
1418        if (group_size == 0)
1419                group_size = 1;
1420
1421        /* sorted by start_index */
1422        fs_for_each_fg(fg, ft) {
1423                if (candidate_index + group_size > fg->start_index)
1424                        candidate_index = fg->start_index + fg->max_ftes;
1425                else
1426                        break;
1427                prev = &fg->node.list;
1428        }
1429
1430        if (candidate_index + group_size > max_fte)
1431                return ERR_PTR(-ENOSPC);
1432
1433        fg = alloc_insert_flow_group(ft,
1434                                     spec->match_criteria_enable,
1435                                     spec->match_criteria,
1436                                     candidate_index,
1437                                     candidate_index + group_size - 1,
1438                                     prev);
1439        if (IS_ERR(fg))
1440                goto out;
1441
1442        if (group_size == ft->autogroup.group_size)
1443                ft->autogroup.num_groups++;
1444
1445out:
1446        return fg;
1447}
1448
1449static int create_auto_flow_group(struct mlx5_flow_table *ft,
1450                                  struct mlx5_flow_group *fg)
1451{
1452        struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1453        int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in);
1454        void *match_criteria_addr;
1455        u8 src_esw_owner_mask_on;
1456        void *misc;
1457        int err;
1458        u32 *in;
1459
1460        in = kvzalloc(inlen, GFP_KERNEL);
1461        if (!in)
1462                return -ENOMEM;
1463
1464        MLX5_SET(create_flow_group_in, in, match_criteria_enable,
1465                 fg->mask.match_criteria_enable);
1466        MLX5_SET(create_flow_group_in, in, start_flow_index, fg->start_index);
1467        MLX5_SET(create_flow_group_in, in, end_flow_index,   fg->start_index +
1468                 fg->max_ftes - 1);
1469
1470        misc = MLX5_ADDR_OF(fte_match_param, fg->mask.match_criteria,
1471                            misc_parameters);
1472        src_esw_owner_mask_on = !!MLX5_GET(fte_match_set_misc, misc,
1473                                         source_eswitch_owner_vhca_id);
1474        MLX5_SET(create_flow_group_in, in,
1475                 source_eswitch_owner_vhca_id_valid, src_esw_owner_mask_on);
1476
1477        match_criteria_addr = MLX5_ADDR_OF(create_flow_group_in,
1478                                           in, match_criteria);
1479        memcpy(match_criteria_addr, fg->mask.match_criteria,
1480               sizeof(fg->mask.match_criteria));
1481
1482        err = root->cmds->create_flow_group(root, ft, in, fg);
1483        if (!err) {
1484                fg->node.active = true;
1485                trace_mlx5_fs_add_fg(fg);
1486        }
1487
1488        kvfree(in);
1489        return err;
1490}
1491
1492static bool mlx5_flow_dests_cmp(struct mlx5_flow_destination *d1,
1493                                struct mlx5_flow_destination *d2)
1494{
1495        if (d1->type == d2->type) {
1496                if ((d1->type == MLX5_FLOW_DESTINATION_TYPE_VPORT &&
1497                     d1->vport.num == d2->vport.num &&
1498                     d1->vport.flags == d2->vport.flags &&
1499                     ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_VHCA_ID) ?
1500                      (d1->vport.vhca_id == d2->vport.vhca_id) : true) &&
1501                     ((d1->vport.flags & MLX5_FLOW_DEST_VPORT_REFORMAT_ID) ?
1502                      (d1->vport.pkt_reformat->id ==
1503                       d2->vport.pkt_reformat->id) : true)) ||
1504                    (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1505                     d1->ft == d2->ft) ||
1506                    (d1->type == MLX5_FLOW_DESTINATION_TYPE_TIR &&
1507                     d1->tir_num == d2->tir_num) ||
1508                    (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE_NUM &&
1509                     d1->ft_num == d2->ft_num) ||
1510                    (d1->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_SAMPLER &&
1511                     d1->sampler_id == d2->sampler_id))
1512                        return true;
1513        }
1514
1515        return false;
1516}
1517
1518static struct mlx5_flow_rule *find_flow_rule(struct fs_fte *fte,
1519                                             struct mlx5_flow_destination *dest)
1520{
1521        struct mlx5_flow_rule *rule;
1522
1523        list_for_each_entry(rule, &fte->node.children, node.list) {
1524                if (mlx5_flow_dests_cmp(&rule->dest_attr, dest))
1525                        return rule;
1526        }
1527        return NULL;
1528}
1529
1530static bool check_conflicting_actions(u32 action1, u32 action2)
1531{
1532        u32 xored_actions = action1 ^ action2;
1533
1534        /* if one rule only wants to count, it's ok */
1535        if (action1 == MLX5_FLOW_CONTEXT_ACTION_COUNT ||
1536            action2 == MLX5_FLOW_CONTEXT_ACTION_COUNT)
1537                return false;
1538
1539        if (xored_actions & (MLX5_FLOW_CONTEXT_ACTION_DROP  |
1540                             MLX5_FLOW_CONTEXT_ACTION_PACKET_REFORMAT |
1541                             MLX5_FLOW_CONTEXT_ACTION_DECAP |
1542                             MLX5_FLOW_CONTEXT_ACTION_MOD_HDR  |
1543                             MLX5_FLOW_CONTEXT_ACTION_VLAN_POP |
1544                             MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH |
1545                             MLX5_FLOW_CONTEXT_ACTION_VLAN_POP_2 |
1546                             MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH_2))
1547                return true;
1548
1549        return false;
1550}
1551
1552static int check_conflicting_ftes(struct fs_fte *fte,
1553                                  const struct mlx5_flow_context *flow_context,
1554                                  const struct mlx5_flow_act *flow_act)
1555{
1556        if (check_conflicting_actions(flow_act->action, fte->action.action)) {
1557                mlx5_core_warn(get_dev(&fte->node),
1558                               "Found two FTEs with conflicting actions\n");
1559                return -EEXIST;
1560        }
1561
1562        if ((flow_context->flags & FLOW_CONTEXT_HAS_TAG) &&
1563            fte->flow_context.flow_tag != flow_context->flow_tag) {
1564                mlx5_core_warn(get_dev(&fte->node),
1565                               "FTE flow tag %u already exists with different flow tag %u\n",
1566                               fte->flow_context.flow_tag,
1567                               flow_context->flow_tag);
1568                return -EEXIST;
1569        }
1570
1571        return 0;
1572}
1573
1574static struct mlx5_flow_handle *add_rule_fg(struct mlx5_flow_group *fg,
1575                                            const struct mlx5_flow_spec *spec,
1576                                            struct mlx5_flow_act *flow_act,
1577                                            struct mlx5_flow_destination *dest,
1578                                            int dest_num,
1579                                            struct fs_fte *fte)
1580{
1581        struct mlx5_flow_handle *handle;
1582        int old_action;
1583        int i;
1584        int ret;
1585
1586        ret = check_conflicting_ftes(fte, &spec->flow_context, flow_act);
1587        if (ret)
1588                return ERR_PTR(ret);
1589
1590        old_action = fte->action.action;
1591        fte->action.action |= flow_act->action;
1592        handle = add_rule_fte(fte, fg, dest, dest_num,
1593                              old_action != flow_act->action);
1594        if (IS_ERR(handle)) {
1595                fte->action.action = old_action;
1596                return handle;
1597        }
1598        trace_mlx5_fs_set_fte(fte, false);
1599
1600        for (i = 0; i < handle->num_rules; i++) {
1601                if (refcount_read(&handle->rule[i]->node.refcount) == 1) {
1602                        tree_add_node(&handle->rule[i]->node, &fte->node);
1603                        trace_mlx5_fs_add_rule(handle->rule[i]);
1604                }
1605        }
1606        return handle;
1607}
1608
1609static bool counter_is_valid(u32 action)
1610{
1611        return (action & (MLX5_FLOW_CONTEXT_ACTION_DROP |
1612                          MLX5_FLOW_CONTEXT_ACTION_ALLOW |
1613                          MLX5_FLOW_CONTEXT_ACTION_FWD_DEST));
1614}
1615
1616static bool dest_is_valid(struct mlx5_flow_destination *dest,
1617                          struct mlx5_flow_act *flow_act,
1618                          struct mlx5_flow_table *ft)
1619{
1620        bool ignore_level = flow_act->flags & FLOW_ACT_IGNORE_FLOW_LEVEL;
1621        u32 action = flow_act->action;
1622
1623        if (dest && (dest->type == MLX5_FLOW_DESTINATION_TYPE_COUNTER))
1624                return counter_is_valid(action);
1625
1626        if (!(action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST))
1627                return true;
1628
1629        if (ignore_level) {
1630                if (ft->type != FS_FT_FDB &&
1631                    ft->type != FS_FT_NIC_RX)
1632                        return false;
1633
1634                if (dest->type == MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE &&
1635                    ft->type != dest->ft->type)
1636                        return false;
1637        }
1638
1639        if (!dest || ((dest->type ==
1640            MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE) &&
1641            (dest->ft->level <= ft->level && !ignore_level)))
1642                return false;
1643        return true;
1644}
1645
1646struct match_list {
1647        struct list_head        list;
1648        struct mlx5_flow_group *g;
1649};
1650
1651static void free_match_list(struct match_list *head, bool ft_locked)
1652{
1653        struct match_list *iter, *match_tmp;
1654
1655        list_for_each_entry_safe(iter, match_tmp, &head->list,
1656                                 list) {
1657                tree_put_node(&iter->g->node, ft_locked);
1658                list_del(&iter->list);
1659                kfree(iter);
1660        }
1661}
1662
1663static int build_match_list(struct match_list *match_head,
1664                            struct mlx5_flow_table *ft,
1665                            const struct mlx5_flow_spec *spec,
1666                            bool ft_locked)
1667{
1668        struct rhlist_head *tmp, *list;
1669        struct mlx5_flow_group *g;
1670        int err = 0;
1671
1672        rcu_read_lock();
1673        INIT_LIST_HEAD(&match_head->list);
1674        /* Collect all fgs which has a matching match_criteria */
1675        list = rhltable_lookup(&ft->fgs_hash, spec, rhash_fg);
1676        /* RCU is atomic, we can't execute FW commands here */
1677        rhl_for_each_entry_rcu(g, tmp, list, hash) {
1678                struct match_list *curr_match;
1679
1680                if (unlikely(!tree_get_node(&g->node)))
1681                        continue;
1682
1683                curr_match = kmalloc(sizeof(*curr_match), GFP_ATOMIC);
1684                if (!curr_match) {
1685                        free_match_list(match_head, ft_locked);
1686                        err = -ENOMEM;
1687                        goto out;
1688                }
1689                curr_match->g = g;
1690                list_add_tail(&curr_match->list, &match_head->list);
1691        }
1692out:
1693        rcu_read_unlock();
1694        return err;
1695}
1696
1697static u64 matched_fgs_get_version(struct list_head *match_head)
1698{
1699        struct match_list *iter;
1700        u64 version = 0;
1701
1702        list_for_each_entry(iter, match_head, list)
1703                version += (u64)atomic_read(&iter->g->node.version);
1704        return version;
1705}
1706
1707static struct fs_fte *
1708lookup_fte_locked(struct mlx5_flow_group *g,
1709                  const u32 *match_value,
1710                  bool take_write)
1711{
1712        struct fs_fte *fte_tmp;
1713
1714        if (take_write)
1715                nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1716        else
1717                nested_down_read_ref_node(&g->node, FS_LOCK_PARENT);
1718        fte_tmp = rhashtable_lookup_fast(&g->ftes_hash, match_value,
1719                                         rhash_fte);
1720        if (!fte_tmp || !tree_get_node(&fte_tmp->node)) {
1721                fte_tmp = NULL;
1722                goto out;
1723        }
1724        if (!fte_tmp->node.active) {
1725                tree_put_node(&fte_tmp->node, false);
1726                fte_tmp = NULL;
1727                goto out;
1728        }
1729
1730        nested_down_write_ref_node(&fte_tmp->node, FS_LOCK_CHILD);
1731out:
1732        if (take_write)
1733                up_write_ref_node(&g->node, false);
1734        else
1735                up_read_ref_node(&g->node);
1736        return fte_tmp;
1737}
1738
1739static struct mlx5_flow_handle *
1740try_add_to_existing_fg(struct mlx5_flow_table *ft,
1741                       struct list_head *match_head,
1742                       const struct mlx5_flow_spec *spec,
1743                       struct mlx5_flow_act *flow_act,
1744                       struct mlx5_flow_destination *dest,
1745                       int dest_num,
1746                       int ft_version)
1747{
1748        struct mlx5_flow_steering *steering = get_steering(&ft->node);
1749        struct mlx5_flow_group *g;
1750        struct mlx5_flow_handle *rule;
1751        struct match_list *iter;
1752        bool take_write = false;
1753        struct fs_fte *fte;
1754        u64  version = 0;
1755        int err;
1756
1757        fte = alloc_fte(ft, spec, flow_act);
1758        if (IS_ERR(fte))
1759                return  ERR_PTR(-ENOMEM);
1760
1761search_again_locked:
1762        if (flow_act->flags & FLOW_ACT_NO_APPEND)
1763                goto skip_search;
1764        version = matched_fgs_get_version(match_head);
1765        /* Try to find an fte with identical match value and attempt update its
1766         * action.
1767         */
1768        list_for_each_entry(iter, match_head, list) {
1769                struct fs_fte *fte_tmp;
1770
1771                g = iter->g;
1772                fte_tmp = lookup_fte_locked(g, spec->match_value, take_write);
1773                if (!fte_tmp)
1774                        continue;
1775                rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte_tmp);
1776                /* No error check needed here, because insert_fte() is not called */
1777                up_write_ref_node(&fte_tmp->node, false);
1778                tree_put_node(&fte_tmp->node, false);
1779                kmem_cache_free(steering->ftes_cache, fte);
1780                return rule;
1781        }
1782
1783skip_search:
1784        /* No group with matching fte found, or we skipped the search.
1785         * Try to add a new fte to any matching fg.
1786         */
1787
1788        /* Check the ft version, for case that new flow group
1789         * was added while the fgs weren't locked
1790         */
1791        if (atomic_read(&ft->node.version) != ft_version) {
1792                rule = ERR_PTR(-EAGAIN);
1793                goto out;
1794        }
1795
1796        /* Check the fgs version. If version have changed it could be that an
1797         * FTE with the same match value was added while the fgs weren't
1798         * locked.
1799         */
1800        if (!(flow_act->flags & FLOW_ACT_NO_APPEND) &&
1801            version != matched_fgs_get_version(match_head)) {
1802                take_write = true;
1803                goto search_again_locked;
1804        }
1805
1806        list_for_each_entry(iter, match_head, list) {
1807                g = iter->g;
1808
1809                nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1810
1811                if (!g->node.active) {
1812                        up_write_ref_node(&g->node, false);
1813                        continue;
1814                }
1815
1816                err = insert_fte(g, fte);
1817                if (err) {
1818                        up_write_ref_node(&g->node, false);
1819                        if (err == -ENOSPC)
1820                                continue;
1821                        kmem_cache_free(steering->ftes_cache, fte);
1822                        return ERR_PTR(err);
1823                }
1824
1825                nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1826                up_write_ref_node(&g->node, false);
1827                rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1828                up_write_ref_node(&fte->node, false);
1829                if (IS_ERR(rule))
1830                        tree_put_node(&fte->node, false);
1831                return rule;
1832        }
1833        rule = ERR_PTR(-ENOENT);
1834out:
1835        kmem_cache_free(steering->ftes_cache, fte);
1836        return rule;
1837}
1838
1839static struct mlx5_flow_handle *
1840_mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1841                     const struct mlx5_flow_spec *spec,
1842                     struct mlx5_flow_act *flow_act,
1843                     struct mlx5_flow_destination *dest,
1844                     int dest_num)
1845
1846{
1847        struct mlx5_flow_steering *steering = get_steering(&ft->node);
1848        struct mlx5_flow_handle *rule;
1849        struct match_list match_head;
1850        struct mlx5_flow_group *g;
1851        bool take_write = false;
1852        struct fs_fte *fte;
1853        int version;
1854        int err;
1855        int i;
1856
1857        if (!check_valid_spec(spec))
1858                return ERR_PTR(-EINVAL);
1859
1860        for (i = 0; i < dest_num; i++) {
1861                if (!dest_is_valid(&dest[i], flow_act, ft))
1862                        return ERR_PTR(-EINVAL);
1863        }
1864        nested_down_read_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1865search_again_locked:
1866        version = atomic_read(&ft->node.version);
1867
1868        /* Collect all fgs which has a matching match_criteria */
1869        err = build_match_list(&match_head, ft, spec, take_write);
1870        if (err) {
1871                if (take_write)
1872                        up_write_ref_node(&ft->node, false);
1873                else
1874                        up_read_ref_node(&ft->node);
1875                return ERR_PTR(err);
1876        }
1877
1878        if (!take_write)
1879                up_read_ref_node(&ft->node);
1880
1881        rule = try_add_to_existing_fg(ft, &match_head.list, spec, flow_act, dest,
1882                                      dest_num, version);
1883        free_match_list(&match_head, take_write);
1884        if (!IS_ERR(rule) ||
1885            (PTR_ERR(rule) != -ENOENT && PTR_ERR(rule) != -EAGAIN)) {
1886                if (take_write)
1887                        up_write_ref_node(&ft->node, false);
1888                return rule;
1889        }
1890
1891        if (!take_write) {
1892                nested_down_write_ref_node(&ft->node, FS_LOCK_GRANDPARENT);
1893                take_write = true;
1894        }
1895
1896        if (PTR_ERR(rule) == -EAGAIN ||
1897            version != atomic_read(&ft->node.version))
1898                goto search_again_locked;
1899
1900        g = alloc_auto_flow_group(ft, spec);
1901        if (IS_ERR(g)) {
1902                rule = ERR_CAST(g);
1903                up_write_ref_node(&ft->node, false);
1904                return rule;
1905        }
1906
1907        fte = alloc_fte(ft, spec, flow_act);
1908        if (IS_ERR(fte)) {
1909                up_write_ref_node(&ft->node, false);
1910                err = PTR_ERR(fte);
1911                goto err_alloc_fte;
1912        }
1913
1914        nested_down_write_ref_node(&g->node, FS_LOCK_PARENT);
1915        up_write_ref_node(&ft->node, false);
1916
1917        err = create_auto_flow_group(ft, g);
1918        if (err)
1919                goto err_release_fg;
1920
1921        err = insert_fte(g, fte);
1922        if (err)
1923                goto err_release_fg;
1924
1925        nested_down_write_ref_node(&fte->node, FS_LOCK_CHILD);
1926        up_write_ref_node(&g->node, false);
1927        rule = add_rule_fg(g, spec, flow_act, dest, dest_num, fte);
1928        up_write_ref_node(&fte->node, false);
1929        if (IS_ERR(rule))
1930                tree_put_node(&fte->node, false);
1931        tree_put_node(&g->node, false);
1932        return rule;
1933
1934err_release_fg:
1935        up_write_ref_node(&g->node, false);
1936        kmem_cache_free(steering->ftes_cache, fte);
1937err_alloc_fte:
1938        tree_put_node(&g->node, false);
1939        return ERR_PTR(err);
1940}
1941
1942static bool fwd_next_prio_supported(struct mlx5_flow_table *ft)
1943{
1944        return ((ft->type == FS_FT_NIC_RX) &&
1945                (MLX5_CAP_FLOWTABLE(get_dev(&ft->node), nic_rx_multi_path_tirs)));
1946}
1947
1948struct mlx5_flow_handle *
1949mlx5_add_flow_rules(struct mlx5_flow_table *ft,
1950                    const struct mlx5_flow_spec *spec,
1951                    struct mlx5_flow_act *flow_act,
1952                    struct mlx5_flow_destination *dest,
1953                    int num_dest)
1954{
1955        struct mlx5_flow_root_namespace *root = find_root(&ft->node);
1956        static const struct mlx5_flow_spec zero_spec = {};
1957        struct mlx5_flow_destination *gen_dest = NULL;
1958        struct mlx5_flow_table *next_ft = NULL;
1959        struct mlx5_flow_handle *handle = NULL;
1960        u32 sw_action = flow_act->action;
1961        int i;
1962
1963        if (!spec)
1964                spec = &zero_spec;
1965
1966        if (!is_fwd_next_action(sw_action))
1967                return _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1968
1969        if (!fwd_next_prio_supported(ft))
1970                return ERR_PTR(-EOPNOTSUPP);
1971
1972        mutex_lock(&root->chain_lock);
1973        next_ft = find_next_fwd_ft(ft, flow_act);
1974        if (!next_ft) {
1975                handle = ERR_PTR(-EOPNOTSUPP);
1976                goto unlock;
1977        }
1978
1979        gen_dest = kcalloc(num_dest + 1, sizeof(*dest),
1980                           GFP_KERNEL);
1981        if (!gen_dest) {
1982                handle = ERR_PTR(-ENOMEM);
1983                goto unlock;
1984        }
1985        for (i = 0; i < num_dest; i++)
1986                gen_dest[i] = dest[i];
1987        gen_dest[i].type =
1988                MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
1989        gen_dest[i].ft = next_ft;
1990        dest = gen_dest;
1991        num_dest++;
1992        flow_act->action &= ~(MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_PRIO |
1993                              MLX5_FLOW_CONTEXT_ACTION_FWD_NEXT_NS);
1994        flow_act->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
1995        handle = _mlx5_add_flow_rules(ft, spec, flow_act, dest, num_dest);
1996        if (IS_ERR(handle))
1997                goto unlock;
1998
1999        if (list_empty(&handle->rule[num_dest - 1]->next_ft)) {
2000                mutex_lock(&next_ft->lock);
2001                list_add(&handle->rule[num_dest - 1]->next_ft,
2002                         &next_ft->fwd_rules);
2003                mutex_unlock(&next_ft->lock);
2004                handle->rule[num_dest - 1]->sw_action = sw_action;
2005                handle->rule[num_dest - 1]->ft = ft;
2006        }
2007unlock:
2008        mutex_unlock(&root->chain_lock);
2009        kfree(gen_dest);
2010        return handle;
2011}
2012EXPORT_SYMBOL(mlx5_add_flow_rules);
2013
2014void mlx5_del_flow_rules(struct mlx5_flow_handle *handle)
2015{
2016        struct fs_fte *fte;
2017        int i;
2018
2019        /* In order to consolidate the HW changes we lock the FTE for other
2020         * changes, and increase its refcount, in order not to perform the
2021         * "del" functions of the FTE. Will handle them here.
2022         * The removal of the rules is done under locked FTE.
2023         * After removing all the handle's rules, if there are remaining
2024         * rules, it means we just need to modify the FTE in FW, and
2025         * unlock/decrease the refcount we increased before.
2026         * Otherwise, it means the FTE should be deleted. First delete the
2027         * FTE in FW. Then, unlock the FTE, and proceed the tree_put_node of
2028         * the FTE, which will handle the last decrease of the refcount, as
2029         * well as required handling of its parent.
2030         */
2031        fs_get_obj(fte, handle->rule[0]->node.parent);
2032        down_write_ref_node(&fte->node, false);
2033        for (i = handle->num_rules - 1; i >= 0; i--)
2034                tree_remove_node(&handle->rule[i]->node, true);
2035        if (fte->dests_size) {
2036                if (fte->modify_mask)
2037                        modify_fte(fte);
2038                up_write_ref_node(&fte->node, false);
2039        } else if (list_empty(&fte->node.children)) {
2040                del_hw_fte(&fte->node);
2041                /* Avoid double call to del_hw_fte */
2042                fte->node.del_hw_func = NULL;
2043                up_write_ref_node(&fte->node, false);
2044                tree_put_node(&fte->node, false);
2045        }
2046        kfree(handle);
2047}
2048EXPORT_SYMBOL(mlx5_del_flow_rules);
2049
2050/* Assuming prio->node.children(flow tables) is sorted by level */
2051static struct mlx5_flow_table *find_next_ft(struct mlx5_flow_table *ft)
2052{
2053        struct fs_prio *prio;
2054
2055        fs_get_obj(prio, ft->node.parent);
2056
2057        if (!list_is_last(&ft->node.list, &prio->node.children))
2058                return list_next_entry(ft, node.list);
2059        return find_next_chained_ft(prio);
2060}
2061
2062static int update_root_ft_destroy(struct mlx5_flow_table *ft)
2063{
2064        struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2065        struct mlx5_ft_underlay_qp *uqp;
2066        struct mlx5_flow_table *new_root_ft = NULL;
2067        int err = 0;
2068        u32 qpn;
2069
2070        if (root->root_ft != ft)
2071                return 0;
2072
2073        new_root_ft = find_next_ft(ft);
2074        if (!new_root_ft) {
2075                root->root_ft = NULL;
2076                return 0;
2077        }
2078
2079        if (list_empty(&root->underlay_qpns)) {
2080                /* Don't set any QPN (zero) in case QPN list is empty */
2081                qpn = 0;
2082                err = root->cmds->update_root_ft(root, new_root_ft,
2083                                                 qpn, false);
2084        } else {
2085                list_for_each_entry(uqp, &root->underlay_qpns, list) {
2086                        qpn = uqp->qpn;
2087                        err = root->cmds->update_root_ft(root,
2088                                                         new_root_ft, qpn,
2089                                                         false);
2090                        if (err)
2091                                break;
2092                }
2093        }
2094
2095        if (err)
2096                mlx5_core_warn(root->dev,
2097                               "Update root flow table of id(%u) qpn(%d) failed\n",
2098                               ft->id, qpn);
2099        else
2100                root->root_ft = new_root_ft;
2101
2102        return 0;
2103}
2104
2105/* Connect flow table from previous priority to
2106 * the next flow table.
2107 */
2108static int disconnect_flow_table(struct mlx5_flow_table *ft)
2109{
2110        struct mlx5_core_dev *dev = get_dev(&ft->node);
2111        struct mlx5_flow_table *next_ft;
2112        struct fs_prio *prio;
2113        int err = 0;
2114
2115        err = update_root_ft_destroy(ft);
2116        if (err)
2117                return err;
2118
2119        fs_get_obj(prio, ft->node.parent);
2120        if  (!(list_first_entry(&prio->node.children,
2121                                struct mlx5_flow_table,
2122                                node.list) == ft))
2123                return 0;
2124
2125        next_ft = find_next_ft(ft);
2126        err = connect_fwd_rules(dev, next_ft, ft);
2127        if (err)
2128                return err;
2129
2130        err = connect_prev_fts(dev, next_ft, prio);
2131        if (err)
2132                mlx5_core_warn(dev, "Failed to disconnect flow table %d\n",
2133                               ft->id);
2134        return err;
2135}
2136
2137int mlx5_destroy_flow_table(struct mlx5_flow_table *ft)
2138{
2139        struct mlx5_flow_root_namespace *root = find_root(&ft->node);
2140        int err = 0;
2141
2142        mutex_lock(&root->chain_lock);
2143        if (!(ft->flags & MLX5_FLOW_TABLE_UNMANAGED))
2144                err = disconnect_flow_table(ft);
2145        if (err) {
2146                mutex_unlock(&root->chain_lock);
2147                return err;
2148        }
2149        if (tree_remove_node(&ft->node, false))
2150                mlx5_core_warn(get_dev(&ft->node), "Flow table %d wasn't destroyed, refcount > 1\n",
2151                               ft->id);
2152        mutex_unlock(&root->chain_lock);
2153
2154        return err;
2155}
2156EXPORT_SYMBOL(mlx5_destroy_flow_table);
2157
2158void mlx5_destroy_flow_group(struct mlx5_flow_group *fg)
2159{
2160        if (tree_remove_node(&fg->node, false))
2161                mlx5_core_warn(get_dev(&fg->node), "Flow group %d wasn't destroyed, refcount > 1\n",
2162                               fg->id);
2163}
2164EXPORT_SYMBOL(mlx5_destroy_flow_group);
2165
2166struct mlx5_flow_namespace *mlx5_get_fdb_sub_ns(struct mlx5_core_dev *dev,
2167                                                int n)
2168{
2169        struct mlx5_flow_steering *steering = dev->priv.steering;
2170
2171        if (!steering || !steering->fdb_sub_ns)
2172                return NULL;
2173
2174        return steering->fdb_sub_ns[n];
2175}
2176EXPORT_SYMBOL(mlx5_get_fdb_sub_ns);
2177
2178struct mlx5_flow_namespace *mlx5_get_flow_namespace(struct mlx5_core_dev *dev,
2179                                                    enum mlx5_flow_namespace_type type)
2180{
2181        struct mlx5_flow_steering *steering = dev->priv.steering;
2182        struct mlx5_flow_root_namespace *root_ns;
2183        int prio = 0;
2184        struct fs_prio *fs_prio;
2185        struct mlx5_flow_namespace *ns;
2186
2187        if (!steering)
2188                return NULL;
2189
2190        switch (type) {
2191        case MLX5_FLOW_NAMESPACE_FDB:
2192                if (steering->fdb_root_ns)
2193                        return &steering->fdb_root_ns->ns;
2194                return NULL;
2195        case MLX5_FLOW_NAMESPACE_SNIFFER_RX:
2196                if (steering->sniffer_rx_root_ns)
2197                        return &steering->sniffer_rx_root_ns->ns;
2198                return NULL;
2199        case MLX5_FLOW_NAMESPACE_SNIFFER_TX:
2200                if (steering->sniffer_tx_root_ns)
2201                        return &steering->sniffer_tx_root_ns->ns;
2202                return NULL;
2203        default:
2204                break;
2205        }
2206
2207        if (type == MLX5_FLOW_NAMESPACE_EGRESS ||
2208            type == MLX5_FLOW_NAMESPACE_EGRESS_KERNEL) {
2209                root_ns = steering->egress_root_ns;
2210                prio = type - MLX5_FLOW_NAMESPACE_EGRESS;
2211        } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX) {
2212                root_ns = steering->rdma_rx_root_ns;
2213                prio = RDMA_RX_BYPASS_PRIO;
2214        } else if (type == MLX5_FLOW_NAMESPACE_RDMA_RX_KERNEL) {
2215                root_ns = steering->rdma_rx_root_ns;
2216                prio = RDMA_RX_KERNEL_PRIO;
2217        } else if (type == MLX5_FLOW_NAMESPACE_RDMA_TX) {
2218                root_ns = steering->rdma_tx_root_ns;
2219        } else { /* Must be NIC RX */
2220                root_ns = steering->root_ns;
2221                prio = type;
2222        }
2223
2224        if (!root_ns)
2225                return NULL;
2226
2227        fs_prio = find_prio(&root_ns->ns, prio);
2228        if (!fs_prio)
2229                return NULL;
2230
2231        ns = list_first_entry(&fs_prio->node.children,
2232                              typeof(*ns),
2233                              node.list);
2234
2235        return ns;
2236}
2237EXPORT_SYMBOL(mlx5_get_flow_namespace);
2238
2239struct mlx5_flow_namespace *mlx5_get_flow_vport_acl_namespace(struct mlx5_core_dev *dev,
2240                                                              enum mlx5_flow_namespace_type type,
2241                                                              int vport)
2242{
2243        struct mlx5_flow_steering *steering = dev->priv.steering;
2244
2245        if (!steering)
2246                return NULL;
2247
2248        switch (type) {
2249        case MLX5_FLOW_NAMESPACE_ESW_EGRESS:
2250                if (vport >= steering->esw_egress_acl_vports)
2251                        return NULL;
2252                if (steering->esw_egress_root_ns &&
2253                    steering->esw_egress_root_ns[vport])
2254                        return &steering->esw_egress_root_ns[vport]->ns;
2255                else
2256                        return NULL;
2257        case MLX5_FLOW_NAMESPACE_ESW_INGRESS:
2258                if (vport >= steering->esw_ingress_acl_vports)
2259                        return NULL;
2260                if (steering->esw_ingress_root_ns &&
2261                    steering->esw_ingress_root_ns[vport])
2262                        return &steering->esw_ingress_root_ns[vport]->ns;
2263                else
2264                        return NULL;
2265        default:
2266                return NULL;
2267        }
2268}
2269
2270static struct fs_prio *_fs_create_prio(struct mlx5_flow_namespace *ns,
2271                                       unsigned int prio,
2272                                       int num_levels,
2273                                       enum fs_node_type type)
2274{
2275        struct fs_prio *fs_prio;
2276
2277        fs_prio = kzalloc(sizeof(*fs_prio), GFP_KERNEL);
2278        if (!fs_prio)
2279                return ERR_PTR(-ENOMEM);
2280
2281        fs_prio->node.type = type;
2282        tree_init_node(&fs_prio->node, NULL, del_sw_prio);
2283        tree_add_node(&fs_prio->node, &ns->node);
2284        fs_prio->num_levels = num_levels;
2285        fs_prio->prio = prio;
2286        list_add_tail(&fs_prio->node.list, &ns->node.children);
2287
2288        return fs_prio;
2289}
2290
2291static struct fs_prio *fs_create_prio_chained(struct mlx5_flow_namespace *ns,
2292                                              unsigned int prio,
2293                                              int num_levels)
2294{
2295        return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO_CHAINS);
2296}
2297
2298static struct fs_prio *fs_create_prio(struct mlx5_flow_namespace *ns,
2299                                      unsigned int prio, int num_levels)
2300{
2301        return _fs_create_prio(ns, prio, num_levels, FS_TYPE_PRIO);
2302}
2303
2304static struct mlx5_flow_namespace *fs_init_namespace(struct mlx5_flow_namespace
2305                                                     *ns)
2306{
2307        ns->node.type = FS_TYPE_NAMESPACE;
2308
2309        return ns;
2310}
2311
2312static struct mlx5_flow_namespace *fs_create_namespace(struct fs_prio *prio,
2313                                                       int def_miss_act)
2314{
2315        struct mlx5_flow_namespace      *ns;
2316
2317        ns = kzalloc(sizeof(*ns), GFP_KERNEL);
2318        if (!ns)
2319                return ERR_PTR(-ENOMEM);
2320
2321        fs_init_namespace(ns);
2322        ns->def_miss_action = def_miss_act;
2323        tree_init_node(&ns->node, NULL, del_sw_ns);
2324        tree_add_node(&ns->node, &prio->node);
2325        list_add_tail(&ns->node.list, &prio->node.children);
2326
2327        return ns;
2328}
2329
2330static int create_leaf_prios(struct mlx5_flow_namespace *ns, int prio,
2331                             struct init_tree_node *prio_metadata)
2332{
2333        struct fs_prio *fs_prio;
2334        int i;
2335
2336        for (i = 0; i < prio_metadata->num_leaf_prios; i++) {
2337                fs_prio = fs_create_prio(ns, prio++, prio_metadata->num_levels);
2338                if (IS_ERR(fs_prio))
2339                        return PTR_ERR(fs_prio);
2340        }
2341        return 0;
2342}
2343
2344#define FLOW_TABLE_BIT_SZ 1
2345#define GET_FLOW_TABLE_CAP(dev, offset) \
2346        ((be32_to_cpu(*((__be32 *)(dev->caps.hca_cur[MLX5_CAP_FLOW_TABLE]) +    \
2347                        offset / 32)) >>                                        \
2348          (32 - FLOW_TABLE_BIT_SZ - (offset & 0x1f))) & FLOW_TABLE_BIT_SZ)
2349static bool has_required_caps(struct mlx5_core_dev *dev, struct node_caps *caps)
2350{
2351        int i;
2352
2353        for (i = 0; i < caps->arr_sz; i++) {
2354                if (!GET_FLOW_TABLE_CAP(dev, caps->caps[i]))
2355                        return false;
2356        }
2357        return true;
2358}
2359
2360static int init_root_tree_recursive(struct mlx5_flow_steering *steering,
2361                                    struct init_tree_node *init_node,
2362                                    struct fs_node *fs_parent_node,
2363                                    struct init_tree_node *init_parent_node,
2364                                    int prio)
2365{
2366        int max_ft_level = MLX5_CAP_FLOWTABLE(steering->dev,
2367                                              flow_table_properties_nic_receive.
2368                                              max_ft_level);
2369        struct mlx5_flow_namespace *fs_ns;
2370        struct fs_prio *fs_prio;
2371        struct fs_node *base;
2372        int i;
2373        int err;
2374
2375        if (init_node->type == FS_TYPE_PRIO) {
2376                if ((init_node->min_ft_level > max_ft_level) ||
2377                    !has_required_caps(steering->dev, &init_node->caps))
2378                        return 0;
2379
2380                fs_get_obj(fs_ns, fs_parent_node);
2381                if (init_node->num_leaf_prios)
2382                        return create_leaf_prios(fs_ns, prio, init_node);
2383                fs_prio = fs_create_prio(fs_ns, prio, init_node->num_levels);
2384                if (IS_ERR(fs_prio))
2385                        return PTR_ERR(fs_prio);
2386                base = &fs_prio->node;
2387        } else if (init_node->type == FS_TYPE_NAMESPACE) {
2388                fs_get_obj(fs_prio, fs_parent_node);
2389                fs_ns = fs_create_namespace(fs_prio, init_node->def_miss_action);
2390                if (IS_ERR(fs_ns))
2391                        return PTR_ERR(fs_ns);
2392                base = &fs_ns->node;
2393        } else {
2394                return -EINVAL;
2395        }
2396        prio = 0;
2397        for (i = 0; i < init_node->ar_size; i++) {
2398                err = init_root_tree_recursive(steering, &init_node->children[i],
2399                                               base, init_node, prio);
2400                if (err)
2401                        return err;
2402                if (init_node->children[i].type == FS_TYPE_PRIO &&
2403                    init_node->children[i].num_leaf_prios) {
2404                        prio += init_node->children[i].num_leaf_prios;
2405                }
2406        }
2407
2408        return 0;
2409}
2410
2411static int init_root_tree(struct mlx5_flow_steering *steering,
2412                          struct init_tree_node *init_node,
2413                          struct fs_node *fs_parent_node)
2414{
2415        int err;
2416        int i;
2417
2418        for (i = 0; i < init_node->ar_size; i++) {
2419                err = init_root_tree_recursive(steering, &init_node->children[i],
2420                                               fs_parent_node,
2421                                               init_node, i);
2422                if (err)
2423                        return err;
2424        }
2425        return 0;
2426}
2427
2428static void del_sw_root_ns(struct fs_node *node)
2429{
2430        struct mlx5_flow_root_namespace *root_ns;
2431        struct mlx5_flow_namespace *ns;
2432
2433        fs_get_obj(ns, node);
2434        root_ns = container_of(ns, struct mlx5_flow_root_namespace, ns);
2435        mutex_destroy(&root_ns->chain_lock);
2436        kfree(node);
2437}
2438
2439static struct mlx5_flow_root_namespace
2440*create_root_ns(struct mlx5_flow_steering *steering,
2441                enum fs_flow_table_type table_type)
2442{
2443        const struct mlx5_flow_cmds *cmds = mlx5_fs_cmd_get_default(table_type);
2444        struct mlx5_flow_root_namespace *root_ns;
2445        struct mlx5_flow_namespace *ns;
2446
2447        if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE &&
2448            (table_type == FS_FT_NIC_RX || table_type == FS_FT_NIC_TX))
2449                cmds = mlx5_fs_cmd_get_default_ipsec_fpga_cmds(table_type);
2450
2451        /* Create the root namespace */
2452        root_ns = kzalloc(sizeof(*root_ns), GFP_KERNEL);
2453        if (!root_ns)
2454                return NULL;
2455
2456        root_ns->dev = steering->dev;
2457        root_ns->table_type = table_type;
2458        root_ns->cmds = cmds;
2459
2460        INIT_LIST_HEAD(&root_ns->underlay_qpns);
2461
2462        ns = &root_ns->ns;
2463        fs_init_namespace(ns);
2464        mutex_init(&root_ns->chain_lock);
2465        tree_init_node(&ns->node, NULL, del_sw_root_ns);
2466        tree_add_node(&ns->node, NULL);
2467
2468        return root_ns;
2469}
2470
2471static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level);
2472
2473static int set_prio_attrs_in_ns(struct mlx5_flow_namespace *ns, int acc_level)
2474{
2475        struct fs_prio *prio;
2476
2477        fs_for_each_prio(prio, ns) {
2478                 /* This updates prio start_level and num_levels */
2479                set_prio_attrs_in_prio(prio, acc_level);
2480                acc_level += prio->num_levels;
2481        }
2482        return acc_level;
2483}
2484
2485static void set_prio_attrs_in_prio(struct fs_prio *prio, int acc_level)
2486{
2487        struct mlx5_flow_namespace *ns;
2488        int acc_level_ns = acc_level;
2489
2490        prio->start_level = acc_level;
2491        fs_for_each_ns(ns, prio) {
2492                /* This updates start_level and num_levels of ns's priority descendants */
2493                acc_level_ns = set_prio_attrs_in_ns(ns, acc_level);
2494
2495                /* If this a prio with chains, and we can jump from one chain
2496                 * (namepsace) to another, so we accumulate the levels
2497                 */
2498                if (prio->node.type == FS_TYPE_PRIO_CHAINS)
2499                        acc_level = acc_level_ns;
2500        }
2501
2502        if (!prio->num_levels)
2503                prio->num_levels = acc_level_ns - prio->start_level;
2504        WARN_ON(prio->num_levels < acc_level_ns - prio->start_level);
2505}
2506
2507static void set_prio_attrs(struct mlx5_flow_root_namespace *root_ns)
2508{
2509        struct mlx5_flow_namespace *ns = &root_ns->ns;
2510        struct fs_prio *prio;
2511        int start_level = 0;
2512
2513        fs_for_each_prio(prio, ns) {
2514                set_prio_attrs_in_prio(prio, start_level);
2515                start_level += prio->num_levels;
2516        }
2517}
2518
2519#define ANCHOR_PRIO 0
2520#define ANCHOR_SIZE 1
2521#define ANCHOR_LEVEL 0
2522static int create_anchor_flow_table(struct mlx5_flow_steering *steering)
2523{
2524        struct mlx5_flow_namespace *ns = NULL;
2525        struct mlx5_flow_table_attr ft_attr = {};
2526        struct mlx5_flow_table *ft;
2527
2528        ns = mlx5_get_flow_namespace(steering->dev, MLX5_FLOW_NAMESPACE_ANCHOR);
2529        if (WARN_ON(!ns))
2530                return -EINVAL;
2531
2532        ft_attr.max_fte = ANCHOR_SIZE;
2533        ft_attr.level   = ANCHOR_LEVEL;
2534        ft_attr.prio    = ANCHOR_PRIO;
2535
2536        ft = mlx5_create_flow_table(ns, &ft_attr);
2537        if (IS_ERR(ft)) {
2538                mlx5_core_err(steering->dev, "Failed to create last anchor flow table");
2539                return PTR_ERR(ft);
2540        }
2541        return 0;
2542}
2543
2544static int init_root_ns(struct mlx5_flow_steering *steering)
2545{
2546        int err;
2547
2548        steering->root_ns = create_root_ns(steering, FS_FT_NIC_RX);
2549        if (!steering->root_ns)
2550                return -ENOMEM;
2551
2552        err = init_root_tree(steering, &root_fs, &steering->root_ns->ns.node);
2553        if (err)
2554                goto out_err;
2555
2556        set_prio_attrs(steering->root_ns);
2557        err = create_anchor_flow_table(steering);
2558        if (err)
2559                goto out_err;
2560
2561        return 0;
2562
2563out_err:
2564        cleanup_root_ns(steering->root_ns);
2565        steering->root_ns = NULL;
2566        return err;
2567}
2568
2569static void clean_tree(struct fs_node *node)
2570{
2571        if (node) {
2572                struct fs_node *iter;
2573                struct fs_node *temp;
2574
2575                tree_get_node(node);
2576                list_for_each_entry_safe(iter, temp, &node->children, list)
2577                        clean_tree(iter);
2578                tree_put_node(node, false);
2579                tree_remove_node(node, false);
2580        }
2581}
2582
2583static void cleanup_root_ns(struct mlx5_flow_root_namespace *root_ns)
2584{
2585        if (!root_ns)
2586                return;
2587
2588        clean_tree(&root_ns->ns.node);
2589}
2590
2591void mlx5_cleanup_fs(struct mlx5_core_dev *dev)
2592{
2593        struct mlx5_flow_steering *steering = dev->priv.steering;
2594
2595        cleanup_root_ns(steering->root_ns);
2596        cleanup_root_ns(steering->fdb_root_ns);
2597        steering->fdb_root_ns = NULL;
2598        kfree(steering->fdb_sub_ns);
2599        steering->fdb_sub_ns = NULL;
2600        cleanup_root_ns(steering->sniffer_rx_root_ns);
2601        cleanup_root_ns(steering->sniffer_tx_root_ns);
2602        cleanup_root_ns(steering->rdma_rx_root_ns);
2603        cleanup_root_ns(steering->rdma_tx_root_ns);
2604        cleanup_root_ns(steering->egress_root_ns);
2605        mlx5_cleanup_fc_stats(dev);
2606        kmem_cache_destroy(steering->ftes_cache);
2607        kmem_cache_destroy(steering->fgs_cache);
2608        mlx5_ft_pool_destroy(dev);
2609        kfree(steering);
2610}
2611
2612static int init_sniffer_tx_root_ns(struct mlx5_flow_steering *steering)
2613{
2614        struct fs_prio *prio;
2615
2616        steering->sniffer_tx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_TX);
2617        if (!steering->sniffer_tx_root_ns)
2618                return -ENOMEM;
2619
2620        /* Create single prio */
2621        prio = fs_create_prio(&steering->sniffer_tx_root_ns->ns, 0, 1);
2622        return PTR_ERR_OR_ZERO(prio);
2623}
2624
2625static int init_sniffer_rx_root_ns(struct mlx5_flow_steering *steering)
2626{
2627        struct fs_prio *prio;
2628
2629        steering->sniffer_rx_root_ns = create_root_ns(steering, FS_FT_SNIFFER_RX);
2630        if (!steering->sniffer_rx_root_ns)
2631                return -ENOMEM;
2632
2633        /* Create single prio */
2634        prio = fs_create_prio(&steering->sniffer_rx_root_ns->ns, 0, 1);
2635        return PTR_ERR_OR_ZERO(prio);
2636}
2637
2638static int init_rdma_rx_root_ns(struct mlx5_flow_steering *steering)
2639{
2640        int err;
2641
2642        steering->rdma_rx_root_ns = create_root_ns(steering, FS_FT_RDMA_RX);
2643        if (!steering->rdma_rx_root_ns)
2644                return -ENOMEM;
2645
2646        err = init_root_tree(steering, &rdma_rx_root_fs,
2647                             &steering->rdma_rx_root_ns->ns.node);
2648        if (err)
2649                goto out_err;
2650
2651        set_prio_attrs(steering->rdma_rx_root_ns);
2652
2653        return 0;
2654
2655out_err:
2656        cleanup_root_ns(steering->rdma_rx_root_ns);
2657        steering->rdma_rx_root_ns = NULL;
2658        return err;
2659}
2660
2661static int init_rdma_tx_root_ns(struct mlx5_flow_steering *steering)
2662{
2663        int err;
2664
2665        steering->rdma_tx_root_ns = create_root_ns(steering, FS_FT_RDMA_TX);
2666        if (!steering->rdma_tx_root_ns)
2667                return -ENOMEM;
2668
2669        err = init_root_tree(steering, &rdma_tx_root_fs,
2670                             &steering->rdma_tx_root_ns->ns.node);
2671        if (err)
2672                goto out_err;
2673
2674        set_prio_attrs(steering->rdma_tx_root_ns);
2675
2676        return 0;
2677
2678out_err:
2679        cleanup_root_ns(steering->rdma_tx_root_ns);
2680        steering->rdma_tx_root_ns = NULL;
2681        return err;
2682}
2683
2684/* FT and tc chains are stored in the same array so we can re-use the
2685 * mlx5_get_fdb_sub_ns() and tc api for FT chains.
2686 * When creating a new ns for each chain store it in the first available slot.
2687 * Assume tc chains are created and stored first and only then the FT chain.
2688 */
2689static void store_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2690                                        struct mlx5_flow_namespace *ns)
2691{
2692        int chain = 0;
2693
2694        while (steering->fdb_sub_ns[chain])
2695                ++chain;
2696
2697        steering->fdb_sub_ns[chain] = ns;
2698}
2699
2700static int create_fdb_sub_ns_prio_chain(struct mlx5_flow_steering *steering,
2701                                        struct fs_prio *maj_prio)
2702{
2703        struct mlx5_flow_namespace *ns;
2704        struct fs_prio *min_prio;
2705        int prio;
2706
2707        ns = fs_create_namespace(maj_prio, MLX5_FLOW_TABLE_MISS_ACTION_DEF);
2708        if (IS_ERR(ns))
2709                return PTR_ERR(ns);
2710
2711        for (prio = 0; prio < FDB_TC_MAX_PRIO; prio++) {
2712                min_prio = fs_create_prio(ns, prio, FDB_TC_LEVELS_PER_PRIO);
2713                if (IS_ERR(min_prio))
2714                        return PTR_ERR(min_prio);
2715        }
2716
2717        store_fdb_sub_ns_prio_chain(steering, ns);
2718
2719        return 0;
2720}
2721
2722static int create_fdb_chains(struct mlx5_flow_steering *steering,
2723                             int fs_prio,
2724                             int chains)
2725{
2726        struct fs_prio *maj_prio;
2727        int levels;
2728        int chain;
2729        int err;
2730
2731        levels = FDB_TC_LEVELS_PER_PRIO * FDB_TC_MAX_PRIO * chains;
2732        maj_prio = fs_create_prio_chained(&steering->fdb_root_ns->ns,
2733                                          fs_prio,
2734                                          levels);
2735        if (IS_ERR(maj_prio))
2736                return PTR_ERR(maj_prio);
2737
2738        for (chain = 0; chain < chains; chain++) {
2739                err = create_fdb_sub_ns_prio_chain(steering, maj_prio);
2740                if (err)
2741                        return err;
2742        }
2743
2744        return 0;
2745}
2746
2747static int create_fdb_fast_path(struct mlx5_flow_steering *steering)
2748{
2749        int err;
2750
2751        steering->fdb_sub_ns = kcalloc(FDB_NUM_CHAINS,
2752                                       sizeof(*steering->fdb_sub_ns),
2753                                       GFP_KERNEL);
2754        if (!steering->fdb_sub_ns)
2755                return -ENOMEM;
2756
2757        err = create_fdb_chains(steering, FDB_TC_OFFLOAD, FDB_TC_MAX_CHAIN + 1);
2758        if (err)
2759                return err;
2760
2761        err = create_fdb_chains(steering, FDB_FT_OFFLOAD, 1);
2762        if (err)
2763                return err;
2764
2765        return 0;
2766}
2767
2768static int init_fdb_root_ns(struct mlx5_flow_steering *steering)
2769{
2770        struct fs_prio *maj_prio;
2771        int err;
2772
2773        steering->fdb_root_ns = create_root_ns(steering, FS_FT_FDB);
2774        if (!steering->fdb_root_ns)
2775                return -ENOMEM;
2776
2777        maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BYPASS_PATH,
2778                                  1);
2779        if (IS_ERR(maj_prio)) {
2780                err = PTR_ERR(maj_prio);
2781                goto out_err;
2782        }
2783        err = create_fdb_fast_path(steering);
2784        if (err)
2785                goto out_err;
2786
2787        maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_TC_MISS, 1);
2788        if (IS_ERR(maj_prio)) {
2789                err = PTR_ERR(maj_prio);
2790                goto out_err;
2791        }
2792
2793        maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_BR_OFFLOAD, 3);
2794        if (IS_ERR(maj_prio)) {
2795                err = PTR_ERR(maj_prio);
2796                goto out_err;
2797        }
2798
2799        maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_SLOW_PATH, 1);
2800        if (IS_ERR(maj_prio)) {
2801                err = PTR_ERR(maj_prio);
2802                goto out_err;
2803        }
2804
2805        /* We put this priority last, knowing that nothing will get here
2806         * unless explicitly forwarded to. This is possible because the
2807         * slow path tables have catch all rules and nothing gets passed
2808         * those tables.
2809         */
2810        maj_prio = fs_create_prio(&steering->fdb_root_ns->ns, FDB_PER_VPORT, 1);
2811        if (IS_ERR(maj_prio)) {
2812                err = PTR_ERR(maj_prio);
2813                goto out_err;
2814        }
2815
2816        set_prio_attrs(steering->fdb_root_ns);
2817        return 0;
2818
2819out_err:
2820        cleanup_root_ns(steering->fdb_root_ns);
2821        kfree(steering->fdb_sub_ns);
2822        steering->fdb_sub_ns = NULL;
2823        steering->fdb_root_ns = NULL;
2824        return err;
2825}
2826
2827static int init_egress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2828{
2829        struct fs_prio *prio;
2830
2831        steering->esw_egress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_EGRESS_ACL);
2832        if (!steering->esw_egress_root_ns[vport])
2833                return -ENOMEM;
2834
2835        /* create 1 prio*/
2836        prio = fs_create_prio(&steering->esw_egress_root_ns[vport]->ns, 0, 1);
2837        return PTR_ERR_OR_ZERO(prio);
2838}
2839
2840static int init_ingress_acl_root_ns(struct mlx5_flow_steering *steering, int vport)
2841{
2842        struct fs_prio *prio;
2843
2844        steering->esw_ingress_root_ns[vport] = create_root_ns(steering, FS_FT_ESW_INGRESS_ACL);
2845        if (!steering->esw_ingress_root_ns[vport])
2846                return -ENOMEM;
2847
2848        /* create 1 prio*/
2849        prio = fs_create_prio(&steering->esw_ingress_root_ns[vport]->ns, 0, 1);
2850        return PTR_ERR_OR_ZERO(prio);
2851}
2852
2853int mlx5_fs_egress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2854{
2855        struct mlx5_flow_steering *steering = dev->priv.steering;
2856        int err;
2857        int i;
2858
2859        steering->esw_egress_root_ns =
2860                        kcalloc(total_vports,
2861                                sizeof(*steering->esw_egress_root_ns),
2862                                GFP_KERNEL);
2863        if (!steering->esw_egress_root_ns)
2864                return -ENOMEM;
2865
2866        for (i = 0; i < total_vports; i++) {
2867                err = init_egress_acl_root_ns(steering, i);
2868                if (err)
2869                        goto cleanup_root_ns;
2870        }
2871        steering->esw_egress_acl_vports = total_vports;
2872        return 0;
2873
2874cleanup_root_ns:
2875        for (i--; i >= 0; i--)
2876                cleanup_root_ns(steering->esw_egress_root_ns[i]);
2877        kfree(steering->esw_egress_root_ns);
2878        steering->esw_egress_root_ns = NULL;
2879        return err;
2880}
2881
2882void mlx5_fs_egress_acls_cleanup(struct mlx5_core_dev *dev)
2883{
2884        struct mlx5_flow_steering *steering = dev->priv.steering;
2885        int i;
2886
2887        if (!steering->esw_egress_root_ns)
2888                return;
2889
2890        for (i = 0; i < steering->esw_egress_acl_vports; i++)
2891                cleanup_root_ns(steering->esw_egress_root_ns[i]);
2892
2893        kfree(steering->esw_egress_root_ns);
2894        steering->esw_egress_root_ns = NULL;
2895}
2896
2897int mlx5_fs_ingress_acls_init(struct mlx5_core_dev *dev, int total_vports)
2898{
2899        struct mlx5_flow_steering *steering = dev->priv.steering;
2900        int err;
2901        int i;
2902
2903        steering->esw_ingress_root_ns =
2904                        kcalloc(total_vports,
2905                                sizeof(*steering->esw_ingress_root_ns),
2906                                GFP_KERNEL);
2907        if (!steering->esw_ingress_root_ns)
2908                return -ENOMEM;
2909
2910        for (i = 0; i < total_vports; i++) {
2911                err = init_ingress_acl_root_ns(steering, i);
2912                if (err)
2913                        goto cleanup_root_ns;
2914        }
2915        steering->esw_ingress_acl_vports = total_vports;
2916        return 0;
2917
2918cleanup_root_ns:
2919        for (i--; i >= 0; i--)
2920                cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2921        kfree(steering->esw_ingress_root_ns);
2922        steering->esw_ingress_root_ns = NULL;
2923        return err;
2924}
2925
2926void mlx5_fs_ingress_acls_cleanup(struct mlx5_core_dev *dev)
2927{
2928        struct mlx5_flow_steering *steering = dev->priv.steering;
2929        int i;
2930
2931        if (!steering->esw_ingress_root_ns)
2932                return;
2933
2934        for (i = 0; i < steering->esw_ingress_acl_vports; i++)
2935                cleanup_root_ns(steering->esw_ingress_root_ns[i]);
2936
2937        kfree(steering->esw_ingress_root_ns);
2938        steering->esw_ingress_root_ns = NULL;
2939}
2940
2941static int init_egress_root_ns(struct mlx5_flow_steering *steering)
2942{
2943        int err;
2944
2945        steering->egress_root_ns = create_root_ns(steering,
2946                                                  FS_FT_NIC_TX);
2947        if (!steering->egress_root_ns)
2948                return -ENOMEM;
2949
2950        err = init_root_tree(steering, &egress_root_fs,
2951                             &steering->egress_root_ns->ns.node);
2952        if (err)
2953                goto cleanup;
2954        set_prio_attrs(steering->egress_root_ns);
2955        return 0;
2956cleanup:
2957        cleanup_root_ns(steering->egress_root_ns);
2958        steering->egress_root_ns = NULL;
2959        return err;
2960}
2961
2962int mlx5_init_fs(struct mlx5_core_dev *dev)
2963{
2964        struct mlx5_flow_steering *steering;
2965        int err = 0;
2966
2967        err = mlx5_init_fc_stats(dev);
2968        if (err)
2969                return err;
2970
2971        err = mlx5_ft_pool_init(dev);
2972        if (err)
2973                return err;
2974
2975        steering = kzalloc(sizeof(*steering), GFP_KERNEL);
2976        if (!steering) {
2977                err = -ENOMEM;
2978                goto err;
2979        }
2980
2981        steering->dev = dev;
2982        dev->priv.steering = steering;
2983
2984        steering->fgs_cache = kmem_cache_create("mlx5_fs_fgs",
2985                                                sizeof(struct mlx5_flow_group), 0,
2986                                                0, NULL);
2987        steering->ftes_cache = kmem_cache_create("mlx5_fs_ftes", sizeof(struct fs_fte), 0,
2988                                                 0, NULL);
2989        if (!steering->ftes_cache || !steering->fgs_cache) {
2990                err = -ENOMEM;
2991                goto err;
2992        }
2993
2994        if ((((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH) &&
2995              (MLX5_CAP_GEN(dev, nic_flow_table))) ||
2996             ((MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_IB) &&
2997              MLX5_CAP_GEN(dev, ipoib_enhanced_offloads))) &&
2998            MLX5_CAP_FLOWTABLE_NIC_RX(dev, ft_support)) {
2999                err = init_root_ns(steering);
3000                if (err)
3001                        goto err;
3002        }
3003
3004        if (MLX5_ESWITCH_MANAGER(dev)) {
3005                if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, ft_support)) {
3006                        err = init_fdb_root_ns(steering);
3007                        if (err)
3008                                goto err;
3009                }
3010        }
3011
3012        if (MLX5_CAP_FLOWTABLE_SNIFFER_RX(dev, ft_support)) {
3013                err = init_sniffer_rx_root_ns(steering);
3014                if (err)
3015                        goto err;
3016        }
3017
3018        if (MLX5_CAP_FLOWTABLE_SNIFFER_TX(dev, ft_support)) {
3019                err = init_sniffer_tx_root_ns(steering);
3020                if (err)
3021                        goto err;
3022        }
3023
3024        if (MLX5_CAP_FLOWTABLE_RDMA_RX(dev, ft_support) &&
3025            MLX5_CAP_FLOWTABLE_RDMA_RX(dev, table_miss_action_domain)) {
3026                err = init_rdma_rx_root_ns(steering);
3027                if (err)
3028                        goto err;
3029        }
3030
3031        if (MLX5_CAP_FLOWTABLE_RDMA_TX(dev, ft_support)) {
3032                err = init_rdma_tx_root_ns(steering);
3033                if (err)
3034                        goto err;
3035        }
3036
3037        if (mlx5_fpga_ipsec_device_caps(steering->dev) & MLX5_ACCEL_IPSEC_CAP_DEVICE ||
3038            MLX5_CAP_FLOWTABLE_NIC_TX(dev, ft_support)) {
3039                err = init_egress_root_ns(steering);
3040                if (err)
3041                        goto err;
3042        }
3043
3044        return 0;
3045err:
3046        mlx5_cleanup_fs(dev);
3047        return err;
3048}
3049
3050int mlx5_fs_add_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3051{
3052        struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3053        struct mlx5_ft_underlay_qp *new_uqp;
3054        int err = 0;
3055
3056        new_uqp = kzalloc(sizeof(*new_uqp), GFP_KERNEL);
3057        if (!new_uqp)
3058                return -ENOMEM;
3059
3060        mutex_lock(&root->chain_lock);
3061
3062        if (!root->root_ft) {
3063                err = -EINVAL;
3064                goto update_ft_fail;
3065        }
3066
3067        err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3068                                         false);
3069        if (err) {
3070                mlx5_core_warn(dev, "Failed adding underlay QPN (%u) to root FT err(%d)\n",
3071                               underlay_qpn, err);
3072                goto update_ft_fail;
3073        }
3074
3075        new_uqp->qpn = underlay_qpn;
3076        list_add_tail(&new_uqp->list, &root->underlay_qpns);
3077
3078        mutex_unlock(&root->chain_lock);
3079
3080        return 0;
3081
3082update_ft_fail:
3083        mutex_unlock(&root->chain_lock);
3084        kfree(new_uqp);
3085        return err;
3086}
3087EXPORT_SYMBOL(mlx5_fs_add_rx_underlay_qpn);
3088
3089int mlx5_fs_remove_rx_underlay_qpn(struct mlx5_core_dev *dev, u32 underlay_qpn)
3090{
3091        struct mlx5_flow_root_namespace *root = dev->priv.steering->root_ns;
3092        struct mlx5_ft_underlay_qp *uqp;
3093        bool found = false;
3094        int err = 0;
3095
3096        mutex_lock(&root->chain_lock);
3097        list_for_each_entry(uqp, &root->underlay_qpns, list) {
3098                if (uqp->qpn == underlay_qpn) {
3099                        found = true;
3100                        break;
3101                }
3102        }
3103
3104        if (!found) {
3105                mlx5_core_warn(dev, "Failed finding underlay qp (%u) in qpn list\n",
3106                               underlay_qpn);
3107                err = -EINVAL;
3108                goto out;
3109        }
3110
3111        err = root->cmds->update_root_ft(root, root->root_ft, underlay_qpn,
3112                                         true);
3113        if (err)
3114                mlx5_core_warn(dev, "Failed removing underlay QPN (%u) from root FT err(%d)\n",
3115                               underlay_qpn, err);
3116
3117        list_del(&uqp->list);
3118        mutex_unlock(&root->chain_lock);
3119        kfree(uqp);
3120
3121        return 0;
3122
3123out:
3124        mutex_unlock(&root->chain_lock);
3125        return err;
3126}
3127EXPORT_SYMBOL(mlx5_fs_remove_rx_underlay_qpn);
3128
3129static struct mlx5_flow_root_namespace
3130*get_root_namespace(struct mlx5_core_dev *dev, enum mlx5_flow_namespace_type ns_type)
3131{
3132        struct mlx5_flow_namespace *ns;
3133
3134        if (ns_type == MLX5_FLOW_NAMESPACE_ESW_EGRESS ||
3135            ns_type == MLX5_FLOW_NAMESPACE_ESW_INGRESS)
3136                ns = mlx5_get_flow_vport_acl_namespace(dev, ns_type, 0);
3137        else
3138                ns = mlx5_get_flow_namespace(dev, ns_type);
3139        if (!ns)
3140                return NULL;
3141
3142        return find_root(&ns->node);
3143}
3144
3145struct mlx5_modify_hdr *mlx5_modify_header_alloc(struct mlx5_core_dev *dev,
3146                                                 u8 ns_type, u8 num_actions,
3147                                                 void *modify_actions)
3148{
3149        struct mlx5_flow_root_namespace *root;
3150        struct mlx5_modify_hdr *modify_hdr;
3151        int err;
3152
3153        root = get_root_namespace(dev, ns_type);
3154        if (!root)
3155                return ERR_PTR(-EOPNOTSUPP);
3156
3157        modify_hdr = kzalloc(sizeof(*modify_hdr), GFP_KERNEL);
3158        if (!modify_hdr)
3159                return ERR_PTR(-ENOMEM);
3160
3161        modify_hdr->ns_type = ns_type;
3162        err = root->cmds->modify_header_alloc(root, ns_type, num_actions,
3163                                              modify_actions, modify_hdr);
3164        if (err) {
3165                kfree(modify_hdr);
3166                return ERR_PTR(err);
3167        }
3168
3169        return modify_hdr;
3170}
3171EXPORT_SYMBOL(mlx5_modify_header_alloc);
3172
3173void mlx5_modify_header_dealloc(struct mlx5_core_dev *dev,
3174                                struct mlx5_modify_hdr *modify_hdr)
3175{
3176        struct mlx5_flow_root_namespace *root;
3177
3178        root = get_root_namespace(dev, modify_hdr->ns_type);
3179        if (WARN_ON(!root))
3180                return;
3181        root->cmds->modify_header_dealloc(root, modify_hdr);
3182        kfree(modify_hdr);
3183}
3184EXPORT_SYMBOL(mlx5_modify_header_dealloc);
3185
3186struct mlx5_pkt_reformat *mlx5_packet_reformat_alloc(struct mlx5_core_dev *dev,
3187                                                     struct mlx5_pkt_reformat_params *params,
3188                                                     enum mlx5_flow_namespace_type ns_type)
3189{
3190        struct mlx5_pkt_reformat *pkt_reformat;
3191        struct mlx5_flow_root_namespace *root;
3192        int err;
3193
3194        root = get_root_namespace(dev, ns_type);
3195        if (!root)
3196                return ERR_PTR(-EOPNOTSUPP);
3197
3198        pkt_reformat = kzalloc(sizeof(*pkt_reformat), GFP_KERNEL);
3199        if (!pkt_reformat)
3200                return ERR_PTR(-ENOMEM);
3201
3202        pkt_reformat->ns_type = ns_type;
3203        pkt_reformat->reformat_type = params->type;
3204        err = root->cmds->packet_reformat_alloc(root, params, ns_type,
3205                                                pkt_reformat);
3206        if (err) {
3207                kfree(pkt_reformat);
3208                return ERR_PTR(err);
3209        }
3210
3211        return pkt_reformat;
3212}
3213EXPORT_SYMBOL(mlx5_packet_reformat_alloc);
3214
3215void mlx5_packet_reformat_dealloc(struct mlx5_core_dev *dev,
3216                                  struct mlx5_pkt_reformat *pkt_reformat)
3217{
3218        struct mlx5_flow_root_namespace *root;
3219
3220        root = get_root_namespace(dev, pkt_reformat->ns_type);
3221        if (WARN_ON(!root))
3222                return;
3223        root->cmds->packet_reformat_dealloc(root, pkt_reformat);
3224        kfree(pkt_reformat);
3225}
3226EXPORT_SYMBOL(mlx5_packet_reformat_dealloc);
3227
3228int mlx5_flow_namespace_set_peer(struct mlx5_flow_root_namespace *ns,
3229                                 struct mlx5_flow_root_namespace *peer_ns)
3230{
3231        if (peer_ns && ns->mode != peer_ns->mode) {
3232                mlx5_core_err(ns->dev,
3233                              "Can't peer namespace of different steering mode\n");
3234                return -EINVAL;
3235        }
3236
3237        return ns->cmds->set_peer(ns, peer_ns);
3238}
3239
3240/* This function should be called only at init stage of the namespace.
3241 * It is not safe to call this function while steering operations
3242 * are executed in the namespace.
3243 */
3244int mlx5_flow_namespace_set_mode(struct mlx5_flow_namespace *ns,
3245                                 enum mlx5_flow_steering_mode mode)
3246{
3247        struct mlx5_flow_root_namespace *root;
3248        const struct mlx5_flow_cmds *cmds;
3249        int err;
3250
3251        root = find_root(&ns->node);
3252        if (&root->ns != ns)
3253        /* Can't set cmds to non root namespace */
3254                return -EINVAL;
3255
3256        if (root->table_type != FS_FT_FDB)
3257                return -EOPNOTSUPP;
3258
3259        if (root->mode == mode)
3260                return 0;
3261
3262        if (mode == MLX5_FLOW_STEERING_MODE_SMFS)
3263                cmds = mlx5_fs_cmd_get_dr_cmds();
3264        else
3265                cmds = mlx5_fs_cmd_get_fw_cmds();
3266        if (!cmds)
3267                return -EOPNOTSUPP;
3268
3269        err = cmds->create_ns(root);
3270        if (err) {
3271                mlx5_core_err(root->dev, "Failed to create flow namespace (%d)\n",
3272                              err);
3273                return err;
3274        }
3275
3276        root->cmds->destroy_ns(root);
3277        root->cmds = cmds;
3278        root->mode = mode;
3279
3280        return 0;
3281}
3282
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.